Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

default · ~99.6k rows (showing the first 120k)

Split (2)

train · ~72.4k rows (showing the first 93.1k)

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1007.0864	1	1007	2010-07-06T11:21:17	FTIR spectral imaging as a probe of ultrasound effect on cells in vitro	[ "physics.bio-ph" ]	Safe and efficient intracellular delivery of genes or drugs is critically important in targeted cancer treatment and gene therapy applications. Ultrasound (US) has been demonstrated to alter the cell membrane permeability due to a biophysical mechanism (Sonoporation) and exploited as a promising non-invasive gene transfer method. The sonoporation process could induce the formation of transient pores without significantly affecting cell viability. This research is aimed at investigating some bioeffects due to Therapeutic Ultrasound (pulsed-1 MHz) which could allow to enhance drugs or genes delivery in a non tumoral cell line. We have used the NIH-3T3 cell line as model system and exposed it to US at two different distances from the source; the effects of this pulsed ultrasonic wave on cells were assessed by Fourier transform infrared (FT-IR) spectroscopic imaging analysis. This technique combined with a focal plane array (FPA) detector has been widely used to study the general biochemical changes in vitro; moreover, the development of FPA detectors and shortening of measurement times specifically for IR imaging, from several hours to few minutes, have made possible to image the distribution of molecular species in biological samples. We have also performed a cytokinesis-block micronucleus (CBMN) assay to reveal the presence or not of micronuclei (named Howell-Jolly bodies) formed during the cell division due to the DNA damage. The results of IR analysis combined with the cytogenetic analysis have shown that these experimental conditions can not cause DNA mutations in the NIH-3T3 cell line. Finally, the comparison between the spectral parameters of the average spectrum extracted from the spectral map and those of the set of all spectra from the spectral map could be limited by the presence of bad pixels inside the map.	physics.bio-ph	physics	Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 FTIR spectral imaging as a probe of ultrasound effect on cells in vitro. L. Di Giambattista(1,2), P. Grimaldi(1), I. Udroiu(3), D. Pozzi(4), G. Cinque(5), M.D. Frogley(5), A. Giansanti(1) and A. Congiu Castellano(1). (1) Physics Department, Sapienza, University of Rome (IT). (2) CISB (Interdepartmental Research Centre for Models and Information Analysis in Biomedical Systems), Sapienza, University of Rome (IT). (3) DIPIA, ISPESL, via Urbana 167, Rome (IT). (4) Experimental Medicine and Pathology Department, Sapienza, University of Rome (IT). (5) Diamond Light Source Ltd, Didcot, Oxfordshire (UK). Keywords: Ultrasound bioeffects – FT-IR Imaging – Cytogenetic effect. Abstract Safe and efficient intracellular delivery of genes or drugs is critically important in targeted cancer treatment and gene therapy applications. Ultrasound (US) has been demonstrated to alter the cell membrane permeability due to a biophysical mechanism (Sonoporation) and exploited as a promising non-invasive gene transfer method. The sonoporation process could induce the formation of transient pores without significantly affecting cell viability. This research is aimed at investigating some bioeffects due to Therapeutic Ultrasound (pulsed-1 MHz) which could allow to enhance drugs or genes delivery in a non tumoral cell line. We have used the NIH-3T3 cell line as model system and exposed it to US at two different distances from the source; the effects of this pulsed ultrasonic wave on cells were assessed by Fourier transform infrared (FT-IR) spectroscopic imaging analysis. This technique combined with a focal plane array (FPA) detector has been widely used to study the general biochemical changes in vitro; moreover, the development of FPA detectors and shortening of measurement times specifically for IR imaging, from several hours to few minutes, have made possible to image the distribution of molecular species in biological samples. We have also performed a cytokinesis-block micronucleus (CBMN) assay to reveal the presence or not of micronuclei (named Howell-Jolly bodies) formed during the cell division due to the DNA damage. The results of IR analysis combined with the cytogenetic analysis have shown that these experimental conditions can not cause DNA mutations in the NIH-3T3 cell line. Finally, the comparison between the spectral parameters of the average spectrum extracted from the spectral map and those of the set of all spectra from the spectral map could be limited by the presence of bad pixels inside the map. List of symbols Amide I (-C=O stretching), proteins A1 Amide II (-N-H bending, -C-N stretching), proteins A2 Binucleated cells BNC Cytokinesis-block micronucleus CBMN Control cells (untreated cells) CTR 4’-6’-diamidino-2-phenylindole DAPI DMEM Dulbecco’s Modified Eagle’s minimum essential Medium Fetal serum bovine FBS Focal plane array FPA Fourier transform infrared FT-IR Hierarchical Cluster Analysis HCA Inter-spectral distance H Intra-cluster heterogeneity h Distance between petri dish and transducer hD Nuclear division index NDI NIH-3T3 Mouse fibroblast cell line Nbs MCT MN MNC PBS PCA PCs PC1 PC2 R1 RMS SONt_d S/N US Number of bad spectra Mercury cadmium telluride Micronucleus Mononucleated cells Dulbecco's Phosphate Buffer Saline Principal Component Analysis Principal components First principal component Second principal component A 1 A 2 I Root Mean Square Sonicated cells (treated cells), t=time and d=distance Signal-to-Noise ratio Ultrasound intensity ratio I Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 1. Introduction In the last years, Ultrasounds have been employed in gene delivery for the advantages over other systems such as virus or nonvirus-mediated systems. Among its advantages there is the fact that it is a less invasive-method, with an a higher delivery efficiency, and more applications and minimal cell death can be achieved. US systems are available for both in vitro and in vivo studies [1-4]. US effects (thermal, cavitation and microstreaming) depends on physical and biological factors, such as frequency, intensity, time exposure, duty cycle, temporal and spatial structure of sound field, the physiological state and the size-volume of a sonicated sample, and external conditions like temperature, pressure. Such a great number of variables complicates the analysis of the phenomena [5]. The focus of this contribution is on the quantitative evaluation of collateral biological effects of the treatment of non tumoral cell lines. We have used FT-IR spectroscopy imaging based on the FPA detector [6,7] to study US-bioeffects on a cellular system in the range of therapeutic ultrasound (1 MHz in pulsed system). This technique has been combined with a cytogenetic analysis (CBMN), that allows to study DNA- damage at chromosomic level. A micronucleus is formed during the metaphase/anaphase transition of mitosis; it may arise from a whole lagging chromosome (aneugenic event leading to chromosome loss) or an acentric chromosome fragment detaching from a chromosome after breakage (clastogenic event) which do not integrate in the daughter nuclei. These micronuclei are also known as Howell-Jolly bodies. In vitro, the analysis of cells in presence of cytochalasin B (added 44 hours after the start of cultivation), an inhibitor of actins, allows to distinguish easily between mononucleated cells which did not divide and binucleated cells which completed nuclear division during in vitro culture; the frequencies of mononucleated cells provide an indication of the background level of chromosome/genome mutations accumulated in vivo. Moreover, the frequencies of binucleated cells with MN allow to measure the damage accumulated before cultivation plus mutations expressed during the first in vitro mitosis. According to the previous test, we have established the following conditions of experimental set-up: the size of the plexiglass tank, the distances and position between the cellular sample and transducer (ultrasound source) within tank , the use of US in pulsed system with 75 % of duty cycle. Finally, we have compared the utility of different methods to evaluate the quality of FT-IR maps and the chemical similarity between the samples. 2. Materials and Methods 2.1 Cell culture In this research, we have used a healthy adherent fibroblast cell line, named NIH-3T3; the NIH-3T3 were grown with a solution of Dulbecco’s Modified Eagle’s minimum essential Medium (DMEM) without calcium with 10% fetal bovine serum (FBS), 1% penicillin-streptomycin and 1% L- glutamine at 37° C in humidified atmosphere containing 95% air and 5% CO2. 2.2 Sample preparation Before the US exposure, the NIH-3T3 were cultured as monolayer on CaF2 windows, pre-treated with polylysine, inside a 35-mm petri dish. The viability of both the untreated (control) and the US- treated (sonicated) cells has been determined by Trypan blue exclusion test (over 90% of viability). After the treatment with US, the control and sonicated cells were fixed in paraformaldehyde (2% Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 for 15 min), washed in Dulbecco's Phosphate Buffer Saline (PBS) and in distilled water to remove PBS residues, and dried in a desiccator. 2.3 Ultrasound exposure set up We have built a plexiglass tank filled with partially degassed water and the US transducer was placed at the bottom. According to preliminary tests, the following parameters were chosen: distances (hD) between the cellular sample in a petri dish and the transducer were 10 and 15 cm (Figure 1), the times of US exposure were 15, 30, 45, 60 minutes. Acoustic pressure at both distances was measured by a needle hydrophone (Precision Acoustics LTD, HP 0.5 mm Interchangeable Probe) with a sensitivity of -272.7dB re 1V/µPa at the frequency of 1 MHz (Table a) [8]. AIR PETRI DISH Figure1. The general scheme of the experimental set-up. 10 15 214.5 ± 21 318.7 ± 32 (6.86 ± 0.9)10-6 Distance (cm) Pressure ± ∆P (Pa) (3.11 ± 0.4)10-6 100% Power with 75% of duty cycle Acoustic Intensity ± ∆I (W/cm2) The US frequency has been set to 1 MHz with 100% of maximum power in pulsed system with 75% duty cycle. Table a. Values of the pressure within the petri dish and acoustic instant intensity at 10 and 15 cm. The temperature inside and outside of the petri dish at two different distances was measured; a maximum increase of 2 °C or less, that can not induce any breaks in the integrity of plasma membrane, was revealed during time of US exposure. In the results, we have indicated the control cells as “CTR” and the sonicated cells as “SONt_d” where t=0,15,30,45,60 minutes and d=10,15 cm. 2.4 Measurement using FPA detector An FTIR microscope (Hyperion 3000) coupled to an Vertex 80v FTIR spectrometer (both from Bruker Optics, Germany) was used to the analysis and was controlled via PC running OPUS-NT software, version 6.5. The microscope was equipped with a computer-controlled x,y stage and the sample area within a perspex box was purged in 99,98% pure nitrogen. The Hyperion 3000 microscope was equipped with a mercury cadmium telluride (MCT) based FPA detector of 64x64 Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 pixels where each pixel corresponded to an area of 40x40 µm, so that the visible field corresponded to an area 170x170 µ m2 with 15x magnification objective. Each pixel corresponds to a single acquired FT-IR spectrum. For IR measurements, the spectrometer was under vacuum to reduce spectral contributions from water vapour and CO2. The IR images were continuously acquired with a spectral resolution of 8 cm-1 and by co-adding 512 scans of absorbance spectra were recorded in the range from 4000 to 900 cm-1. Every cellular sample was measured on CaF2 window in Trasmission mode and a background spectrum/ image of a “blank” area on the same CaF2 window was recorded before each cellular sample measurement in order to account for temporal variations of water vapour and CO2 levels. Three images were acquired for each CaF2 window, resulting in 30 images with 4096 spectra for the whole experiment. The acquisition time for each FTIR image was approximately 7 minutes. To study the chemical map after a pre-processing analysis, the spectral data were processed via the rubberband method baseline correction (64 baseline points). The whole spectral region (4000-900 cm-1) was normalized through vector normalization in the spectral range of the Amide I (1595-1800 cm-1) band and we have recorded spectra only in the spectral region of proteins and nucleic acids regions, from 1800 to 900 cm-1. The molecular maps were created from these spectra by plotting different spectral parameters, such as intensities peaks ratio, as a function of x-y pixel position. 2.5 Data pre-processing for image analysis and statistical analysis The resulting FT-IR images were pre-processed by different sofwares (OPUS 6.5, OriginPro 8.0, Mathematica 7.0) and were analysed in proteins and nucleic acids regions (1800-900 cm-1). To remove poor quality spectra, the data sets were subjected to a quality test such as Signal-to-Noise ratio (S/N) and statistical test such as Principal Component Analysis (PCA). In the (S/N) test, the signal S in each of the 4096 pixels was evaluated as the maximum in the frequency region of the Amide I band (1595-1800 cm-1), while the noise N was calculated as the standard deviation (RMS) in the spectral range 1800-1900 cm-1. In order to identify poor quality pixels, those with a Signal-to-Noise ratio S/N smaller than 100 were labeled as bad pixels and the corresponding spectra, bad spectra. The threshold set at 100 was chosen in accordance to [9]. The number of bad spectra inside a map is denoted with Nbs. The maps having more than 1000 bad pixels have been discarded. This acceptance threshold for the number of bad pixels was determined by noting that the dispersion of the spectroscopic parameter R1 (defined later) around the mean value increased significantly for Nbs > 1000. In particular, bad spectra tend to populate the low-R1 tail of the distribution, that is thus significantly skewed towards the left for large values of Nbs. On the contrary, the R1 distribution of good spectra tends to be Gaussian. We also show how poor spectra can be singled out using PCA. PCA is a multivariate method where a complex data set containing P (each spectrum of a single map) different sets of Y variables (the absorbance at wavenumbers from 1800 to 900 cm-1) is transformed into a smaller set of new indipendent variables or principal components (PCs), which maximise the variance of the original data set and are actually unrelated (whereas the original, untransformed variables may have been correlated to some extent). Therefore, the new reference system identified by the PCs is expressed as a linear combination of the original data set. PCs are computed hierarchically, with the first PC (PC1) accounting for the maximum amount of variance and the others for the subsequent maximum residual variance. The coordinates of the spectra in the new reference system are called scores and the coefficients of the linear combination describing each PC, i.e. the weights of the original variables on each PC, are called loadings. Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 After the data pre-processing on the quality of FT-IR maps, a second statistical test known as Hierarchical Cluster Analysis (HCA), was introduced to evaluate the chemical similarity between control and sonicated maps of cellular samples; we have extracted the average spectrum of each map for control and sonicated cells, and then the average spectrum was processed with a baseline, smooth and vector normalization at Amide I as described in section 2.4. To calculate the inter-spectral distance or similarity distance (H) is used the squared euclidean distance while the intra-cluster heterogeneity or cluster method (h) is assessed by Ward's minimum- variance (OPUS 6.5 software). In the analysis of FT-IR spectroscopic data combined with cytogenetic-results, we have introduced the Fisher statistical test (F-test), which allows to establish if any linear correlation exists between any two variables of the pool. Finally, the statistical significance in the CBMN assay was evaluated by Dunnet’s test. 2.6 Cytokinesis-block micronucleus (CBMN) assay The control and sonicated cellular samples were treated with 6 µg/ml cytochalasin B. The cells were sampled 24 h after addition of cytochalasin B by centrifugation for 5 min; then, hypotonic treatment consisted of careful resuspension of the cellular samples in 5 ml hypotonic saline (75 mM KCl). Immediately after addition of hypotonic solution, the cells were collected and they fixed in Carnoy's fixative (3:1 methanol / acetic acid). Finally, the cells were transferred onto pre-cleaned slides and were stained with 10 µg/ml of 4’-6’- diamidino-2-phenylindole (DAPI) in antifade solution (Vector Laboratories). For each concentration of a test compound 500 binucleated cells (BNC) were evaluated. Finally, the Nuclear Division Index (NDI), a parameter of cellular mitogen response and cytogenetic effect of US, was evaluated for each sample according to Eastmond and Tucker (1989): NDI = (21[ M + M (3)2 + M (4)3 + M )]4 N where M1, M2, M3, M4 indicate the number of mono-, bi-, tri- and quadrinucleate cells and N is the total number of counted viable cells. For the scoring of micronuclei the following criteria were adopted from Fenech et al. (2003). 3. Results and Discussion 3.1 Spectra pre-processing From the results of the S/N quality test, we have established the number of bad spectra and have evaluated their weight in each data set. We report, as an example, the distribution of S/N values in two maps, one of the control cells and one of the sonicated cells (SON45_15), that has shown a larger number of bad spectra (Figure 2). Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 Figure 2. Distribution of the S/N ratio vs. Pixel (0-4096): on the left, control cells (4051 good spectra and 44 bad spectra), and on the right, sonicated cells for 45 minutes at 15 cm (3344 good spectra and 752 bad spectra). By Figure 2, we extract a histogram where we also visualize the total number of pixels of the map for control and sonicated (SON45_15) cells corresponding to the values of S/N ratio. Figure 3. The total number of Pixels corresponding to S/N values: on the left, control cells and on the right, sonicated cells for 45 minutes at 15 cm; the good spectra are shown in blue and the bad spectra are shown in red. We have not discarded the FT-IR maps with Nbs<1000 to monitor the presence of major number of empty spaces due to the US that perturbed the spatial cellular distribution on the slide forming the islands of cellular monolayers. As shown in Figure 4, we have compared the number of bad spectra vs. time of FT-IR maps between the control and sonicated cells at two distances from the US source. Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 Maps 10cm 0 10 20 30 40 50 60 Time[min] Maps 15cm s b N s b N 1000 900 800 700 600 500 400 300 200 100 0 1000 900 800 700 600 500 400 300 200 100 0 0 10 20 30 40 50 60 Time[min] Figure 4. The number of bad spectra (Nbs) vs. time of US exposure within the maps reported at the 10 and 15 cm from the US source. Using the PCA statistical tool, large spectral data were reduced into a small number of indipendent variables known as principal components (PC1, PC2, PC3 etc.); contributions of these components to a given spectra are called scores. The score of the principal components is one of the parameters widely used for classification. In this research, we have employed PCA method to discriminate poor quality spectra from good quality ones within the FT-IR maps. The PCA eigenvalue plot (data not reported) has shown that the PCs which have the most significant information, were only the first and second principal component (PC1, PC2) for all spectra in the FT-IR maps. In Figure 5, we have reported an example of PCA results where the information due to the CTR map and SON45_15 map is compressed in the first and second principal component of PCA (see the percentages of the variance in the caption of Figure 5). The PCA analysis of SON45_15 map shows as the majority of the bad spectra (red points) were distributed towards the negative values of PC2 and were dissimilar from the good spectra (black points); in the CTR map, the bad spectra (red points) were similar between them and were distribuited as the good spectra (black points) towards the positive value of PC1. Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 Figure 5. The PCA results (black points=good spectra; red points=bad spectra) obtained for CTR (top) and SON45_15 (bottom) FT-IR map. In the CTR sample, the values of PC1 and PC2 are 69.6% and 16.2%, respectively; in the SON45_15 sample, PCs values are different from CTR sample (PC1=79.7% and PC2=11.1%). 3.2 CBMN assay Using the cytogenetic test, we have found that the sonicated cells have reported a micronuclei frequency higher than the control cells at both distances. As described in Figure 6, we have observed a statistically significant value (p = 0.045) only for the sonicated cells for 60 minutes at 15 cm. Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 CBMN assay (15cm) 6 5 4 3 2 1 0 C N B 0 0 5 / C N M * 1.66 1.64 1.62 1.60 1.58 1.56 r e t e m a r a p I D N CBMN assay (15cm) 0 10 20 30 40 50 60 0 10 20 30 40 50 60 Time[min] Time[min] Figure 6. On the left, the histogram describes the value of micronucleated cells (MNC) / 500 binucleated (BNC) cells for control and sonicated cells at 15 cm from the source; significance compared to controls: *p < 0.045. On the right, the histogram shows the NDI parameter vs. time; no significant differences are discovered. Finally, no significant differences between the control and sonicated cells for the values of NDI parameter were detected for both distances. 3.3 FTIR Spectral Imaging FTIR spectral imaging enables the determination of the distribution of several molecules of interest; we have studied the chemical map and the average spectrum due to the FT-IR imaging through the analysis of a spectroscopic parameter belonging to the proteins and nucleic acids regions (1800-900 cm-1): R 1 = A 1 A 2 I I where AnI (n=1,2) parameters indicate the intensity of Amide I and Amide II peaks respectively and finally, the R1 parameter indicates the ratio due to the intensity of Amide I and Amide II peaks. The intensities of IR peaks provide quantitative analysis about sample contents, depending on the nature of molecular structure, their bonds, and their environment. The Amide I (A1) and Amide II (A2) bands are centered in the control spectrum at 1653 cm-1 and 1543 cm-1 respectively. The shape of the Amide I band is influenced by the overall composition in the secondary structure of the samples; the relative contributions fall in the following spectral regions: α-helix between 1645-1662 cm-1, β-sheets between 1613-1637 cm-1, β-turns between 1662-1682 cm-1 and random coil between 1637-1645 cm-1. Before showing the R1 parameter analysis, we have applied the Hierarchical Cluster Analysis method to investigate the relationships among spectra for all samples in the spectral range (1800- 900 cm-1) at both distances from the US source. Through this statistical test, the average spectrum extracted from the control and sonicated maps was classified within their respective type cluster according to biochemical similarities. As described in Figure 7, the maximum heterogeneity level in the proteins and nucleic acids regions was H=1.38 at 15 cm distance; while the maximum heterogeneity within clusters was h=0.14 at both distances. Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 HCA Proteins and Nucleic Acids Regions 15cm H=1.38, h= 0.14 HCA Proteins and Nucleic Acids Regions 10cm H=0.45, h= 0.14 Figure 7. HCA results show the cluster members according to the biochemical similarities for proteins and nucleic acids regions at 10 and 15 cm from the US source. The spectral comparison through HCA test is consistent with the results of S/N quality test and PCA analysis obtained from the FT-IR maps; HCA results were also confirmed by the analysis of R1 parameter for all the average spectra. As shown in Figure 8, we have analysed the Amide I / Amide II (parameter A 1 A 2 I ratio that represents an indirect measure of DNA content, due to the carbonyl group from the bases and the spectral changes in the range of 1245-960 cm-1 that suggests conformational changes and/or rearrangement of existing nucleic acid structures [14]. ) intensity R 1 = I Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 3.4 3.2 3.0 2.8 2.6 2.4 2.2 2.0 1.8 1.6 1.4 1.2 1.0 0.8 3.4 3.2 3.0 2.8 2.6 2.4 2.2 2.0 1.8 1.6 1.4 1.2 1.0 1 R 1 R R from chemical map at 10 cm 2 R from average spectrum at 10 cm 2 0 10 20 30 40 50 60 Time[min] R R from chemical map at 15 cm 1 from average spectrum at 15 cm 1 0 10 20 30 40 50 60 Time[min] Figure 8. Amide I / Amide II intensity ratio vs. time of US exposure at 10 cm (top) and 15 cm (bottom) from chemical map and average spectrum is reported, showing small differences between the values of chemical map with those of average spectrum. The DNA content is detectable by IR spectroscopy mainly when the cells are in the S phase of the cell cycle because the DNA is so tightly packed in the nucleus in the G1 and G2 phases that it appears opaque to IR radiation [15-18]. Assuming that the cells are into the S phase, the small spectral changes in R1 ratio, observed in our experiments and confirmed by HCA for the proteins and nucleic acids regions, could correspond to a better detection of DNA content. In the following Figure 9, we have reported the comparison between the chemical maps of R1 parameter for CTR and SON45_15 sample, where it is present a considerable difference between the number of bad spectra; moreover, the spectral profiles of CTR and SON45_15 average spectrum were compared. Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 CTR average spectrum SON45_15 average spectrum A 1I ] . u . a [ S B A A 2I 1800 1750 1700 1650 1600 1550 1500 Wavenumber[cm-1] Figure 9. On the top, the chemical maps of R1 parameter for control (left) and sonicated cells for 45 minutes at 15 cm (right); on the bottom, the overlap of average spectrum for CTR and SON45_15 samples, showing the spectral shape of the Amide I and Amide II bands. In general, the use of FT-IR imaging or spectroscopy only, without other techniques, does not allow to understand if the nature of the spectral changes in the R1 parameter, due to DNA unpacking effects [19], is linked directly with cell division or the apoptotic process. Therefore, we have used the micronuclei test to check a correlation between the results of chemical map and the Nuclear Division Index (NDI) parameter. Figure 10 shows the correlation between the R1 parameter and the NDI at 15 cm distance from the US source; this correlation has a R-value of about 0,85 with a probability of about 99% (Fisher test) for the linear significance. At 10 cm distance from the US source (data not shown), the degree of correlation decreases to a R- value of about 0,70 with a probability of about 98% (Fisher test). Thus, the small spectral changes in R1 ratio are coherent with the results of micronuclei test: the Ultrasound does not cause the loss of structural and functional integrity of DNA at both distances of US exposure. Di Giambattista L. et al - Biophys. & Bioeng. Letters (2009), Vol 2, Nr 2 Chemical map (15cm) Linear Fit R=85% 3.5 3.0 2.5 1 R 2.0 1.5 1.0 1.595 1.600 1.605 1.610 NDI Figure 10. Correlation between the Amide I / Amide II intensity ratio and the NDI parameter with a R-value of about 0,85 and a probability of about 99% (Fisher test). 4. Conclusions The information obtained in this study can be utilized as experimental basis to detect the ultrasonic effects in vitro. Through the results of FT-IR spectral imaging correlated with applied pression and a cytogenetic test, we have found no damage on functionality and structure of DNA (R1 parameter) at 10 and 15 cm from the US source. Therefore, the Ultrasound at 1 MHz frequency does not cause DNA mutations differently from known environmental agents (ultraviolet light, nuclear radiation or certain chemicals). In the pre-processing analysis, the (S/N) ratio was evaluated to determine the number of bad spectra within the maps and the maps with a number of bad spectra over 1000 were discarded. We have also used a second method, PCA, to evaluate the distribution of bad spectra inside the map. We have used the HCA test to study the chemical similarity between the control and the sonicated samples and we have found correspondence between the results of this test and the variation in R1 parameter, i.e. maps with similar average values of R1 result (1800-900 cm-1) to be clustered together. From spectral data in the proteins and nucleic acids regions, we have not found a perfect agreement between the results of the chemical map and the average spectrum. This could be due first to the fact that the average value of R1 is different from the ratio of the average values of A1I and A2I . Moreover, the R1 –value obtained from the map is actually the mode of the distribution of R1 –values, and this concides with the mean only if the distribution is gaussian; this distribution is not gaussian due to the presence of the bad pixels. Other experiment are in progress to verify whether the ultrasonic source can produce a transient and/or permanent phenomenon of sonoporation on plasma membrane useful for transfection process and to study the size of pores and the membrane recovery time due to different kinetics for small and large molecules in situ. Acknowledgments The authors are grateful to Dr. K. Wehbe for their supporting the FTIR-spectral imaging at Beamline B22 FTIR-microspectroscopy of the Diamond Light Source (Oxfordshire). We want also to thank Dr. S. Gaudenzi, Dr. A. Bedini, Dr. C. Giliberti and Dr. R. Palomba for discussions and help in the experiments. Finally, we are grateful to Dr. M. Lattanzi for his assistance with Mathematica software and Dr. S. Belardinelli for his assistance in the laboratory experiments. REFERENCES 1. Newman C.M., Lawrie A., Brisken A.F., Cumberland D.C. Ultrasound gene therapy: On the road from concept to reality. Echocardiography 2001 (18):339–347. 2. Bao S., Thrall B.D., Miller D.L. Transfection of a reporter plasmid into cultured cells by sonoporation in vitro. Ultrasound Med. Biol., 1997 (23):953–959. 3. Greenleaf W.J., Bolander M.E., Sarkar G., Goldring M.B., Greenleaf J.F., Artificial cavitation nuclei significantly enhance acoustically induced cell transfection. Ultrasound Med. Biol., 1998 (24):587–595. 4. Koch S., Pohl P., Cobet U., Rainov N.G., Ultrasound enhancement of liposome-mediated cell transfection is caused by cavitation effects. Ultrasound Med. Biol., 2000 (26):897–903. 5. Chun-Yen L., Chia-Hsuan W., Chia-Chun C. and Pai-Chi L., Quantitative realtions of acoustic inertial caviatation with sonoporation and cell viability. Ultrasound Med. Biol., 2006 (32):1931–1941. 6. Lasch P., Naumann D., FT-IR microspectroscopic imaging of human carcinoma thin sections based on pattern recognition techniques. Cell. Mol. Biol., 1998 (44):189–202. 7. Kohler A. , Bertrand D., Martens H., Hannesson K., Kirschner C., Ofstad R., Multivariate image analysis of a set of FTIR microspectroscopy images of aged bovine muscle tissue combining image and design information. Anal. Bioanal. Chem., 2007 (389):1143–1153. 8. Lewin P.A., Lypacewicz G., Bautista R., Devaraju V., Sensitivity of ultrasonic hydrophone probes below 1 MHz. Ultrasonics, 2000 (38):135–139. 9. Griffiths P.R., De Haseth J.A., Fourier Transform infrared spectrometry, John Wiley and Sons, NY, 1986. 10. Eastmond D.A., Tucker J.D., Identification of aneuploidy-inducing agents using cytokinesis-blocked human lymphocytes and anti-kinetochore antibody. Environ. Mol. Mutagen., 1989 (13): 34-43. 11. Fenech M., Chang W.P., Kirsch-Volders M., Holland N., Bonassi S., Zeiger E., HUman MicronNucleus project. HUMN project: detailed description of the scoring criteria for the cytokinesis-block micronucleus assay using isolated human lymphocyte cultures. Mutat. Res., 2003 (534):65-75. 12. Zwaal R. F.A., J. Schroit A., Pathophysiologic implications of membrane phospholipid asymmetry in blood cells. Blood, the J. A. Soc. Hemat., 1997 (89):1121-1132. 13. Takashi K., Eiichi K. , Effect of Free Radicals Induced by Ultrasonic Cavitation on Cell Killing. Int. J. Rad. Biol., 1988 (54):475-486. 14. Benedetti E., Bramanti E., Rossi I., Determination of the relative amount of nucleic acids and proteins in leukemic and normal lymphocytes by means of FT-IR microspectroscopy. Appl. Spectrosc. 1997 (51):792-797. 15. Conti L., Grimaldi P., Udroiu I., Bedini A., Giliberti C., Palomba R., Congiu Castellano A., Effects induced in cells by ultrasound revealed by ATR-FTIR spectroscopy. Vibrat. Spectrosc., 2010 (52):79-84. 16. Diem M. S., Boydston S., Chiriboga L., Infrared spectroscopy of cells and tissues: shining light onto a novel subject. Appl. Spectrosc., 1999 (53):148A-161A. 17. Boydston-White S., Gopen T., Houser S., Bargonetti J., Diem M. S., Infrared spectroscopy of human tissue. V. Infrared spectroscopic studies of myeloid leukemia (ML-1) cells at different phases of the cell cycle. Biospectroscopy, 1999 (5):219-227. 18. Wang J. S., Shi J. S., Xu Y. Z., Duan X. Y., Zhang L., Wang J., Yang L. M., Weng S. F., Wu J. G., FT-IR spectroscopic analysis of normal and cancerous tissues of esophagus. World J. Gastroenterol, 2003 (9):1897-1899. 19. Gasparri F., Muzio M., Monitoring of apoptosis of HL60 cells by Fourier-transform infrared spectroscopy. Biochem. J., 2003 (369):239-248.
1505.02609	1	1505	2015-05-11T13:38:30	Label-free characterization of white blood cells by measuring 3D refractive index maps	[ "physics.bio-ph" ]	The characterization of white blood cells (WBCs) is crucial for blood analyses and disease diagnoses. However, current standard techniques rely on cell labeling, a process which imposes significant limitations. Here we present three-dimensional (3D) optical measurements and the label-free characterization of mouse WBCs using optical diffraction tomography. 3D refractive index (RI) tomograms of individual WBCs are constructed from multiple two-dimensional quantitative phase images of samples illuminated at various angles of incidence. Measurements of the 3D RI tomogram of WBCs enable the separation of heterogeneous populations of WBCs using quantitative morphological and biochemical information. Time-lapse tomographic measurements also provide the 3D trajectory of micrometer-sized beads ingested by WBCs. These results demonstrate that optical diffraction tomography can be a useful and versatile tool for the study of WBCs.	physics.bio-ph	physics	Label-free characterization of white blood cells by measuring 3D refractive index maps Jonghee Yoon1, Kyoohyun Kim1, HyunJoo Park1, Chulhee Choi2, Seongsoo Jang3, and YongKeun Park1,* 1Department of Physics, Korea Advanced Institute of Science and Technology, Daejeon 305-701, South Korea 2Department of Bio and Brain Engineering, Korea Advanced Institute of Science and Technology, Daejeon 305-701, South Korea 3Department of Laboratory Medicine, University of Ulsan, College of Medicine and Asan Medical Center, Seoul 138-736, Republic of Korea. [email protected] Abstract: The characterization of white blood cells (WBCs) is crucial for blood analyses and disease diagnoses. However, current standard techniques rely on cell labeling, a process which imposes significant limitations. Here we present three- dimensional (3D) optical measurements and the label-free characterization of mouse WBCs using optical diffraction tomography. 3D refractive index (RI) tomograms of individual WBCs are constructed from multiple two-dimensional quantitative phase images of samples illuminated at various angles of incidence. Measurements of the 3D RI tomogram of WBCs enable the separation of heterogeneous populations of WBCs using quantitative morphological and biochemical information. Time-lapse tomographic measurements also provide the 3D trajectory of micrometer-sized beads ingested by WBCs. These results demonstrate that optical diffraction tomography can be a useful and versatile tool for the study of WBCs. 1. Introduction White blood cell (WBC) residue throughout the body plays various roles in defending the host from invaders and abnormal cells [1]. With regard to immunology, the characterization of WBCs is important for understanding the pathophysiology of many diseases, including autoimmune diseases [2], neurodegenerative diseases [3], and cancer [4]. WBCs are classified into several subsets according to their morphologies and roles in the immune system. To classify and analyze WBCs, several optical imaging methods are extensively used. For example, light microscopy with Giemsa staining, which is the standard method in clinical practice, visualizes the characteristic features of the cytoplasm and nuclei of WBCs [5, 6]. However, Giemsa staining requires chemical fixation procedures which limit live cell analysis, and only 2D images can be obtained. Confocal fluorescence microscopy enables the 3D structural images of living WBCs at a high resolution and with high molecular specificity, without physical sectioning [7, 8]. However, chemical staining procedures or genetic modifications are invasive methods, inevitably presenting significant drawbacks such as phototoxicity and photobleaching. However, measuring the refractive indices (RIs) of biomolecules, i.e., intrinsic optical properties describing light-matter interactions, circumvents the aforementioned limitations in the study of WBCs. Quantitative phase imaging (QPI) techniques have been introduced to visualize cells and tissues, exploiting RIs as imaging contrasts [9-11]. Exploiting the principle of light interference, QPI techniques can quantitative and non-invasively measure the RI information of samples. Recently, QPI techniques have been widely applied to study the pathophysiology of various biological samples, including red blood cells [12-21], neurons [22, 23], cancer cells [24], and phytoplankton [25]. Thus far, although various efforts for the imaging WBCs using QPI techniques have been proposed [26-30], previous works have been limited to 2D imaging. Earlier research has reported morphological and biochemical changes of WBCs during WBC-mediated cytotoxicity [26, 27], pathogen infection [28, 30], and differentiation [29] by measuring the RIs of samples. Methods which use the transport of intensity equation also enable the measurement of quantitative phase information by numerical calculations, and these methods have been used to investigate WBCs [31, 32]. Although these previous approaches have demonstrated the potential of 2D QPI techniques to study WBCs, 3D tomographic measurements of WBCs have not been investigated. All previous attempts have been limited to the topographic 2D imaging of WBCs, while 3D quantitative morphological and biochemical analyses of WBCs have not been attempted. This limitation in 3D imaging is unfortunate because QPI has much to offer in research related to WBCs given its unique quantitative imaging contrast and the non-invasiveness of these processes. Furthermore, considering the complex subcellular structures of WBCs and their importance in cell classification and immunology, there is strong motivation to extend this technology to the 3D RI tomographic imaging of WBCs. Here, we report the optical measurement of the 3D RI distributions of individual mouse lymphocytes and macrophages via optical diffraction tomography (ODT). High-resolution 3D RI tomograms of individual WBCs are constructed from multiple 2D optical phase delay images of the samples illuminated with various angles of incidence [33]. 3D RI tomogram provides quantitative morphological and biochemical information about WBCs, including the cellular dry mass, the dry mass density, volume, surface area, and sphericity. We demonstrate that 3D RI tomography enables the separation of heterogeneous populations of WBCs using quantitative structural and biochemical information. Moreover, time-lapse tomographic measurements of individual WBCs are shown to be capable of precisely visualizing the 3D trajectories of micrometer-sized beads ingested by macrophages, from which the viscoelastic properties of local cytoplasm can be investigated. The measured 3D RI tomograms of individual WBCs clearly demonstrate that ODT can be a useful and versatile tool for the study of WBCs. 2. Methods 2.1 Experimental setup The optical setup is presented in Fig. 1(a). Complex optical fields of a sample, containing both the amplitude and phase images, are recorded using a Mach-Zehnder interferometric microscope for various illumination angles [34, 35]. A laser beam from a diode-pumped solid state laser (λ = 532 nm, 100 mW, Shanghai Dream Laser Co., Shanghai, China) is split into two arms by a beam splitter (BS1). One arm is used as a reference beam, and the other arm is tilted by a dual-axis scanning galvanometer (GVS012, Thorlabs, Newton, NJ, USA) for varying the angle of the illumination beam impinging to the sample. Hologram of the sample is generated by interference of two beams, which is recorded by a high-speed CMOS camera (1024 PCI, Photron USA Inc., San Diego, CA, USA) with a frame rate of 1,000 Hz. Typically, 300 holograms of the sample, illuminated by plane waves with various illumination angles (-70° to 70° at the sample plane), are recorded for reconstructing one RI tomogram. Details about the experimental setup used to measure complex optical fields can be found in the literature [33]. (a) Laser P BS1 GM (b) L1 L2 Camera BS2 L3 CL OL L4 Hologram (a.u.) 2 Amplitude Phase M2 M1 L6 L5 0 (c) (rad) 4 z x RI tomogram n 1.41 1.34 y 0 x 5 μm y z Fig. 1. Schematic of the experimental ODT setup and the procedure of a 3D RI tomogram reconstruction. (a) A Mach-Zehnder interferometric microscope equipped with a 2D scanning galvanometer-based mirror. BS1–2, beam splitters; GM, galvano mirror; OL, objective lens; CL, condenser lens; M1–2, mirrors; P, pinhole; L1–6, lenses. (b) Holograms are recorded with various illumination angles (top) and the retrieved amplitude and the phase images corresponding to a hologram at a specific illumination angle (bottom). Inset: zoomed-in view of spatially modulated interference patterns. Scale bar, 5 μm. (c) Cross-sectional slices of a RI tomogram of a WBC. Scale bar, 5 μm. 2.2 Tomogram reconstruction From the measured multiple 2D complex amplitude images of a sample, a 3D RI tomogram of individual samples is reconstructed via the ODT algorithm (Fig. 1(b)), which is analogous to 3D computed tomography in X-ray. First, the complex optical fields are extracted from measured holograms using a field retrieval algorithm [36]. Multiple complex amplitude images obtained with various illumination angles are 2D Fourier transformed. Then the spectral information are mapped onto a surface, so-called Ewald sphere, in 3D Fourier space. Finally, 3D RI tomogram is reconstructed by applying 3D inverse Fourier transformation to the mapped 3D Fourier space. Due to the limited numerical aperture (NA) of the used imaging system, there exist missing spectral information. To fill this missing information, Gerchberg-Papoulis algorithm based on a non-negativity constraint was used [37]. The theoretical lateral and axial resolution of the reconstructed tomogram is 111 and 354 nm, respectively, which was calculated from the maximum range of the Fourier spectra [34]. The lateral and axial resolution was experimentally measured as 373 and 496 nm, respectively, by analyzing the edge of the reconstructed tomograms of polystyrene beads. The detailed reconstruction process including a MatLabTM code can be found in our previous work [33]. 2.3 Cell preparation All experiments used 7- to 10-week-old male Balb/c mice (Orient Bio Inc., Gapyeong, Korea). Lymphocytes and macrophages were collected from mice peripheral blood and peritoneal cavity, respectively. Peripheral blood obtained from the heart of euthanized mice was added to heparin (10 U/ml). Heparinized blood was diluted with an equal volume of phosphate-buffered saline (PBS, Welgene Inc., Gyeongsan, Korea) and layered on 3 ml of lymphocyte separation medium (MP Biomedicals, Irvine, CA, USA) in a 15 ml conical tube. This solution was centrifuged at 400 g at room temperature for 20 min to separate lymphocytes from red blood cells. Lymphocytes layer was collected and washed 2 times with PBS. The cells were resuspended in Dulbecco's modified Eagles' medium (DMEM, Gibco, Big Cabin, OK, USA) supplemented with 10% heat-inactivated fetal bovine serum (FBS). To isolate macrophages, ice cold PBS (with 3% FBS) was injected into the peritoneal cavity of euthanized mice. Injected fluid was collected from peritoneal cavity and centrifuged at 250 g for 8 min. The cells were maintained in DMEM for 2 days before experiment [38]. Macrophage and lymphocyte were sandwiched between two cover slips and imaged at room temperature. For phagocytosis analysis, 1 μm polystyrene beads (89904, Sigma- Aldrich, St. Louis, USA) were added to culture medium of cultured macrophages. Measurements were performed 1 hour after beads addition. The sample preparation procedures and the methods were approved by the Institutional Review Board (KA-2015-03). 3. Results and discussions 3.1 Label-free 3D RI imaging of lymphocytes and macrophages We initially measured the 3D RI tomograms of individual lymphocytes and macrophages (See Methods). Figures 2(a, c) present cross-sectional images of 3D RI tomograms on various axial planes: 1 μm below the focal plane (left column), the focal plane (center column), and 1 μm above the focal plane (right column) along the axial axis. The RI distribution map clearly exhibits the plasma membrane shapes and intracellular structures of a lymphocyte and a macrophage. (a) z= -1 μm z= 0 μm z= +1 μm 1.4 (b) (c) 1.34 (d) 1.4 1.34 Fig. 2. 3D visualization of WBCs. Cross-sectional slices of the RI distribution of (a) a lymphocyte and (c) a macrophage at various axial planes. Scale bar, 4 µm. Rendered iso-surfaces of the RI map of (b) a lymphocyte and (d) a macrophage. For visualization purposes, the threshold iso-surfaces of the plasma membrane (white) and the internal structures (cyan, red) are set to 1.345, 1.375, and 1.39, respectively. Scale bar, 3 μm. The lymphocyte shows a round cell morphology, small cytoplasmic volume, and one large nucleus with high RI values (Fig. 2(a)). Although the macrophage contains one nucleus, it shows a larger cytosolic volume compared to the lymphocyte as well as multiple vacuoles in the cytoplasm (Fig. 2(b)). 3D rendered iso-surfaces of the RI values of a lymphocyte and a macrophage respectively visualize 3D morphology of plasma membrane (white) and intracellular organelles (cyan and red), depending on RI values (Figs. 2(b), (d)). For visualization purposes, the threshold iso-surfaces of the plasma membrane (white) and the internal structures (cyan, red) are set to 1.345, 1.375, and 1.39, respectively. The structural information of the 3D RI tomogram are comparable to images obtained by confocal fluorescence microscopy [7] or electron microscopy [39]. 3.2 Quantitative characterizations of WBCs using 3D RI tomograms Measurements of the 3D RI distribution of WBCs can provide morphological (cellular volume, surface area, and sphericity) and biochemical (dry mass and dry mass density) information about individual WBCs. To demonstrate the quantitative imaging capability, we measured the 3D RI tomograms of lymphocytes and macrophages and retrieved morphological and biochemical information pertaining to individual WBCs. Despite the fact that lymphocytes and macrophages are subtypes of a mononuclear cell, they have distinct morphologies and various roles in the immune system. These results are presented in Fig. 3. (a) ) L f ( e m u o v r a u l l l l e C 2000 1000 0 (b) 1500 ) 2 m µ ( a e r a e c a f r u S 1000 500 0 * (c) y t i c i r e h p S 1.0 0.75 0.5 * Lymphocyte Macrophage Lymphocyte Macrophage Lymphocyte Macrophage (d) ) L d / g ( y t i s n e d s s a m y r D 20 10 0 * (e) ) g p ( s s a m y r D 400 300 200 100 0 * Lymphocyte Macrophage Lymphocyte Macrophage Fig. 3. Quantified morphological and biochemical information about individual lymphocytes (n = 29) and macrophages (n = 22). (a) Cellular volume, (b) surface area, (c) sphericity index, (d) dry mass density, and (e) dry mass. Each symbol represents an individual cell measurement and the horizontal black line indicates the mean value with the vertical line of standard deviation. The symbol * indicates a p-value of < 0.001 in the comparison of the lymphocytes and macrophages according to Student’s t test. In order to obtain the morphological information, the iso-surfaces retrieved from the measured 3D RI tomograms are used. The cellular volume V was obtained by integrating voxels which have RI values higher than that of surrounding media in the measured 3D RI tomograms. The mean values of the cellular volumes of individual WBCs are 150.85 ± 31.68 fL (n = 29) and 1106.08 ± 334.28 fL (n = 22), for lymphocytes and macrophages, respectively (Fig. 3(a)). The cell surface area S was obtained by calculating the surface area of the RI iso-surface corresponding to the cell boundaries. The mean values of the cellular surfaces areas are 144.76 ± 21.88 µm2 and 695.84 ± 192.91 µm2 for lymphocytes and macrophages, respectively (Fig. 3(b)). From the measured S and V values, the sphericity SI can be directly calculated, as follows: SI π= ( 1/3 6 V )2/3 S . (1) The SI provides a dimensionless measure of how spherical an object is; The SI of a perfect sphere is 1 and the SI of a plane is 0. The calculated mean values of the SI are determined to be 0.94 ± 0.03 and 0.76 ± 0.10 for lymphocytes and macrophages, respectively (Fig. 3(c)). Measurements of 3D RI tomograms provide quantitative information about local concentrations of non-aqueous biomolecules such as intracellular proteins inside cells. From the average RI values in the cytoplasmic areas of WBCs, the mean dry mass density ρ was calculated. The RI value of the cytoplasm is proportional to the concentration of non-aqueous molecules in the cytoplasm (mostly proteins) according to the relationship n = n0 + αρ [40], where n is the refractive index of protein, n0 is the refractive index of the surrounding medium, and α is the specific refractive index increment (RII) of a protein species. Most proteins have similar RII values [41, 42]; thus, we used the specific RII value of 0.2 mL/g for the calculations in this study. The total cellular dry mass m of a WBC [43-45] was obtained by integrating the local dry mass density ρ over the cell volume V. The mean values of the dry mass density are 18.78 ± 1.97 g/dL and 14.64 ± 1.90 g/dL, and the mean values of the dry mass are 28.50 ± 8.05 pg and 163.16 ± 58.75 pg for lymphocytes and macrophages, respectively (Figs. 3(d), (e)). The majority of the measured lymphocytes were round shapes whereas the macrophages displayed various non-spherical morphologies. The cytoplasm of a lymphocyte is mainly composed of one large nucleus with high RI values, but not in a macrophage. Thus, the cellular dry mass densities of macrophages are significantly lower than those of lymphocytes (Fig. 3(d)). Although the cellular dry mass densities are low in macrophages, the total dry masses of macrophages are approximately four times higher than those of lymphocytes due to their greater cell volumes (Figs. 3(d), (e)). 3.3 Heterogeneous populations even in isolated lymphocytes Heterogeneous populations in isolated lymphocytes were observed in the measured 3D RI tomograms of WBCs. Figures 4(a, b) present quantitative analyses of representative small and large lymphocyte, which results are consistent with previous reports [46, 47]. Previous studies report that lymphocytes can be distinguished into small and large lymphocytes, whereas previous approaches only considered the 2D mean diameters D of cells obtained with light microscopes [46, 47]. The values of the 2D mean diameters obtained in this study are 6.29±0.27 µm and 8.33±0.97 µm for small and large lymphocytes, respectively. These 2D mean diameters are in good agreement with those in previous reports [46, 47]. In order to demonstrate the capability of the present approach, we retrieved morphological and biochemical information pertaining to both small and large lymphocytes (Figs. 5(b)-(f)). The cellular volume and surface area of the large lymphocytes are 2.10 times larger and 1.62 times larger than those of the small lymphocytes, respectively (Figs. 5(b), (c)). However, both cases demonstrate similar spherical shapes and similar RI values. For a further demonstration of the applicability of the present method, we investigated morphological alternation in lymphocytes due to immunogenic stimulation. In order to stimulate lymphocytes, we used lipopolysaccharide (LPS, L3012, Sigma-Aldrich, St. Louis, USA), which the most abundant component in bacterial cell walls. It is known that a treatment of LPS induces functional and morphological changes in immune cells [48, 49]. Isolated lymphocytes were stimulated by a LPS treatment (10 ng/mL) for three hours before measurements were taken. These results are presented in Fig. 4(c) and Fig. 5. The LPS-treated lymphocytes exhibit significant alterations in their morphologies Sphere-like lymphocytes are flattened (the mean SI value decreases from 0.94±0.03 to 0.92±0.05), and the formation of multiple filopodia is clearly shown. We observed that the cellular volume and dry mass of lymphocytes increase upon a LPS treatment, which is consistent with the findings in previous reports [50, 51]. (a) Small lymphocyte (b) Large lymphocyte (c) LPS-treated lymphocyte V= 126.9 fl m= 24.01 pg ρ= 18.92 g/dL S= 125.07 μm2 SI = 0.98 D= 5.09 μm V= 498.55 fl m= 89.02 pg ρ= 17.86 g/dL S= 319.59 μm2 SI = 0.95 D= 9.89 μm V= 289.74 fl m= 54.07 pg ρ= 18.66 g/dL S= 269.9 μm2 SI = 0.78 Fig. 4. Quantitative analyses of representative lymphocytes in heterogeneous populations. 3D rendered RI surfaces of (a) a small lymphocyte, (b) a large lymphocyte, and (c) a LPS-treated lymphocytes. The cellular volume V, cellular dry mass m, dry mass density ρ, surface area S, sphericity SI, and 2D mean diameter D are specified. Scale bar, 2 μm. (a) (b) (c) ) m µ ( r e t e m a D i (d) y t i c i r e h p S * Small Large 12 10 8 6 4 1.0 0.9 0.8 0.7 * Small Large LPS 800 600 400 200 0 24 22 20 18 16 14 ) L f ( e m u o v r a u l l l l e C (e) ) L d / g ( y t i s n e d s s a m y r D ) 2 m µ ( a e r a e c a f r u S (f) ) g p ( s s a m y r D * Small Large LPS 400 300 200 100 0 150 100 50 0 Small Large LPS Small Large LPS Small Large LPS Fig. 5. Heterogeneous populations in isolated lymphocytes. Quantified morphological and biochemical information about small (n=21), large (n=9), and LPS-treated lymphocytes (n=22). (n = 29) and macrophages (n = 22). (a) Diameter, (b) cellular volume, (c) surface area, (d) sphericity, (e) dry mass density, and (f) dry mass. Each symbol represents an individual cell measurement and the horizontal black line indicates the mean value with the vertical line of standard deviation. The symbol * indicates a p-value of < 0.001 in the comparison of the lymphocytes and macrophages according to Student’s t test. 3.4 3D dynamics of micrometer-sized beads ingested by a macrophage To demonstrate the dynamic 3D imaging capability, we emulated the phagocytosis of macrophages using micrometer-sized beads and measured time-lapse dynamic 3D RI tomograms of macrophages with engulfed microspheres. Phagocytosis is a crucial mechanism in the immune system; macrophages identify, track, and ingest harmful pathogens or abnormal cells, eliminating them to protect the host [52, 53]. We added 1 μm of polystyrene beads to a culture medium of macrophages for 1 hour before the measurement (see Methods). Figure 6(a) clearly displays three ingested beads inside a macrophage, as indicated by the RI value of polystyrene (approx. 1.41) inside the cytoplasm. In order to study the temporal dynamics of the ingested beads within a macrophage, a series of 3D RI tomograms was obtained every 10 s. All three microspheres exhibit movement on the x-y plane and in the axial direction (Movie 1). From the series of 3D RI tomograms, the 3D trajectories of each of the beads were retrieved, as presented in Fig. 6(b). The tracking of the microspheres inside living cells provides valuable information about the viscoelastic properties of the cell cytoplasm. Most previous approaches used 2D projected trajectories of target objects in order to assess micro-rheological properties of cells. However, the 3D tracking of objects is technically challenging; holographic field measurements, 3D confocal microscopy, or multiple projection processes with various views have been exploited. Herein, we demonstrate that the dynamic 3D tomographic measurements of cells with engulfed microspheres readily enables an investigation of the viscoelastic properties of the cytoplasm of cells while also visualizing the overall cell shape. This is possible because the quantitative and non-invasive capabilities of the present method, i.e., engulfed microspheres with a specific RI, can be readily identified in measured 3D RI tomograms. The movements of ingested beads may appear to be random Brownian motion, but particle movements within living cells show active non-equilibrium dynamics [54, 55]. To the analyze movements of engulfed particles within a macrophage further, we calculated the mean square displacements (MSDs). A double-logarithmic plot of the MSDs of three beads as a function of the lag time τ is shown in Fig. 6(c), with fitted slopes for both the fast dynamics (τ = 10−40 s) and slow dynamics (τ = 50−100 s). In the fast dynamics, or short timescales, all of the beads show slopes smaller than 1.0, indicating sub-diffusive motion. However, one of the beads exhibits super-diffusive motion (slope = 1.49) in the slow dynamics, suggesting active transport of the bead within the macrophage (b) z Bead 1 Initial position Final position (a) y (c) 1.42 Bead 2 z Bead 1 Bead 2 Bead 3 10 μm 1.34 Bead 3 z x x x y y ) 2 m µ ( 〉 ) τ ( 2 r ∆ 〈 -1 10 0.77 1.49 0.88 Bead 1 Bead 2 Bead 3 0.87 0.64 0.61 101 τ (s) 102 Fig. 6. Time-lapse measurement of the 3D RI tomograms and 3D trajectories of polystyrene beads engulfed by a macrophage. (a) Cross-sectional slice of a 3D RI tomogram of a macrophage on the focal plane. Three engulfed beads are indicated. (b) 3D trajectories of polystyrene beads in (a). The initial and final positions of each bead are indicated by the black and red circles, respectively. (c) MSD of the beads in (b) as a function of the lag time. Dashed lines indicate the fitted slope. 4. Conclusion Here, we present the label-free quantitative RI tomographic imaging of WBCs and the characterization of individual WBCs. We used an interferometric microscope based on Mach-Zehnder interferometry equipped with a 2D galvano mirror. Using this setup, 3D RI maps of individual WBCs are constructed from multiple 2D optical field images obtained with various illumination angles. We demonstrate that 3D RI tomograms of individual WBCs provide quantitative information about the structural and biochemical characteristics of WBCs, including the cellular volumes, surface areas, sphericities, dry mass densities, and total dry mass. In addition, the present approach was applied to an investigation of subpopulations of lymphocytes and to morphological alternations in LPS-treated lymphocytes at the individual cell level. Furthermore, we found that dynamic measurements of 3D RI tomograms of WBCs can be utilized for the study of phagocytosis and of the viscoelasticity of WBSs using a model of engulfed microspheres. The 3D trajectories of engulfed polystyrene microspheres were obtained from dynamic 3D RI tomograms, and these dynamics were quantified by calculating the MSD. Because RI is an intrinsic optical parameter of a material, the present approach does not require the use of exogenous labeling agents, staining procedures, or genetic modifications. This label-free capability of RI measurements can offer long- term observation of individual cells without phototoxicity and photobleaching processes, which can perturb certain cellular functions. More importantly, this RI, an intrinsic optical contrast, was measured quantitatively according to the principles of interferometry or holography. We demonstrated the potentials of the quantitative imaging capability by investigating the quantitative characteristics of the measured 3D RI tomograms. We undertook dynamic measurements of 3D RI tomograms at a speed of 0.1 Hz. However, the acquisition, reconstruction, and visualization speed can be enhanced significantly using a graphics processor unit (GPU) and a sparse sampling method [56]. Recently, our group successfully measured 3D RI tomograms at 60 Hz [57]. From a technical perspective, the present approach can be expanded by combining it with other imaging modalities, including fluorescence imaging [26] and light polarization [58-60]. Furthermore, current optical bright-field microscopes can be converted for ODT purposes using a recently developed QPI unit [61, 62], which will expand the applicability of the present method in immunology without the need for complicated optical alignments. One of the limitations of the present approach is the limited molecular specificity. Although the molecular specificity of label-free RI imaging may not be as high as its fluorescence-labeled counterparts, spectroscopic RI measurements may be used to separate molecules spectrally with various optical dispersions [63-67]. In sum, we envision that ODT will be a useful tool in the study of immunology given its ability to investigate the precise morphologies of cells and subcellular organelles, including nucleoli, and to provide detailed biochemical information. The present approach can be readily applied to the study of WBCs in various disease conditions. For example, leukemia results in very high number of WBCs and induces abnormalities in the morphologies and functions of WBCs [68, 69]. 3D RI tomograms of individual WBCs can be utilized to detect abnormalities in WBCs and to diagnose leukemia in the early stage of the disease. Acknowledgments This work was supported by KAIST, National Research Foundation (NRF) of Korea (2012R1A1A1009082, 2013K1A3A1A09076135, M3C1A1-048860, 2013M3C1A3063046, NRF-2012-M3C1A1-048860, 2013R1A1A3011886, 2014M3C1A3052537), and KUSTAR-KAIST project. KHK is supported by Global Ph.D. Fellowship from NRF. 2. References and links 1. L. Balagopalan, E. Sherman, V. A. Barr, and L. E. Samelson, "Imaging techniques for assaying lymphocyte activation in action," Nature Reviews Immunology 11, 21-33 (2010). J. C. Edwards and G. Cambridge, "B-cell targeting in rheumatoid arthritis and other autoimmune diseases," Nature Reviews Immunology 6, 394-403 (2006). 3. M. T. Heneka, M. P. Kummer, and E. Latz, "Innate immune activation in neurodegenerative disease," Nature Reviews Immunology 14, 463-477 (2014). 4. W. Zou and N. P. Restifo, "TH17 cells in tumour immunity and immunotherapy," Nature Reviews Immunology 10, 248-256 (2010). 5. C. E. Grossi, S. R. Webb, A. Zicca, P. M. Lydyard, L. Moretta, M. C. Mingari, and M. Cooper, "Morphological and histochemical analyses of two human T-cell subpopulations bearing receptors for IgM or IgG," The Journal of experimental medicine 147, 1405-1417 (1978). 6. R. M. Steinman and Z. A. Cohn, "Identification of a novel cell type in peripheral lymphoid organs of mice I. Morphology, quantitation, tissue distribution," The Journal of experimental medicine 137, 1142-1162 (1973). 7. R. Scott Brock, X.-H. Hu, D. A. Weidner, J. R. Mourant, and J. Q. Lu, "Effect of detailed cell structure on light scattering distribution: FDTD study of a B-cell with 3D structure constructed from confocal images," Journal of Quantitative Spectroscopy and Radiative Transfer 102, 25-36 (2006). 8. M. Nieminen, T. Henttinen, M. Merinen, F. Marttila–Ichihara, J. E. Eriksson, and S. Jalkanen, "Vimentin function in lymphocyte adhesion and transcellular migration," Nature cell biology 8, 156-162 (2006). 9. G. Popescu, Quantitative Phase Imaging of Cells and Tissues (McGraw-Hill Professional, 2011). 10. K. Lee, K. Kim, J. Jung, J. Heo, S. Cho, S. Lee, G. Chang, Y. Jo, H. Park, and Y. Park, "Quantitative phase imaging techniques for the study of cell pathophysiology: from principles to applications," Sensors 13, 4170-4191 (2013). 11. M. K. Kim, "Principles and techniques of digital holographic microscopy," SPIE Reviews 1, 018005-018005-018050 (2010). 12. Y. Kim, H. Shim, K. Kim, H. Park, S. Jang, and Y. Park, "Profiling individual human red blood cells using common-path diffraction optical tomography," Scientific reports 4(2014). 13. H. Byun, T. R. Hillman, J. M. Higgins, M. Diez-Silva, Z. Peng, M. Dao, R. R. Dasari, S. Suresh, and Y. Park, "Optical measurement of biomechanical properties of individual erythrocytes from a sickle cell patient," Acta biomaterialia (2012). 14. Y. Park, C. A. Best, T. Auth, N. S. Gov, S. A. Safran, G. Popescu, S. Suresh, and M. S. Feld, "Metabolic remodeling of the human red blood cell membrane," Proceedings of the National Academy of Sciences 107, 1289 (2010). 15. Y. Park, C. A. Best, K. Badizadegan, R. R. Dasari, M. S. Feld, T. Kuriabova, M. L. Henle, A. J. Levine, and G. Popescu, "Measurement of red blood cell mechanics during morphological changes," Proceedings of the National Academy of Sciences 107, 6731 (2010). 16. Y. Park, C. A. Best, T. Kuriabova, M. L. Henle, M. S. Feld, A. J. Levine, and G. Popescu, "Measurement of the nonlinear elasticity of red blood cell membranes," Physical Review E 83(2011). 17. Y. Park, M. Diez-Silva, G. Popescu, G. Lykotrafitis, W. Choi, M. S. Feld, and S. Suresh, "Refractive index maps and membrane dynamics of human red blood cells parasitized by Plasmodium falciparum," Proceedings of the National Academy of Sciences 105, 13730 (2008). 18. G. Popescu, Y. Park, W. Choi, R. R. Dasari, M. S. Feld, and K. Badizadegan, "Imaging red blood cell dynamics by quantitative phase microscopy," Blood cells, molecules & diseases 41, 10-16 (2008). 19. B. Rappaz, A. Barbul, Y. Emery, R. Korenstein, C. Depeursinge, P. J. Magistretti, and P. Marquet, "Comparative study of human erythrocytes by digital holographic microscopy, confocal microscopy, and impedance volume analyzer," Cytometry Part A 73, 895-903 (2008). 20. N. T. Shaked, L. L. Satterwhite, M. J. Telen, G. A. Truskey, and A. Wax, "Quantitative microscopy and nanoscopy of sickle red blood cells performed by wide field digital interferometry," J Biomed Opt 16, 030506 (2011). 21. H. Park, S.-H. Hong, K. Kim, S.-H. Cho, W.-J. Lee, Y. Kim, S.-E. Lee, and Y. Park, "Characterizations of individual mouse red blood cells parasitized by Babesia microti using 3-D holographic microscopy," arXiv preprint arXiv:1505.00832 (2015). 22. M. Mir, T. Kim, A. Majumder, M. Xiang, R. Wang, S. C. Liu, M. U. Gillette, S. Stice, and G. Popescu, "Label-free characterization of emerging human neuronal networks," Scientific reports 4(2014). 23. P. Jourdain, N. Pavillon, C. Moratal, D. Boss, B. Rappaz, C. Depeursinge, P. Marquet, and P. J. Magistretti, "Determination of transmembrane water fluxes in neurons elicited by glutamate ionotropic receptors and by the cotransporters KCC2 and NKCC1: a digital holographic microscopy study," The Journal of Neuroscience 31, 11846-11854 (2011). 24. Y. Kim, H. Shim, K. Kim, H. Park, J. H. Heo, J. Yoon, C. Choi, S. Jang, and Y. Park, "Common-path diffraction optical tomography for investigation of three-dimensional structures and dynamics of biological cells," Optics express 22, 10398-10407 (2014). 25. S. Lee, K. Kim, A. Mubarok, A. Panduwirawan, K. Lee, S. Lee, H. Park, and Y. Park, "High-Resolution 3-D Refractive Index Tomography and 2-D Synthetic Aperture Imaging of Live Phytoplankton," Journal of the Optical Society of Korea 18, 691-697 (2014). 26. Y. Park, G. Popescu, K. Badizadegan, R. R. Dasari, and M. S. Feld, "Diffraction phase and fluorescence microscopy," Opt Express 14, 8263-8268 27. T. A. Zangle, D. Burnes, C. Mathis, O. N. Witte, and M. A. Teitell, "Quantifying biomass changes of single CD8+ T cells during antigen specific 28. A. E. Ekpenyong, S. M. Man, S. Achouri, C. E. Bryant, J. Guck, and K. J. Chalut, "Bacterial infection of macrophages induces decrease in refractive 29. K. J. Chalut, A. E. Ekpenyong, W. L. Clegg, I. C. Melhuish, and J. Guck, "Quantifying cellular differentiation by physical phenotype using digital holographic microscopy," Integr. Biol. 4, 280-284 (2012). microscopy," J Biomed Opt 16, 036004-036004-036007 (2011). for quantitative phase imaging," Optics letters 35, 447-449 (2010). 30. S. Lee, Y. R. Kim, J. Y. Lee, J. H. Rhee, C.-S. Park, and D. Y. Kim, "Dynamic analysis of pathogen-infected host cells using quantitative phase 31. S. S. Kou, L. Waller, G. Barbastathis, and C. J. Sheppard, "Transport-of-intensity approach to differential interference contrast (TI-DIC) microscopy 32. C. Zuo, Q. Chen, W. Qu, and A. Asundi, "Noninterferometric single-shot quantitative phase microscopy," Optics letters 38, 3538-3541 (2013). 33. K. Kim, H. Yoon, M. Diez-Silva, M. Dao, R. R. Dasari, and Y. Park, "High-resolution three-dimensional imaging of red blood cells parasitized by Plasmodium falciparum and in situ hemozoin crystals using optical diffraction tomography," J Biomed Opt 19, 011005-011005 (2014). 34. V. Lauer, "New approach to optical diffraction tomography yielding a vector equation of diffraction tomography and a novel tomographic microscope," (2006). cytotoxicity," PloS one 8, e68916 (2013). index," Journal of biophotonics 6, 393-397 (2013). Journal of Microscopy 205, 165-176 (2002). Optics express 17, 266-277 (2009). 35. Y. Sung, W. Choi, C. Fang-Yen, K. Badizadegan, R. R. Dasari, and M. S. Feld, "Optical diffraction tomography for high resolution live cell imaging," 36. S. K. Debnath and Y. Park, "Real-time quantitative phase imaging with a spatial phase-shifting algorithm," Optics letters 36, 4677-4679 (2011). 37. R. Gerchberg, "Super-resolution through error energy reduction," Journal of Modern Optics 21, 709-720 (1974). 38. A. Ray and B. N. Dittel, "Isolation of mouse peritoneal cavity cells," Journal of visualized experiments: JoVE (2010). 39. J. Savill, A. Wyllie, J. Henson, M. Walport, P. Henson, and C. Haslett, "Macrophage phagocytosis of aging neutrophils in inflammation. Programmed cell death in the neutrophil leads to its recognition by macrophages," Journal of Clinical Investigation 83, 865 (1989). 40. R. Barer and S. Tkaczyk, "Refractive index of concentrated protein solutions," Nature 173, 821-822 (1954). 41. K. E. Handwerger, J. A. Cordero, and J. G. Gall, "Cajal bodies, nucleoli, and speckles in the Xenopus oocyte nucleus have a low-density, sponge-like structure," Molecular biology of the cell 16, 202-211 (2005). 42. M. Mir, Z. Wang, Z. Shen, M. Bednarz, R. Bashir, I. Golding, S. G. Prasanth, and G. Popescu, "Optical measurement of cycle-dependent cell growth," Proceedings of the National Academy of Sciences 108, 13124-13129 (2011). 43. R. Barer, "Determination of dry mass, thickness, solid and water concentration in living cells," Nature 172, 1097-1098 (1953). 44. T. A. Zangle and M. A. Teitell, "Live-cell mass profiling: an emerging approach in quantitative biophysics," Nature methods 11, 1221-1228 (2014). 45. G. Popescu, Y. Park, N. Lue, C. Best-Popescu, L. Deflores, R. R. Dasari, M. S. Feld, and K. Badizadegan, "Optical imaging of cell mass and growth dynamics," American Journal of Physiology-Cell Physiology 295, C538-C544 (2008). 46. T. Timonen and R. Herberman, "Characteristics of human large granular lymphocytes and relationship to natural killer and K cells," The Journal of 47. A. Szenberg and N. Warner, "Quantitative Aspects of the Simonsen Phenomenon: I. The Role of the Large Lymphocyte," British journal of experimental medicine 153, 569-582 (1981). experimental pathology 43, 123 (1962). 48. A. Ngkelo, K. Meja, M. Yeadon, I. Adcock, and P. A. Kirkham, "LPS induced inflammatory responses in human peripheral blood mononuclear cells is mediated through NOX4 and GiÎ±dependent PI-3kinase signalling," J Inflamm (Lond) 9, 1-1 (2012). 49. M. Guha and N. Mackman, "The phosphatidylinositol 3-kinase-Akt pathway limits lipopolysaccharide activation of signaling pathways and expression of inflammatory mediators in human monocytic cells," Journal of Biological Chemistry 277, 32124-32132 (2002). 50. R. A. Miller, S. Gartner, and H. S. Kaplan, "Stimulation of mitogenic responses in human peripheral blood lymphocytes by lipopolysaccharide: serum and T helper cell requirements," The Journal of Immunology 121, 2160-2164 (1978). 51. S. Vogel, M. Hilfiker, and M. Caulfield, "Endotoxin-induced T lymphocyte proliferation," The Journal of Immunology 130, 1774-1779 (1983). 52. D. G. Russell, B. C. VanderVen, S. Glennie, H. Mwandumba, and R. S. Heyderman, "The macrophage marches on its phagosome: dynamic assays of phagosome function," Nature Reviews Immunology 9, 594-600 (2009). 53. P. J. Murray and T. A. Wynn, "Protective and pathogenic functions of macrophage subsets," Nature Reviews Immunology 11, 723-737 (2011). 54. D. Wirtz, "Particle-tracking microrheology of living cells: principles and applications," Annual review of biophysics 38, 301-326 (2009). 55. C. P. Brangwynne, G. H. Koenderink, F. C. MacKintosh, and D. A. Weitz, "Intracellular transport by active diffusion," Trends in cell biology 19, 423- 56. K. Kim, K. S. Kim, H. Park, J. C. Ye, and Y. Park, "Real-time visualization of 3-D dynamic microscopic objects using optical diffraction tomography," 427 (2009). Optics Express 21, 32269-32278 (2013). 57. K. Kim, J. Yoon, and Y. Park, "Simultaneous 3D visualization and position tracking of optically trapped particles using optical diffraction 58. Y. Kim, J. Jeong, J. Jang, M. W. Kim, and Y. Park, "Polarization holographic microscopy for extracting spatio-temporally resolved Jones matrix," 59. Z. Wang, L. J. Millet, M. U. Gillette, and G. Popescu, "Jones phase microscopy of transparent and anisotropic samples," Optics letters 33, 1270-1272 tomography," Optica 2, 343-346 (2015). Optics Express 20, 9948-9955 (2012). (2008). 60. T. Colomb, F. Dürr, E. Cuche, P. Marquet, H. G. Limberger, R.-P. Salathé, and C. Depeursinge, "Polarization microscopy by use of digital holography: application to optical-fiber birefringence measurements," Applied optics 44, 4461-4469 (2005). 61. K. Lee and Y. Park, "Quantitative phase imaging unit." 62. K. Kim, Z. Yaqoob, K. Lee, J. W. Kang, Y. Choi, P. Hosseini, P. T. So, and Y. Park, "Diffraction optical tomography using a quantitative phase 63. J. Jung, K. Kim, H. Yu, K. Lee, S. Lee, S. Nahm, H. Park, and Y. Park, "Biomedical applications of holographic microspectroscopy [Invited]," Applied imaging unit," Opt Lett 39, 6935-6938 (2014). optics 53, G111-G122 (2014). encoded light (TROVE)," Nature Photonics 7, 300-305 (2013). in red blood cells," Optics Express 20, 9673-9681 (2012). 64. B. Judkewitz, Y. M. Wang, R. Horstmeyer, A. Mathy, and C. H. Yang, "Speckle-scale focusing in the diffusive regime with time reversal of variance- 65. Y. Jang, J. Jang, and Y. Park, "Dynamic spectroscopic phase microscopy for quantifying hemoglobin concentration and dynamic membrane fluctuation 66. M. Rinehart, Y. Zhu, and A. Wax, "Quantitative phase spectroscopy," Biomed. Opt. Express 3, 958-965 (2012). 67. Y. Park, T. Yamauchi, W. Choi, R. Dasari, and M. S. Feld, "Spectroscopic phase microscopy for quantifying hemoglobin concentrations in intact red blood cells," Optics letters 34, 3668-3670 (2009). medicine 71, 633-637 (2004). 68. N. S. Majhail and A. E. Lichtin, "Acute leukemia with a very high leukocyte count: confronting a medical emergency," Cleveland clinic journal of 69. M. Mohty, D. Jarrossay, M. Lafage-Pochitaloff, C. Zandotti, F. Brière, X.-N. de Lamballeri, D. Isnardon, D. Sainty, D. Olive, and B. Gaugler, "Circulating blood dendritic cells from myeloid leukemia patients display quantitative and cytogenetic abnormalities as well as functional impairment," Blood 98, 3750-3756 (2001).
1310.5601	1	1310	2013-10-21T15:30:26	Neuronal Growth as Diffusion in an Effective Potential	[ "physics.bio-ph" ]	Current understanding of neuronal growth is mostly qualitative, as the staggering number of physical and chemical guidance cues involved prohibit a fully quantitative description of axonal dynamics. We report on a general approach that describes axonal growth in vitro, on poly-D-lysine coated glass substrates, as diffusion in an effective external potential, representing the collective contribution of all causal influences on the growth cone. We use this approach to obtain effective growth rules that reveal an emergent regulatory mechanism for axonal pathfinding on these substrates.	physics.bio-ph	physics	Neuronal Growth as Diffusion in an Effective Potential Daniel J. Rizzo1, James D. White1,2, Elise Spedden1, Matthew R. Wiens1, David L. Kaplan2, Timothy J. Atherton1, Cristian Staii1,* 1. Department of Physics and Astronomy, Center for Nanoscopic Physics, Tufts University, 2. Department of Biomedical Engineering, Tufts University, Medford, MA 02155 Medford, MA 02155 [] Corresponding Author: Prof. C. Staii, E-mail: [email protected] Keywords: Neuron, Axonal Growth, Fokker-Planck, Stochastic Process Submitted to Physical Review E 1 Abstract Current understanding of neuronal growth is mostly qualitative, as the staggering number of physical and chemical guidance cues involved prohibit a fully quantitative description of axonal dynamics. We report on a general approach that describes axonal growth in vitro, on poly-D- lysine coated glass substrates, as diffusion in an effective external potential, representing the collective contribution of all causal influences on the growth cone. We use this approach to obtain effective growth rules that reveal an emergent regulatory mechanism for axonal pathfinding on these substrates. 2 The basic working unit of the nervous system is the neuron, a highly specialized cell consisting of three main structural components (Fig. 1a): the cell body (soma), several branching dendrites, and a single long axon that transmits electrical impulses to other neurons. In the developing nervous system each newly formed neuron extends an axon, which navigates through a complex and changing environment to reach dendrites from other target neurons and subsequently form functional connections called synapses [1]. Axonal guidance is governed primarily by the growth cone, a complex sensing unit located at the distal tip of the axon (Fig. 1a) that responds to a host of biophysical, chemical and mechanical cues. The high motility of the growth cone is based upon its cytoskeleton, a dynamic and flexible biopolymer network made from actin and microtubule filaments and their associated regulatory molecules. Collectively, these control the growth cone shape and its mechanical stability, mediate its sensing, guide the intracellular transport of various biomolecules, and direct axonal elongation [1]. It is now generally accepted that axons do not simply rely on an intrinsic “program of directions” that would uniquely specify each one of the billions of connections that form the neuronal architecture, being described instead by a set of general rules that apply across a large number of neurons and patterns of connections [2]. Finding the fundamental physical principles that govern the development of connections and communications between neurons is one of the key problems in biological physics. The main challenge faced when studying these processes in vivo lies both in the complex and highly controlled structure of neuronal matter in the nervous system, as well as in the complexity of the interplay between different environmental cues. Therefore, an alternative approach [3, 4] is often used to uncover the basic rules that underlie the formation of functional neuronal connections. Within this approach one aims to 3 create a simplified neuronal growth environment in vitro and to systematically investigate the effect that various cues: chemotactic, biochemical, mechanical, and topographical, have on the formation of neuronal networks. These studies show that physical stimuli (gradients of various molecular species, stiffness of the growth substrate, traction forces generated during axonal extension etc.) play a key role in the wiring of the nervous system [3-9]. However, our current understanding of neuronal growth is mostly qualitative, the vast complexity of the parameter space still prohibiting fully quantitative predictions of outcomes from given initial conditions such as: geometry of the neuronal circuit, type of biochemical cues on the growth substrate, topography or mechanical properties of the substrate. A general description of axonal growth must take into account the inherent stochastic nature of this process due to different chemotactic signals, internal biochemical reactions and the randomness of external signals of various strengths [5, 7, 10-20]. Much of the previous work on stochastic effects has focused on describing axonal movement in the context of either intercellular diffusion of known neuron growth factors [5, 11-14], or intracellular events, such as polymerization of cytoskeletal structures [15-17], production of substrate adhesions sites [6], and transmembrane receptor activity [18]. In these cases, a model is motivated by a known underlying mechanism, whose validity is then tested against experimental data. Conversely, stochastic models have been recently introduced in order to provide purely phenomenological descriptions of axonal growth in some special cases, such as edge movement of the growth cone [10] or growth on asymmetrical surfaces [19]. In this paper, we present a general framework, based on the Fokker-Planck (FP) equation, to quantitatively describe and predict axon growth dynamics for cortical neurons (obtained from rat embryos) cultured on poly-D-lysine coated glass surfaces, through systematic measurements 4 of axon growth velocity (throughout the paper referring specifically to the time-derivative of the axon arclength). We show that on these surfaces the axonal growth is governed by an effective potential, which incorporates all of the causal influences on the growth cone, and determines the evolution of its velocity distribution function. We find that axonal growth is not governed by simple diffusion, being instead described by a Laplace velocity distribution. The resulting time- dependent solutions of the 1-dimensional FP equation are used to extract an effective diffusion coefficient for axonal elongation rates. We demonstrate that this general model can be used to quantitatively describe the long-term behavior of growth cone dynamics and to predict experimental outcomes. All measurements were performed on day 18 embryonic rat cortical neurons, cultured on poly-D-lysine (PDL) coated glass substrates. After plating, cells were allowed to incubate at 37°C for 8, 15, 19, 26, 33, or 46 hours. Samples are then removed from the incubator, loaded into the BioHeater™ Closed Fluid Cell, and imaged under bright-field using the optical stage of the MFP-3D-BIO Atomic Force Microscope (AFM, Asylum Research) [9, 19] (Fig. 1). The sample is then imaged every 5 minutes for a total period of 20 minutes (Fig. 2b,c). This time interval was chosen in order to allow enough time for typical axon growth to exceed the precision of our axon measurement (~0.1µm), while being short enough to accurately approximate an instantaneous velocity. The Nikon NIS-Elements Basic Research software [9, 19] is used to measure the change in axon length, yielding 3 consecutive values of the 5-minute average axon growth velocity. While Fig. 1 shows an example of axon extension, retraction (i.e. negative velocity) and zero-growth rates are also observed. 5 FIG. 1. (a) Cultured rat cortical neuron labeled with main structural components. (b), (c) Examples of axonal arclength measurements. Measurements are performed before (b) and after (c) a 5-minute wait period. The average velocity is determined from the difference between L1 and L2 divided by the 5- minute time interval. The growth description is presented as a one-dimensional probability distribution of the measured velocity. Axon elongation rates are collected for each incubation period, resulting in 6 distinct time-lapse sets for the average growth velocity, each set containing between 140 and 250 data points. These are combined to form a single time-independent distribution containing a total of 986 observations (Fig. 2a). We note that the measured velocities follow a Laplace distribution, 6 and not a Gaussian distribution that would be expected if the growth cone moved according to a Brownian random walk [21]. We show this by calculating the relative likelihood that observed data was produced by a Gaussian vs. Laplace distribution, i.e. the ratio of the products of the respective functions (Gaussian vs. Laplace) evaluated for all data points shown in Fig. 2a. The relative likelihood is found to be extremely small (~10-60) demonstrating that the Laplace distribution is definitively favored. To gain a clearer insight into the growth dynamics that lead to the observed velocity distribution we construct a general model using the 1-dimensional Fokker-Planck equation: (1) where is the probability distribution for velocity, is the effective potential that governs the evolution of this distribution, and D is the diffusion coefficient in velocity space. We note that in general, equation (1) could also incorporate drift terms to account for biochemical or mechanical interactions between neurons, or between neurons and the growth substrate. For example, in our previous work we have used a constant velocity drift term to quantify axonal bias imparted by mechanical interactions between the growth cone and asymmetric growth surfaces [19]. The potential V(v) in eqn. 1 can be obtained from the time-integrated data: first we find the stationary solution ps(v) by setting the time derivative to zero and then solve equation (1) for the potential V(v): (2) where N is a normalization constant. Hence, for a constant D the effective potential that governs the axonal growth may be deduced directly from the observed time-independent distribution ps(v) as displayed in Fig. 2. 7 FIG. 2. (a) Time-integrated data of axon elongation rates. Fitting a Laplace distribution yields the mean velocity = 5.1 μm/hr and scale parameter 1/κ = 16.7 μm/hr. (b) V-shaped velocity-potential obtained from the probability distribution shown in (a) using equation (2). Error bars represent a 95% confidence interval of a binomial distribution. The time-integrated data shown in Fig. 2a follows a Laplace distribution: (3) with the parameter values: κ = 0.06 hr/µm and = 5.1 µm/hr. From eqns. (2) and (3) we obtain an effective potential: (4) 8 The normalization constant N in Eq. 4, may be ignored, as it has no bearing on the final form of the normalized solution p(v,t). We note that a simple random walk would result in a parabolic potential [21]. The rather uncommon V-shape of the effective potential provides insight into an effective mechanism governing axonal growth on PDL coated glass surfaces. The constant absolute value of the slope of the potential for all , regardless of how far it may be from the preferred , indicates that a constant effective drift force, corresponding to the derivative of the potential, “pushes” elongation rates toward the preferred value. That is, the axon appears to have a bimodal tendency to restore the growth velocity from “too fast” or “too slow” regimes. We speculate that this behavior may be a result of the previously observed bistable process involved in the growth of leading edge lamellipodia which act as the sensors of the growth cone [10]. We now turn to the time-evolution of the growth process. From the experimentally determined time-independent potential , one may solve for the time-dependent distribution by substituting eqn. (4) into eqn. (1), and selecting an appropriate initial condition. A well-known method [21] is used to determine the time-dependent solution by transforming (1) into a Schrödinger-like equation of the new function governed by a new potential : (5) (6) (7) 9 Inserting (4) into (7), one finds that for a V-shaped potential, the corresponding Schrödinger potential will be . Thus, one may obtain the solution of the Fokker- Planck equation for our system from the solution of the Schrödinger equation in the elementary delta-function potential well. The solution of the corresponding Schrödinger equation has a single bound state, , and a continuum of eigenstates sorted by parity [21]: (8) (9) While the bound state has an eigenvalue of 0, the other eigenvalues are: where k is a real number. Each unbound state has the typical exponential time-dependence: . Given our choice of initial conditions (described below) and eqn. (6), the coefficients of the eigenfunction expansion are determined, which in turn yield the final solution upon summation and integration over k. Together, these determine the final solution of the FP equation (Fig. 3) [22]. 10 FIG. 3. Theoretical time-evolution of axonal velocity distributions that follow an initial truncated Laplace distribution (dotted line) defined in the range (a) [-2.5,12.5], and (b) (-∞, -2.5)∪(12.5, ∞). All units are in µm/hr. These initial distributions evolve in time toward the time-independent Laplace distribution given by eqn. (2) (represented by the continuous curves in (a) and (b), respectively). The time evolution is governed by the Fokker-Planck equation; two intermediate states are shown as dotted-dashed lines in (a) and (b). The diffusion coefficient for axonal growth velocity is determined by fitting these solutions to the corresponding initial subsets in the experimental time-lapse data (see Fig. 4). Inspection of time-dependent data reveals that for all incubation periods (i.e. 8, 15, 19, 26, 33, or 46 hours), the velocity distribution is better described by the time-independent solution (3), than any time-dependent solution with physically reasonable initial conditions [22]. This shows that for each incubation period the overall velocity distributions are stationary over the timeframe of the observation (20 minutes). However, for any individual neuron, axon velocities might vary considerably among the 3 time points measured within the 20-minute observation timeframe, showing that measureable evolution does take place over a timescale of minutes. To quantify the dynamics on these shorter timescales, time-lapse velocity measurements for all 11 neurons are combined to form two sets of 3 time-dependent distributions. One set incorporates only those neurons whose initial velocity falls in the range of [-2.5,12.5] µm/hr (Fig. 4b), and the other only those that do not fall in this range. In so doing, we limit our observation to two non- overlapping, non-equilibrium subpopulations of neurons, the evolution of which may be quantified with our solutions (Fig. 3). A best-fit value of D for both experimental subsets is determined by selecting that D which maximizes the likelihood of measuring the given data sets, while the FWHM of the peak of the likelihood is used to calculate an error on D [22]. With this method, the joint fit of all incubation periods yields an effective value for the diffusion coefficient of D = µm2/hr3. FIG. 4. Plot of initial measurements for all neurons from all incubation periods (a) before and (b) after limiting observation to specific subset of data, with velocities in the interval [-2.5,12.5] µm/hr, indicated with arrows. (c) The distribution of velocities for this subset of neurons measured after 5 and (d) 10 minutes, showing evolution toward a Laplace stationary distribution. Similar behavior is observed for the complimentary data set. 12 To our knowledge, this is the first instance in which a diffusion (or random motility) coefficient for growth cones has been measured. Although the above analysis outlines the measurement of velocity-space diffusion, we have also determined the position-space diffusion coefficient from axon length data. For regular diffusion, where the effective potential in the Fokker-Planck equation is and the velocity distribution is Gaussian, the mean square displacement (MSD) grows linearly with time , where the proportionality constant Dp is the position diffusion (i.e. random motility) coefficient [23, 24]. We established, using a Monte Carlo simulation of an ensemble of random walkers where the Gaussian distribution was replaced with a Laplace distribution, that the linear time dependence of the MSD is expected to remain linear as well for the Laplace distribution. Using experimental data from all incubation periods, it was found that mean-squared axon length indeed has linear time dependence, with a diffusion coefficient: Dp = (15 ±1) µm2/hr [22]. A simple dimensional analysis argument demonstrates the consistency of the position and velocity space measurements. One may derive a characteristic position-space diffusion coefficient from the parameters in the Fokker-Planck equation κ and D, by recognizing that these define a characteristic velocity of and a time scale of experimental determined value of κ and D, we have that: Substituting . has a value Dp ~ 6 µm2/hr, which is comparable to the value obtained from the MSD analysis. We note that this value is also comparable to the diffusion coefficients measured for human peritoneal mesothelial cells in vitro [23], and about two orders of magnitude smaller than the values obtained for glioma cells, which are known to exhibit abnormally high motility rates [24]. 13 In this paper, we have obtained effective rules for axonal growth on poly-D-lysine coated glass surfaces that extend beyond that of simple diffusion. Using the Fokker-Planck equation, we quantify the bimodal growth behavior of axons on these surfaces as diffusion in a V-shaped potential. This potential represents the collective contribution of all causal influences on the growth cone, and reveals an emergent growth rate regulatory mechanism for neurons on PDL coated surfaces. We emphasize that the symmetric V-shaped potential that governs the axonal growth rates on PDL coated glass surfaces, in the absence of other chemotactic sources is not universal, and does not describe neuronal growth in general. Other types of potentials are expected to be found on different growth environments and/or on different types of surfaces, such as asymmetric and/or spatially dependent potentials resulting from chemotactic sources, from growth on asymmetric or mechanically heterogeneous surfaces etc. Future studies that apply Fokker-Planck formalism (including possible drift terms and spatially dependent probability distributions) for studying growth dynamics under well-defined external conditions, such as neuron growth under controlled geometries on protein-patterned substrates, and on substrates of different stiffness or topography, may therefore provide a general framework for quantitative description of the role that different types of environmental cues have on axonal growth. Future studies might also provide new insight into the intracellular mechanisms that lead to the emergent growth rules observed in these controlled experiments, by correlating the growth potential with forces and stresses generated during growth (measured for e.g. by Traction Force Microscopy), as well as with cytoskeletal rearrangements measured using fluorescent techniques. Acknowledgement: The authors thank Dr. Steve Moss’s laboratory (Tufts Center for Neuroscience) for providing embryonic rat brain tissues. The authors gratefully acknowledge 14 financial support for this work from National Science Foundation (NSF-CBET 1067093) and Tufts University. Author Contributions: TJA, DLK, and CS planned and supervised the research. DJR, JDW, MW and ES cultured the cells and measured neuronal growth. DJR, TJA and CS analyzed the data and developed the theoretical model for axonal growth. All authors contributed to writing and revising the manuscript, and agreed on its final contents. References [8] [9] [4] [1] [2] [3] L. A. Lowery and D. Van Vactor, Nat. Rev. Mol. Cell. Biol. 10, 332 (2009). H. D. Simpson, D. Mortimer, and G. J. Goodhill, Curr. Top. Dev. Biol. 87, 1 (2009). C. Staii, C. Viesselmann, J. Ballweg, J. C. Williams, E. W. Dent, S. N. Coppersmith, and M. A. Eriksson, Langmuir 27, 233 (2011). C. Staii, C. Viesselmann, J. Ballweg, L. Shi, G. Y. Liu, J. C. Williams, E. W. Dent, S. N. Coppersmith, and M. A. Eriksson, Biomaterials 30, 3397 (2009). [5] G. J. Goodhill and J. S. Urbach, J. Neurobiol. 41, 230 (1999). [6] M. Vanveen and J. Vanpelt, J. Theor. Biol. 159, 1 (1992). [7] Y. E. Pearson, E. Castronovo, T. A. Lindsley, and D. A. Drew, Bull. Math. Biol. 73, 2837 (2011). D. Koch, W. J. Rosoff, J. Jiang, H. M. Geller, and J. S. Urbach, Biophys. J. 102, 452 (2012). E. Spedden, J. D. White, E. N. Naumova, D. L. Kaplan, and C. Staii, Biophys. J. 103, 868 (2012). [10] T. Betz, D. Lim, and J. A. Kas, Phys. Rev. Lett. 96, 098103 (2006). [11] G. J. Goodhill, Trends Neurosci. 21, 226 (1998). [12] G. J. Goodhill and H. Baier, Neural Comput. 10, 521 (1998). [13] H. G. Hentschel, and A. van Ooyen, Proc. Biol. Sci. 266, 2231 (1999). [14] R. Segev, and E. Ben-Jacob, Neural Networks 13, 185 (2000). [15] H. M. Buettner, Cell. Motil. Cytoskeleton 32, 187 (1995). [16] D. J. Odde, E. M. Tanaka, S. S. Hawkins, and H. M. Buettner, Biotechnol. Bioeng. 50, 452 (1996). [17] M. E. Robert and J. D. Sweeney, J. Theor. Biol. 188, 277 (1997). [18] H. Meinhardt, J Cell Sci 112, 2867 (1999). [19] R. Beighley, E. Spedden, K. Sekeroglu, T. Atherton, M. C. Demirel, and C. Staii, Appl. Phys. Lett. 101, 143701 (2012). [20] A. Granato and J. Van Pelt, Dev. Brain Res. 142, 223 (2003). [21] H. Risken, The Fokker-Planck Equation: Methods of Solution and Applications (Springer, Berlin, 1996), 2nd ed. [22] See Supplemental Material at [URL inserted by publisher] for further explanation of solutions, fitting and simulation. [23] P. K. Maini, D. L. McElwain, and D. I. Leavesley, Tissue Eng. 10, 475 (2004). 15 [24] K. R. Swanson, Math. Comput. Model. 47, 638 (2008). 16 Supplemental Material for Neuronal Growth as Diffusion in an Effective Potential Daniel J. Rizzo1, James D. White1,2, Elise Spedden1, Matthew R. Wiens1, David L. Kaplan2, Timothy J. Atherton1, Cristian Staii1, 1. Department of Physics and Astronomy, Center for Nanoscopic Physics, Tufts University, Medford, MA 02155 2. Department of Biomedical Engineering, Tufts University, Medford, MA 02155 [*] Corresponding Author: Prof. C. Staii, E-mail: [email protected] Keywords: Neuron, Axonal Growth, Fokker-Planck, Stochastic Process Submitted to Physical Review E 17 In this supplemental material, we outline the calculation of the time-dependent solutions of the Fokker-Planck equation for a V-shaped potential (S1), and describe how this solution is used to find a best-fit value for the velocity-space diffusion coefficient (S2). We then present simulation results that demonstrate the linear behavior of mean square displacement (MSD) with time for random-walkers whose velocities are drawn from a Laplace distribution. These are presented alongside experimental time-dependent MSD data (S3). Finally, we present experimental data that shows the general behavior of axon velocities that lead to Laplace distributions for each incubation period (S4). 18 S1. Time-Dependent Fokker-Planck Solution in a V-shaped Potential Fig 2a (main text) shows that the time-independent experimental data is well described by a Laplace distribution. From this, we have derived a V-shaped potential using the Fokker- Planck Equation (Fig. 2b). Given that the shape of the potential is known, a time-dependent solution can be determined. To do this, a well-known method is used that transforms the Fokker-Planck equation of the function into a Schrodinger-like equation of the function [1]. The relationship between these functions is: (SE1) where is the time-independent distribution given by equation (3) in the main text. As stated in the main text, solving for in a V-shaped potential is equivalent to solving for in an elementary Dirac Delta function potential well, using the transformation (SE1). Thus, the eigenstates of are: (SE2) (SE3) (SE4) where (SE2) has an eigenvalue λ=0, and the states (SE3) and (SE4) have a continuum of real eigenvalues defined as: 19 (SE5) where k is a real number. These solutions define the usual time-dependence: (SE6) As explained in the main text, the two non-intersecting, non-equilibrium initial conditions for p(v,t), with the corresponding initial q(v,t) eigenfunctions (see SE1) are given by: (SE7) (SE8) (SE9) (SE10) In order to determine the final solution, the coefficients for the eigenfunction expansion must be determined for each state. This is done by projecting each set of eigenstates onto the respective initial conditions, where the subscript i =1, 2 denotes which of the two initial 20 conditions are being used: (SE10) (SE11) (SE12) Thus, the final form of the function can be determined by integrating over all continuous eigenstates, and adding the one discrete state ψ0: (SE13) The final solution for is determined by plugging (SE13) into (SE1) (with given by equation (3) in the main text). 21 S2. Maximum Likelihood Fitting Procedure and Errors For N measured velocities vi, each with known associated time ti and probability distribution , the likelihood function is defined as: The value of the diffusion coefficient D that maximizes L is the best-fit value. The reported error on this value is the FWHM of the peak in the likelihood function. Here, the natural log of the likelihood function is presented for all data combined, and the ln[2] is subtracted from the maximum to find the FWHM. FIG. S1. Maximum likelihood fitting for D. 22 S3. Determination of MSD with time for Laplace Velocity Distribution We have performed a Monte-Carlo simulation of an ensemble of random walkers whose velocities are drawn from both Gaussian and Laplace velocity distributions for a given diffusion coefficient, and distribution width. FIG. S2. Monte Carlo simulation of mean square displacements for random walkers drawn from Gaussian (blue) and Laplace (red) distributions. FIG. S3. Experimentally determined mean square displacement vs. time for the ensemble of growth cones. The error bars represent the standard error on the MSD for each time point. The slope for the Gaussian MSD is equal to 2Dp (see main text). Given the observed similarity 23 in slope between Laplace and Gaussian cases (Fig. S2), the experimental value of Dp is extracted from the MSD assuming a Gaussian random walk. From the slope of the linear fit in Fig. S3 we obtain: Dp = (15±1) μm2/hr, where the reported uncertainty is the error in the linear fit. 24 S4. Time-Dependent Distributions by Incubation Period Incubation Period Slope of V(v)/D (hr) Mean/Minimum (μm/hr) (hr/μm) 8 15 19 26 33 46 4.7 4.5 -9.5 7.3 6.3 10.3 0.08 0.10 0.05 0.04 0.08 0.05 FIG. S4. Top: velocity distributions at different incubation periods. Bottom: table showing the velocities at which the V-shape potential has a minimum, and the gradient (slope) of the potential for each incubation period. 25 When attempting to fit a time-dependent solution to the data sets shown in Fig. S4 it was determined that they were better described by the time-independent solution then any time- dependent solution. To quantify this, the initial condition was assumed to be δ(v), which is a physically reasonable assumption given that no axons are growing during plating. Given this condition, the likelihood (example shown below) as a function of the fit parameter Dt (i.e. the product between diffusion coefficient and time) is observed to plateau at high values of Dt, without the presence of a local maximum (i.e. best-fit parameter Dt). This can be explained by noting that the time-dependent solution approaches the time-independent solution for large Dt, and thus the plateau is the result of the time-dependent solution effectively reaching the equilibrium or stationary state. The fact that the likelihood is at its highest value at this point reveals that this stationary solution is a better fit than any previous time point, i.e. for each incubation period the overall velocity distributions are stationary over the timeframe of the observation (20 minutes). [1] H. Risken, The Fokker-Planck Equation: Methods of Solution and Applications (Springer, Berlin, 1996), 2nd edn. 26
1605.04041	1	1605	2016-05-13T04:30:55	Schistosoma mansoni cercariae exploit an elastohydrodynamic coupling to swim efficiently	[ "physics.bio-ph", "physics.flu-dyn", "q-bio.QM" ]	The motility of many parasites is critical for the infection process of their host, as exemplified by the transmission cycle of the blood fluke Schistosoma mansoni. In their human infectious stage, immature, submillimetre-scale forms of the parasite known as cercariae swim in freshwater and infect humans by penetrating through the skin. This infection causes Schistosomiasis, a parasitic disease that is comparable to malaria in its global socio-economic impact. Given that cercariae do not feed and hence have a finite lifetime of around 12 hours, efficient motility is crucial for the parasite's survival and transmission of Schistosomiasis. However, a first-principles understanding of how cercariae swim is lacking. Via a combined experimental, theoretical and robotics based approach, we demonstrate that cercariae propel themselves against gravity by exploiting a unique elastohydrodynamic coupling. We show that cercariae beat their tail in a periodic fashion while maintaining a fixed flexibility near their posterior and anterior ends. The flexibility in these regions allows an interaction between the fluid drag and bending resistance: an elastohydrodynamic coupling, to naturally break time-reversal symmetry and enable locomotion at small length-scales. We present a theoretical model, a 'T-swimmer', which captures the key swimming phenotype of cercariae. We further validate our results experimentally through a macro-scale robotic realization of the 'T-swimmer', explaining the unique forked-tail geometry of cercariae. Finally, we find that cercariae maintain the flexibility at their posterior and anterior ends at an optimal regime for efficient swimming, as predicted by our theoretical model. We anticipate that our work sets the ground for linking the swimming of S. mansoni cercariae to disease transmission and enables explorations of novel strategies for Schistosomiasis prevention.	physics.bio-ph	physics	Schistosoma mansoni cercariae exploit an elastohydrodynamic coupling to swim efficiently Deepak Krishnamurthy1, Georgios Katsikis1, Arjun Bhargava2 & Manu Prakash3∗ 1Department of Mechanical Engineering, 2Department of Applied Physics 3Department of Bioengineering, Stanford University, Stanford, CA ∗To whom correspondence should be addressed; E-mail: [email protected] Keywords Motility, Parasites, Swimming, Hydrodynamics, Low Reynolds number, Schistosomiasis Abstract The motility of many parasites is critical for the infection process of their host, as exemplified by the transmission cycle of the blood fluke Schistosoma mansoni [1]. In their human infectious stage, immature, submillimetre-scale forms of the parasite known as cercariae swim in freshwater and infect human hosts by penetrating through the skin [1, 2]. This infection causes Schistosomiasis, a parasitic disease that is compa- rable to malaria in terms of global socio-economic impact [3, 4]. Given that cercariae do not feed and hence have a finite lifetime of around 12 hours [5, 6], efficient motility is crucial for the parasite's survival and transmission of the Schistosomiasis disease. However, a first-principles understanding of how cercariae swim is completely lacking. Via a combined experimental, theoretical and robotics based approach - we demon- strate that cercariae efficiently propel themselves against gravity by exploiting a unique elastohydrodynamic coupling. We show that cercariae beat their tail in a periodic fashion while maintaining a fixed flexibility near their posterior and anterior ends. The flexibility in these regions allows an interaction between the fluid drag and bending resistance - an elastohydrodynamic coupling, to naturally break time-reversal symmetry and enable locomotion at small length-scales [7]. We present a theoretical model, a 'T-swimmer', which captures the key swimming phenotype of cercariae. We further validate our results experimentally through a macro-scale robotic realization of the 'T-swimmer', explaining the unique forked-tail geometry of cercariae. Finally, we find that cercariae maintain the flexibility at their posterior and anterior ends at an optimal regime for efficient swimming, as predicted by our theoretical model. We anticipate that our work sets the ground for linking the swimming of S. mansoni cercariae to disease transmission and enables explorations of novel strategies for Schistosomiasis control and prevention. 1 Introduction In their natural habitat of rivers and ponds, S. mansoni cercariae are released from in- termediate snail hosts and aggregate below the water surface to increase their chances of finding a human host [8, 9] (Fig. 1a). The cercariae are around 500 µm long and consist of an anterior head, a slender tail and a posterior bifurcation of the tail called the 'furca', henceforth termed the fork (Fig. 1b) [10, 11]. Being negatively buoyant, they need to swim against gravity in order to stay near the water surface, which they accomplish by using their fork [10, 11]. The fork is an uncommon appendage, not found in any well-studied swimming microorganisms such as bacteria [12], algae or ciliates [13, 14]. Also, unlike the aforemen- tioned microorganisms, cercariae do not feed [5, 6]. Hence their swimming efficiency is an important determinant of their ability to survive longer to find and infect humans. For such an important human parasite, while a body of work has focused on the behavioral [1, 2] and qualitative aspects [10, 15] of cercarial motility, the swimming mechanics of cercariae remains unknown. Macroscale swimming behaviour To mimic the natural habitat of cercariae we conducted laboratory experiments with multiple freely swimming specimens in a 4 cm × 1 cm × 1.4 mm chamber which is much larger than their own size (≈ 500 µm). We observed that cercariae display an intermittent swimming behaviour consisting of three modes [10, 11] (Fig. 1 b): A tail-first mode in which they swim up against gravity with their fork fully extended (Fig. 1 b (i)), a free-sinking mode on account of being negatively buoyant where the fork is partially extended (Fig. 1b (ii)) and a head-first mode which is used to penetrate through the skin, where the fork is folded back (Fig. 1b (iii)). We analysed trajectories (n = 2774 segments; Methods) of individual cercariae and obtained the relative time durations spent in each mode. We found that cercariae spend a majority of their time in the free-sinking mode (83 % of the total time) 2 Figure 1 Macroscale swimming behaviour of Schistosoma mansoni cercariae. a. Artistic rendering (not-to-scale) of a typical transmission site where cercariae (immature, human infectious of the parasites) emerge into freshwater from snails and aggregate near the water surface [8]. b, Artistic rendering of the three distinct modes displayed by cercariae in their intermittent swimming behaviour termed (i) tail-first mode and (ii) free-sinking mode and (iii) head-first mode. Arrows indicate the direction of motion. Scale bar 100 µm. c. Snapshot of cercariae swimming in a chamber (Methods) showing tail-first swimming (red) and sinking (blue) trajectories for 10 minutes. The resulting tracks of organisms have a columnar nature with very little drift in the horizontal directions. This is due to the fact that cercariae are bottom-heavy swimmers (Supplementary Fig. S2) and undergo a gravitational realignment such that their longitudinal axis is aligned with the vertical with the head pointing down. Scale bar is 1 mm. 3 with a mean sinking speed (cid:104)Vsinking(cid:105) = 0.1 mm s−1 followed by the tail-first swimming mode (17 %) with (cid:104)Vtail−f irst(cid:105) = 0.7 mm s−1. The head-first mode is often triggered by chemical cues [16, 17, 18] which are absent in our current experiments, leading to very few observed trajectories (< 0.1%). Despite the fact that cercariae spend longer periods of time freely sinking, we expect that most of the energy consumption occurs during the tail-first mode [5], since this involves active swimming. Thus the swimming efficiency of the tail-first mode directly affects the ability of cercariae to aggregate near the water surface to increase chances of encountering a human host. Also, given that the swimming gait of cercariae in the tail-first mode is distinct from that of well-studied microorganisms [14], it is of fundamental interest to understand its mechanics. Swimming kinematics: Tail-first swimming To study the tail-first swimming mode we conducted high-speed imaging experiments (Meth- ods) on individual cercariae. We observed that cercariae swim using a periodic gait, wherein they beat their tail from side-to-side with a frequency f in the range 15 − 20Hz (Fig. 2a). The plane in which the tail beats varies much slower (≈ 250 ms) than a swimming cycle (tcycle = 1/f ≈ 50 ms), hence justifying a two-dimensional description of the swimming motion. For understanding the hydrodynamics of cercarial swimming, we estimated the relevant Reynolds number Re = (cid:104)Vtail−f irst(cid:105)L/ν ≈ 0.3, where L ≈ 500 µm is the organism length and ν = 10−6 m2s−1 is the kinematic viscosity of the ambient fluid (water). This Re < 1 indicates the relative dominance of viscous over inertial forces in the swimming. To con- firm this we measured the flow-field around the organism using Particle-Image-Velocimetry (PIV) (Methods) and observed regions of recirculating fluid (vortices) which remain adja- cent to the organism (Fig. 2d and e). The flow field exhibits stagnation points that reverse orientation in each cycle. The vorticity (a measure of local fluid rotation rate [19]) shows peaks at the surface of the organism, with no vortex shedding: a characteristic feature of 4 Shape kinematics of S. mansoni cercariae swimming in the tail-first mode. a, Figure 2 Extracted dynamic shapes from high speed videos of cercariae at different time-points during a swim- ming cycle (Methods). The trajectory of the head is colour-coded based on the instantaneous velocity magnitude along the local tangent (s) to the mean swimming trajectory shown as a solid black line in the last panel. b, Illustration of kinematic parameters: the angle at the tail-head joint φth shown in blue, the angle at the tail-fork joint φtf shown in red and an effective bending angle of the tail φt shown in orange. c, Time series plot of the tangential head velocity along s (left axis) demonstrates forward motion although mean-curvature of the tail (right axis) is symmetric, seemingly appearing to be a reciprocal stroke. Gray regions (1 → 2, 3 → 4 etc) correspond to power-strokes where the head of the organism has a tangential velocity in the mean swimming direction, while white regions (2 → 3, 4 → 1(cid:48) etc) are recovery-strokes where the head moves in the opposite direction. Time is normalized by 1/f , where f = 15.6 Hz is the beat frequency. d, e Snapshots of the transient flow velocity exhibiting reversing stagnation points. Vorticity fields obtained using particle-image-velocimetry (PIV) during the power-stroke and recovery-stroke, respectively. The regions of vorticity remain attached to the organism and are not shed. f, Phase plot of φth and φtf as a function of φt clearly shows a finite area hence re- vealing the role of tail-fork and tail-head joints in producing a non-reciprocal swimming stroke, required for low-Reynolds-number swimming. Scale bars are 100 µm. 5 flow dominated by viscous forces [20]. Based on nature of the vorticity field and other flow metrics (Supplementary Fig. S1a, S1b, S1c) we concluded that inertial forces are negligible in cercarial swimming. We therefore studied the hydrodynamics of cercarial swimming using purely viscous or Stokes flow theories which correspond to the limit Re = 0 [19]. To achieve locomotion, cercariae, like any swimmer dominated by viscosity, must change their shape in a manner that breaks time-reversal symmetry [7, 13]. Using image analysis we extracted the dynamic shape of the organism (Fig. 2 a) and observed that their periodic gait consists of successive power and recovery strokes that break time-reversal symmetry. During a power stroke, the tail bends from a straightened state to its maximally curved state while the fork sweeps out an arc, moving almost perpendicular to its longitudinal axis (sequence 1 → 2 in Fig. 2a, c, d and f). The fork's motion is directed towards the head of the organism, generating a net thrust force that pulls the head forward along the mean swimming direction s (sequence 1 → 2 in Fig. 2 a, c). During a recovery stroke the tail returns to its straightened state and the fork moves parallel to its own longitudinal axis (sequence 2 → 3 in Fig. 2 a, c, e and f) pushing the head backwards. Owing to the different orientation of the fork, this backward displacement is smaller than that during the power stroke thus leading to a cumulative displacement of the head along s. The head of the organism additionally generates thrust by breaking time-reversal-symmetry (sequence 1 and 3 in Fig. 2c and f) in a manner similar to conventional Stokesian swimming strategies which involve waving an elastic arm [21]. At low Re the fluid drag on a body is only weakly anisotropic as compared to the same body at high Re. For slender bodies, such as the fork in cercariae (aspect ratio ≈ 20), the drag at low Re is around twice as large when the body moves perpendicular to its longitudinal axis as compared to parallel to this axis [22, 23]. By moving the fork perpendicular and parallel to its axis during the power and recovery strokes, respectively, cercariae seem to maximally utilize this weak anisotropy. The kinematics of the fork and head in relation to the tail, as quantified by the respective 6 joint angles (Fig. 2b), highlight the non-reciprocal nature of the swimming gait which breaks time-reversal symmetry (Fig. 2f). Interestingly, the tail bends symmetrically and does not contribute to the breaking of time-reversal symmetry (right axis of Fig. 2 c). Therefore cercarial locomotion crucially depends on the degree of freedom at the tail-fork and tail-head joints (Fig. 2b). This degree of freedom appears to be a result of increased local flexibility in the joints due to smooth-muscle mediated constrictions in the transverse dimension of the organism [10, 24]. Theoretical models for swimming cercariae To further understand the swimming mechanism and exactly how these joint angles evolve over time, we hypothesize the simplest passive control strategy where a balance between elastic bending stiffness and viscous drag forces dictates the exact kinematics of cercarial swimming - an elastohydrodynamic effect, rather than active muscular control. We conjec- ture that cercariae maintain a fixed local stiffness at these joints using smooth musculature, only modifying the joint stiffness when they change their swimming mode. Our hypothe- ses are supported by earlier studies, which demonstrate that the striated muscles that are capable of rapid beating are confined to the tail, and not present in the fork [11, 25]. To test our hypotheses we developed a theoretical model: a 'T-swimmer', consisting of three, linear, rigid links (Fig. 3a (ii)). The first two links correspond to the fore and aft ends of the head-tail portion in cercariae, and third link which is attached transversely at its mid-point, corresponds to the fork (Fig. 3a (i) and (ii)). In our geometry, as a starting point, we neglect the head in cercariae and also the effects of gravity, to make the simplest possible swimming model (Fig. 3a (ii)). As the reader will quickly notice, our 'T-swimmer' is inspired by Purcell's three-link-swimmer [7], often referred to as the simplest low-Reynolds-number swimmer since it requires only two actively actuated joints. In contrast to Purcell's swim- mer, in the T-swimmer only a single longitudinal joint is actively actuated (red dot in Fig. 3a (ii)), to model the bending of the tail of cercariae, while the transverse joint is passive 7 Figure 3 Theoretical 'T-swimmer' and scaled-up robotic model for swimming S. mansoni cercariae a, Schematic of (i) cercariae swimming tail-first and (ii) proposed 'T-swimmer' model. (iii) Photograph of a macroscale, self-propelled, 'T-swimmer' robot (Methods) designed to swim in a high- viscosity fluid (corn syrup, viscosity µ ≈ 8 P a s) chosen to be dynamically similar (Rerobot ≈ 0.2) to swimming cercariae. The joint between the longitudinal links (red dot in (ii)) is active and actuated periodically with a given amplitude A and frequency f . The transverse joint is assumed to be a passive linear torsional spring (depicted as a black spiral) with stiffness Γtf to model the flexibility of the tail-fork joint in cercariae. The red and blue arrows in (iii) indicate the active and passive joints, respectively in the robot. b, c Phase plots of the joint angles φtf and φt for a T-swimmer model (b) and robot (c) for a range of Γtf showing the non-reciprocal nature of the swimming cycle, where Γtf is Γtf normalized by the torque scale µf l3 c . The arrows indicate the direction of phase trajectories. d, e Plots of the average swimming speed for the T-swimmer model (d) and robot (e) as a function of Γtf for different actuation amplitudes A. Both the model (d) and robot (e) swim with the passive joint preceding the active joint, and the average swimming speed shows a single maximum at an O(1) value of Γtf , highlighting an optimal value of torsional stiffness for a given 'T-swimmer'. The horizontal lines in e indicate measured swimming speeds for a robot for a free-joint (dashed lines) and fixed-joint (dotted lines) which are an order of magnitude smaller than the peak values. Error bars correspond to standard-deviations over different experiments. f, Snapshots of final positions after 60 s of swimming for T-swimmer robots with a range of Γ (i)-(v) and frequency maintained at ≈ 0.4Hz. The white dashed line denotes the starting point of the robots. The free (i) and fixed joint (ii) robots show relatively small displacements. 8 and assumed to be a linear torsional spring (blue spiral in Fig. 3a (ii)). The parameters of the T-swimmer include the lengths of the links (l1, l2, l3) and their transverse dimensions (r1, r2, r3); the actuation amplitude of the active joint A, the driv- ing frequency f and the torsional stiffness of the passive joint Γ (Fig. 3a (ii)). We non- dimensionalized the system using the half-length of the organism lc = L/2 and the time scale tc = 1/f . The force and torque scales follow the viscous scaling Fc = µf l2 c and τc = µf l3 c , respectively [19]. Therefore the key dynamical parameter of the system is the dimensionless torsional stiffness Γ = Γ/(µf l3 c ). To specify the hydrodynamic forces and torques on the T-swimmer we used local slender- body-theory [22, 23] and solved the resulting equations of motion numerically (Supplemen- tary Information section 2.5). Our results show that the T-swimmer swims with the passive joint preceding the active one, similar to cercariae swimming tail-first. In contrast, a slight variation of the classic Purcell's swimmer, which has one active and one passive joint, swims in an opposite direction to the T-swimmer, with the active joint preceding the passive one [26]. Our results demonstrate that a simple transition from a longitudinal to a transverse link, either of which are attached via a torsional spring, results in a reversal in swimming direction - a viable strategy for any organism trying to reverse swimming directions from head-first to tail-first. For validating our theory we also performed scaled-up experiments on a centimetre-scale (Lrobot ≈ 10 cm) robot mimicking our T-swimmer (Fig. 3a (iii)). The robot was immersed in a corn-syrup medium to achieve dynamically similar conditions (Rerobot ≈ 0.2) to swimming cercariae (Methods). We found that both the T-swimmer (Fig. 3b) and the robot (Fig. 3c), reproduce the non-reciprocal swimming gait of cercariae thereby supporting our hypothesis (Fig. 2f). For both the T-swimmer and the robot, we varied Γ and A and found that in the upper and lower limits of joint stiffness (Γ (cid:29) 1 and Γ (cid:28) 1), the gait becomes reciprocal (Fig. 3b, c) and the swimming speed averaged over > 60 cycles is negligible (Fig. 3d, e, f). This is explained by the fact that for Γ (cid:29) 1 the swimmer has a single degree of freedom 9 since the tail-fork joint is rigid, and for Γ (cid:28) 1, there is again a single effective degree of freedom due to the torque free condition at the tail-fork joint. For both the T-swimmer and the robot, there is an intermediate value of torsional stiffness Γ that optimizes the average swimming speed (Fig. 3d, e and f). This happens due to resonant interaction between the driving time scale (tf orcing = 1/f ) and the intrinsic relaxation time of the torsional spring (trelaxation = µl3 c /Γ) i.e 1/f ∼ µl3 c /Γ. Comparison of models and biological experiments As a next extension to the three-link T-swimmer, we added a fourth link to account for the thrust generated by the head in cercariae and also included the effects of gravity (Supple- mentary Information section 2.2). This fourth link is attached longitudinally via a torsional spring to model the flexibility of the tail-head joint (Fig. 4a). The resulting model can now be compared to experimental measurements on S. mansoni cercariae. To make this comparsion, we need the torsional spring stiffness of the two joints (Γth and Γtf ) in a living (swimming) cercariae. We developed a novel in situ measurement using high-speed imaging data (Fig. 2) and hydrodynamic theory. To do this, first we used slender- body-theory [19] to estimate the instantaneous torques at the two joints τ tf and τ th from the translational and angular velocities of the fork and head (Fig. 4b and Supplementary Information section 2.3). We then investigated the relationship between these torques and the corresponding joint angles φtf and φth and found that they are linearly correlated (Fig. 4c). This linearity implies that the two joints are indeed well-approximated as linear torsional springs, serving as a posteriori verification of our T-swimmer model. Finally we obtained the torsional spring stiffness of the two joints as the slopes of the linear fits. By comparing our four-link T-swimmer with experiments on S. mansoni cercariae, we find that the theoretical four-link T swimmer indeed captures the swimming displacement per cycle (Fig. 4d) and the time evolution of the non-reciprocal joint angle kinematics very well (Fig. 4e). A crucial connection to disease transmission lies in the swimming efficiency of this 10 Figure 4 Comparison between four-link 'T-swimmer' and swimming S. mansoni cercariae.a, Schematic of a four-link swimmer model overlayed on an image of a cercariae. The flexibility at the tail-fork and tail-head joints in cercariae is modelled via linear torsional springs depicted as black spirals at the respective joints. b, Free-body-diagrams of the head and fork showing translational and angular velocities and the resulting hydrodynamic torque about the respective joints. This torque is balanced instantaneously by the mechanical torques at the joints, thus allowing in situ means of estimating these torques (Methods and Supplementary Information section 2.3). Scale bar 100µm. c, Plot of the estimated torques at the tail-head (blue triangles) and tail-fork (red circles) joints as a function of the corresponding joint angles φth and φtf . The slope of the respective linear fits (R-squared tail-head 0.7 and tail-fork 0.8) shown as solid blue and dashed red lines gives the joint stiffnesses (Γth and Γtf ) in live, swimming cercariae. d, Time series plot of displacements of the head along s and n for the four-link T-swimmer with estimated joint stiffnesses (black solid and dashed curves) compared with experimental measurements (red circles and green triangles). e, Phase plot between φth, φtf and φt for the four-link T-swimmer, shown as blue and red lines, and for cercariae shown as blue triangles and red circles. f, g Contour plots of average swimming speed (f) and swimming efficiency (g) as a function of the joint stiffnesses Γtf and Γth show a single optimal point (black crosses). The red dots indicate the values estimated for cercariae showing that they optimize their joint stiffnesses to swim efficiently. 11 human parasite. Next, we estimate the cercarial swimming efficiency using our T-swimmer model. To measure how effectively cercariae can swim up the water column, we defined this efficiency as the ratio of average distance covered relative to the average viscous power dissipation in the fluid during a swimming cycle ηdist = Xstroke/Pdiss,stroke (Supplementary Information section 2.5). By exploring a range of values in the (Γtf , Γth) plane we find a single optimal point for both the swimming speed (Fig. 4 f) and efficiency (Fig. 4 g). Interestingly, we observed that the joint spring stiffnesses estimated for cercariae (red dots in Fig. 4 f and g) lies very close to this optimal peak. This suggests that cercariae maintain the stiffness of these two joints at the optimal range required for efficient swimming. This tuning may have resulted from strong evolutionary pressures on the parasite since the host-seeking processes is energy constrained [5, 6]. Outlook Our work points to an unusual elastohydrodynamic effect in cercarial swimming, where the tail provides the energy for propulsion but no thrust while tail-head and tail-fork joints act as passive torsional springs providing all the thrust. This passive control strategy is minimalistic in nature where the only control lies in tuning the flexibility of the respective joints. Furthermore, cercariae are the first reported organisms to use both a longitudinal (the head) and transverse element (the fork) [14, 27] to generate a net propulsion from bending waves propagating in opposite directions (due to the tail). Since current mass drug administration strategies have significant limitations [28], mech- anistically linking biophysics of human-seeking parasites such as S. mansoni to disease trans- mission in ecological field conditions could inspire a new paradigm of environmental control of this neglected tropical disease. The limited energy reserves of the cercarial stage of this parasite might be its Achilles heel. Being of a higher density than water, simply targeting the swimming motility could impair the parasites ability to aggregate near the air-water interface, thereby reducing transmission rates. Towards this, our study of the swimming 12 mechanics has given the first quantitative estimate of the swimming efficiency, and allows an understanding of how perturbations to various parameters affect the motility of cercariae. Acknowledgements We thank all members of Prakash lab for fruitful discussions. DK is supported by a Stanford Bio-X Bowes fellowship. GK was supported by scholarships from the Onassis and the A.G. Leventis Foundation. MP is supported by a Multidisciplinary University Research Initiatives (MURI) grant. We thank Dr. Judy Sakanari of UCSF for providing lab space and organisms. We thank Mattias Lanas for the scientific illustrations of cercariae in their natural habitat. References [1] Wilfried Haas. PHYSIOLOGICAL ANALYSIS OF CERCARIAL BEHAVIOR. 78(2):243–255, 1992. [2] Wilfried Haas. Parasitic worms: strategies of host finding, recognition and invasion. Zoology (Jena, Germany), 106(4):349–64, January 2003. [3] Peter J Hotez and Aruna Kamath. Neglected tropical diseases in sub-saharan africa: review of their prevalence, distribution, and disease burden. PLoS Negl Trop Dis, 3(8):e412, 2009. [4] Peter J Hotez and Alan Fenwick. Schistosomiasis in Africa: an emerging tragedy in our new global health decade. PLoS neglected tropical diseases, 3(9):e485, January 2009. [5] J Ruth Lawson and RA Wilson. The survival of the cercariae of schistosoma mansoni in relation to water temperature and glycogen utilization. Parasitology, 81(02):337–348, 1980. [6] P. J. Whitfield, a. Bartlett, N. Khammo, and R. H. Clothier. Age-dependent survival and infectivity of Schistosoma mansoni cercariae. Parasitology, 127(1):29–35, July 2003. 13 [7] Edward M. Purcell. Life at low Reynolds number. Americal Journal of Physics, 45(1):3– 11, 1977. [8] Wilfried Haas, Bernadett Beran, and Christina Loy. Selection of the host's habitat by cercariae: from laboratory experiments to the field. Journal of Parasitology, 94(6):1233– 1238, 2008. [9] C Combes, A Fournier, H Mon´e, and A Th´eron. Behaviours in trematode cercariae that enhance parasite transmission: Patterns and processes. Parasitology, 109:S3, 1994. [10] W. Hohorst & H. DR AGER G. Graefe. Forked Tail of the Cercaria of Schistosoma mansonia Rowing Device. 215:207–208, 1967. [11] CJ Nuttman. The Structure and Behaviour of the Cercaria of Schistosoma Mansoni. PhD thesis, University of York, 1975. [12] Howard C Berg. Random walks in biology. Princeton University Press, 1993. [13] Eric Lauga and Thomas R Powers. The hydrodynamics of swimming microorganisms. Reports on Progress in Physics, 72(9):96601, 2009. [14] Ganesh Subramanian and Prabhu R Nott. The Fluid Dynamics of Swimming Microor- ganisms and Cells. Journal of the Indian Institute of Science, 91(3):283–314, 2012. [15] DAP Bundy. Swimming behaviour of the cercaria of transversotrema patialense. Para- sitology, 82(02):319–334, 1981. [16] Wilfried Haas, Simone Haeberlein, Sabina Behring, and Eveline Zoppelli. Schistosoma mansoni: human skin ceramides are a chemical cue for host recognition of cercariae. Experimental parasitology, 120(1):94–7, September 2008. [17] Sebastian Brachs and Wilfried Haas. Swimming behaviour of Schistosoma mansoni cercariae: responses to irradiance changes and skin attractants. Parasitology research, 102(4):685–90, March 2008. 14 [18] W Haas, K Grabe, C Geis, T Pach, K Stoll, M Fuchs, B Haberl, and C Loy. Recog- nition and invasion of human skin by Schistosoma mansoni cercariae: the key-role of L-arginine. Parasitology, 124(Pt 2):153–167, 2002. [19] L Gary Leal. Advanced transport phenomena: fluid mechanics and convective transport processes, volume 7. Cambridge University Press, 2007. [20] Josu´e Sznitman, Xiaoning Shen, Raphael Sznitman, and Paulo E Arratia. Propulsive force measurements and flow behavior of undulatory swimmers at low reynolds number. Physics of Fluids (1994-present), 22(12):121901, 2010. [21] Chris H. Wiggins and Raymond E. Goldstein. Flexive and propulsive dynamics of elastica at low reynolds number. Phys. Rev. Lett., 80:3879–3882, Apr 1998. [22] R G Cox. The motion of long slender bodies in a viscous fluid part 1. general theory. Journal of Fluid mechanics, 44(04):791–810, 1970. [23] G K Batchelor. Slender-body theory for particles of arbitrary cross-section in Stokes flow. Journal of Fluid Mechanics, 44(03):419–440, 1970. [24] J F Reger. Studies on the fine structure of cercarial tail muscle of Schistosoma sp. (Trematoda). Journal of ultrastructure research, 57(1):77–86, October 1976. [25] CJ Nuttman. The fine structure and organization of the tail musculature of the cercaria of schistosoma mansoni. Parasitology, 68(02):147–154, 1974. [26] Emiliya Passov and Yizhar Or. Dynamics of purcell's three-link microswimmer with a passive elastic tail. The European Physical Journal E, 35(8):1–9, 2012. [27] Christopher Brennen. Locomotion of flagellates with mastigonemes. Journal of Mechanochemistry and Cell Motility, 3(3):207–217, 1975. 15 [28] Donato Cioli, Livia Pica-Mattoccia, Annalisa Basso, and Alessandra Guidi. Schistoso- miasis control: praziquantel forever? Molecular and biochemical parasitology, 195(1):23– 29, 2014. [29] William Thielicke and Eize J Stamhuis. Pivlab-time-resolved digital particle image ve- locimetry tool for matlab. Published under the BSD license, programmed with MATLAB, 7(0.246):R14, 2010. [30] James Lighthill. Flagellar hydrodynamics. Siam Review, 18(2):161–230, 1976. 16 Supplementary Information for "Schistosoma mansoni cercariae exploit an elastohydrodynamic coupling to swim efficiently" 1 Methods 1.1 Experiments on Live Cercariae 1.1.1 Organism preparation Live organisms were obtained from UCSF (Judy Sakanari Lab) and experiments were con- ducted there. Experiments on vertical swimming of cercariae were carried out in a flow chamber with height 4 cm, width 1 cm and thickness 1.4 mm. Standard soft lithography techniques were used to create Polydimethysiloxane (PDMS) chambers of the required di- mensions which were then bonded on to corona treated glass slides (Electron Microscopy Sciences) of dimensions (75mm × 51mm × 1.2mm). One side of the resulting flow chamber was left exposed to air so that when the chamber was partially filled, an air-water interface was present. This setup mimics the natural conditions in which cercariae swim vertically in the water column. Cercariae freshly shed from a snail were washed and a fixed volume of snail-conditioned-water (100µl) containing the organism was introduced into the chamber so that around 75% of the chamber height was full. 1.1.2 Imaging system For imaging the swimming of cercariae, the flow chamber was mounted statically on a op- tics board( Thor Labs) with a plane mirror as a backing. A reflection microscopy platform was constructed using a modular infinity microscope system (Applied Scientific Intrumenta- tion), along with warm white light (500 mW LED source ,Thor Labs, MWHCL3) channeled through a fibre optic cable and combined with a collimator lens and beam-splitter (Edmund 17 Optics, 50R/50T plate). Infinity-corrected, long-working-distance objectives (Mitutoyo Cor- poration) were used to achieve a magnification range of 1× to 50×, with a corresponding resolving power of 11µm to 0.5µm. Image acquisition, at up to 2000 fps, was done through a high-speed camera (Phantom v12.10). The entire reflection microscopy system including the high-speed camera was mounted on a x − y − z stage to provide translation capability along all three axes. 1.1.3 Swimming trajectory and Particle-Image-Velocimetry For obtaining bulk-swimming trajectories, measurements were made using the reflection mi- croscopy system at 100 fps at 2× magnification. This allowed a large enough field-of-view (circular with diameter 12mm, ≈ 30 body lengths) so that bulk swimming characteristics could be observed yet the resolution allowed detection of the precise shape of individual cer- cariae. The resolution of the shape of the organism was important to classify the trajectories into different swimming modes. For detailed analysis of the swimming kinematics of individual organisms, data was ob- tained at 1000 fps at 10× and 20× magnification. This allowed a swimming stroke to be resolved into ≈ 50 time points which was enough to resolve the detailed motions of the head, tail and fork during a swimming stroke, which were obtained using an in-house image pro- cessing tool. For measuring the flow-field around the organisms, the snail-conditioned-water was seeded with 2 µm beads (Fluoresbrite, calibration grade) and images were acquired at 1000 fps at 10× and 20× magnification. From these images the shape of the organism was extracted using an in-house image processing tool and applied as a dynamic mask to ex- clude regions containing the organism. Velocity fields were obtained from this post-processed image using an open-source Particle-Image-Velocimetry tool implemented in MATLAB [29]. 18 1.2 Scaled-up experiments An autonomous macro-scale robot was designed and built to fully explore the novel 'T- swimmer' geometry proposed in the paper. The robot had three links in the form of rect- angular plates fabricated from acrylic sheets of thickness 1.5 mm. Of these two were lon- gitudinal links while one was transverse resulting in a three-link 'T-swimmer'. The height of the links was 5 cm and the total length of the robot was ≈ 10 cm. The joint between the longitudinal links is actively driven by a micro-metal gear motor (100:1 medium-power with dual-shaft, Pololu Robotics and Electronics) controlled by circuits and power systems which are onboard the robot to avoid the undesirable effects of external wires or control lines. A closed-loop control circuit was developed using a LB 1836 IC motor driver, Texas Instruments functioning as a half H-bridge to change the direction of rotation of the motor. To precisely measure the angular displacement of the motor shaft an optical encoder system (Pololu Robotics and Electronics) was attached to the auxiliary shaft of the motor and its output fed to a differential comparator (LM 311, Texas Instruments). This resulted in an angular resolution of 0.72◦. The motors and control circuit were interfaced using a PICAXE 08M2 micro-controller which allowed programmability of the amplitude of the active joint. The motor and control circuits were independently powered by 3.7 V Lithium-Polymer bat- teries (Tenergy Corporation). The motor, control electronics and batteries were housed in a water-tight box in the distal longitudinal link of the swimmer. The stiffness of the passive 'T-joint' was modified by using torsional springs (180◦ steel music wire, McMaster-Carr). By using a number of springs in parallel and also by using individual springs of varying stiffness, around two orders of magnitude in torsional stiffness of the passive joint was achieved. The 'T-swimmer' robot was placed in a tank of dimensions (90cm × 45cm × 40cm) filled with light corn syrup (Karo). The rheological properties of the corn syrup were measured using a high-resolution rheometer (TA Instrument ARES-G2 ) and was found to be Newto- nian over a shear rate of 0.1 − 10 s−1 with a viscosity of µ ≈ 8P as at 20◦C. The robot was 19 made neutrally buoyant and stable by suitably adding ballast so that the centre-of-gravity was close to the centre-of-buoyancy. The robot thus swims in a horizontal plane and gravity does not play a role in the motion. The microcontroller system was pre-programmed and the robot was placed near the middle of the tank. An overhead camera (Canon EOS Rebel T5) was set up to record the motion of the robot. To capture the shape kinematics colored markers were placed at different points on the robot (Fig. 3 f). The resulting videos (Fig. 3 f) were processed to obtain the x − y locations of the markers. This allowed us to derive the both the overall kinematics of the robot as well as the kinematics of the joint angles (as shown in Fig. 3 c, e). 1.3 Image Processing For the analysis of the grayscale video data from the experiments with live cercarie, two codes were used. The first code was used to track the trajectories of multiple cercarie swimming in the vertical chambers (Fig. 1 c). This code was written in MATLAB and utilized an open-source code (MATLAB Particle Tracking Code by Daniel Blair and Eric Dufresne) to track the cercarie. The post-processing of the tracking data, including the identification of swimming modes, was done using a custom MATLAB code. To identify the swimming mode for each tracked cercarie, the instantaneous velocity was calculated and the swimming mode was determined based on the sign and the magnitude of that velocity. The second code was used to study single cercarie (Fig. 2) by identifying their distinct parts, including the head, the fork, and the tail, in order to track their positions and shapes as cercarie swim over time. The head and the fork were tracked by a user-defined mask that was manually set for the first frame of each video sequence. For the subsequent frames the head and the fork were tracked automatically using a brightness threshold. The code was written in MATLAB and fit ellipsoids on the head and the fork to extract the coordinates of their centroids, their major and minor axes and the orientation angles for each frame. The tail of the cercarie was extracted by subtracting the head and the fork from the whole body of the organism 20 that was also tracked automatically using a brightness threshold. To extract the curvature and the intersection of the angles of the tail at joints with the head and fork, the tail was skeletonized and a spline was fit along its length. For the analysis of the RGB video data from the scaled-up experiments with the macroscopic robot, a custom code was written in MATLAB. The code used the K-means method to obtain the locations and extract the coordinates of the coloured markers placed at different points of the robot. 2 Additional Discussion 2.1 Fluid Dynamical Regime for Cercariae In this section we establish that the flow around swimming cercariae, and hence the swimming mechanics, is dominated by viscous rather than inertial effects. To discuss the hydrodynamics of the ambient fluid, the relevant dimensionless number is the Reynolds number (Re), which is defined as: Re = U L ν , (1) where U is the swimming speed of the organism, L is the characteristic length scale, and ν = 10−6m2s−1 the kinematic viscosity of water which is the ambient fluid. Low-Reynolds- numbers (or Stokesian) flows are characterized by the relative dominance of viscous stresses compared to inertial stresses in the flow. For cercariae, the Reynolds number based on the maximum measured swimming speed of 1.5mm/s and maximum half length of the organism L = 250µm (from the tip of the head to the base of the fork, with the main tail fully extended), is around 0.3. In the near-field of the organism, a second Reynolds number may be defined based on the beating frequency of the tail as: (2) Ref = f Ld ν 21 (a) (b) (c) a, Fluid velocity magnitude (normalized by the velocity scale lcf ) in a region surrounding Figure S1 a freely swimming cercaria. b, Decay of the fluid velocity magnitude normal to the axis of the tail (n), ensemble averaged over different spatial locations along the tail and time instants, plotted as a function of normalized distance away from the swimmer surface (r/L), where L is the length of the organism. r = 0 corresponds to the surface of the swimmer. The solid black curve is the theoretical prediction for an undulatory swimmer at low-Reynolds-numbers [30]. c, Fourier transform in time of the average fluid velocity magnitude in a region surrounding the swimmer. The extent of this region is around 2 swimmer lengths. where f is the beat frequency of the tail whose maximum value from experiments is around 20Hz, and d is the tail diameter which is measured to be 20µm. This Reynolds-number then 22 Įŝg0.25 mm00.511.52radial distance along a normal (r/L) 00.20.40.60.811.2Fluid velocity magnitudeV/Vmax (ExperimentsTheory (exp(-2 π r/ λ)050100150200Frequency (Hz)00.010.020.030.040.050.06Amplitude (Y(f))15.63 Hz60.55 Hz31.25 Hzhas the value: Ref = 20(250 × 10−6)20 × 10−6 1 × 10−6 = 0.1 (3) Both these Reynolds numbers are higher than typically studied Stokesian micro-swimmers such as bacteria and algae which is around Re ∼ O(10−4)[13], and fall near the upper-bound of what is usually regarded the Stokesian regime. Hence it is worth investigating in detail if the flow-field and hence swimming mechanics is qualitatively dominated by viscous or inertial effects. To do so we consider several metrics based on the measured velocity field around freely swimming cercariae. We first consider the spatial decay of the fluid velocity magnitude in a direction normal to the swimmer surface (see Fig. S1a), ensemble averaged over different time and spatial locations along the tail of the swimmer (Fig. S1b). The values are normalized by the mean of the maximum fluid velocity magnitude for each time instant and spatial location. The resulting data is seen to collapse onto a single curve which compares well to an exponential decay of the form exp(−2πr/λ), where r is the spatial distance away from the swimmer surface and λ is the characteristic wavelength of the undulations, which for the case of cercariae is given by λ = 2ltail. This is because the tail bends in a symmetric fashion with the tail-fork and tail-head joints acting like nodes. This exponential decay is a characteristic feature of Stokesian flow around an undulatory swimmer [20, 30]. Fig. S1c shows the Fourier transform of the spatially-averaged fluid velocity magnitude as a function of time. The fundamental frequency of the signal is seen as a sharp peak at f ≈ 15Hz which is the beat frequency of the organism in this particular experiment, while the higher frequency peaks occur at approximately 2f and 4f . This characteristic nature of the power spectrum with sharp peaks at f , 2f and 4f , points to the flow around the organism adjusting instantaneously to the motion of the organism. This linear response of the fluid to the imposed boundary motion is a typical feature of low-Reynolds-number flows. The peak at 2f is attributed to the swimming gait consisting of beating on the dorsal and ventral side of the organism. On the other hand, the 4f peak is likely due to the motion 23 a, Snapshots at 1 s intervals of a cercaria passively sinking and reorienting due to gravity. Figure S2 b, A schematic depiction of the location of the center-of-gravity and centre-of-buoyancy both of which, by symmetry, lie on the central axis, separated by a distance Lg. c, A reduced-order model to allow calculation of Lg based on a balance of gravitational and viscous torques. d, Experimental measurements and theoretical prediction of angle of the longitudinal axis with the vertical during passive sinking. Inset shows snapshots of the model at 1 s intervals. of the fork, which, during a swimming stroke, undergoes four distinct motions (two power and two recovery strokes in alternating fashion). That this is due to the presence of the fork is seen by comparing the spectrum for cercariae to that of C. elegans [20]. This organism being a simple undulatory swimmer with no fork shows peaks at f and 2f but no peak at 4f . In summary, all the metrics for the fluid flow surrounding the organism presented in the main paper and also above point to the underlying Stokesian flow physics which governs the swimming mechanics of cercariae. 2.2 Estimating the location of the centre-of-gravity and centre-of-buoyancy for cercariae Cercariae are negatively-buoyant, bottom-heavy swimmers. To model their swimming it is important to know their excess density compared to water and also the realigning torque due 24 θgxCOGxCOBLggxCOGxCOBLgθFgFb0123456Time51015202530354045θExperimentModelabcdt=01 s2 s3 s4 s5 s1 st=02 s3 s4 s5 sto gravity. In this section, we use trajectories of cercariae passively sinking and reorienting due to gravity to estimate this excess density as well as locations of centre-of-gravity (COG) and centre-of-buoyancy (COB) in cercariae. This in turn we use to estimate the gravitational torque. Since there is no available information regarding the mass or density of individual anatom- ical parts of cercariae, we first restrict ourselves to solving for the average density of the organism (defined as mass of organism by total volume) based on the sinking speeds of the cercariae. Further, we assume each part of the cercarial anatomy has a uniform density. This is a reasonable assumption since each of these sub-divisions have a distinct set of mus- cle groups and smaller scale anatomical structures unique to that portion [24]. It is possible to show that assumption of uniform density for each segment of the cercaria precludes the effect of gravity in the bending of the tail-fork and tail-head joints. The shape of a cercariae sinking passively with the fork partially extended is shown in Fig. S2a, and the reduced-order-model of the same using slender rods is shown in Fig. S2b. In trajectories where the cercaria sank while oriented parallel to gravity, there is no reorientation and we can simply equate the net buoyant force to that due to the total hydrodynamic drag giving: (ρcerc − ρf )Vcercg = µUsink(CL,headlhead + CL,tailltail + CN,f orklf ork) (4) where ρcerc and ρf are the densities of the organism and ambient fluid, respectively; Vcerc is the total volume of a cercariae while sinking passively estimated from the measured dimensions assuming a circular cross-section throughout its body, g = 9.81 m s−2 is the acceleration due to gravity, Usink = 0.13 mm s−1 is the average sinking speed (over > 1000 trajectory segments). The CL and CN are the drag coefficients of the different parts of the cercaria which are modeled as slender rods (see Section 2.5). This gives an estimate for the excess density as ρcerc − ρf ≈ 70 kg m−3. Next we estimate the location of the COG of the organism. The COB of a submerged object in a fluid of homogeneous density is just the centroid. The location of the COB is 25 a, Snapshot of a cercaria swimming freely with reduced-order representations of the head, Figure S3 tail and fork overlaid. These representations allows position and shape information to be extracted from the high-speed videos. The head (blue) and fork (red) have positions (denoted by blue and red dots) and orientations given by the centroids and major axes of the respective ellipses. The shape of the tail is extracted by fitting a spline curve along its central axis (solid orange line). The bending of the tail φt and the tail-head and tail-fork angles φth and φtf are also shown. b, Representation of the kinematics of the head and fork along with the hydrodynamic and mechanical torques at the joints. The head and fork are considered to be slender-rods with equivalent aspect ratio. ph is the orientation vector along the major axis of the head. thus known given the dimensions of the organism. Let Lg be the distance between this COB and the unknown location of the COG. The reorientation of cercaria happens due to a balance between the gravitational torque ρf VcercgLgsin(θ), where θ is the angle made by the longitudinal axis of the organism with the vertical (see Fig. S2b) and the viscous torque. The net viscous torque is computed for the shape in Fig. S2c using the slender-body drag coefficients for the head, tail and fork segments (also see Section 2.5), with the simplification that the swimmer is rigid and passive. The resulting dynamics of the orientation in terms of θ as a function of time were compared with experimental measurements for different values of Lg and the value that minimized the error between the two was found to be Lg = 8.3±1.1µm (Fig . S2d). 26 φtfφtfφtxhxfTail-head jointTail-fork jointτtf, hydτtf, jointτth, hydτth, jointωfωhvhphab2.3 Estimating bending stiffness from high-speed kinematics data In this section we provide details of the procedure to estimate the bending stiffness of the tail- head and tail-fork joints in cercariae . High-speed videos of individual cercariae swimming were analysed using an in-house image processing code (Methods). This analysis separated the cercariae into three parts namely the head, tail and fork. The head and fork were extracted by fitting an equivalent ellipse to the respective part of the organism (Fig. S3a). A spline fit was used to extract the shape of the tail. The two ellipses and the spline gave us all the necessary morphological information about the organism at each time instant. From this data we can extract the relevant angles at the tail-fork and tail-head joints as shown in Fig. S3a. The velocities of the centroid of the head and fork (assumed to be the centroids of the equivalent ellipses) are extracted using a central-difference of their position data. Addition- ally we calculated the normal component of the velocity vh.n of the head normal to its major axis as shown in Fig. S3b. We assume that the head and fork are slender bodies of aspect ratio h/f = lh/f , where lh/f and rh/f are the length and semi-minor axis of the head and rh/f fork, respectively, giving h ≈ 9 and f ≈ 19. The large values of the aspect ratio justifies the use of slender body theory. With this assumption we can estimate the net hydrodynamic torque at the tail-head joint as τ th,hyd = −CNh(l2 h/2) ph ∧ (vh.n)n − Crhl3 hωh, where ph is the orientation unit vector along the major axis of the head (Fig. S3 b). Here CN is the resistance coefficient for motion of a slender body of length l normal to its length and is given by CN = 4πµ/ln(), and Cr is the resistance coefficient for solid body rotation about an axis normal to the slender body axis and is given by Cr = πµ/(3ln()), where µ is the fluid viscosity. Similarly the hydrodynamic torque at the tail-fork joint is give by τ f,hyd = −Crf l3 f ωf . The quasi-steadiness of Stokes flow implies that the instantaneous net torque on the head and fork must vanish [19]. Thus the junction of the head and fork with the tail must supply an equivalent and opposite mechanical bending moment to counterbalance the hydrodynamic 27 torques on the head and fork calculated above. This implies τ th/tf,hyd = −τ th/tf,joint. Thus we can estimate this bending moment at the joint. Independently, we have measured the bending angles φth and φtf at the two joints. A phase plot of the calculated torques as a function of the bending angles is given in the main paper in Fig. 4c. The linear response of the torque to the bending angle points to the elastic nature of the joint and justifies modelling them as linear torsional springs. 2.4 Swimmer Kinematics for Horizontally Swimming Cercariae Here we present data on cercariae swimming in the horizontal plane near the interface. This occurs when cercariae continue swimming even after reaching the interface. In this case, gravity is normal to the swimming direction and pointing out of the plane as shown in Fig. S4a (since the imaging is upwards). The kinematics of the organism in terms of the head trajectory and angles at the head and fork joints were found to be similar to the case of vertical swimming (Fig. 2). Hiwever, the average swimming speed was found to be 3 body lengths/s which is larger than for the vertical swimming case since gravity is acting in a direction perpendicular to the mean swimming direction. Following the method described in Section 2.3, we can estimate the torsional stiffnesses at the two joints in dimensionless terms to be Γtf = 4.2 and Γth = 8.3 which are similar to that obtained for the vertical swimming values. Also the average bending angle amplitude of the tail is found to be A ≈ 150◦. Inputting these parameters into the four-link-swimmer model (as described in Section 2.5), we obtain the theoretical predictions for the swimmer trajectory and the joint angle kinematics which are plotted in Fig. S4 for 12 swimming cycles. As in the vertical swimming case there is very good agreement between the theoretical and experimental trajectories and also the average swimming speed (Fig. S4b). There is good agreement also in the phase plot of the joint angles shown in Fig. S4c. The deviation between theory and experiments for the tail-fork angle is possibly due to the fact that the organism is swimming in a curved trajectory (Fig. S4a) which is seemingly due to the 28 a, The trajectory of the head of the cercariae swimming in a horizontal plane. The s Figure S4 and n directions are defined as along the tangent and normal to the local average swimming direction, respectively. b, Displacement of swimmer along s and n. The symbols (red circles and green triangles) represent tangential and normal displacements, respectively, from experimental measurements. The black solid and dashed lines represent tangential and normal displacements, respectively, from the theory. c, Phase plot of the joint angles φth and φtf as a function of the tail bending angle φt for experiments (symbols) and theory (solid lines). 29 024681012Time-101234DisplacementTangential displacement (model)Normal displacement (model)Tangential displacement (biological experiments)Normal displacement (biological experiments)-200-1000100200Bending angle of tail (degrees)-100-50050100150Joint angle (degrees)Tail and fork angle φtf (model)Tail and head angle φth (model)Tail and fork angle φtf (experiments)Tail and head angle φth (experiments)90 degreesbcagt=0t=0.6sŝĮSchematic of the three-link swimmer ('T-swimmer') showing the coordinate system used Figure S5 in the formulating the model. The positions of the links (blue) and joints (gray) are shown. The joint between the longitudinal links is active (red dot), while the 'T-joint' is a passive linear torsional spring (blue spiral). The direction of gravity is along the negative y axis. observed asymmetry in the tail-fork angle about the equilibrium value of 90◦. 2.5 Formulation of Theoretical Model for Swimming Cercariae Here we formulate the three-link and four-link swimmer models described in the paper. The four-link model, being more representative of actual cercariae, is used to compare theoretical predictions with experiments. However, here we only formulate the model for the simpler three-link swimmer, the extension of which to a four-link-swimmer is straightforward. The model parameters for the three-link swimmer consists of the link lengths l1, l2 and l3 and transverse dimensions r1, r2 and r3; the amplitude of actuation of the active joint A and the torsional stiffness of the passive joint Γ. The swimmer is immersed in an ambient fluid of viscosity µ and density ρf . Following from the low-Reynolds-number hydrodynamics that governs the swimming, the problem can be made dimensionless using the following scales: the length scale lc is taken as half the swimmer length (the swimmer length is defined as the distance from the tip of the head to the base of the fork when the tail is relaxed and straight); the time scale tc = 1/f which is the inverse of the beat frequency; the typical force and torque scales are µf l2 c and µf l3 c , respectively. Following normalization with the 30 Φ1Φ2ߠyxx1x2=xbx3xJ1, xJ2y0x0gaforementioned scales, the system is governed by the dimensionless link lengths l1, l2 and l3; aspect ratios 1, 2 and 3; and the actuation amplitude A. The only non-geometric parameter is the dimensionless torsional stiffness of the 'T-joint' which is given by Γ = Γ/(µf l3 c ). Following the approach of [26] (see also supplementary information), we write down the equations of motion governing the dynamical system consisting of the three-links. We specify a body-fixed reference frame whose origin is assumed to be the centre of link 2 (Fig. S5), with the x axis aligned with the longitudinal axis of the link, which is oriented at an angle θ with the lab-fixed frame of reference (Fig. S5). The position and orientation of the body- fixed coordinate system is measured relative to a lab-fixed coordinate system whose origin is taken as X0 = 0 (Fig. S5). The positions of the links are given by the positions of their centers denoted by xi. The links are assumed rigid and the joints are assumed to have only a rotational degree of freedom. The angles between the links, at the two joints are denoted by φ1 and φ2. For the case of the three-link swimmer φ1 = φt and φ2 = φtf , as defined in the main paper. Thus the position, orientation and shape of the swimmer is entirely defined by the state-vector {Xb, Φ}, where Xb = {xb, θ} is the body position vector and Φ = {φ1, φ2} shape vector. Since the swimmer is restricted to motions on a two-dimensional plane, the positions vectors are understood to mean x = {x, y, 0}. Therefore, the body-velocity of the swimmer is defined as Vb = {vb, ωb}, where vb = xb, ωb = θ and the shape-velocity is given by Φ = { φ1, φ2}, where the 'dot' refers to a time derivative. Note that the velocity vectors, for instance vb (denoted by bold symbols) correspond to elements in R3, unless specified otherwise. The angular velocity vectors have the form ω = ωez, which is of the form of a scalar times a constant unit vector since the axis of rotation is always along ez. Due to the rigid nature of the links and the continuity of translational velocities at the joints, we can write the translational and angular velocities of each link, denoted respectively by xi and ωi, in terms of the translational velocity Vb of the body-fixed frame and shape 31 velocity Φ of the joint angles: xi = vb + ωb ez ∧ (xi − xb) + ωi = ωb ez + Lij ez φj 2(cid:88) j=1 2(cid:88) j=1 Lij φj ez ∧ (xi − xJ j ) (5) (6) where ez is the unit vector in the z direction, xJ j are the positions of the joints and Lij is a matrix which depends on the topology of the swimmer and is 1 if the velocity of ith link depends on rotation about the jth joint and is zero otherwise. For the case of a three-link  . ∧ 1 0 0 0 0 1 swimmer, with two joints and link 2 as the body-frame this is given by L = denotes the vector cross-product. Equations (5) and (6) can be written in matrix form as: X = AVb + B Φ, (7) where X = [ X1, X3] and Xi = { xi, ωi}. A and B, are, respectively, the matrices connecting the body-velocity and shape-velocity of the swimmer to the velocity of the ith X2, link. Once the kinematics relations between the links have been specified, we now specify the forces and torques on the links due to the fluid. Let Fi = {fi, mi} and F = [F1, F2, F3]. From the linearity of the Stokes equations which govern the fluid flow around the swimmer at low-Reynolds-numbers, there exists a linear relationship between the velocities of the links and the resulting hydrodynamic forces [19] which gives: F = −R X, (8) where R is the grand-resistance-matrix and in this work it is assumed to be block-diagonal, 32 consisting of the resistance matrices Ri of each of the links. Such a block-diagonal structure arises since hydrodynamic interactions are neglected at leading order. Thus R is given by:  R = R1 0 0 R2 0 0 0 0 R3  (9) For slender bodies this is true to O(1/ln()), where = l/r is the aspect ratio (length to transverse dimension) of the slender body. Given the high amplitude beating of the tail in cercariae, this assumption is not strictly valid but provides a valuable leading order estimate. Extending the analysis to the next order in 1/ln() using slender-body-theory [22] to capture hydrodynamic interactions between the links is straightforward and will be considered in a future work. Local slender-body-theory gives the net force on a slender rod of length l and radius r at a location s along its axis as f (s) = −CN vn(s) − CLvs(s), where vn(s) and vn(s), denote the velocity in the direction normal and tangential to the axis of the slender body. CN and CL are the drag coefficients in the normal and tangential directions and for the limit of very slender bodies ( = l/r (cid:29) 1), they are given by CL = 2π/ln() and CN = 4π/ln(). The torque on a slender rod due to its angular velocity ω is given by τ = −CN ωl3/12. We can thus write the resistance matrix for a linear, rigid link as:  sin αi 2 2 + CN CL cos αi − sin αi cos αi( CN CL − 1) − sin αi cos αi( CN CL − 1) CN CL cos αi 2 + sin αi 2 Ri = CLli 0 0  0 0 CN CL l2 i 12 (10) Here αi are the orientation angles of each link relative to the body-frame orientation. For the 'T-swimmer', with link-2 as the body frame, α1 = θ − φ1, α2 = θ and α3 = θ + φ2 (see Fig. S5). The net force on the swimmer and net moments at the location of the body-frame are 33 given by fb =(cid:80)3 i=1 fi and mb =(cid:80)3 form this is given by: Fb =  = AT F  fb mb i=1(mi ez + (xi− xb)∧ fi), where mb = mb ez. In matrix (11) where the T denotes the matrix transpose. A consequence of Stokes flow is that the net hy- drodynamic force and torque on a self-propelled swimmer are instantaneously zero. However, in our case, since we are trying model cercariae which are negatively buoyant, bottom-heavy swimmers, there exists an additional force and torque due to gravity. Taking the direction of gravity to be along the negative y axis (see Fig. S5), the net buoyant force and torque on the swimmer are F g = [fg, mg], where fg = {0, (ρcerc − ρf )Vcercg, 0} is the resultant force due to gravitational and buoyant effects and mg = mg ez = ρf Vcercgey ∧ (xCOG − xCOB) sin θ is the torque on the swimmer due to the centre-of-gravity and centre-of-buoyancy of the swimmer being separated (see Section 2.2). Using this expression for F g, the net force and torque balance for the swimmer gives: Fb + F g = 0 − ATR(AVb + B Φ) + F g = 0 Simplifying and inverting the relation to solve for Vb, we have: Vb = G−1 bb (−Gbs Φ + F g) (12) (13) (14) where Gbb = ATRA and Gbs = ATRB. This expression thus gives the velocity of the body- frame attached to the swimmer at link-2 based on velocities of the joint angles. For the purposes of modelling a swimmer with a passive joint of fixed torsional stiffness, it is necessary to express the RHS of Equation (14) in terms of the internal torques at the joints. To do this, we consider a force and torque balance on a section of the swimmer which is divided at the jth joint. Thus the force and torque at the joint denoted by f J j and mJ j are then 34 balanced by the hydrodynamic forces and torques on the remaining portion of the swimmer. As per our convention earlier, the links that affect a given joint j is given by Jj = {i Lij = 1}. The forces and torques at the joints can be written in terms of the bending moments j ) ∧ fi). Note that the effect of gravity does not enter into bending moments at the joints assuming each j = − (cid:80) j = − (cid:80) and forces on the links as, f J (mi + (xi − xJ fi and mJ i∈Jj i∈Jj link is homogeneous in density, and hence has a coincident COG and COB (an assumption we make for cercariae as well (see Section 2.2). Let τ = [τ1, τ2] be the vector containing the internal torques at the joints. It is straight- forward to show that τ = −BT F , where B is the same matrix appearing in Equation (14). Substituting for F from Equation (8), we get: τ = BTR(AVb + B Φ) Substituting for Vb from Equation (14), we get: τ = (Gss − GT bsG−1 bb Gbs) Φ, (15) (16) where Gss = BTRB. To arrive at Equation (16) we have used GT bsG−1 bb Fg = 0 since grav- itational buoyant forces do not contribute to bending moments at the joints as per our assumptions. Inverting this relation we obtain Φ = Hτ , where H = (Gss − GT bb Gbs)−1. We can now write the body velocity of the swimmer in terms of the torques at the joints as: bsG−1 Vb = G−1 bb (−GbsHτ + F g) (17) For our three-link swimmer, we specify the active joint to be undergoing a sinusoidal motion at a frequency f and of the form φ1 = A sin (2πf t). The passive joint, on the other hand, experiences a restoring torque due to the torsional spring when the angle deviates from the equilibrium angle, that is given by τ2 = Γ(φ0 − φ2) where for the 'T-swimmer' the 35 equilibrium angle φ0 = π/2. We can calculate the torque required at the active link to produce the sinusoidal motion above by solving Equation (16) for τ1. Doing so we get:  (2πA cos (2πf t)−H12Γ(π/2−φ2)) H11 Γ(π/2 − φ2)  τ = (18) Equation (17) is now solved with τ as specified by Equation (18) by integrating forward in time using an adaptive Runge-Kutta solver. An important parameter is the swimming efficiency which we define as: ηdist = Xcycle Pdiss,cycle , (19) where Xcycle is the displacement of the swimmer body-frame (link-2 in Fig. S5) averaged motions of the swimmer. This is given by Pdiss,cycle = −(cid:82) tcycle over a swimming cycle and Pdiss,cycle is the net energy dissipated per cycle due to all the V TRV dt, F · V dt =(cid:82) tcycle 0 0 where tcycle = 1/f is the time to complete one swimming cycle. 36
1909.01429	1	1909	2019-09-03T20:13:35	Statistical physics and mesoscopic modeling to interpret tethered particle motion experiments	[ "physics.bio-ph", "cond-mat.soft" ]	Tethered particle motion experiments are versatile single-molecule techniques enabling one to address in vitro the molecular properties of DNA and its interactions with various partners involved in genetic regulations. These techniques provide raw data such as the tracked particle amplitude of movement, from which relevant information about DNA conformations or states must be recovered. Solving this inverse problem appeals to specific theoretical tools that have been designed in the two last decades, together with the data pre-processing procedures that ought to be implemented to avoid biases inherent to these experimental techniques. These statistical tools and models are reviewed in this paper.	physics.bio-ph	physics	Statistical physics and mesoscopic modeling to interpret tethered particle motion experiments Manoel Manghi,1, ∗ Nicolas Destainville,1, † and Annael Brunet2, ‡ 1Laboratoire de Physique Th´eorique (IRSAMC), Universit´e de Toulouse, CNRS, UPS, France 2Department of Molecular Medicine, Institute of Basic Medical Sciences, Faculty of Medicine, University of Oslo, 0317 Oslo, Norway (Dated: September 5, 2019) Tethered particle motion experiments are versatile single-molecule techniques enabling one to address in vitro the molecular properties of DNA and its interactions with various partners involved in genetic regulations. These techniques provide raw data such as the tracked particle amplitude of movement, from which relevant information about DNA conformations or states must be recovered. Solving this inverse problem appeals to specific theoretical tools that have been designed in the two last decades, together with the data pre-processing procedures that ought to be implemented to avoid biases inherent to these experimental techniques. These statistical tools and models are reviewed in this paper. I. INTRODUCTION The main advantage of single-molecule techniques over traditional bulk experiments is the possibility to disentangle sample heterogeneity and to gain insight into subpopulation properties. The tethered particle motion (TPM[104]) single-molecule technique has been developed in the early 1990's [1, 2] to detect and quantify conformational changes of biopolymers induced by their interaction with other molecular partners or changes in their environment [3 -- 5]. It consists in tracking the Brownian motion of a nano-particle (tens to hundreds of nanometers in diameter) attached to a glass surface by a biopolymer such as a DNA molecule and measuring the particle amplitude of movement and its changes when experimental conditions are modified. TPM experiments do not require expensive experimental set-ups, in particular because the particle is tracked by an optical microscope. This explains why numbers of experimental groups adopt this technique to investigate the effects of agents (e.g., enzymes, drugs, ions, or more generally pH, ionic strength or temperature) acting on the characteristics of the tethering polymer, such as its persistence length, its conformation, or its denaturation properties. A variant of TPM is Tethered fluorophore motion (TFM) [6, 7]. It uses the same principles as TPM but employs a fluorophore (and sometimes two [8]) in place of the particle. TFM can thus be combined with fluorescence techniques such as Forster resonance energy transfer. However, TFM is limited in observation time because of fluorophore photobleaching [7]. Optical and magnetic tweezers [9, 10] are another class of powerful tools to investigate the elastic properties of DNA molecules. Optical tweezers [11, 12] rely on a focused laser beam to provide an attractive force on the order of the pico-Newton (pN) to manipulate micrometric particles. Magnetic tweezers [13, 14] consist of two permanent magnets producing a horizontal magnetic field at the location of a magnetic particle. In both cases, as in TPM, the particle is attached to a biopolymer tether, itself grafted to the glass surface. Magnetic torque tweezers (MTT) are an extension of conventional magnetic tweezers where a cylindrical magnet creates a vertical magnetic field and permits to apply both forces and torques. At zero turn the particle is at its rotational equilibrium position and the tethered DNA is torsionally relaxed. After applying turns the DNA molecule is twisted, which gives access to the torsional elastic properties of DNA and also its non-linear response when twist is converted into writhe through the creation of superhelical DNA regions [15]. As compared to optical or magnetic tweezers, no external force is applied to the particle in TPM [4], if not the weak repulsion exerted by the glass surface on the polymer and the bead, in the tens of fN range [16]. Studying the biopolymer in quasi-force-free conditions enables one to tackle its equilibrium properties, as well as reaction rates between two (or more) states such a assembly/disassembly rates of DNA constructs or binding/unbinding rates of enzymes on a tensionless DNA [17]. These different techniques are thus complementary. Recent improvements of TPM rely on multiplexing, hundreds of DNA-bead complexes being positioned in a con- trolled manner by soft nano-lithography and monitored in parallel. The ensuing technique is called high-throughput ∗Electronic address: [email protected] †Electronic address: [email protected] ‡Electronic address: [email protected] 9 1 0 2 p e S 3 ] h p - o i b . s c i s y h p [ 1 v 9 2 4 1 0 . 9 0 9 1 : v i X r a 2 TPM (htTPM) [18]. By drastically reducing acquisition time as compared to anterior setups where the molecules where observed one by one, multiplexing gives access to highly refined statistics allowing one to distinguish between closely related conformations [19 -- 23]. Dealing with these refined statistics justifies the development of improved statistical tools and modeling to interpret them, as detailed in this Review. From a biological perspective, single-molecule techniques have enabled many research groups to decipher key mech- anisms at play in cells, starting with the pioneering paper by Schafer and coworkers in 1991, where the progressive shortening of the tether gave access to the processivity of immobilized RNA polymerases [1]. The estimated value, even though rather rough at this time, was of a dozen of bps/s (see also Ref. [2]), in satisfying agreement with mea- surements in solution. Later, using a similar strategy, the RuvAB-directed branch migration of individual Holliday junctions [24] was measured, also on the order of 10 to 20 bps/s, as well as its dependence on the construct sequence. Several other examples of biological applications can be found in previous review articles [5, 25] or will be discussed in the present Review. The interpretation of these experiments not only depends on accurate measurements but also on adequate and reliable physical models able to account for sometimes very weak and subtle effects. It allows one to identify the relative roles of the intrinsic parameters of the system. The bead size, the tether length, the surface state of the substrate and the solvent are all likely to play pivotal roles in this context, as well as the acquisition rate of the camera used to track the particle. Their different contributions must be precisely quantified, as discussed below. This Review article can be seen as a companion article to the one by Jean-Fran¸cois Allemand, Catherine Tardin and Laurence Salom´e in the same issue [23]. It gives additional details about physical, mathematical and algorithmic issues related to TPM and the ways to tackle them. Even though they are not the main focus of this Review, optical and magnetic tweezers or AFM often come as complementary tools to study single-DNA molecules under force and/or torque. They also need theoretical and algorithmic tools for the interpretation and modeling of experiments. The last decade has witnessed their rapid development, and the reader can for example refer to Refs. [26 -- 33] for further detail. When the connection with TPM experiments is meaningful, we shall discuss some works dealing with force and torque experiments in the present Review. II. MODELING SINGLE DNA-MOLECULE EXPERIMENTS AND THEIR DYNAMICS Several coarse-grained models have been developed in the past decades to model a single DNA molecule. These models are either numerical and/or analytical, the simplest one being the Gaussian chain, for which the end-to-end distance is Ree ≡ (cid:112)(cid:104)R2(cid:105) (cid:39) (bL)1/2, a valid expression[105] as soon as the DNA contour length L is much larger than the Kuhn length b = 2(cid:96)p. The DNA persistence length is (cid:96)p = κ/(kBT ) where κ is the bending modulus and is approximately 50 nm for double-stranded DNA in physiological conditions [19]. The associated numerical model is the bead-spring model which is easy to implement. It consists in modeling the DNA by N beads, whose diameter is equal to the Kuhn length, connected by springs, in Brownian dynamics or Monte Carlo simulations. Although this model is central to understand polymer properties at large scales [34], it is not adapted to single-DNA molecule experiments, which are interested in DNA lengths from few hundreds to few thousand base-pairs, on the order of the Kuhn length. Moreover the bead has a radius much larger than the dsDNA radius (cid:39) 1 nm. It should be noted that at large scales, DNA is in good solvent conditions, i.e. the DNA is swollen compared to the Gaussian chain. The end-to-end distance is now Ree (cid:39) (v/b)1/5L3/5 (for L (cid:29) b) where the excluded volume v depends on the salt concentration. The most adapted model is the worm-like (or semi-flexible) chain model because it reveals the mechanical and statistical properties that are probed in single-molecule experiments, without describing the DNA structure in detail. This model considers the chain as a homogeneous stiff rod, the bending energy of which leads to a short-ranged tangent-tangent correlation, (cid:104)t(s) · t(0)(cid:105) = e−s/(cid:96)p where s is the curvilinear index along the chain and t(s) the normalised tangent vector. The end-to-end distance is then given by the Kratky-Porod result [35] (cid:18) L (cid:96)p (cid:19) ee (cid:39) 2(cid:96)2 R2 p − 1 + e−L/(cid:96)p (1) ee (cid:39) L2, when L (cid:28) (cid:96)p and the Gaussian chain, R2 the chain connectivity and an additional bending energy Eb = κ(cid:80)N−1 ee (cid:39) 2L(cid:96)p, when which yields the two good limits of the rigid rod, R2 L (cid:29) (cid:96)p. The associated discretised numerical model is a bead-spring model with a large spring stiffness to enforce i=1 (1 − cos θi) where κ = (cid:96)pkBT is the bending modulus and θi the angle between two consecutive links. In this discrete worm-like chain model, the successive beads are free to rotate with respect to each other (which is equivalent to set the torsional modulus C to 0). Note that in (cid:80)N−1 the torque experiments, torsion must also be taken into account: the twist angle φi is defined between two consecutive base-pairs (or by defining a material frame for each bead) and the associated elastic energy of a torsional spring writes, i=1 (φi − φ0)2, where φ0 = 0.62 rad is the equilibrium DNA twist in the limit of large C (valid for DNA), Et = C 2 in physiological conditions). 3 Depending on the experimental setup, possible boundary conditions and/or interactions between the DNA and external objects might be considered. In TPM, one DNA end is tethered to a substrate, which is usually modeled as a freely rotative joint. The other DNA end is attached to a spherical particle. The DNA-particle link is also treated as freely rotative joint, except in torsion experiments. The glass coverslip is treated as a hard wall boundary condition limiting the motion of the DNA and the particle to the upper half plane. Although easy to implement numerically, the hard wall condition and the large particle size modify the equilibrium statistics of the DNA in TPM experiments. The amplitude of movement of the particle is defined as σ = particle position parallel to the coverslip. Note that without loss of generality, we have set here (cid:104)r(cid:107)(s)(cid:105) = 0. The amplitude of movement σ can be estimated analytically only in the limit of flexible DNA (L (cid:29) (cid:96)p) [16, 17, 36], and some interpolation formulas have been proposed in the semi-flexible regime [37]. (cid:113)(cid:104)r2(cid:107)(cid:105) where r(cid:107) is the two-dimensional Concerning the dynamical properties of the DNA, the relaxation time (see next Section) is also modified by the experimental setup. In particular, the no-slip boundary condition enforced by the presence of the coverslip slows down the DNA -- particle dynamics. Hence in the numerical simulations, the hydrodynamics interactions induced by the wall are encoded using Fax´en's law prescribing how the diffusion coefficient of both the DNA molecule and the particle is reduced close to the wall, in order to satisfy the no-slip condition for the solvent velocity field at the wall [17]. FIG. 1: A TPM numerical model: the DNA molecule is modeled as a polymer chain made of N connected beads (various colours), anchored to the coverglass (in blue) at one extremity and to the tracked particle (in red) at the other end. The 2D position r(cid:107) of the particle center is represented by the polar coordinates ρ and ϕ in the (xOy) plane. III. DEALING WITH STATISTICAL AND SYSTEMATIC ERRORS We first introduce the different experimental times of interest, here and in the sequel. The camera acquisition period is denoted by Tac. It will play an important role when dealing with the blurring effect below. It typically ranges from few ms for fast acquisition devices [38] to few tens of ms at video rate. The camera exposure time is Tex ≤ Tac, whilst in general Tex = Tac. It must be shorter than the characteristic physical times of interest in the experiment in order to have access to all relevant events. As for the trajectory duration, it must be long as possible in order to reduce statistical uncertainties. Since the tethered particle is not subject to photobleaching, the trajectory can be recorded for several minutes in TPM, which is a great advantage in terms signal-to-noise ratio as compared to single-fluorophore techniques. One of the objectives here is to measure the useful correlation (or relaxation) time τ of the 2D tethered-particle position r(cid:107). It sets the typical time needed for the particle to explore its configuration space. It is defined through the auto-correlation function (2) where the average (cid:104). . .(cid:105) is taken along the trajectory. For sake of simplicity, we assume that the time correlation function has the form C(t) = C(0)e−t/τm . This expression is exact only in the case of a quadratic confining potential. In the case of a more complicated confining potential induced by the polymer tether, it is only an approximation C(t) = (cid:104)r(cid:107)(s + t) · r(cid:107)(s)(cid:105) − (cid:104)r(cid:107)(s)(cid:105)2 ROρϕxyz4 because the auto-correlation function is a sum of decreasing exponentials and τm is then associated with the slowest diffusion mode, which dominates at long times [35]. In practice, we shall see below that the measured correlation time τm is slightly larger than the real one τ because of blurring effets that we aim at quantifying in the present section. A. Cleaning data from spurious points A first data preprocessing step is essential to minimize the contributions of unwanted biases and experimental variability. First of all, it is essential to deal with the measurement heterogeneity induced by undesirable artefacts due, e.g., to ill-assembled objects. For all single molecule approaches involving anchored DNA molecules, the substrate surface state is crucial [4]. This becomes even more critical when studying dynamically acting proteins or molecules binding on the DNA, and especially for AFM [39] or tweezer [40] manipulations. Such approaches require that the DNA molecules binds to substrate surface without modifying physiological functions and properties, and that only desired anchoring is realized, without any sort of spurious secondary attachments [17]. When advanced experimental protocols are not sufficient to prevent those artefacts, statistical tools offer an alter- native way to do so during the data preprocessing step. In TPM experiments, malformed DNA-particle complexes, e.g. with two grafted DNA molecules instead of an expected unique one would interfere with main population of interest and hamper the authentic experimental noise. Applying a selecting filter based on the asymmetry factor of the 2D trajectories allows one to select only well-defined tethered DNA/particle complexes as mentioned in the article by Allemand et al. in this issue [23]. An asymmetry factor (or aspect ratio) larger than 1.35 is assumed to be associated with two DNA tethers cross-linked to the same tracked particle [19]. However, this criterion may appear to be insufficient to deal with spurious non-specific binding of some particles to the coverslip. One must get rid of trajectories laying in the far tails of the amplitude of movement distribution. Hence, if a single population is expected, for example when extracting a bending angle or the persistence length (see Section IV C below), trajectories with amplitude of movement outside the interval (mean ±2.5 standard deviations) were discarded in Refs [19, 20, 22]. Extracting the particle displacement-time properties provide an alternative and complementary objective crite- rion for classifying the trajectories into the main population and outliers. The correlation time for each 2D TPM- trajectories, τ , is determined as described above. The resulting data for the global DNA population is expected to behave as a single population for a given DNA state. Based on that, and as above for amplitude of movements, a trajectory is declared as an outlier if its τ lays in the far tails of this distribution and is then sorted out. In practice, sorting out points deviating by more than 1 or 2 standard deviations from the average value is also a reasonable criterion [20]. Based on the same principle for filtering experimental data, Schickinger et al. [8] exploit the average dwell-time. It is determined in each specific state, unbound and bound DNA, for each particle. Comparing bound vs. unbound dwell times reveals multiple data point clusters and provided a criterion of selection. Considerable discrepancy with the main populations leads to the discrimination of outliers. More generally, all single-molecule approaches come with their own specific protocol-induced defects, which need to be taken into account in order to clean data from outliers before confronting them to statistical analysis and theoretical modeling. This is even more true when using high-throughput approaches that provide high sampling levels and allow to define precisely the main conformation populations. B. Subtracting instrumental drift Once data have been cleaned from outliers, the first systematic error to correct comes from the instrumental drift, due, inter alia, to thermal expansion of the observation setup [17]. Along a given trajectory, the DNA approximate anchoring point at each time t0 is determined by averaging the particle position over an interval of duration Tav (typically 1 or 2 s, sometimes even more [41]) centered at t0, and then subtracted from r(cid:107). If Tav is chosen to be much larger than the measured relaxation time τm, this anchoring point is determined with a good accuracy. This sets (cid:104)r(cid:107)(cid:105) (cid:39) 0, as desired. Note that subtraction of drift induces non-trivial systematic anti-correlations at short times t (cid:28) Tav. Indeed, let us assume that the measured time correlation function in absence of drift also has the form Cm(t) = Cm(0)e−t/τm . Then subtraction of drift modifies it to (cid:18) (cid:19) Cm(t) Cm(0) = 1 + 2 τm Tav e−t/τm − 2 τm Tav (3) at first order in 2τm/Tav (cid:28) 1 [17]. This will have to be taken into account when measuring τm below. (cid:34) (cid:18) τ (cid:19)2(cid:16) (cid:17)(cid:35)−1/2 C. Correcting blurring effect 5 Another important source of systematic error in TPM comes from the finiteness of the camera exposure time Tex. An image in fact represents the optical signal averaged over a time interval of duration Tex. In Refs. [17, 37, 42, 43], the ensuing time-averaging (or blurring) effect in single-particle tracking experiments was investigated. It occurs whenever the trajectory of the tracked particle is confined in a bounded domain, not only in TPM but also when tracking plasma membrane constituants, for instance. Indeed, in the extreme case where the frame exposure time would be much larger than the system auto-correlation time, itself inversely proportional to the domain area (see below), the measured particle position during this period would remain very close to the anchoring point, giving the erroneous impression that the amplitude of motion is much smaller than its actual value. However, we shall see that when it is not too strong, this effect can efficiently be corrected. By using Eq. (3), the measured correlation function Cm(t) is first fitted to obtain the measured correlation time τm, the only free parameter in this expression (the value of Tav has been chosen from the beginning). It can be proven that the real correlation time τ is given by (4) which remains a correct approximation while τm ≥ 2 Tex/3 [17, 42]. From this value, the diffusion domain size can now be corrected. It is characterized by the trajectory standard deviation σ ≡(cid:113)(cid:104)r2(cid:107)(cid:105), also called "amplitude of movement", τ (cid:39) τm − Tex 3 measured on a sufficiently long interval in order to accurately sample configurations [38] (see Section V A). If σm is its measured value, then the real one is recovered in its turn from σ (cid:39) σm − 2 τ Tex 1 − e− Tex τ (5) For example, if Tex = 2τ then σm (cid:39) 0.75σ [42]. In the case where Tex (cid:28) τ , we naturally get σ (cid:39) σm. Note that the particle diffusion coefficient D depends on both σ and τ , through D = Const. σ2/τ , where Const. depends on the domain geometry [44]. The measured value Dm can be substantially different from the actual one D, e.g., Dm (cid:39) 0.34D if Tex = 2τ . It is necessary to correct both σ and τ thanks to the above formulae to get the correct value of D. Tex 2 . Incorrectly dealing with this blurring phenomenon can have dramatic effects when varying the experimental tem- Indeed, the water viscosity ηw(T ) decreases rapidly with increasing T . This leads to a 4-fold fall of perature T . τ ∝ D−1 ∝ ηw(T )/T when T grows from 15 to 70◦C [22]. Even though larger than Tex at low T , τ likely becomes comparable to or smaller than τ at high T , then requiring correction of the blurring phenomenon. This issue has previously led to erroneous conclusions about denaturation profiles of DNA as observed by TPM [22, 45] (see also Section IV C 4 below). D. Estimating error bars The last step is to determine average values and associated error bars. The most straightforward way is to estimate the single average of an observable O over the distribution of the set of N data point as its mean µO ≡ (cid:104)O(cid:105). Statistical fluctuations are estimated through the the variance σ2O ≡ (cid:104)O2(cid:105) − (cid:104)O(cid:105)2, and the standard deviation σO. Assuming √ statistical independence of samples, the error bar on µO is then σO/ N (68% confidence interval) or twice this value (95% confidence). More advanced methods can be used as the jackknife, used in Refs. [46 -- 48], or the bootstrap, used in Refs. [19 -- 22, 49], to estimate the error bar. Both are resampling methods. In the jackknife, one considers N resampled sets of data, each containing all but one of the original data points. The bootstrap uses M sets of data, each containing N data points obtained by random sampling, performed by Monte Carlo, of the original set of N points. During the Monte Carlo sampling with replacement, the probability that a data point is selected is 1/N . The number of bootstrap samplings, M , should be chosen to be large enough so that the average bootstrap sampling is reproducible with sufficient accuracy. The jackknife approach leads to identical results each time it is run on the same set of data, which is not true for bootstrap. IV. SOLVING THE INVERSE PROBLEM "Solving the inverse problem" can be generically stated as calculating, from a set of experimental measurements, the causes that produced them. The aim is to gain insight into the physical properties of a system by indirect measurements or conversely to set up a predictive model that can reproduce observations. In the present context, coupling theory and experiments offers appropriate tools to probe the intrinsic physical properties of the DNA macromolecule: for instance, DNA persistence length or its local defects, from the apparent end-to-end distance of the polymer as accessible from TPM. 6 A. Analytical approaches and their limits The exact equilibrium distribution, p(r(cid:107)), of the amplitude of movement of the TPM particle can only be computed in the rigid and flexible limits [16, 17]. Using the mirror reflection argument, the probability distribution in the Gaussian (flexible) regime and in the limit where the particle radius is large, R (cid:29)(cid:112)L(cid:96)p, is given by [17]: (cid:115) r(cid:107)(cid:113) r2(cid:107) + R2 3 πL(cid:96)p pG(r(cid:107)) ≈ (cid:20) (cid:16) (cid:113) (cid:17)(cid:21) exp − 3 4L(cid:96)p r2(cid:107) + 2R2 − 2R r2(cid:107) + R2 . (6) The intermediate semi-flexible regime, L (cid:39) (cid:96)p, of interest in TPM experiments, is well described by the worm-like chain model. For instance, in the experiments described in Ref. [17], DNA lengths vary between 400 and 2080 bp, which corresponds to 2 < L/(cid:96)p < 14. This model can be tackled analytically [50, 51] but the boundary conditions must be handled with care and the final step should be solved numerically, which does not provide any analytical formula for the probability distribution with fitting parameters. Moreover, it should be kept in mind that real chains are self-avoiding and that the presence of the labelling particle renders the problem even more intricate. For instance, the effect of the excluded volume of the particle is to widen and shift to large r(cid:107) the Gaussian distributions. It then becomes analytically intractable for finite chains, but it can be tackled numerically by Monte Carlo or Brownian dynamics simulations (see Section IV B). The over-damped dynamics of the DNA in the flexible regime is controlled by the Rouse time τ(cid:107) = N R2 G/D0 where RG is the radius of gyration and D0 = kBT /(6πη(cid:96)p) is the diffusion coefficient of a monomer sphere of radius (cid:96)p in a liquid of viscosity η [17, 52]. However, when the TPM particle is grafted at one end, the relaxation time and the diffusion coefficient measured by tracking the particle dynamics are features of the dynamics of the whole DNA -- particle complex. Using the Langevin equation, it is shown that the diffusion coefficient Dc is given by: Dc = DpartDDNA Dpart + DDNA (7) where Dpart = kBT /(6πηR). Thus the particle does not slow down the complex provided that Dpart (cid:29) DDNA. Knowing Dpart, the value of DDNA is inferred from the measurement of Dc. Note however that these analytical considerations do not allow us to compare with TPM experiments due to (i) the fact that the chain is in the semiflexible regime; and (ii) the neglect of hydrodynamics interactions in this approach. B. Numerical simulations Several types of numerical simulations have been developed to circumvent the above limitations. To test both the dynamics and the equilibrium properties in the TPM setup geometry, the adequate numerical methods are Brownian dynamics simulations and kinetic Monte Carlo simulations. In Brownian dynamics simulations, the evolution of each sphere position ri(t) is governed by an iterative Langevin equation (discrete time step δt and discrete time variable n = t/δt) ri(n + 1) = ri(n) − D0 ∇ri U (n) + 2 D0 ξi(n), (8) where the rescaled random displacement has variance unity (cid:104)ξi(n) · ξj(m)(cid:105) = 3 δijδnm. The rescaled bare diffusion coefficient D0 = D0δt/a2 is the diffusion constant in an unbounded space in units of the particle radius a and time step δt. For sufficient numerical accuracy the usual choice is D0 = 10−3 -- 10−5. The dimensionless potential U = U/(kBT ) is the sum of stretching and bending potentials described in Section II. The excluded volume interaction is modeled i<j[(b/ri − rj)12 − 2(b/ri − rj)6 + 1] valid for separation ri− rj < 2 and b = R + 1 for j = N and b = 2 otherwise. Since the polymer motion is limited to the upper half-plane z > 0, we use the reflection boundary condition: if a sphere intersects the substrate, its height zi is replaced by its mirror image zrefl by a repulsive, truncated Lennard-Jones potential ULJ =(cid:80) N = 2(R − a) − zN for the particle). i = 2a − zi (zrefl (cid:113) 7 FIG. 2: Experimental (τ exp(cid:107) for different DNA lengths, L, and particle radius, R, in linear-log coordinates. Inset: Ratio τ exp(cid:107) as above, in log-log coordinates. The dashed line shows the best linear regression, with slope −0.355. Taken from Ref. [17]. , solid symbols) and numerical (τ sim(cid:107) , using z-dependent diffusion coefficients, open symbols) relaxation times /τ sim(cid:107) versus L/R with the same symbols In Monte Carlo simulations, at each step (MCStep) δt, a bead is chosen uniformly at random among the N + 1 possible ones (monomer spheres and labelling particle). Then a random move δr is attempted for this bead, uniformly in a ball of center 0 and radius Rb. One shows that in this case (cid:104)δr2(cid:105) = 3R2 b /5. This quantity must be equal to 10D0δt. 6D0δt, where D0 is the diffusion coefficient of the spherical bead, depending on its diameter, which sets Rb = Interactions between adjacent beads are treated via the potential U , whereas interactions between non-adjacent beads are of hard core nature, like surface-bead interactions: whenever a move would lead to the penetration of a bead into an other one or the surface, it is rejected. The physical time is incremented of δt following each Monte Carlo Sweep (MCS, equal to a sequence of N + 1 MCSteps). A simulation snapshot is shown in Fig. 1 at equilibrium. Note that the choice of excluded volume interaction in Monte Carlo simulations saves computational time as compared to the calculation of truncated Lennard-Jones potentials used in Brownian Dynamics. √ If one is only interested in equilibrium properties of the conformation of the DNA -- particle complex, i.e. The advantage of these simulations is to give access to dynamical properties. From this viewpoint, Brownian Dynamics and kinetic Monter Carlo simulations are equivalent in the small δt limit. As an example, experimental and numerical values of the relaxation times τ (extracted from C(t) defined in Eq. (2)) are shown in Figure 2 [17]. Experimental and numerical values are found in good agreement, with ratios of experimental to numerical values varying from 0.5 to 2 (shown in the inset). the probability distribution p(r(cid:107)) or its standard deviation σ for various apparent DNA contour lengths L, a faster numerical method is to compute DNA -- particle conformations by exact Monte Carlo sampling. This method has been developed by Segall and collaborators [16, 37] and used in subsequent works [7, 19]. It consists in generating labeled DNA as a random walk of N steps with a bending energy Eb (defined in Section II) by step. The angles θi between successive links are randomly chosen through a probability distribution in agreement with the Boltzmann weight at equilibrium ∝ exp[−Eb/(kBT )]. The starting point is the bead tethered to the substrate, and at each step, self-intersecting trajectories (resp. trajectories intersecting the substrate) are discarded to take into account intra-chain excluded volume interactions (resp. repulsive interactions with the substrate). Then statistical averages are computed. This numerical method made possible comparisons with approximate analytical expressions, and the finding of very accurate interpolation functions for σ(L) [37]. Moreover, it has been used in Ref. [19] to obtain a graphical reference for σ in function of the dsDNA persistence length (cid:96)p and therefore to study quantitatively the variation of (cid:96)p with the salt concentration in the solution (see Section IV C 3). C. Examples Weak magnitude changes in DNA conformation can be difficult to detect by TPM. Coupling multiplexed single- molecule experiments, statistical physics and mesoscopic modeling allows one to detect and investigate narrow changes in the amplitude of movement σ(t). Local modifications along the DNA molecule, such as kink, protein binding, loop formation, or more global effects due to a change in the surrounding environment, such as viscosity, pH, ionic strength expsimL/Ror temperature, will impact the physical properties of the DNA molecule. Changes in the distribution of DNA molecule conformations induce a transition in the apparent contour length, a variable directly accessible through single-molecules techniques. We now review representative examples. 8 1. Intrinsic curvature angle in TPM Local bending of the DNA double helix axis can be induced by either the binding of proteins [4, 40, 53] (Figure 1, right) or by specific sub-sequences. Specific sub-sequences, such as short A-tracts composed of a succession of adenines on the same strand, can locally change the bio-molecule mechanical properties, which is measured through small local bends. Different single-molecule techniques can give access to quantitative measurement of local bending, including AFM, fluorescence spectroscopy, tweezers and TPM [40, 54]. Joint theory and simulation establish the adapted formalism to explore the effect of a local bend in TPM experi- ments. Using the worm-like chain model to describe a local bending deformation, modeled as a kink of angle θ (cid:54)= 0 located at distance (cid:96) from one end, the end-to-end distance Ree is given by a modified version of Eq. (1) [19], the so-called kinked worm-like chain model: R2 ee = 2(cid:96)2 p − 2 + e−(cid:96)/(cid:96)p + e−(L−(cid:96))/(cid:96)p + cos(θ) 1 − e−(cid:96)/(cid:96)p − e−(L−(cid:96))/(cid:96)p + e−L/(cid:96)p (9) (cid:20)(cid:18) L (cid:96)p (cid:19) (cid:16) (cid:17)(cid:21) From a numerical perspective, simulated TPM is adapted to model locally bent DNA by incorporating the preferred angle θ between three successive beads into the bead chain. Only the inserted sub-sequence is expected to induce an intrinsic curvature, the remainder of the DNA sequence is assumed to be a random one, without any intrinsic curvature. For 575-bp-long DNA molecule, an angle of θ = π would induce a decrease of the apparent DNA contour length of ≈ 30% (absolute reduction of ≈ 35nm) for a bend located in the middle of the DNA compared to DNA without any curvature. In contrast, the variation is of ≈ 15% (absolute reduction of ≈ 15nm) for a the same bending site now placed at 1/3 from the DNA free end. HtTPM experiments were performed on a set of designed DNA-molecules containing increasing number of A-tracts, from 0 to 7 [20]. Then, the apparent end-to-end distance of the entire DNA molecule was compared to the predictions of the kinked worm-like chain model to extract the bending angle. The discrepancy between experiments and the analytical expression pointed out the fact that the statistical model did not properly incorporate some biological heterogeneity, notably the non-zero intrinsic curvature spread over the whole molecule. Taking it into account leads to a better extrapolation of data. This highlights a fact often neglected in theoretical approaches: the zero-temperature limit of real DNA is not a straight, regular helix, but rather a slowly meandering one related to the sequence. The analytical models can also consider separately the cases of a bend with a fixed angle and of a local flexible hinge with no spontaneous curvature. Those two distinct causes similarly affect the apparent end-to-end distance measured by TPM and are not easy to disentangle. To account for these two mechanical modifications simultaneously, more precise theoretical developments are needed. 2. DNA looping in TPM DNA looping is a common phenomenon, useful to gene regulation in both prokaryotes and eukaryotes. Probing protein binding and the induced loop formation ([25] and references therein) helps to resolve more precisely the geometry of the DNA-protein complexes involved in many biological processes. TPM measurements can be used to unravel the structure of the loop. Moreover, by knowing the length of the loop and the positions of protein-binding sites, the change in apparent end-to-end distance measured by TPM can be used to infer geometrical or conformational alteration of the looped protein-DNA structure. Using statistical mechanics models based upon elastic interactions in small DNA molecules, Biton et al. [55] and Johnson et al. [41] probed the interactions of DNA molecules with Lac repressor proteins. They performed TPM measurements to extract the distribution and changes in the apparent length of tethered DNA in function of the operators. Due to the symmetry of the two identical dimeric arms of the Lac repressor, different operators can bind to each arm with different possible orientations, yielding distinct loop types, either parallel or anti-parallel. Monte Carlo simulations were performed [55], where the DNA molecule was modeled as a necklace of rigid beads separated by rigid cylinders, which takes into account the fluctuations of the Lac repressor-mediated DNA looped segment. Then, the looping probabilities for a specific DNA loop configuration were calculated from the simulations. This information gave access to the probability of a loop of a particular length and a given topology. Comparing computational results to TPM experimental results enabled the authors to identify the looped state topology. In addition, simulations were also used to estimate the dissociation constants associated with the binding of the Lac repressor to each of the operators. This numerical study [55] suggests different looping probabilities for anti-parallel and parallel loop types formed along a 1632 bp-long DNA molecule, to which a particle of radius 160 nm is attached. The looping probabilities for the anti-parallel loop between binding site centers located at base-pairs 444 and 1044 (resp. 1344) are 4-fold (resp. 6-fold) higher than those for the parallel loop. It also suggests that the anti-parallel loop is entropically favored. 9 3. Effects of salt on elastic properties The amplitude of movement σ not only depends on the physical properties of the DNA-particle system, but also on the physicochemical properties of the surrounding solution. (cid:80) i z2 The stiffness properties of nucleic acids molecules depend on the solution ionic strength, defined as I = 1 2 i ci where zi and ci are respectively the valency and the concentration of ion i, that induces screening of the electrostatic repulsion between the negatively charged phosphate groups along the sugar-phosphate backbones. Single-molecule techniques permit to characterize changes of the polyelectrolyte mechanical properties induced by changes in ionic strength as pioneered by Lambert et al. [56]. Studies based on high-throughput TPM investigated the dependence of the persistence length on salt concentration for monovalent (Li+, Na+, K+) and divalent (Mg2+, Ca2+) metallic valent ions [19, 57]. In the first study [19], after correcting the blurring effect using Eq. (4), the end-to-end distance of the tethered 1201 and 2060 bp-long dsDNA was observed to decrease as a function of the ionic strength I (ranging from 10 mM to 3 M), as expected. A TPM coarse-grained model, taking into account excluded volume interactions, was used to extract (cid:96)p from measurements by resolving the inverse problem (described in Section IV B). It was observed that (cid:96)p varies from 30 to 55 nm over the large range of ionic conditions, comparable to previous experimental results. Two main models were used to fit the data: the Odijk-Skolnick-Fixman model valid at large I, relying on a mean field approach valid at low values of the electrostatic potential [58, 59]; the Odijk-Skolnick-Fixman-Manning model where the above approach was corrected at low I accounting for the Manning condensation of a few counterions that decreases the effective charge along the DNA [60]. These models could not account quantitatively for the whole experimental data set obtained with Na+ or Mg2+. The more recent Manning model [61] with internal electrostatic stretching force due to the repulsion of the charges along the polyelectrolyte well fitted the entire range of I for the Na+ case only. The second experimental study [57] refined the physical understanding of the phenomenon by exploring the role played by ion size and a larger range of I (from 0.5 mM to 6 M). The experimental protocol was also improved because pH was observed to decrease significantly in phosphate buffer when ions were added. A 4-(2- hydroxyethyl)- 1-piperazineethanesulfonic acid (HEPES) buffer was used instead and a slower decrease of the apparent length of the dNA molecule was now observed as comparer to the previous study. Then, the extracted persistence length of the DNA could be quantitatively described by more sophisticated theoretical approaches, the Netz-Orland [19, 62] for divalent ions and Shen-Trizac [63] for monovalent ones. These theories, by including non-linear electrostatic effects and the finite DNA radius, can account for the observed behavior of (cid:96)p over the whole I range. Interestingly, the metallic ion size does not influence the persistence length in contrast to alkyl ammonium monovalent ions at high I [57]. A recent work also probed the salt dependence of the torsional stiffness of DNA by multiplexed MTT [64]. Few different Na+ monovalent concentrations, 20, 100 and 500 mM, and a combination of 100 mM of Na+ and 10 mM of Mg2+ were tested. The extension-rotation and torque-rotation curves were collected. The effective torsional stiffness, Ceff , was determined by fitting the linear torque-rotation regime for each of the ionic strength conditions over various stretching forces. At high stretching forces (f > 6 pN), when stretching forces suppress bending along the DNA, the intrinsic torsional stiffness is independent of salt concentration. However, at small stretching forces, Ceff increases when the ionic strength increases. Coupling MTT measurement and simulation of the twistable worm-like chain model, permits to examine in more detail the torsional persistence length of the DNA molecule [65]. The discrepancy between experimental and numerical measurements underlay that bending and twisting are intrinsically coupled in DNA molecule because of the difference between the major and minor groves. The bending elasticity is not isotropic anymore. This effect can be implemented in the alternative twistable worm-like chain elastic model proposed by Marko and Siggia [15], where twist-bend coupling is fully taken into account. The systematic deviations of the twist response of dsDNA investigated by magnetic tweezers experiments with the numerical model reported in previous studies could be explained by taking into account this direct coupling between twist and bend deformations. 4. Effects of temperature -- DNA denaturation 10 Due to base-pairing and stacking energies on the order of the thermal energy kBT , DNA flexibility is strongly dependent not only on the ionic strength, but also on the temperature T . It affects the cohesive interactions between the DNA bases as well as the contribution of the chain configurational entropy in the free energy [66]. From a biological perspective, various species live in extreme environments and are subjected either to high temperatures or to large temperature fluctuations. This emphasizes the importance of knowing how DNA structure, properties and protein-DNA interactions are affected by temperature. To this purpose temperature-controlled TPM studies were performed during the last decade [22, 45]. This technique allowed one to explore the temperature-dependence of the apparent DNA persistence length (cid:96)p. In the measured temperature ranges, from 23 to 52◦C in Ref. [45] and from 15 to 75◦C in Ref. [22], a correlation between the increase of T and the decrease of the apparent end-to-end distance of the DNA molecule was revealed. Driessen et al. [45] used a numerical procedure, by solving the Langevin equation, Eq. (8), to model the Brownian motion of the tethered particle. The simulated results in function of the DNA persistence length were fitted with a quadratic function in order to extract the relation between (cid:96)p and the amplitude of movement σ. This empirical equation was used to extrapolate the persistence length from the values of σ measured in TPM experiments. As expected, (cid:96)p is slightly dependent on the AT/GC base-pair composition of the DNA. More surprisingly, this study revealed that the intrinsic flexibility of dsDNA strongly and linearly depends on temperature in a range well below the DNA melting temperature, an effect much stronger than expected. Brunet et al. [22] investigated further the same question by coupling htTPM experiments and Monte Carlo simula- tions (Section IV B). The extracted values of (cid:96)p showed a slower decrease of the amplitude of movement as compared to the previous study. Considering the changes in buffer viscosity with T , the authors put forward that the detec- tor time-averaging blurring effect (Section III C) needed to be cautiously corrected. The observed decrease of the apparent end-to-end distance of the tethered DNA well below the DNA melting temperature is mainly due to this effect. After carefully correcting the TPM measurements from Ref. [45], the variation of (cid:96)p = κ(T )/(kBT ) with T is sharper, in much better agreement with the expected dependency of the bending modulus κ with the temperature at physiological salt conditions. Up to 60◦C, the extracted values of (cid:96)p display a temperature dependency that can be associated with an intact dsDNA molecule, without a significant fraction of denaturated base-pairs. Additional work focussed on the temperature dependence of the response of DNA to torsion [67]. This study combined single-molecule magnetic tweezers measurements with all-atom molecular dynamics and coarse-grained simulations. DNA extension-rotation curves were measured over a temperature range of 24 to 42◦C. Increasing temperature systematically shifted the extension-rotation curves to a negative number of turns. In other words, the point where the DNA molecule is torsionally relaxed changes linearly with temperature. Measurements show that the temperature-dependent helical change is not force-dependent for stretching forces < 1 pN (the overall extension- rotation response of DNA is symmetric around zero turn at low force, see below). Averaging over small forces gives the DNA helical twist constant ∆Tw(T ) = (−11.0 ± 1.2)◦/(◦C.kbp), in agreement with anterior studies. Using the oxDNA coarse-grained model, describing DNA as two inter-twined strings of rigid nucleotides, a 600 bp-long DNA molecule is simulated including Debye-Huckel screened electrostatic interactions, at I = 150 mM to match experimental conditions. Simulated temperatures ranged from 27 to 67◦C. The mean twist angle for zero torque decreased as when the temperature was raised. Linear fit over temperature yielded the slope ∆Tw(T ) = (−6.4 ± 0.2)◦/(◦C.kbp), smaller than the experimental value. Additionally, this question was addressed in all-atom simulations of a 33 bp mixed DNA sequence with explicit water molecules and ions at five different temperatures ranging from 7 to 47◦C. The twist angle linearly decreases in the range, DNA-twist changes are equal to ∆Tw(T ) = (−11.1 ± 0.3)◦/(◦C.kbp), in close agreement with experimental values. The discrepancy in coarse-grained simulations suggests an important role of structural local changes along the DNA molecule, only taken into account in the all-atom simulation. Coupling experimental results with theoretical predictions, this work suggested that the temperature- dependent change in twist is predominantly due to partial and local loss of hydrogen bonds over the DNA backbone that is not correctly considered in coarse-grained models. DNA structure modifications can be attained not only through temperature changes as discussed above, but also by applying torque and/or force with tweezers. At low longitudinally applied forces (f < 1 pN), both strong over- and under-twisting lead to the formation of plectonemic supercoils. At higher forces, plectonemic supercoils are formed under positive torque, but the two DNA single strands separate locally (denaturation bubbles) when a sufficient negative twist is applied [26]. The consequence of a force applied longitudinally without any external torque is more subtle. Analytical models are essential tools to determine the nature of the DNA overstretched state observed for f above ≈ 60 − 70 pN. Since its initial discovery in 1996 [11, 68] a debate has arisen as to whether this overstretched state is a new S-DNA form or more simply a denaturated state, i.e. a large denaturation bubble if the two ends are closed or peeled ssDNA if one end is open [69]. New experiments have then been done to probe (i) the impact of the experimental conditions of attachement on the coverslip and the bead [69], (ii) the effect of the NaCl salt 11 (b) (a) FIG. 3: (a) Force-extension curve for a ssDNA. Data (symbols) are taken from Hugel et al. [74]. The black solid curve corresponds to a fit using the discrete worm like chain interpolation with the non-linear bond elasticity, Eq. (A2) (the red one corresponds to discrete version of the Marko Siggia interpolation, Eq. (A1)). The parameters values are: L = 3.40 µm, κ = 1.5, a = 0.20 nm. (b) Force-extension curve for a poly(dG-dC) dsDNA. Data (blue symbols) are taken from Rief et al. [75]. Solid curves correspond to the discrete worm-like chain interpolation for B-DNA (red), S-DNA (green) and with non-linear extensibility for ssDNA (pink). The black curve corresponds to Eq. (A3), where linear stretching is included as shown by the blue curve for pure B-DNA. The red symbols correspond to the semi-analytical calculation using transfer matrix. Parameters values are: LB = 0.14 µm, κB = 147, γ = 1.89, κS = κBS = 3.8, EB = 1200 pN. Inset: Fraction of base-pairs ϕS in the S state vs. force, and Ising correlation function 1 − (cid:104)σiσi+1(cid:105) (dashed curve). Taken from Ref. [27]. concentration [70, 71], and (iii) the influence the base content [72, 73], on the overstretched states. It has been clearly shown (see for instance Refs. [27, 69]) that the knowledge of the different formulas that fit these three possible states are central in the interpretation of data. Although fitting the transitions itself needs complicated theories such as the one presented in A leading to the fit shown in Figure 3, simpler formulas such as Eq. (A1) with fewer fitting parameters (the DNA length L, its persistence length (cid:96)p) or Eq. (A2) for ssDNA stretching allows one to undoubtedly recognize the overstretched state. This issue is an example where analytical approaches cannot be replaced by numerical simulations since a good numerical model would necessitate both the structural details of the double helix and a dsDNA length L between 0.2 and 4 µm (i.e. 600 to 10000 bp). An attempt has been done using the oxDNA code [76] for a 100 bp dsDNA but the S-DNA state has not been observed. V. DYNAMICALLY DETECTING TWO (OR MORE) DISTINCT STATES Detecting dynamical configurational changes of DNA molecules is challenging in many situations of interest in genetic regulation. As already stressed in Section IV C 2, protein-induced DNA looping is a paradigmatic mechanism that has drawn much attention during the last 25 years [25], because single-molecule techniques have enabled various research groups to shed light not only on looping thermodynamics of different molecular systems but also on their kinetics. As TPM minimizes mechanical constraints on DNA and proteins, it can give access to kinetics at the molecular scale, with high time-resolution. In 1995, lactose repressor-mediated loop formation and disassembly were kinetically monitored for the first time [3]. In this case, the total DNA molecule was L = 1150 bps long, the polystyrene particle radius was R = 115 nm, the two DNA sites that are bound when the loop is formed were ∆s (cid:39) 300 bps away, and the repressor concentration was 1 nM. The looping and disassembly lifetimes were found to be very long and both on the order of 100 s. These values were later refined [77]. Systematic exploration of the effect of the distance ∆s on looping time was performed in Ref. [8]. In 2006, the IS911 transpososome assembly was analyzed by following a similar strategy [53]. During the last decade, the bridging activity of site-specific recombinases could also be studied by TPM or TFM by employing a construct where the synapse assembly also reduces the apparent length of the DNA molecule by forming a loop [5 -- 7, 21, 78]. Further addition of sodium dodecyl sulfate (SDS) allows one to assess whether strand exchanged occurred or not within the synapse. Other two-state system kinetics have also been thoroughly studied by TPM. When protein binding provokes a DNA 0200400600800100012000.00.51.01.52.02.53.03.5fpNzΜm12 bend, it is detected as an effective shortening [5, 53, 54, 78]. Nucleosome assembly in eukaryotes also leads to a shorter apparent length [79]. Furthermore, TPM can be used to probe the kinetics of (single) secondary bonds, which can, e.g., transiently form between the particle and the substrate [80]. TPM is well adapted to follow dynamics of two-state systems when the dwell-times in both states are on the same order of magnitude, in other words when their free energies are comparable. In this section, we explain how recent theoretical developments likely improve the indirect measurement accuracy of the transition rates between the two (or more) states with TPM. A. Thresholding and correlation time We illustrate the concept of thresholding [17, 81, 82] (Figure 4) in the case of DNA looping between two specific operators and mediated by DNA-binding proteins [3, 8, 41, 53, 77]. The average dwell-times in the unlooped and looped states are respectively denoted by τLF and τLB. By definition, the transition rates between these two states are τ−1 LF and τ−1 LB . These quantities depend on the binding energy of the DNA-protein complex, on the protein concentration, and on the DNA elastic properties [77]. We assume that the slowly diffusing bead does not significantly alter looping kinetics if it is sufficiently small, as discussed in Ref. [17] and in Section IV A. The most basic idea is to start from the fact that the amplitude of movement σ of the tethered particle is smaller in the looped state. Hence the plot of σ (or the variance σ2) in function of time will display an alternance of time intervals where σ is small and large, as displayed in Figure 4. On must define a threshold, denoted here by σc, below which the molecule is considered to be looped, and above which it is unlooped. The value of σc can be set by examining the bimodal distribution of amplitudes of movement [53, 77]. One can for example set it at half-amplitude between the two maxima of the bimodal distribution. Since σ2 ≡ (cid:104)r2(cid:107)(cid:105), the average (cid:104). . .(cid:105) must be estimated on a (sliding) time-window, the duration of which, also called the averaging time, is denoted here by Tav. From a signal-processing perspective, this averaging scheme corresponds to window-filtering, i.e. convolution of r2(cid:107)(t) with a window function. One might choose to switch to more elaborate exponential or Gaussian filters [77, 83, 84], without gaining a significant advantage, however. FIG. 4: Thresholding: The plots of the variance σ2 (in nm2) versus time for a simulated TPM experiment of a DNA molecule of length L = 798 bp, a particle of radius R = 20 nm and three averaging times Tav = 3 (yellow), 30 (red) and 300 ms (black). The vertical lines indicate the transition events (looping or unlooping) that are forced in the simulation [17]. The horizontal solid lines show the average values of σ2 in the looped (bottom) and unlooped (top) states. The dashed horizontal line shows the threshold value σc separating these two states for detection purposes. Detecting close-lying transitions, as in the right part of the figure, is the most critical issue of thresholding together with detection of false transition events (see text). Taken from Ref. [17]. One of the main difficulties comes from the fact that the probability distributions of σ for the looped and unlooped states can overlap substantially when using a too short averaging time Tav [85], as illustrated in Figure 4. However, only events occurring at a time-scale larger than Tav can be detected. Efficient thresholding thus relies on a compromise between a large value of Tav needed to estimate at best the amplitude of movement (and avoid at best false detections), and a short value needed to get the best time resolution (and minimize missed transition events) [17, 83]. In particular, 01234567t (s)0100002000030000!r2 (nm2)the measured values of the rate constants can depend significantly on the window size [85] because the rate is the inverse of the average dwell-time in a state, and measured dwell-times are bounded below by the window size. Consequently, the Tav must be chosen consistently with the dwell-times, typically Tav < τLF, τLB, even though some improvements can substantially correct for missed events [83]. A lower bound on Tav comes from the correlation time τ introduced in Section III C, as discussed in Ref. [17]. Let us assume for simplicity that the two states have comparable correlation times τ . Their amplitudes of movement are σ1 < σ2. We introduce the parameter λ = 1 − (σ1/σ2)2. It was demonstrated [17] that the minimal averaging time Tav needed to resolve them with good accuracy (i.e. with few false detections of transitions) is typically equal to τ /λ2. All in all, optimal Tav must satisfy 13 τ λ2 < Tav < τLF, τLB. B. Hidden Markov chains (10) Hidden Markov modeling (HMM) combined with a maximum-likelihood approach can be used to determine the numerical values of model parameters such as the transition rates. In contrast to thresholding, no windowing is required, nor the prior selection of a threshold. HMM, initially developed by mathematicians [86], has nowadays plenty of applications in various fields of science. It has been adapted to TPM in 2007 by Beausang et al. [85, 87]. The underlying idea is that the system under consideration can be modeled by a Markov chain [88], the states of which are not directly observed in the experiment, i.e., they are "hidden". Here we again illustrate these ideas in the case of DNA looping, even though it can be generalized, e.g., to protein- binding. The hidden state, denoted by q(t), can be "looped" or "unlooped" DNA. We again look for the average dwell-times τLF and τLB. The most basic idea [85] would be to consider that this two-state system is governed by a two-state Markov chain with a 2 × 2 transition matrix [88]. However, this naive approach fails because it ignores the fact that transition rates between both states depend on the polymer conformation: for example, looping is forbidden when the polymer is too stretched. As a consequence, the Markov chain must also take into account the chain configuration, for example through the 2D position of the particle, r(cid:107), itself governed by an over-damped Langevin equation in a harmonic potential. The system state as it appears in the Markov chain is now (q(t), r(cid:107)(t)). Once an experimental time series (q(t), r(cid:107)(t))t has been recorder, the idea is then to calculate the likelihood that it is observed for a given pair of dwell-times (τLF, τLB). This can be done with the standard tools of probability theory. Then the dwell-time values maximizing this likelihood are considered to be the most probable ones. This procedure has been tested on numerically generated trajectories for which the dwell times were exactly known. It was able to correctly recover these values, up to statistical error bars. However, in spite of its conceptual simplicity, the practical implementation of the method relies on some strong approximations about the looping process, as stated by the authors themselves [85]. For example, looping is allowed if and only if the particle excursion (cid:107)r(cid:107)(t)(cid:107) is smaller than a threshold ρmax. This is modeled by a crude Θ step-function in the Markov chain. In addition, a simplification is "to ignore the unobserved height variable z [above the glass substrate plane], in effect treating the bead motion as diffusion in two dimensions." This is again an issue when deciding whether looping is possible or not for a given value of (cid:107)r(cid:107)(t)(cid:107) because z can be large even though (cid:107)r(cid:107)(t)(cid:107) is small. "A better analysis might treat z as another hidden (unobserved) variable." Finally, as in the thresholding approach, the polymer is considered to be in quasi-equilibrium in both the looped and unlooped state, from which transition probabilities are inferred. This assumption is only valid if the free-energy wells around each state are sufficiently deep. A refinement of the HMM method relies on variational Bayesian inference, first used in the study of Lac repressor- mediated looping [41] that we have already mentioned in Section IV C 2. Bayesian inference is able to determine not only the most likely model parameters but also the most likely number of hidden states, that was fixed a priori in the above HMM method, by observing the number of peaks in the probability distribution of σ. The improved ability of this method to resolve close-lying transitions (see Figure 4) has also been tested in this work. For instance, two states separated by an amplitude of movement σ of 40 nm could be resolved by their technique at a mean lifetime of about 0.5 s. In contrast, lifetimes of more than 4 s were necessary for states to be resolvable by simple thresholding. The reason for this difference is that Bayesian inference does not require time-filtering as evoked above. When the free energies of the states are significantly different, one state is favoured with respect to the other. For example, let us assume that once the loop is formed, it is very stable because breaking it would require a free energy ∆F (cid:29) kT . Then loop opening events become rare and cannot be observed in practice. It becomes interesting to apply a force f (with the help of tweezers) in the pN range on the tethered particle because the associated potential energy difference between both states can compensate ∆F and make them roughly equiprobable, thus allowing one 14 to observe more frequent opening events. Indeed, the dwell-times are known to depend exponentially on the applied force [89]. Beyond looping, this approach has been successfully used in combination with HMM analysis to study single- nucleosome unwrapping [90]. Since DNA is wrapped in about two turns around a histone octamer, three distinct conformations can be observed: fully wrapped, about one turn unwrapped and fully unwrapped. Applying a 2.5 pN force allows one to observe transitions between the two first states, and at a stretching force of 6 pN, the second turn unwraps. Detailed information about nucleosome unwrapping such as bending angles of nucleosomal DNA or dwell-times could be extracted from these experiments. Zero-force dwell-times can then be extrapolated from these measurements. This method can in principle be used to detect any transient DNA-protein complex formation and dwell-times. Similar approaches have been used to study DNA hairpins opening/closure [91] and DNA G-quadruplex unfolding/refolding [92]. VI. CONCLUSION This Review has illustrated in many situations the powerful capabilities of TPM experiments coupled to theoretical and/or numerical modeling to give access to quantities of interest in both biological and biophysical contexts. We have identified two levels of difficulties that must be overcome in order to infer the physical parameters of interest with a good accuracy. First, raw data must be carefully processed thanks to well-established protocols in order to deal with both statistical and systematic sources of errors. The existence of outliers seems inherent to single molecule experiments because it is both difficult to prepare samples with 100 % of identical molecules and to graft all molecules in ideal conditions limiting unwanted non-specific interactions. This is in part solved thanks to nano-lithography techniques, but not entirely. After having discarded outliers, the most critical systematic effect to deal with is the blurring effect inherent to the finiteness of the detector exposure. We have proposed an efficient way to solve this issue through simple inversion formulae, as summarized in Eqs. (4) and (5). Once the data have been corrected from these sources of error, solving the inverse problem enables one to infer the DNA state or the physical parameters of interest. This concerns not only quantities measured at thermodynamic equilibrium (e.g., elastic parameters C and (cid:96)p or intrinsic bending angles), but also out-of-equilibrium properties of great biological interest such as transition rates (e.g. binding/unbinding or looping/unlooping rates). In the latter case, we have shown that hidden Markov chain approaches are promising even though they are more complex to implement than simple thresholding. In all cases, the underlying quantitative model must rely on solid physical grounds, appealing to polymer and elasticity theory, and out-of-equilibrium statistical mechanics. We have listed several successes of such approaches in this Review. However, to our point of view, few issues remain to be solved in the future in order to provide a fully operational tool to biophysicists and biologists. We have explained that intrinsic curvature modifies the amplitude of movement in a way that can be quantified with good accuracy. However, weak intrinsic curvature is spread all over the molecule, and it is not only localized at specific high-curvature loci. This "quenched" disorder likely modifies the amplitude of movement in a systematic but ill-controlled manner. This phenomenon should be quantified, for example by using more sophisticated mesoscopic models fully taking account such subtle effects [93]. When a region of DNA is modified or affected in any way, for example through binding of a protein, it can bear In not only a different spontaneous curvature that one wants to quantify, but also a different bending modulus. particular, DNA flexibility is a sequence-dependent quantity [94]. Disentangling sequence-dependent spontaneous curvature and sequence-dependent elastic properties in single-molecule experiments is another issue that will require extensive modeling work in the future. As far as dynamical properties are concerned, we have just stressed that methods relying on hidden Markov chains are quite promising. However, to our knowledge, no systematic quantification of their capabilities, in the spirit of Eq. (10), has been performed so far. Probability theory together with numerical modeling should be able to give definitive and robust conclusions on the strengths and limits of these approaches. Filling this gap seems important to us in order to eventually provide an easy-to-use tool to experimentalists. Acknowledgments We warmly thank our colleagues Catherine Tardin and Laurence Salom´e for fruitful discussions and sound advices during the writing of this Review. We are also tributary to the Universit´e Toulouse III-Paul Sabatier and the Centre National de la Recherche Scientifique (CNRS). Appendix A: Analytical modeling of stretching experiments 15 The classical model for semi-flexible polymer stretching has been developed by Marko and Siggia [95] using the continuous worm-like chain model by developing a formula that interpolates between the exact results in the limits of low (f (cid:28) kBT /(cid:96)p (cid:39) 0.08 pN) and strong forces (f (cid:29) kBT /(cid:96)p): (cid:20) z L (cid:21) f = kBT (cid:96)p − 1 4 + 1 4(1 − z/L)2 . (A1) This formula has been successfully used to fit the force-extension curves for f (cid:46) 65 pN for a λ-phage dsDNA (see the review [96] and references therein). One technical difficulty in fitting these curves is to set the origin. This might be correlated to the fact that experimentally the tethers are not stretched exactly in the direction perpendicular to the coverslip [97]. For larger forces, the DNA internal structure starts to come into play (see below). However this type of approach remains valid for very high stretching of ssDNA [75] provided that both the discrete nature of the chain is taken into account [98] and non-linear stretching terms are included [74]. Fits of force-extension curves of ssDNA up to 1200 pN, as shown in Figure 3a, have been nicely fitted using the modified formula that includes both terms [27]: (cid:19) (cid:18) (cid:19)1/2 −(cid:112) (cid:35) 1 + 4κ2 (A2) (cid:34) (cid:18) (cid:18) + f = kBT a z L (1 + Unl(f )) 3 1 − u(κ) 1 + u(κ) − 1√ 1 + 4κ2 + 1 [1 − z(1 + Unl(f ))/L]2 + 4κ2 where κ = κ/(kBT ) is the bending modulus in units of kBT (cid:39) 4×10−21 J (at room temperature), u(κ) = coth(κ)−1/κ, and Unl(f ) = 1.172777 f − 3.731836 f 2 + 4.118249 f 3 (where f is in units of 10 nN) [74]. The fit leads to the bending modulus value κ(ssDNA) = 1.5 kBT and an effective monomer length a = 0.20 nm (see Figure 3a). Unexpectedly this value of a is much smaller than the distance between two consecutive bases in ssDNA ass (cid:39) 0.7 nm. This result suggests that the number of degrees of freedom increases by a factor 3.5 for strong forces [27], a result that has already been observed for peptides [99]. When strong forces are applied to dsDNA, experiments show a sharp, few picoNewtons wide, cooperative over- stretching transition at a given critical force of around 60 -- 80 pN , accompanied by a sudden 70% increase of the contour length [11, 68]. It corresponds to a transition from the B-form to a new form of unstacked DNA remaining in a duplex form (S-DNA form for Stretched). A second transition is observed at stronger forces [75] consistent with a peeling of one strand from the other strand. The critical force and therefore the selection between these two transitions depends on the DNA sequence and the salt concentration [70, 72, 100]. Using a Ising-Heisenberg coupled model [101 -- 103], an analytical formula has been derived [27] which allows us to fit the first transition: (cid:18) (cid:19) (cid:19) (cid:18) 1 z aBN = F EB − 1 1 + 2αB (cid:104)σiσi+1(cid:105) − 1 ϕB + γ 1 − 1 2αS κB − κBS 4 2αB κBS + F/2 + αB ϕS + γ 2αS κS − κBS κBS + γF/2 + αS (cid:19) (A3) where B and S refer to the DNA state, F = aBf /(kBT ), κBS is the bending modulus at the BS domain wall, γ = aS/aB, EB = aBEB/(kBT ) where EB is the stretching modulus, αB = (κBF + F 2/4)1/2 and αS = [κSγF +(γF )2/4]1/2. The fraction of base-pairs in the B (respectively S) state are ϕB (respectively ϕS = 1−ϕB). Finally (cid:104)σiσi+1(cid:105) is the two-point correlation function of the effective Ising model which is non-zero only close to the transition [27]. An example of a fit of the force-extension curve for a poly(dG-dC) DNA is shown in Figure 3b. The inset shows the variation of the fraction of base pairs in the S state and the Ising correlation function as a function of the applied force. Eq. (A3) has also been used to fit the S-DNA to ssDNA transition for poly(dG-dC) where the linear stretching term is replaced by the non-linear stretching one for ssDNA [27]. References [1] D.A. Schafer, J. Gelles, M.P. Sheetz, R. Landick, Transcription by single molecules of RNA polymerase observed by light microscopy, Nature (1991) 352, 444-448. 16 [2] H. Yin, R. Landick, J. Gelles, Tethered particle motion method for studying transcript elongation by a single RNA polymerase molecule. Biophys. J. (1994) 67, 2468-2478. [3] L. Finzi, J. Gelles, Measurement of lactose repressor-mediated loop formation and breakdown in single DNA molecules. Science (1995) 267, 378-380. [4] G. Zocchi, Analytical assays based on detecting conformational changes of single molecules, ChemPhysChem (2006) 7, 555-560. [5] H.F. Fan, C.H. Ma, M. Jayaram, Single-molecule tethered particle motion: stepwise analyses of site-specific DNA recom- bination, Micromachines (2018) 9, 216. [6] J.N.M. Pinkney, P. Zawadzki, J. Mazuryk, L.K. Arciszewska, D.J. Sherratt, A.N. Kapanidis, Capturing reaction paths and intermediates in Cre-loxP recombination using single-molecule fluorescence. Proc. Natl. Acad. Sci. U.S.A. (2012) 109, 20871-20876. [7] P.F.J. May, J.N.M. Pinkney, P. Zawadzki, G.W. Evans, D.J. Sherratt, A.N. Kapanidis, Tethered fluorophore motion: studying large DNA conformational changes by single-fluorophore imaging, Biophys. J. (2014) 107, 1205-1216 ; Biophys. J. (2015) 109, 457. [8] M. Schickinger, M. Zacharias, H. Dietz, Tethered multifluorophore motion reveals equilibrium transition kinetics of single DNA double helices, Proc. Natl. Acad. Sc. U.S.A. (2018) 115, E7512-E7521. [9] C. Bustamante, Z. Bryant, S.B. Smith, Ten years of tension: single-molecule DNA mechanics, Nature (2003) 421, 423-427. [10] K.C. Neuman, A. Nagy, Single-molecule force spectroscopy: optical tweezers, magnetic tweezers and atomic force mi- croscopy, Nature Methods (2008) 5, 491-505. [11] S.B. Smith, Y. Cui, C. Bustamante, Overstretching B-DNA: the elastic response of individual double-stranded and single-stranded DNA molecules, Science (1996) 271, 795-799. [12] M.D. Wang, H. Yin, R. Landick, J. Gelles, S.M. Block, Stretching DNA with optical tweezers, Biophys. J. (1997) 72, 1335-1346. [13] S.B. Smith, L. Finzi, C. Bustamante, Direct mechanical measurement of the elasticity of single DNA molecules by using magnetic beads, Science (1992) 258, 1122-1126. [14] C. Gosse, V. Croquette, Magnetic tweezers: micromanipulation and force measurement at the molecular level, Biophys J.(2002) 82, 3314-3329. [15] J.F. Marko, E.D. Siggia, Bending and twisting elasticity of DNA, Macromolecules (1994) 27, 981-988. [16] D.E. Segall, P.C. Nelson, R. Phillips, Volume-exclusion effects in tethered-particle experiments: bead size matters. Phys. Rev. Lett. (2006) 96, 088306. [17] M. Manghi, C. Tardin, J. Baglio, P. Rousseau, L. Salom´e, N. Destainville, Probing DNA conformational changes with high temporal resolution by Tethered Particle Motion, Phys. Biol. (2010) 7, 046003. [18] T. Pl´enat, C. Tardin, P. Rousseau, L. Salom´e, High-throughput single-molecule analysis of DNA-protein interactions by tethered particle motion, Nucleic Acids Res. (2012) 40, e89. [19] A. Brunet, C. Tardin, L. Salom´e, P. Rousseau, N. Destainville, M. Manghi, Dependence of DNA persistence length on ionic strength of solutions with monovalent and divalent salts: A joint theory-experiment study. Macromolecules (2015) 48, 3641-3652. [20] A. Brunet, S. Chevalier, N. Destainville, M. Manghi, P. Rousseau, M. Salhi, L. Salom´e, C. Tardin, Probing a label-free local bend in DNA by single molecule tethered particle motion, Nucleic Acids Res. (2015) 43, e72. [21] F. Fournes, E. Crozat, C. Pages, C. Tardin, L. Salom´e, F. Cornet, P. Rousseau, FtsK translocation permits discrimination between an endogenous and an imported Xer/dif recombination complex, Proc. Natl. Acad. Sci. U.S.A. (2016) 113, 7882- 7887. [22] A. Brunet, L. Salom´e, P. Rousseau, N. Destainville, M. Manghi, C. Tardin, How does temperature impact the conformation of single DNA molecules below melting temperature?, Nucleic Acids Res. (2018) 46, 2074-2081. [23] J.-F. Allemand, C. Tardin, L. Salom´e, Parallelized DNA tethered bead measurements to scrutinize DNA mechanical structure, Methods (2019), this issue. [24] C. Dennis, A. Fedorov, E. Kas, L. Salom´e, M. Grigoriev, RuvAB-directed branch migration of individual Holliday junctions is impeded by sequence heterology, EMBO J. (2004) 23, 2413-2422. [25] C. Tardin, The mechanics of DNA loops bridged by proteins unveiled by single-molecule experiments, Biochimie (2017) 142, 80-92. [26] I.D. Vilfan, J. Lipfert, D.A. Koster, S.G. Lemay, N.H. Dekker, Magnetic tweezers for single-molecule experiments. In Handbook of Single-Molecule Biophysics (pp. 371-395). Springer, New York, 2009. [27] M. Manghi, N. Destainville, J. Palmeri, Mesoscopic models for DNA stretching under force: new results and comparison with experiments, Eur. Phys. J. E (2012) 35, 110. [28] I. Heller, T.P. Hoekstra, G.A. King, E.J.G. Peterman, G.J.L. Wuite, Optical tweezers analysis of DNA-protein complexes, Chem. Rev. (2014) 114, 3087-3119. [29] K. Truex, Hoi Sung Chung, John M. Louis, and William A. Eaton, Testing landscape theory for biomolecular processes with single molecule fluorescence spectroscopy, Phys. Rev. Lett. (2015) 115, 018101. [30] D. Dulin, T. J. Cui, J. Cnossen, M.W. Docter, J. Lipfert, N.H. Dekker, High spatiotemporal-resolution magnetic tweezers: calibration and applications for DNA dynamics, Biophys. J. (2015) 109, 2113-2125. [31] R. Sarkar, V.V. Rybenkov, A guide to magnetic tweezers and their applications, Front. Phys. (2016) 4,48. [32] Lin CT., Ha T. Probing single helicase dynamics on long nucleic acids through fluorescence-force measurement. In Gen- nerich A. (eds) Optical tweezers. Methods in Molecular Biology vol. 1486. Humana Press, New York, 2017. [33] F. Kriegel, N. Ermann, J. Lipfert, Probing the mechanical properties, conformational changes, and interactions of nucleic 17 acids with magnetic tweezers, J. Struct. Biol. (2017) 197, 26-36. [34] P.G. de Gennes, Scaling concepts in polymer physics, Cornell University Press, Ithaca, New York, 1979. [35] M. Doi, S.F. Edwards, The Theory of Polymer Dynamics, Oxford University Press, Oxford,1986. [36] M. Lindner, G. Nir, S. Medalion, H.R.C. Dietrich, Y. Rabin, Y. Garini, Force-free measurements of the conformations of DNA molecules tethered to a wall, Phys. Rev. E. (2011) 83, 011916. [37] P.C. Nelson, C. Zurla,D. Brogioli,J.F. Beausang, L. Finzi, D.D. Dunlap, Tethered particle motion as a diagnostic of DNA tether length, J. Phys. Chem. B (2006) 110, 17260-17267. [38] S. Kumar, C. Manzo, C. Zurla, S. Ucuncuoglu, L. Finzi, D. Dunlap, Enhanced tethered-particle motion analysis reveals viscous effects, Biophys. J. (2014) 106, 399-409. [39] T. Ando, T. Uchihashi, N. Kodera, High-speed AFM and applications to biomolecular systems, Annual review of bio- physics (2013) 42, 393-414. [40] A. Gietl, D. Grohmann, Modern biophysical approaches probe transcription-factor-induced DNA bending and looping, Biochem. Soc. Trans. (2013) 41, 368-373. [41] S. Johnson, J.W. van de Meent, R. Phillips, C.H. Wiggins, M. Lind´en, Multiple LacI-mediated loops revealed by Bayesian statistics and tethered particle motion, Nucleic Acids Res. (2014) 42, 10265-10277. [42] N. Destainville, L. Salom´e, Quantification and correction of systematic errors due to detector time-averaging in single molecule tracking experiments, Biophys. J. (2006) 90, L17-L19. [43] K.B. Towles, J.F. Beausang, H.G. Garcia, R. Phillips, P.C. Nelson, First-principles calculation of DNA looping in tethered particle experiments, Phys. Biol. (2009) 6, 025001. [44] T. Bickel, A note on confined diffusion, Physica A (2007) 377, 24-32. [45] R.P.C. Driessen, G. Sitters, N. Laurens, G.F. Moolenaar, G.J.L. Wuite, N. Goosen, R.T. Dame, Effect of temperature on the intrinsic flexibility of DNA and its interaction with architectural protein, Biochemistry (2014) 53, 6430-6438. [46] A. Imparato, F. Sbrana, M. Vassalli, Reconstructing the free-energy landscape of a polyprotein by single-molecule exper- iments, EPL (2008) 5, 58006. [47] C.K.P. Loong, H.-X. Zhou, P.B. Chase, Persistence length of human cardiac α-tropomyosin measured by single molecule direct probe microscopy, PloSOne (2012) 6, e39676. [48] R. Tapia-Rojo, C. Marcuello, A. Lostao, C. G´omez-Moreno, J. J. Mazo, F. Falo, A physical picture for mechanical dissociation of biological complexes: from forces to free energies, Phys. Chem. Chem. Phys. (2017) 6, 4567-4575. [49] S. LB. Kenig, M. Hadzic, E. Fiorini, R. Borner, D. Kowerko, W. U. Blanckenhorn, R. K. Sigel, BOBA FRET: Bootstrap- based analysis of single-molecule FRET data, PloS One (2013) 12, e84157. [50] J. Samuel, S. Sinha, Elasticity of semiflexible polymers, Phys. Rev. E (2002) 66, 050801(R). [51] S. Stepanow, and G. Schutz, The distribution function of a semiflexible polymer and random walks with constraints, EPL-Europhys. Lett. (2002) 60, 546-551. [52] H. Qian, A mathematical analysis for Brownian dynamics of a DNA tether, J. Math. Biol. (2000) 41, 331-340. [53] N. Pouget, C. Turlan, N. Destainville, L. Salom´e, M. Chandler, IS911 transpososome assembly as analysed by tethered particle motion, Nucleic Acids Res. (2006) 34, 4313. [54] S.F. Tolic-Norrelykke, M.B. Rasmussen, F.S. Pavone, K. Berg-Sorensen, L.B. Oddershede, Stepwise bending of DNA by a single TATA-box binding protein, Biophys. J. (2006) 90, 3694-3703. [55] Y. Y. Biton, S. Kumar, D. Dunlap, D. Swigon, Lac repressor mediated DNA looping: Monte Carlo simulation of con- strained DNA molecules complemented with current experimental results, PloS One (2014), 9, e92475. [56] M. Newby Lambert, E. Vocker, S. Blumberg, S. Redemann, A. Gajraj, J.C. Meiners, N.G. Walter, Mg2+-induced com- paction of single RNA molecules monitored by tethered particle microscopy, Biophys. J. (2006) 90, 3672-3685. [57] S. Guilbaud, L. Salom´e, N. Destainville, M. Manghi, C. Tardin, Dependence of DNA persistence length on ionic strength and ion type, Phys. Rev. Lett. (2019) 122, 028102. [58] T. Odijk, Polyelectrolytes near the rod limit, J. Polym. Sci. (1977) 15, 477. [59] J. Skolnick, M. Fixman, Electrostatic persistence length of a wormlike polyelectrolyte, Macromolecules (1977) 10, 944. [60] G. S. Manning, A procedure for extracting persistence lengths from light-scattering data on intermediate molecular weight DNA, Biopolymers (1981) 20, 1751. [61] G.S. Manning, The contribution of transient counterion imbalances to DNA bending fluctuations, Biophys. J. (2006) 90, 3208. [62] R.R. Netz, H. Orland, Variational charge renormalization in charged systems, Eur. Phys. J. E (2003) 11, 301-311. [63] E. Trizac, T. Shen, Bending stiff charged polymers: The electrostatic persistence length, EPL-Europhys. Lett. (2016) 116, 18007. [64] F. Kriegel, N. Ermann, R. Forbes, D. Dulin, N.H. Dekker, J. Lipfert, Probing the salt dependence of the torsional stiffness of DNA by multiplexed magnetic torque tweezers, Nucleic Acids Res. (2017) 45, 5920-5929. [65] S.K. Nomidis, F. Kriegel, W. Vanderlinden, J. Lipfert, E. Carlon, Twist-bend coupling and the torsional response of double-stranded DNA, Phys. Rev. Lett. (2017) 118, 217801. [66] M. Manghi, J. Palmeri, N. Destainville, Coupling between denaturation and chain conformations in DNA: stretching, bending, torsion and finite size effects, J. Phys.: Condens. Matter (2009) 21, 034104. [67] F. Kriegel, C. Matek, T. Drsata, K. Kulenkampff, S. Tschirpke, M. Zacharias, F. Lankas, J. Lipfert, The temperature dependence of the helical twist of DNA, Nucleic Acids Res. (2018) 46, 7998-8009. [68] P. Cluzel, A. Lebrun, C. Heller, R. Lavery, J.-L. Viovy, D. Chatenay, F. Caron, DNA: an extensible molecule, Science (1996) 271, 792-794. [69] X. Zhang, H. Chen, S. Le, I. Rouzina, P.S. Doyle, J. Yan, Revealing the competition between peeled ssDNA, melting 18 bubbles, and S-DNA during DNA overstretching by single-molecule calorimetry, Proc. Natl. Acad. Sci. U.S.A. (2013) 110, 3865-3870. [70] X. Zhang, H. Chen, H. Fu, P.S. Doyle, J. Yan, Two distinct overstretched DNA structures revealed by single-molecule thermodynamics measurements, Proc. Natl. Acad. Sci. U.S.A. (2012) 109, 8103-8108. [71] X. Zhang, Y. Qu, H. Chen, I. Rouzina, S. Zhang, P.S. Doyle, J. Yan, Interconversion between Three Overstretched DNA Structures, J. Am. Chem. Soc. (2014) 136, 16073-16080. [72] H. Fu, H. Chen, X. Zhang, Y. Qu, J.F. Marko, J. Yan, Transition dynamics and selection of the distinct S-DNA and strand unpeeling modes of double helix overstretching, Nucleic Acids Res. (2011) 39, 3473-3481. [73] N. Bosaeus, A.H. El-Sagheer, T. Brown, S.B. Smith, B. Akerman, C. Bustamante, B. Nord´en, Tension induces a base- paired overstretched DNA conformation, Proc. Natl. Acad. Sci. U.S.A. (2012) 109, 15179-15184. [74] T. Hugel, M. Rief, M. Seitz, H.E. Gaub, R.R. Netz, Highly Stretched Single Polymers: Atomic-Force-Microscope Exper- iments Versus Ab-Initio Theory, Phys. Rev. Lett. (2005) 94, 048301. [75] M. Rief, H. Clausen-Schaumann, H.E. Gaub, Sequence-dependent mechanics of single DNA molecules, Nat. Struct. Biol. (1999) 6, 346-349. [76] F. Romano, D. Chakraborty, J.P.K. Doye, T.E. Ouldridge, A.A. Louis, Coarse-grained simulations of DNA overstretching, J. Chem. Phys. (2013) 138, 085101. [77] F. Vanzi, C. Broggio, L. Sacconi, F.S. Pavone, Lac repressor hinge flexibility and DNA looping: Single molecule kinetics by tethered particle motion, Nucleic Acids Res. (2006) 34, 3409-3420. [78] C.T. Diagne, M. Salhi, E. Crozat, L. Salom´e, F. Cornet, P. Rousseau, C. Tardin, TPM analyses reveal that FtsK contributes both to the assembly and the activation of the XerCD-dif recombination synapse, Nucleic Acids Res. (2014) 42, 1721-1732. [79] H.F. Fan, Z.N. Liu, S.Y. Chow, Y.H. Lu, H. Li, Histone chaperone-mediated nucleosome assembly process, PLoS One (2015) 10, e0115007. [80] K.E. Merkus, M.W.J. Prins, C. Storm, Single-bond association kinetics determined by tethered particle motion: concept and simulations, Biophys. J. (2016) 111, 1612-1620. [81] S. Dixit, M. Singh-Zocchi, J. Hanne, G. Zocchi, Mechanics of binding of a single integration-host-factor protein to DNA, Phys. Rev. Lett. (2005) 94, 118101. [82] N. Laurens, S.R.W. Bellamy, A.F. Harms, Y.S. Kovacheva, S.E. Halford, G.J.L. Wuite, Dissecting protein-induced DNA looping dynamics in real time, Nucleic Acids Res. (2009) 37, 5454-5464. [83] F. Vanzi, L. Sacconi, F.S. Pavone, Analysis of kinetics in noisy systems: application to single molecule tethered particle motion, Biophys. J. (2007) 97, 21-36. [84] D. Colquhoun, F.J. Sigworth, Fitting and statistical analysis of single-channel records, in: B. Sakmann, E. Neher (Eds.), Single-Channel Recording, Plenum Press, New York, 1983, pp. 191-263. [85] J.F. Beausang, P.C. Nelson, Diffusive hidden Markov chain model characterization of DNA looping dynamics in tethered particle motion, Phys. Biol. (2007) 4, 205-219. [86] L.E. Baum, T. Petrie, Statistical inference for probabilistic functions of finite state Markov chains, The Annals of Math- ematical Statistics (1966) 37 1554-1563. [87] J.F. Beausang, C. Zurla, C. Manzo, D. Dunlap, L. Finzi, P.C. Nelson, DNA looping kinetics analyzed using diffusive hidden Markov model, Biophys. J. (2007) L64-L66. [88] G. Grimmett, D. Stirzaker, Probability and random processes, Clarendon Press, Oxford, 1982. [89] E.A. Evans, K. Ritchie, Dynamic strength of molecular adhesion bond, Biophys. J. (1997) 72, 1541-1555. [90] M. Kruithof, J. van Noort, Hidden Markov analysis of nucleosome unwrapping under force, Biophys. J. (2009) 96, 3708- 3715. [91] M. Manghi, N. Destainville, Physics of base-pairing dynamics in DNA, Physics Reports (2016) 631, 1-41. [92] H. You, S. Guo, S. Le, Q. Tang, M. Yao, X. Zhao, J. Yan, Two-state folding energy determination based on transition points in nonequilibrium single-molecule experiments, J. Phys. Chem. Lett. (2018) 9, 811-816. [93] M. Zuiddam, R. Everaers, H. Schiessel, Physics behind the mechanical nucleosome positioning code, Phys. Rev. E. (2017) 96, 52412. [94] S. Geggier, A. Vologodskii, Sequence dependence of DNA bending rigidity, Proc. Natl. Acad. Sci. U.S.A. (2010) 107, 15421-15426. [95] J.F. Marko, E.D. Siggia, Stretching DNA, Macromolecules (1995) 28, 8759-8770. [96] C. Bustamante, S.B. Smith, J. Liphardt, D. Smith, Single-molecule studies of DNA mechanics, Curr. Opin. Struct. Biol. (2000) 10, 279-285. [97] C. Ray, J.R. Brown, B.B. Akhremitchev, Correction of systematic errors in single-molecule force spectroscopy with polymeric tethers by atomic force microscopy, Phys. Chem. B (2007) 111, 1963-1974. [98] J. Kierfeld, O. Niamploy, V. Sa-yakanit, R. Lipowsky, Stretching of semiflexible polymers with elastic bonds, Eur. Phys. J. E (2004) 14, 17-34. [99] F. Hanke, A. Serr, H.J. Kreuzer, R.R. Netz, Stretching single polypeptides: The effect of rotational constraints in the backbone EPL-Europhys. Lett. (2010) 92, 53001. [100] H. Fu, H. Chen, J.F. Marko, J. Yan, Two distinct overstretched DNA states, Nucleic Acids Res. (2010) 38, 5594-5600. [101] C. Storm, P.C. Nelson, The bend stiffness of S-DNA, EPL-Europhys. Lett. (2003) 62, 760-766. [102] J. Palmeri, M. Manghi, N. Destainville, Thermal denaturation of fluctuating DNA driven by bending entropy, Phys. Rev. Lett. (2007) 99, 088103. [103] J. Palmeri, M. Manghi, N. Destainville, Thermal denaturation of fluctuating finite DNA chains: The role of bending rigidity in bubble nucleation, Phys. Rev. E (2008) 77, 011913. [104] Principal abbreviations used in this work: TPM: tethered particle motion; htTPM: high-throughput TPM; MTT: magnetic [105] The average (cid:104). . .(cid:105) is an ensemble average over realizations. When dealing with experiments, it will become an average torque tweezers; AFM: atomic force microscopy; HMM: hidden Markov model. over time, assuming the validity of the ergodic theorem. 19
1609.03212	2	1609	2017-03-19T10:08:13	Collective motion of groups of self-propelled particles following interacting leaders	[ "physics.bio-ph", "physics.soc-ph" ]	In order to keep their cohesiveness during locomotion gregarious animals must make collective decisions. Many species boast complex societies with multiple levels of communities. A common case is when two dominant levels exist, one corresponding to leaders and the other consisting of followers. In this paper we study the collective motion of such two-level assemblies of self-propelled particles. We present a model adapted from one originally proposed to describe the movement of cells resulting in a smoothly varying coherent motion. We shall use the terminology corresponding to large groups of some mammals where leaders and followers form a group called a harem. We study the emergence (self-organization) of sub-groups within a herd during locomotion by computer simulations. The resulting processes are compared with our prior observations of a Przewalski horse herd (Hortob\'agy, Hungary) which we use as results from a published case study. We find that the model reproduces key features of a herd composed of harems moving on open ground, including fights for followers between leaders and bachelor groups (group of leaders without followers). One of our findings, however, does not agree with the observations. While in our model the emerging group size distribution is normal, the group size distribution of the observed herd based on historical data have been found to follow lognormal distribution. We argue that this indicates that the formation (and the size) of the harems must involve a more complex social topology than simple spatial-distance based interactions.	physics.bio-ph	physics	Collective motion of groups of self-propelled particles following interacting leaders B. Ferdinandya,b,∗, K. Ozogánya,c, T. Vicseka,c aDepartment of Biological Physics, Eötvös Loránd University, H-1117 Pázmány Péter bMTA-ELTE Comparative Ethology Research Group, Hungarian Academy of Science and sétány 1/A, Budapest, Hungary Eötvös Loránd University, H-1117 Pázmány Péter sétány 1/C, Budapest, Hungary cMTA-ELTE Statistical and Biological Physics Research Group, Hungarian Academy of Science and Eötvös Loránd University, H-1117 Pázmány Péter sétány 1/A, Budapest, Hungary Abstract In order to keep their cohesiveness during locomotion gregarious animals must make collective decisions. Many species boast complex societies with multiple levels of communities. A common case is when two dominant levels exist, one corresponding to leaders and the other consisting of followers. In this paper we study the collective motion of such two-level assemblies of self-propelled particles. We present a model adapted from one originally proposed to describe the movement of cells resulting in a smoothly varying coherent motion. We shall use the terminology corresponding to large groups of some mammals where leaders and followers form a group called a harem. We study the emergence (self-organization) of sub-groups within a herd during locomotion by computer simulations. The resulting processes are compared with our prior observations of a Przewalski horse herd (Hortobágy, Hungary) which we use as results from a published case study. We find that the model reproduces key features of a herd composed of harems moving on open ground, including fights for followers between leaders and bachelor groups (group of leaders without followers). One of our findings, however, does not agree with the observations. While in our model the emerging group size distribution is normal, the group size distribution of the observed herd based on historical data have been found to follow lognormal distribution. We argue that this indicates that the formation (and the size) of the harems must involve a more complex social topology than simple spatial- distance based interactions. Keywords: hierarchy collective motion, SPP model, collective motion of groups, ∗Corresponding author Email address: [email protected] (B. Ferdinandy) Preprint submitted to Elsevier March 21, 2017 1. Introduction and motivation Living in social structures with multiple levels of hierarchy is widespread in the animal kingdom [1, 2]. Examples range across several taxa, beginning with humans and primates [3, 4], through elephants [5], to whales [6, 7] and equids [8, 9]. There are numerous examples of subgroups forming around a single individual. For example groups may emerge around a matriarch from her descendants, like in african elephants [5], sperm whales [7], and killer whales [6]. Alternatively a reproductive unit may form around a breeding male with several breeding females and their young as in Przewalski horses [10] and plains zebras [9]. These breeding units can sometimes also include non-breeding males as well, like in hamadryas baboons [4] or geladas [11]. Our aim in the current study was to examine the way in which such a two- level hierarchy may spontaneously emerge in a group and what implications that hierarchy might have for the collective motion of the group. Our motivation and empirical basis was the collective motion of a Przewalski horse herd in light of group formation within the herd, aided by observations made in [12] at the Hortobágy National Park in Hungary. As mentioned before, the Przewalski horse herd is split into harems, organized around a breeding male, with several breeding females and their young offspring. So-called bachelor groups, which consist of males that do not have there own harem are also present [13]. It should be noted, that although zebra harems form herds in the wild and have a very similar social structure to the Przewalski horse, the Przewalski herd at Hortobágy is only semi-wild as it lives in a bounded environment, which may force them into a herd. Although this has not been studied thoroughly, park officials reported, that the initial population did not form a herd, which only appeared after the growth of population density. Both the collective motion of several different species of animals [14], and the emergence of hierarchy within the social system of the Przewalski horse [12] have already been modelled. Conversely, the collective motion of animals that are hierarchically organized into subgroups within a larger group have not been modelled. Thus, we aimed to construct a model of group formation and collective motion of a herd composed of sub-groups as a self-propelled particle (SPP) model in two dimensions, where we identified leaders forming harems, and followers making up these harems. Leadership is a common concept invoked to explain coordinated group move- ments. In ungulates this is often attributed to a single individual. A recent study [15] raises interesting questions about the validity of such a concept, based on ob- servation of two groups of 12 and 6 Przewalski horses. Using different definitions of leadership (moving first, moving in front, or eliciting joining to movement), no individuals that could be consistently classified as a leader were identified. Some limitations to that study are that several types of movements were not measured. In addition movements in the breeding season were also not mea- sured; this was deemed problematic because in the breeding season the stallions directly elicit movements of their harems away from other stallions. Also, due to methodological reasons, only short periods of the day were observed. It has 2 been shown in some cases that in the same group different type of leadership hierarchies might arise in different contexts [16, 17], and there are examples in nature where certain individuals in animal groups consistently act as leaders, for example in zebras and dolphins [18, 19]. These imply that although attributing leadership to a single individual might not be applicable in all circumstances, it does have explanatory power in a wide range of scenarios. As such it stands to reason that conceptualizing the division between leaders and followers di- chotomously helps simplify modelling at a minor cost. Simplifying modelling is helpful in the initial understanding of the type of collective movement we analysed in the present article. Nonetheless this recent study has implications that warrant further field studies of leadership in animals. It would be partic- ularly illuminating to have detailed and continuous data on the movement of large groups of Przewalski horses, which is not easy task, since we do not know of any herd, where attaching measurement devices to the animals is allowed by officials. A substitute for real wild horses could be domestic and feral horses, on which studies have also been carried out [20, 21]. Interestingly, these studies concentrate on movement initiation and not the collective motion itself, showing a somewhat different point-of-view in physicists and biologists. Herein we consider an earlier SPP model of collective cell movement [22] and extend it with a two level hierarchy by introducing two distinct types of particles (i.e. leaders and followers) while simultaneously attempting to limit the increase in the number of parameters. In contrast with [12] the group formation is not driven by the environment of the herd, but by interactions dynamically evolving during the collective motion of the individuals. While formulating the model, we concentrated on mimicking the movements of Przewalski horses. While this specificity adds some complexity to our model, relative to what is usual in statistical physics, it is mostly related to nuances in movement and does not play a major factor in group formation, which was our main focus. Our study could have potential implications for understanding how and why group formation occurs in nature, how group formation affects the system in which it is happening and the rules governing collective motion in a two-level system. Inferring the universalities and the particulars of the different kind of mechanisms, could potentially be used to artificially control both living and human-made systems, such as domestically kept horse herds or flocks of drones. 2. Model The model is based on [22] in which a model was developed to depict the collective motion of cells. We modified this model to accommodate two types of SPP-s (leaders and followers), asymmetric interactions and group formation rules. While extending this model we aimed at minimizing the number of extra parameters. Compared with the usual SPP models the model of [22] gives smoother results due to intrinsic relaxation times. We choose parameters that allow the development of motion that resembles the movements of a herd made of harems as close as possible within the framework of the model. We provide a graphical overview of the model in Figure 1 and an introduction here. 3 The movements of the horses in the model are confined to a square area, large compared to the size of the herd, representing the herding area available to them (Figure 1 boundary). Periodic boundary conditions were not considered, first, because it is not realistic, and second, because it does not make sense in a co- moving herd to conceptualize that the front may interact with the rear. Also, we introduce a tendency for horses that stray too far from the herd to head back while still going in the general direction of the herd's (Figure 1 a)). All horses may follow all other horses, but the strength of the interaction depends on the types and orientations of the SPP-s in question. Given, that it is plausible that leaders must also pay attention to followers, they will fol- low followers too, but to a much smaller extent than the other way around. Although the interactions taking place are based on metric distances, we in- troduce a directedness, meaning that a horse will follow the ones in front of it more than the ones behind it. Several types of interaction modes have been suggested in modelling collective motion. Early models used a simple metric distance, e.g., interacting with anybody nearer than a given distance [23]. Later topological distances were introduced, e.g., interacting with a fixed number of nearest neighbours [24]. Recently it has been proposed that the most biolog- ically correct interaction ranges should be based on visual perception [25]. In our case, vision plays little part as equine vision is near 360◦ [26] and neither the distances within the herd nor the density of the herd imply that occlusion would have a major effect on interactions. As such, the effect of following the ones in front, rather than the ones behind is related more to the logic of not turning around if there are others heading in the same direction as oneself. Leaders who acquire followers (i.e., a harem), will stay farther away from other leaders than if they were without followers (Figure 1 b) and c)). Harems are established based on spatial distance, but followers will gradually belong more and more to the leader they follow, making it easier for them to stay close, because of the stronger and slightly longer distance interactions with their leader than with another leader (Figure 1 d)). Our model starts from randomized initial positions and velocities, without followers being assigned to any leader, thus all followers find groups and leaders at the same time. Our model forgoes the introduction of complex social rules by using only spatial interactions as described above and not taking into consid- eration that in reality, a new horse would be introduced to a herd already split into harems. On the other hand, taking the latter into consideration would not allow for the study of emergent group formation. 2.1. Formal model description We have NL number of leaders and NF number of followers (the list of parameters can be found in Table 1). The 2-dimensional motion of the horse i ∈ {1,N = NL + NF} is described by the overdamped dynamics dri(t) dt = v0 i ni(θi) + Fint(rij, ϕij) + Fcom(¯r − ri, ¯v) + Fwall(ri, vi) + ξ (1) N(cid:88) j=1 j(cid:54)=i 4 Figure 1: Graphical overview of the model depicting a small herd inside the boundary with various parts of Fint,r(ri, rj ) and Fcom shown. Radii are drawn to scale (cf. Table 1 for actual values), and the herd is magnified from within the boundary to show the forces. Solid arrows depict direction of forces, dashed arrows depict actual velocities. The following details are included: a) a horse farther from the center of mass than the given boundary (large green circle centred on the center of mass) will move towards the herd but also in the direction the herd is going, b) & c) leaders without groups can go closer to each other than to a leader with a group, while followers can go even closer to a leader, d) the attraction radius of the follower-leader interactions is generally smaller than that of the leader-leader interactions, but it is increased when interacting with the leader of the follower's group. where t is time, ri is the position of and vi is the velocity of horse i, v0 is a preferred speed which differs for leaders and followers, ni is a unit vector characterized by the angle θi, Fint is a pairwise interaction with rij = ri − rj and ϕij being the angle between ri− rj and vi, Fcom is a global force dependant on the position (¯r) and the velocity (¯v) of the center of mass of the herd, Fwall is the force acting at the boundaries and ξ is a vector whose components are delta-correlated white noise terms with zero mean. The direction of the self-propelling velocity ni(t), described by the angle θi(t), attempts to relax to vi(t) = dri(t)/ dt with a relaxation time τi: (cid:21) · ez , (2) dθi(t) dt = 1 τi arcsin (cid:20)(cid:18) (cid:19) ni(t) × vi(t) vi(t) 5 ''edge''ofherdfromCoMradiusofrepulsionradiusof(directiondependent)adhesionincreasedradiusb)c)d)a)boundaryleaderwithfollowersleaderwithoutfollowersfolloweringroupfollowernotingroupcenterofmassforcevelocity''edge''ofherdfromCoMradiusofrepulsionradiusof(directiondependent)adhesionincreasedradiusb)c)d)a)boundaryleaderwithfollowersleaderwithoutfollowersfolloweringroupfollowernotingroupcenterofmassforcevelocitywhere ez is a unit vector orthogonal to the plane of motion, and τi differs for leaders and followers. This relaxation provides smooth transitions of the ni(t) desired velocities. The value of τ was chosen larger for leaders than followers, implying that leaders are harder to "convince" than followers to change direc- tions, but our results are not sensitive to changes in τ. The Fint(rij, ϕij) force that carries the direct interaction between the horses can be split into the product of a spatial part (Fint,r(rij)), and a coefficient part (Fint,ϕ(ϕij), the latter being dependent on the angle of the direction of horse j from horse i and the direction of the velocity of horse i. The spatial part consists of a pair-wise, asymmetrical force, the direction of which lies on the line passing through the center of masses of the interacting horses and the magnitude of which is the function of the distance rij between the horses [22]. The actual form of the force depends on the type of horses involved: Fint,r(rij) = FLL(rij), FFL(rij), FLF(rij), FFF(rij), if i and j are both leaders, if i is a follower and j is leader, if i is a leader and j is a follower, if i and j are both followers. (3) For all four cases there are two radii defined, RAT which is the range of attraction, and a smaller radii REX, which is the range of repulsion, and also a distance L, which defines a distance inside RAT but outside of REX, splitting the force into four parts depending on distance, namely a repulsive, an attractive and two non-interacting regimes, with different coefficients for all four types of interaction in both the interacting regimes (F AT for the attractive and F EX for the repulsive), thus having 8 radii with 8 coefficients and 4 distances. On the example of FLF(rij) the equations look like this (leader-leader and follower- leader interactions are slightly different):   FLF(ri, rj) = eij × rij−REX LF REX LF , rij−REX LF LF −REX RAT LF −LLF F EX LF 0, F AT LF 0, rij < REX LF , REX LF < rij < REX LF + LLF, LF + LLF ≤ rij ≤ RAT , REX LF , RAT LF < rij, (4) where eij = (ri − rj)/rij. The non-interacting part between REX and RAT was chosen to be very small its only function being is to remove some "vibrations" that arise at such low densities, when a horse is on the edge of the attractive and repulsive regimes. The form of the force is one of the simplest ways to define gradually growing forces based on distances and the values of the specific parameters were chosen to imitate that leaders with harem wish to protect their followers from other leaders, while bachelor leaders themselves can create groups. In the cases of leader-leader (FLL) and follower-leader (FFL) interaction this picture is slightly changed due to the formation of groups. Followers will develop 6 a certain amount of affinity to leaders who are close by, that increases in strength when they are close to the leader and decreases when they are farther away from the leader. Each follower keeps track of time spent near each leader with the quantities Dij ∈ [0,∞], which follow the simple dynamics dDij dt = −1, 0, +1,  This is then translated into an affinity rij ≤ RAT LF , rij > RAT LF and Dij > 0, LF and Dij ≤ 0. rij > RAT (cid:17) − 0.5  + 1, (cid:16)−Dij 1 Aij = 2A 1 + exp τA (5) (6) where τA is the characteristic time of affinity increase and A is a constant. The form of Eq. 6 was chosen so that Aij goes smoothly from 1 → A + 1 as Dij goes from 0 → ∞. This effectively changes the parameters in Eq. 4 (but not in Eq. 5!) for the FFL case from F AT LF . This allows a follower to split farther from the leader it belongs to, without leaving the harem, thus introducing more consistency into the group compositions. FL → AijF AT FL and from RAT LF → AijRAT The definition of groups is based on the values Dij. Every follower is con- sidered to be in the group of the leader for which the value of Dij is largest for the given follower. The leader–leader interaction differs in one aspect if either of the participating leaders have a group, by effectively increasing the repulsive radius REX LL of both leaders fivefold when interacting with each other. As such two leaders can be close to each other only if they don't each have their own groups. This is reminiscent of the distinction between bachelor groups, where males are close together and harems, where the males are farther apart. The velocity dependent part is the same for both leaders and followers: Fint,ϕ(ϕij) = −1 1 + exp(−4(ϕij − π 2 )) + 1, (7) which effectively means, that a horse will pay more attention to horses that are in front of it, rather than those that are behind it. The form was chosen because of the saturation properties. The total interaction is thus Fint(rij, ϕij) = Fint,ϕ(ϕij)Fint,r(rij). (8) The force Fcom keeps the herd roughly together, since if one strays farther than Rcom from the center of mass of the herd it will experience the force (cid:18) ¯r − ri (cid:19) Fcom(¯r − ri, ¯v) = Fcom ri − ¯r − Rcom Rcom ¯r − ri + β ¯v ¯v , (9) where β is parameter that tunes how much the horse is guided in the direction the center of mass is heading and Fcom is the overall strength of the force. Since 7 Rcom is relatively large this force is usually inactive, but will smoothly guide a lost horse back into the herd (adopted from [27]). The force Fwall sets the boundary conditions. The herd is confined to a square area defined by the length D. This box is impenetrable and horses cannot leave it. For the herd to approach this hard boundary in a realistic way, there is a characteristic distance Rwall where the force Fwall is turned on: (cid:18) (cid:20) (cid:18) Rwall − diw (cid:19)(cid:21) (cid:19)(cid:18)vi · nw (cid:19) vi · tw , (10) Fwall(ri, vi) = Fwall 2 sin π Rwall − 1 2 + 1 where diw is the distance of the horse from the boundary, nw is the normal vector of the boundary and tw is the tangent vector of the boundary, driving the horses smoothly along the wall (adopted from [28] and [29]). Initially both leaders and followers are evenly distributed over a square with a linear size of 500, with velocities also randomly distributed. 2.2. Parameters Going, in a naïve way, from the one-type-particle model of [22] to the two- type-particle model would increase the number of required parameters from 14 to 30 (some parameters are doubled and some are increased fourfold given every possible combination of the particles). By considering that some of these are unnecessary to duplicate (or make four of) our model has 23 parameters. Of these only 7 are relevant in the sense that the formation of meaningful groups is sensitive to their value (parameters that would destroy cohesion even in a one-type-particle model were not taken into account), not considering the size of the herd. A parameter was considered relevant if an increase by twofold or a decrease by half resulted in 0.1% of followers not being in a group on average (this is less than one per a realization of the model). For a complete list of parameters see Table 1. Parameters were chosen so that cohesive movement occurs and that group formation happens. Except for cases where there was a reason to do otherwise, parameters that could be different for leaders and follower were kept the same. The distances were chosen based on observations, the coefficients of the various forces were chosen so that the phenomenology of the movements resembles that of a real herd. The leaders are slightly faster than followers so that they are able to stay in front of their harem. It must be noted, that in many cases, leaders in real-life examples may not be at the front of their group, but rather at the side or behind; we elected to use the leading- from-front paradigm for the purpose of simplicity. Other choices pertaining to parameter value selection have been mentioned in the previous section describing the model. 3. Results Our model, with the given parameters, produces a cohesive and ordered motion of the entire herd, while forming groups around leaders and also bachelor groups from group-less leaders. This is in qualitative agreement with the actual 8 variable description default value approx. dimensions A τA F AT FL RAT LL REX LL R*EX LL , 5REX LL F AT LL , F AT LF NL NF relevant variables affinity of followers for leaders characteristic time of affinity strength of F-L attraction radius of L-L attraction radius of L-L repulsion – strength of L-L and L-F attraction number of leaders number of followers 1.3 500 0.03 200 15 75 0.01 25 175 irrelevant variables F AT FF F EX LL v0 L v0 F τL τF ξ LLL, LLF, LFL, LFF RAT LF , RAT REX LF , REX LF , F EX F EX Rcom Fcom β Fwall Rwall D FL , RAT FF FL , REX FF FL , F EX FF strength of F-F attraction strength of L-L repulsion velocity of leaders velocity of followers L velocity relaxation time F velocity relaxation time strength of the noise non-interaction distances radii of attraction radii of repulsion strength of repulsion radius of the cohesion force strength of the cohesion force cohesion force parameter strength of boundary repulsion distance of boundary repulsion linear size of bounding box 0.0002 2 1 0.9 3 1 0.5 1 50 5 5 250 2.5 0.01 3 200 10000 1.3 218 s 0.0125 m/s 36 m 2.7 m 13.6 m 0.0042 m/s 25 175 0.000083 m/s 0.83 m/s 0.416 m/s 0.375 m/s 1.31 s 0.44 s 0.21 m/s 0.18 m 9.1 m 0.9 m 2.08 m/s 45.5 m 1 m/s 0.01 1.25 m/s 36.4 m 1800 m Table 1: Table of the parameters of the model grouped according to relevancy in group formation. L and F abbreviate leader and follower respectively. The approximate proper dimensions are based on a comparison with observed horses, see Section 3.3 for details. 9 Figure 2: Starting from a uniform random distribution of positions and velocities (left side) the herd forms groups and exhibits ordered motion (right side). Blue dots represent leaders and red dots represent followers (see Supplementary video 2 for a video example). observed herd moving on an open plane and as an interesting extra phenomenon, our model also includes "fights" between leaders for followers. By "fights" we mean a situation where two or more leaders without groups get extremely close to one or more followers and after a short time, one of the leaders "wins", i.e. a follower is ascribed to be in the leader's group for long enough for it to chase away the other leaders (see Supplementary video 1). We find that the forming of groups within the herd causes cohesiveness to drop compared to a case without groups. We also find, that in accordance with but with a greater precision than the previous study, the group size distribution of the horses living in the Hortobágy National Park is lognormal. In contrast to this, the current model, based solely on spatial interactions, gives a normal distribution, which implies that spatial interactions alone are not enough to produce the observed group structure. 3.1. Cohesiveness of the herd Starting from uniform random initial positions and velocities of the individ- uals, after sufficient time, the model develops ordered motion throughout the herd while forming groups and thus arriving at a structured and co-moving herd (see Figure 2 and Supplementary video 2). We assume that during collective migration the horses cannot stop, thus there are two phases of ordered move- ment: translational movement, and collective rotation about the – otherwise slowly moving – center of mass. Indeed, when it is not possible to stop (e.g. due to fear), but is not feasible or desirable to move the herd as a whole, herding mammals have been observed to rotate around a common point. To measure translational cohesiveness we use the following translational order parameter Φt = 1 N (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) N(cid:88) i=1 10 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) , vi vi (11) t=100t=30000Figure 3: The herd as a whole either exhibits an ordered translational motion or rotates around a slowly drifting center of mass. This two different type of motion can be distinguished due to the values of the translational (a) and rotational (b) order parameters. The plots are from the same specific run of the model, with the curves smoothed by a window of ∆t = 1000. The spikes during the translational phase are caused by the confining wall (see Supplementary video 4 for a sample of an interaction with the wall). and to measure the rotational cohesiveness we introduce the following rota- tional order parameter N(cid:88) P i=1 Φc = 1 N (cid:19) (cid:18) vi vi , (12) ¯r. where P denotes projection onto the normal of the line going through ri and Going from a totally disordered translational movement to totally ordered translational movement Φt will grow from 0 to 1, while Φc will move from -1 to 1, as the system moves from a totally ordered rotation around the center of mass in one direction, through no collective rotation to totally ordered rotation in the other direction. We find that the system, with parameters given in Table 1 switches between two modes, one of ordered rotation and one of ordered translational motion (see Figure 3 and Supplementary video 3 for an example of a transition from rotational to translational motion). Since the horses in the model do not have the capacity to stop, in an event of indecision about the direction to move they must rotate about a common axis, namely the center of mass. By averaging over the full length of 1000 runs in total, we find that the rotations have no specific direction, as expected (Mann-Whitney U-test, p = 0.96 on left-right similarity). Calculating the pair-correlation function (13) for the leaders in the normal scenario (i.e. where followers are present) and in the scenario where followers are missing, we find that the main structure of ρ(r) = (cid:104)δ(r − ri)(cid:105), 11 01234500.20.40.60.81⋅105𝑡Φu012345−1−0.500.51⋅105𝑡Φu(a)(b)Figure 4: The pair-correlation for a herd composed of leaders and followers, but only calculated on the leaders (solid line), and in the case where only leaders are present (striped line). In the latter case the distances are scaled with R∗EX LL , to compensate for the effect of no leader having a group (leaders without followers can be closer to each other than ones with followers). The main structure of the herd, even with followers, is set by the leaders, but the presence of followers slackens the rigidity of this structure. LL /REX the herd is given by the leaders, and introducing followers only slightly loosens this (aside from the fact that it increases the distances between the leaders, see Figure 4). We also investigated the effect of introducing followers among the leaders on the order parameter of the translational movement. Comparing Φt (calculated using only the velocities of the leaders in two cases, one where there are only leaders and one where there are also followers) we find that order is decreased when allowing for followers and forming of groups (see Table 2). This loss in the efficiency of the movement of the herd as a whole points to benefits gained from social groupings outside the paradigm of simple locomotion. w/o followers with followers (cid:104)Φt(cid:105) 0.866 ± 0.016 0.608 ± 0.018 (cid:104)Φt(cid:105) (only leaders) – 0.633 ± 0.017 (cid:104)Φc(cid:105) −0.002 ± 0.011 0.025 ± 0.020 (cid:104)Φc(cid:105) (only leaders) – 0.026 ± 0.021 Table 2: The translational and rotational order parameters averaged over 120 simulations with standard errors. The duration of the runs were many times longer than the stabilization of groups. The first row is from simulations where only leaders were present, the second row is the full model with followers. In this case the averages were calculated on the whole herd as well as on the leaders only. Adding followers and thus moving in groups decreases the order of translational movement, implying that group formation has benefits other than increased herd cohesion. Although the herd would rotate often, as expected, there is no specific direction of the rotation (Mann-Whitney U-test on 1000 runs, where the simulations was terminated at a time not long after stabilization of groups yields a p = 0.96 on left-rigth similarity). 12 020040060080000.20.4𝑟𝜌(𝑟)leaders&followersonlyleaders3.2. Group size distribution Starting from a uniform random spatial distribution and group-less state, the model, after sufficient time, will produce co-moving groups based on the relative positions of leaders and followers. The emerging group size distribution is normal, although some leaders and followers may not belong to a group. We define groups by the highest (non-zero) Dij values of the followers, i.e a group consists of the leader and the followers with their highest Dij rating corresponding to this leader. This effectively means that groups are formed by followers spending the most time with a specific leader. The group size distribution rapidly reaches a close-to-final state and after some time relaxes to the final state (see Figure 5). We show the transition by creating a histogram of the group sizes at regular intervals during a simulation and taking the sum of the differences of each respective bin of the histogram in two consecutive measurements, averaged over 1000 independent simulations . Figure 5: The groups size distribution quickly stabilizes as it is shown by the plot of ∆. To calculate ∆ we create a histogram of the group sizes at regular intervals during a simulation and take the sum of the differences of each respective bin of the histogram in two consecutive measurements. Each point is averaged over 1000 independent simulations. On Figure 6 we show a comparison of the simulated distribution with real data obtained from a Przewalski horse herd (see [12] for details). Since harem sizes gradually change over time among the horses, the real data has been im- proved by taking into account historical harem size distributions, showing a more clear lognormal distribution than in the previous study of [12]. In this previous study a network model was formulated to account for the lognormal distribution of the group sizes, while the current model, based on purely spatial interactions was not able to reproduce this. This indicates that at this level of complexity, it is not possible to reduce social interactions to spatial interactions. 13 01230246⋅104𝑡ΔFigure 6: Comparison of the group size distribution in the model with an empirical one (the group size distribution of Przewalski horses living in the Hortobágy National Park). The empirical distribution follows a clearer lognormal distribution than in [12] due to the incorporation of historical data. The distribution obtained from the model is close to normal and is mean-fitted to the empirical distribution. We attribute the difference to the fact that the social interactions of horses are too complex to capture in purely spatial interactions. 3.3. Dimension scales Horses usually travel by walking, which is roughly around 1.5 km/h based on our aerial observations averaged over several minutes. In this model v0 L is the corresponding parameter of the walking speed. To compare our model's length scale with that of reality we have calculated the pair-correlation function of the of the wild horses by using aerial pictures of the real herd and that of the herd in our model and compared the first peaks. This roughly equates the arbitrary length unit of our model to 0.18 m in reality (c.f. REX LF in Table 1). From this we can calculate that the arbitrary time unit of our model is roughly equal to 0.44 s. This puts τL and τF at about reaction time (0.5 − 1.5 s), τA to about 3 and a half minutes, and the emergence of a coherent collective motion, with stable harems to slightly less than 10 minutes. Since τL and τF both characterize a fast cognitive process it is not unrealistic that the characteristic times are on the scale of reaction times. Since in wild horses the groups do not form from randomly distributed individuals spontaneously, but rather evolve in an already laid down social context, the time needed for group formation is not readily comparable to that of the real herd. On the other hand, for a group of 200 unfamiliar individuals, where leaders are already appointed and everybody is actually already moving, the 10 minutes seems like a reasonable time for group formation (the authors' personal experience with spontaneous group formation in human groups of comparable sizes would allow for even longer times). 14 05101520253000.020.040.060.080.1groupsizeratiofitofempiricalempiricalsimulation4. Conclusions As the only truly wild horse in the world, the Przewalski horses, now mostly living in relatively easily accessible nature reserves, have drawn considerable at- tention. Both their collective movements [30] and the formation of their harems have attracted interest [12]. However, the unique collective motion displayed by this species, as a large herd consisting of cohesive harems moving together in a coordinated way, has not been modelled to date. Our model, adapted from a model designed for cells, is able to qualitatively reproduce the motion of a wild horse herd moving on an open plain, along with formation of groups consisting of one leader and some followers and bachelor groups (group of leaders without followers), with a roughly adequate correspon- dence of dimension scales. During the analysis of the behaviour of the model we found three interesting phenomena. First, the herd in our model will at times rotate around its center of mass, while we have not observed the horses to circle, many animals do. This is the direct effect of the fact, that in our model the individuals are unable to stop. Indeed, animals that do rotate along a common axis are usually also unable to stop (e.g. flying animals) or is infeasible or dangerous for them to stop. Although some efforts have already been made to model the stopping of a group of animals [27, 31], we suggest further investigations into a model, that would allow for not only the stopping of, but also for the resuming of locomotion. Second, the translational order parameter is decreased when we introduce followers among the leaders, thus the considered grouping process within the herd effectively reduces locomotion efficiency. In many systems the interactions during motion that give rise to collective motion is for the sake of more efficient locomotion of the group as a whole, but the harem formation within a herd is first and foremost due to reproductive reasons. Thus it is not surprising that the reproductive benefits might outweigh the slight decrease in locomotive efficiency. Third, the results obtained from our model are not in agreement with the observed group size distribution of the herd that motivated our work (the latter being a lognormal while the former being a normal distribution). Our sim- ple model operates solely with interactions based on spatial distances, while group-forming processes in real societies have many complex attributes, thus deviations from the exact features of the empirical population is expected. On the other hand, the collective motion in many species can be described by purely distance-based interactions, making the exact nature of these deviations non- trivial. Consequently, we propose further investigations of collectively moving systems to find the properties that allow for the spatial formulation of interac- tions within the system. It can be supposed that in systems where individuals are interchangeable (in the meaning that individual recognition during the mo- tion is not feasible), like a group of cells, ants or a flock of starlings, considering only distance-based interactions is enough to reproduce the observed collective motion pattern, but in animals living in structured social systems (and main- taining an individual recognition), like horses, social factors are much more im- portant during interactions than actual distances, thus interchangeability might 15 be one such property. 5. Acknowledgements We are grateful to the Directorate of Hortobágy National Park for providing us with their data as well as authorizing our observations. This work was supported by the Hungarian Academy of Sciences (grant numbers MTA 01 031 and 2011TKI552). References References [1] C. C. Grueter, B. Chapais, D. Zinner, Evolution of multilevel social systems in nonhuman primates and humans., Int J Primatol 33 (5) (2012) 1002– 1037. doi:10.1007/s10764-012-9618-z. URL http://dx.doi.org/10.1007/s10764-012-9618-z [2] C. C. Grueter, I. Matsuda, P. Zhang, D. Zinner, Multilevel societies in Introduction to the special issue., Int J primates and other mammals: Primatol 33 (5) (2012) 993–1001. doi:10.1007/s10764-012-9614-3. URL http://dx.doi.org/10.1007/s10764-012-9614-3 [3] J. Abegglen, On Socialization in Hamadryas Baboons: a field study., Buck- nell University Press, Lewisburg, 1984. [4] H. Kummer, Social Organisation of Hamdryas Baboons. A Field Study., University of Chicago Press, Chicago, 1968. [5] G. Wittemyer, I. Douglas-Hamilton, W. Getz, The analysis of the processes socioecology creating multitiered so- (2005) 1357 – 1371. elephants: structures, Animal Behaviour 69 (6) of cial doi:http://dx.doi.org/10.1016/j.anbehav.2004.08.018. URL S0003347205000667 http://www.sciencedirect.com/science/article/pii/ [6] R. Baird, The killer whale: Foraging specializations and group hunting., in: J. Mann, R. Connor, P. Tyack, H. Whitehead (Eds.), Cetacean Societies, University of Chicago Press, Chicago, 2000, p. 127–153. [7] H. Whitehead, R. Antunes, S. Gero, S. Wong, D. Engelhaupt, L. Rendell, Multilevel societies of female sperm whales (physeter macrocephalus) in the atlantic and pacific: Why are they so different?, International Journal of Primatology 33 (5) (2012) 1142–1164. doi:10.1007/s10764-012-9598-z. URL http://dx.doi.org/10.1007/s10764-012-9598-z 16 [8] C. Feh, B. Munkhtuya, S. Enkhbold, T. Sukhbaatar, Ecology and so- cial structure of the gobi khulan equus hemionus subsp. in the gobi b national park, mongolia, Biological Conservation 101 (1) (2001) 51 – 61. doi:http://dx.doi.org/10.1016/S0006-3207(01)00051-9. URL S0006320701000519 http://www.sciencedirect.com/science/article/pii/ [9] D. Rubenstein, M. Hack, Natural and sexual selection and the evolution of multi-level societies: insights from zebras with comparisons to primates., in: P. Kappeler, C. van Schaik (Eds.), Sexual Selection in Primates: New and Comparative Perspectives, Cambridge University Press, New York, 2004, p. 266–279. [10] L. Boyd, K. Houpt, Przewalski's Horses, State University of New York Press, New York, 1994. [11] R. Dunbar, P. Dunbar, Social dynamics of gelada baboons., Contrib Pri- matol 6 (1975) 1–157. [12] K. Ozogány, T. Vicsek, Modeling the emergence of modular leadership hierarchy during the collective motion of herds made of harems, Journal of Statistical Physics (2014) 1–19doi:10.1007/s10955-014-1131-7. URL http://dx.doi.org/10.1007/s10955-014-1131-7 [13] E. Hartmann, E. Søndergaard, L. J. Keeling, Keeping horses in groups: A review, Applied Animal Behaviour Science 136 (2–4) (2012) 77 – 87. doi:http://dx.doi.org/10.1016/j.applanim.2011.10.004. URL S0168159111003091 http://www.sciencedirect.com/science/article/pii/ [14] T. Vicsek, A. Zafeiris, Collective motion, Physics Reports 517 (3–4) (2012) 71 – 140, collective motion. doi:http://dx.doi.org/10.1016/j. physrep.2012.03.004. URL S0370157312000968 http://www.sciencedirect.com/science/article/pii/ [15] M. Bourjade, B. Thierry, M. Hausberger, O. Petit, Is leadership a reliable concept in animals? an empirical study in the horse, PLoS ONE 10 (5) (2015) 1–14. doi:10.1371/journal.pone.0126344. URL http://dx.doi.org/10.1371/journal.pone.0126344 [16] I. Watts, B. Pettit, M. Nagy, T. B. de Perera, D. Biro, Lack of experience- based stratification in homing pigeon leadership hierarchies, Royal Society Open Science 3 (1). arXiv:http://rsos.royalsocietypublishing.org/ content/3/1/150518.full.pdf, doi:10.1098/rsos.150518. URL http://rsos.royalsocietypublishing.org/content/3/1/150518 [17] M. Nagy, G. Vásárhelyi, B. Pettit, I. Roberts-Mariani, T. Vicsek, D. Biro, Context-dependent hierarchies in pigeons, Proceedings of the National Academy of Sciences 110 (32) (2013) 13049–13054. 17 [18] L. C. David Lusseau, The emergence of unshared consensus decisions in bottlenose dolphins, Behavioral Ecology and Sociobiology 63 (7) (2009) 1067–1077. URL http://www.jstor.org/stable/40295396 [19] I. R. Fischhoff, S. R. Sundaresan, J. Cordingley, H. M. Larkin, M.-J. Sell- ier, D. I. Rubenstein, Social relationships and reproductive state influence leadership roles in movements of plains zebra, equus burchellii, Animal Behaviour 73 (5) (2007) 825–831. [20] L. Briard, C. Dorn, O. Petit, Personality and affinities play a key role in the organisation of collective movements in a group of domestic horses, Ethology 121 (9) (2015) 888–902. doi:10.1111/eth.12402. URL http://dx.doi.org/10.1111/eth.12402 [21] K. Krueger, B. Flauger, K. Farmer, C. Hemelrijk, Movement initiation in groups of feral horses, Behavioural Processes 103 (2014) 91 – 101. doi:http://dx.doi.org/10.1016/j.beproc.2013.10.007. URL S0376635713002222 http://www.sciencedirect.com/science/article/pii/ [22] B. Szabó, G. J. Szöllösi, B. Gönci, Z. Jurányi, D. Selmeczi, T. Vicsek, Phase transition in the collective migration of tissue cells: Experiment and model, Phys. Rev. E 74 (2006) 061908. doi:10.1103/PhysRevE.74.061908. URL http://link.aps.org/doi/10.1103/PhysRevE.74.061908 [23] T. Vicsek, A. Czirók, E. Ben-Jacob, I. Cohen, O. Shochet, Novel type of phase transition in a system of self-driven particles, Physical review letters 75 (6) (1995) 1226. [24] M. Ballerini, N. Cabibbo, R. Candelier, A. Cavagna, E. Cisbani, I. Giar- dina, V. Lecomte, A. Orlandi, G. Parisi, A. Procaccini, et al., Interaction ruling animal collective behavior depends on topological rather than metric distance: Evidence from a field study, Proceedings of the national academy of sciences 105 (4) (2008) 1232–1237. [25] A. Strandburg-Peshkin, C. R. Twomey, N. W. Bode, A. B. Kao, Y. Katz, C. C. Ioannou, S. B. Rosenthal, C. J. Torney, H. S. Wu, S. A. Levin, et al., Visual sensory networks and effective information transfer in animal groups, Current Biology 23 (17) (2013) R709–R711. [26] D. E. Brooks, A. Matthews, Equine ophthalmology, Veterinary ophthal- mology 2 (1999) 1165–1274. [27] K. Bhattacharya, T. Vicsek, Collective decision making in cohesive flocks, doi:10.1088/1367-2630/12/9/ NEW JOURNAL OF PHYSICS 12. 093019. 18 [28] C. Virágh, G. Vásárhelyi, N. Tarcai, T. Szörényi, G. Somorjai, T. Nepusz, T. Vicsek, Flocking algorithm for autonomous flying robots, Bioinspiration & Biomimetics 9 (2) (2014) 025012. URL http://stacks.iop.org/1748-3190/9/i=2/a=025012 [29] C. Virágh, G. Vásárhelyi, N. Tarcai, T. Szörényi, G. Somorjai, T. Ne- pusz, T. Vicsek, Corrigendum: Flocking algorithm for autonomous fly- ing robots (2014 bioinspir. biomim. 9 [http://dx.doi.org/10.1088/1748- 3182/9/2/025012] 025012 ), Bioinspiration & Biomimetics 9 (4) (2014) 049501. URL http://stacks.iop.org/1748-3190/9/i=4/a=049501 [30] M. Bourjade, B. Thierry, M. Maumy, O. Petit, Decision-making in prze- walski horses (equus ferus przewalskii) is driven by the ecological con- texts of collective movements, Ethology 115 (4) (2009) 321–330. doi: 10.1111/j.1439-0310.2009.01614.x. URL http://dx.doi.org/10.1111/j.1439-0310.2009.01614.x [31] B. Ferdinandy, K. Bhattacharya, D. Ábel, T. Vicsek, Landing together: How flocks arrive at a coherent action in time and space in the presence of perturbations, Physica A 391 (4) (2012) 1207–1215. doi:doi:10.1016/j. physa.2011.10.010. 19
1412.5343	2	1412	2016-06-17T09:29:26	Transport of organelles by elastically coupled motor proteins	[ "physics.bio-ph", "cond-mat.soft", "cond-mat.stat-mech", "q-bio.SC" ]	Motor-driven cargo transport is a complex phenomenon where multiple motor proteins attached on to a cargo engage in pulling activity, often leading to tug-of-war, displaying bidirectional motion. However, most mathematical and computational models ignore the details of the motor-cargo interaction. Here, we study a generic model in which N motors are elastically coupled to a cargo, which itself is subjected to thermal noise in the cytoplasm and to an additional external applied force. The motor-hopping rates are chosen to satisfy detailed balance with respect to the energy of elastic stretching. With these assumptions, an (N+1)-variable master equation is constructed for dynamics of the motor-cargo complex. By expanding the hopping rates to linear order in fluctuations in motor positions, we obtain a linear Fokker-Planck equation. The deterministic equations governing the average quantities are separated out and explicit analytical expressions are obtained for the mean velocity and diffusion coefficient of the cargo. We also study the statistical features of the force experienced by an individual motor and quantitatively characterize the load-sharing among the cargo-bound motors. The mean cargo velocity and the effective diffusion coefficient are found to be decreasing functions of the stiffness. While increase in the number of motors N does not increase the velocity substantially, it decreases the effective diffusion coefficient which falls as 1/N asymptotically. We further show that the cargo-bound motors share the force exerted on the cargo equally only in the limit of vanishing elastic stiffness; as stiffness is increased, deviations from equal load sharing are observed. Numerical simulations agree with our analytical results where expected. Interestingly, we find in simulations that the stall force of a cargo elastically coupled to motors is independent of the stiffness.	physics.bio-ph	physics	EPJ manuscript No. (will be inserted by the editor) 6 1 0 2 n u J 7 1 ] h p - o i b . s c i s y h p [ 2 v 3 4 3 5 . 2 1 4 1 : v i X r a Transport of organelles by elastically coupled motor proteins Deepak Bhata and Manoj Gopalakrishnan Department of Physics, Indian Institute of Technology Madras, Chennai 600036, India (e-mail: [email protected]) Received: date / Revised version: date Abstract. Motor-driven intracellular transport is a complex phenomenon where multiple motor proteins simultaneously attached on to a cargo engage in pulling activity, often leading to tug-of-war, displaying bidirectional motion. However, most mathematical and computational models ignore the details of the motor-cargo interaction. A few studies have focused on more realistic models of cargo transport by including elastic motor-cargo coupling, but either restrict the number of motors and/or use purely phenomenological forms for force-dependent hopping rates. Here, we study a generic model in which N motors are elastically coupled to a cargo, which itself is subjected to thermal noise in the cytoplasm and to an additional external applied force. The motor-hopping rates are chosen to satisfy detailed balance with respect to the energy of elastic stretching. With these assumptions, an (N + 1)− variable master equation is constructed for dynamics of the motor-cargo complex. By expanding the hopping rates to linear order in fluctuations in motor positions, we obtain a linear Fokker-Planck equation. The deterministic equations governing the average quantities are separated out and explicit analytical expressions are obtained for the mean velocity and diffusion coefficient of the cargo. We also study the statistical features of the force experienced by an individual motor and quantitatively characterize the load-sharing among the cargo-bound motors. The mean cargo velocity and the effective diffusion coefficient are found to be decreasing functions of the stiffness. While increase in the number of motors N does not increase the velocity substantially, it decreases the effective diffusion coefficient which falls as 1/N asymptotically. We further show that the cargo-bound motors share the force exerted on the cargo equally only in the limit of vanishing elastic stiffness; as stiffness is increased, deviations from equal load sharing are observed. Numerical simulations agree with our analytical results where expected. Interestingly, we find in simulations that the stall force of a cargo elastically coupled to motors is independent of the stiffness of the linkers. PACS. 05.40.Fb random walks -- 05.40.-a stochastic processes -- 87.16.Nn motor proteins -- 05.10.Gg Langevin method 1 Introduction Motor-protein based cargo transport is a mechanism which governs the spatial organisation of organelles like endo- some, vesicles, mitochondria etc. inside a eukaryotic cell. A few proteins belonging to dynein and kinesin families are known to be involved in transporting the cargoes on mi- crotubule filaments [1,2,3]. Using the structural polarity of the microtubule, dyneins move toward the minus end of a microtubule, while most kinesins move toward the plus end [4]. In many cases, the cargo is driven by multiple motor proteins leading to increased stall force and trans- port over longer distances [5,6]. Due to the involvement both dyneins and kinesins, motion of cargo is found to be bidirectional in some cases [7,8]. Experiments have shown evidence for tug-of-war mediated mechanical interaction between two teams of opposing motors (dyneins and ki- nesins) leading to bidirectional motion of cargoes [9,10]. a Present address: International Centre for Theoretical Sci- ences, Survey No. 151, Shivakote, Hesaraghatta Hobli, Ben- galuru North 560089, India (e-mail:[email protected]) Mechanical interaction between motors of the same polar- ity is less understood. A few experiments have given insights about the na- ture of interaction between multiple motor proteins by coupling them artificially through a DNA scaffold [11,12]. In [11], DNA coupled by two kinesins effectively behaved as though the single motor attachment state dominated the motility. This led to the conclusion that asynchronous stepping of kinesins show dominant negative interference. While the average velocity of the coupled system was al- most same as that of single motor, the run length was observed to be much larger for coupled motors. In an- other interesting experiment reported in Furuta et al.[12], multiple motors were made to attach on a DNA scaffold with controlled separations between them. The velocity and the stall force of the DNA scaffold was found to be affected by changes in the spatial separation between the motors. These observations clearly indicate the presence of chemical or mechanical interaction between similar mo- tors. 2 Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins gated and it is shown that the mean velocity of the cargo is independent of the stiffness, whereas the mean runtime decreases with increase in stiffness. A discrete state tran- sition rate model employed in [24] showed slight reduction in two motor run length and velocity at larger stiffnesses, due to increase in the strain force. In a more recent study, Berger et al. [25] identified four distinct transport regimes in a system of two elastically coupled motor proteins. In [28], the dynamics of the cargo coupled to a single motor protein was studied using a novel coarse-graining approach based on separation of time-scales between the cargo and motor. A very recent study by Bouzat[29] showed that a model in which motors experience history-dependent forces captures some of the experimental observations bet- ter than the previously studied models. Some of the formalisms developed in the above men- tioned analytical studies were applied to simple cases with at most two motor proteins driving a cargo [24,25,26, 28], while many other studies are by and large compu- tational in nature [19,20,22,23,29]. A general theoretical framework to study the motion of a cargo driven by ar- bitrary number of elastically bound motors has not been developed. We develop such a treatment in this paper. We study the motion of a cargo pulled by N elastically cou- pled motor proteins in a viscous medium, subject to ther- mal noise and an external applied force. Starting from the complete master equation, we extract deterministic dynamical equations for the averages and a linear Fokker- Planck equation(LFPE) for the fluctuations. We study the effects of elastic stiffness and the number of motors on var- ious statistical properties such as the drift and diffusion of the cargo, force experienced by each motor protein and its deviation from the mean field approximation. Finally, we carry out computer simulations in order to verify our analytical results and mostly observe good agreement be- tween the two wherever expected. Among the limitations of our study, spontaneous and load-induced detachment of motors from the filament is completely neglected. Also, truncation of the expansion of the master equation leads to incorrect prediction on the variation of stall force with stiffness. These limitations will be partially addressed in a future publication. The paper is organised as follows. In sec.2, we describe our model and set up the compete master equation, which is the separated into deterministic "macroscopic" equa- tions for the averages and a LFPE for fluctuations. We derive various properties of the cargo transport such as the average cargo velocity and effective diffusion coefficient of the motor-cargo complex in sec.2.1. One of the important ingredients in the model, the elastic stretching energy de- pendence of motor jumping rates, is discussed in sec.2.2. In sec.2.3, we apply the formalism developed in sec.2.1 to a cargo driven by N identical motors. In sec.2.4, we de- scribe the computer simulations techniques employed in our study and in sec.2.5 we verify all the results obtained in sec.2.3 using computer simulations. We compare predic- tions of our model with experiments in sec.3. Finally, in sec.4 we summarize our results and discuss some of their implications. Fig. 1. The figure shows a set of motor proteins are pulling a cargo against an external load f. As each motor protein is at a different spatial separation from the cargo, individual mo- tors experiences different stretching forces. In particular, the leading motor experiences a large force and a lagging motor experience a smaller force. Motor proteins and motor-driven membranous organelles like endosomes, mitochondria etc. are soft molecules with stiffness typically of the order of a few pNnm−1 [1,13,14] Since motor proteins exert forces of a few pN on these organelles, the elastic nature of motor-cargo assembly is likely to be an influential factor in the transport process. Early models on unidirectional and bidirectional trans- port have generally ignored the details of interaction be- tween similar motor proteins in a team [15,16,17,18]. In these models, external force on the cargo is assumed to be shared equally among the like motors, akin to a "mean- field" approximation [19]. It is, however, plausible that stretching of the elastic motor-cargo linkers generates ten- sion on the motors even in the absence of any external force on the cargo. Further, in the presence of external force, the force experienced by an individual motor could differ from the mean-field value. As depicted in fig.1, the elastic stretching force experienced by a motor on a cargo depends on its spatial location which changes with time, and is, therefore, a fluctuating quantity. Several studies have appeared in the last few years, based (mostly) on Hookean spring-like interaction between motor and the cargo [19,20,21,22,23,24,25,26,27,28,29]. In Kunwar et al. [20], along with elastic interaction among the motors, non-linear force-velocity relations are assigned to different motors. Average run length of multiple motor driven cargo in the presence of external force is studied computationally at different motor stiffnesses. A more sys- tematic semi-analytic study done by Materassi et al.[21] reproduced the results in [20]. It was shown by Kunwar et al. [19] that, compared to the mean-field model in which load is assumed to be shared equally among the motors of same directionality, a stochastic load sharing model based on elastic motor-cargo interaction can explain uni- directional transport more reliably. They found, however, that neither model is consistent with all the experimen- tal observations of bidirectional transport. In a computa- tional study, Bouzat and Falo [22] have shown that the stall force of the cargo-motor assembly is larger for non- interacting motors than motors interacting through elas- tic strain force. In another study by Bouzat and Falo [23], tug-of-war between multiple opposing motors is investi- Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins 3 where γ is the drag coefficient, ζ(t) is Gaussian white noise: hζ(t)i = 0 and hζ(t)ζ(t′)i = Dδ(t − t′), where D = 2kBT /γ following the fluctuation-dissipation theo- rem. Note that the sign of f is positive when the force acts in the +x0 direction; hence an "opposing force" will have − sign in this convention. n (∆n) and W − The hopping rates of the individual motors are modi- fied by the presence of the elastic stretching energy. Let us denote by W + n (∆n) the single motor hop- ping rates for the forward (xn → xn + 1) and backward (xn → xn−1) transitions respectively, as shown in fig.2(b). Then, the complete equation for the probability distribu- tion P (x; t), that describes the stochastic dynamics of the motor-cargo system is written as: ∂P (x; t) ∂t − N = Xn=1(cid:2)(E+ ∂x0 " N Xi=1 ∂ κi γ n − 1)W − n P + (E− n − 1)W + n P(cid:3) ∂x0# (2) 2ℓ2(cid:19) ∂P ∆i + f ℓγ! P −(cid:18) D where E+ n and E− n are a set of N raising and lowering operators, defined through the relations E+ P (x0, x1, .., xn+1, ..), and similarly, E− P (x0, x1, .., xn − 1, ..)[32]. n P (x0, x1, .., xn, ..) = n P (x0, x1, .., xn, ..) = 2.1 Expansion of the master equation In eq.2, we expand all the variables xn about their average values xn(t) as follows: xn = xn(t) + ηn 0 ≤ n ≤ N, (3) where ηn are fluctuations about the averages, hence hηni = 0 by construction. The dynamics may be described now in terms of the variables ηn, the probability distribu- tion of which is defined as Π(η; t) ≡ P (x; t) so that ∂P (x; t) ∂t = ∂Π(η; t) ∂t − dxn dt ∂Π ∂ηn . (4) N Xn=0 With the shift of variables in eq.3, the operators E+ n and E− n in eq.2 admit the Taylor expansion [32] E± n = ∞ Xm=0 (±1)m m! ∂m ∂ηm n . (5) Insertion of eq.5 along with eq.4 into eq.2 yields the following Kramers-Moyal expansion in terms of the vari- ables ηn: ∂Π ∂t − N Xn=0 xn ∂Π ∂ηn = N Xn=0 ∞ Xm=1h − (1) + 1 (2m − 1)! ∂2m−1 ∂η2m−1 n (VnΠ) 1 (2m)! ∂2m ∂η2m n (DnΠ)i, (6) (a) (b) Fig. 2. (a) Cartoon of multiple elastic motor proteins (shown as springs) coupled to a cargo performing one-dimensional Brownian motion on a microtubule. (b) A motor at xn jumps forward with a rate W + n (∆n) and backward with a rate W − n (∆n). The corresponding reverse transition rates, with starting points at xn + 1 and xn − 1 are, respectively, given by W − n (∆n + 1) and W + n (∆n − 1). 2 A generic model for a cargo elastically coupled to N motor proteins A motor protein moving on a microtubule filament moves in a sequence of jumps of fixed length (usually, although dynein is known to take variable-sized jumps in response to load [30,31]) from one monomer to the next . For a single free motor protein, let ℓ be the jump length, wn and vn be the forward and backward hopping rates re- spectively. A motor protein is imagined to be bound to the cargo using a spring of stiffness κn, which is an ap- proximation to the motor-cargo linker in our model. We consider a cargo pulled by N such motor proteins simul- taneously, as depicted in fig.2(a). In reality, the stiffness κn could also have contribution from the elastic nature of membranous cargo, which we have not taken separately in to account. Let x0 and xn (1 ≤ n ≤ N ) be the positions of the cargo and the n'th motor on the filament respectively, expressed in units of jump length ℓ. For the sake of later convenience, we use the convention x ≡ {x0, x1....xN }. If ∆n = xn − x0 is the instantaneous separation between n'th motor and the cargo, then the instantaneous elastic force on the n'th motor is fn = −ℓκn∆n and so the elas- tic force on the cargo due to all the motors at different i=1 κi∆i. For a cargo bound to N motor proteins and acted on by an external force f , the over-damped Langevin equation is written as [22,28] locations on the filament is fc = ℓPN ℓ x0 = ℓ N Xi=1 κi∆i γ + f γ + ζ(t), 4 Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins where Vn = W + n − W − n ; Dn = W + n + W − n , (7) By using the definition of hηpηqi given in eq.13, we obtain the dynamics of the correlation hηpηqi from eq.12 as follows (see Appendix B for the derivation): for 1 ≤ n ≤ N , while V0 = κi γ N Xi=1 ∆i + f ℓγ ; D0 = D ℓ2 . (8) To proceed further, we carry out Taylor expansion of Vn and Dn in powers of the fluctuations ηn, leading to Vn(x) = V n +Xk Dn(x) = Dn +Xk α(1) nk ηk +Xk,l nk ηk +Xk,l β(1) α(2) nklηkηl + .. β(2) nklηkηl + .. (9) N dhηpηqi dt = Xn=0hDnδpnδqn + α(1) pn hηqηni + α(1) qn hηpηnii . (15) The results obtained in eqs.14 and 15 are very general and valid for arbitrary number of motors of either direc- tionality. 2.2 Energy-dependence of motor hopping rates We will now make specific choices for the functional form of motor hopping rates in the model. Let E(x) = (ℓ2/2)Pn κn∆2 be the total energy of the system in a configuration x ≡ {x0, x1, ..xn, ..xN }. We define the local energy differences n where we have defined (a) the "macroscopic" variables Vn ≡ Vn(x) ; Dn ≡ Dn(x) (10) ε± n = ±[E(..xn ± 1, ..) − E(..xn, ..)] = κnℓ2 2 [2∆n ± 1](16) and (b) Taylor coefficients of order r (r ≥ 1) ; β(r) nkl.. = 1 r! . ∂rDn ∂xk∂xl...(cid:12)(cid:12)(cid:12)(cid:12)x=x α(r) nkl.. = 1 r! ∂rVn ∂xk∂xl...(cid:12)(cid:12)(cid:12)(cid:12)x=x It can be shown that the coefficients α(r) (11) nkl.. and β(r) nkl.. are proportional to (βκℓ2)r (see Appendix A), and there- fore, if κ is sufficiently small, higher order terms may be neglected in the expansion in eq.9. For ℓ = 8nm, this re- quires κ ≪ 0.06pN/nm, far smaller than the estimated value for kinesin (0.3 pN/nm) [13,14]. Nonetheless, for the sake of making further analytic progress, we proceed with this "weak coupling" approximation. When the expansion in eq.9 is thus truncated at r = 1, we obtain the LFPE ∂Π ∂t ≃ N Xn=0 ∂ α(1) ∂ηn " xn − V n −Xk nk ηk! Π# Xn=0 ∂2Π ∂η2 n Dn 1 2 + N . (12) The averages and the correlation functions of ηn are defined as follows: hηpi =Z ηpΠdη ; hηpηqi =Z ηpηqΠdη. (13) In order to satisfy the condition hηpi = 0, we put the convective term (corresponding to first order derivative with respect to ηn) in eq.12 to zero (see Appendix B), thereby arriving at the "macroscopic equations" [32]: xn = Vn ; (0 ≤ n ≤ N ). (14) for a certain motor n at position xn, corresponding to a single hop to its right or left. Then, we propose that the energy-dependent forward and backward hopping rates in- troduced earlier follow a local detailed balance condition, i.e., n (∆n) W + n (∆n + 1) W − = wn vn exp(−βε+ n ). (17) where β = (kBT )−1. The dynamic quantities that char- acterise the transport process, e.g. the instantaneous ve- locity of the motor and its diffusion coefficient depend on the difference and sum of the forward and backward rates respectively (eq.7). Hence, our results for these quantities will depend on the specific forms for the forward and back- ward rates satisfying eq.17. We consider a set of rates of the following form: W + W − n (∆n) = wn exp(cid:2)−βε+ n θn(cid:3) n (∆n) = vn exp(cid:2)βε− n (1 − θn)(cid:3) (18) with 0 ≤ θn ≤ 1. This model has been studied exten- sively in the literature [28,33,34,35,36]. In some of these studies [33,34], in the exponent, θn is accompanied by the external force exerted on the motor, whereas in our model (also in [35,36]) it is accompanied by the energy differ- ence ε±. It is suggested in [34] that, θn determines the location of the transition state in the periodic free-energy landscape in which the motor takes steps. In the next section, we will extract analytical expres- sions for several quantities of interest for a cargo driven by a set of identical motors (wn = w, vn = v, κn = κ, θn = θ) using the formalism developed so far. When motors are identical, the mean separation between motor and the Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins 5 0 2 . q e RHS (0,0) LHS ∆ Fig. 3. The functions on left hand side (LHS) and right hand side (RHS) of the eq.20 are shown in this figure schematically. The solution to eq.20 is the value of ∆ at which LHS cuts the RHS in the figure (dashed line). cargo, ∆ ≡ ∆n, is the same for all the motors. Also, Vn = V1 and Dn = D1 ∀ n ≥ 2. Further, the rates in eq.18 depends only on the separation ∆n = xn − x0, therefore by the definition of Taylor coefficients in eq.11, α(1) n0 = −α(1) nm = 0 if n, m ≥ 1 and n 6= m. Finally, α(1) γ . We use these quantities in the next section to analytically de- termine different properties characterising the dynamics of the motor-cargo complex. nn ≡ α for all n ≥ 1. However α(1) 00 = −N α(1) 0n = −N κ 2.3 Results for a cargo driven by identical motor proteins (i) Average cargo velocity: For identical motors, since the hopping rates and stiffness of all the motors are equal, the mean separation between motor and the cargo is the same for all motors: ∆ ≡ ∆n. Therefore, from eqs.1, 8 and 10, the average cargo velocity is given by Vavr ≡ ℓV0 = N κℓ γ ∆ + f γ , (19) In the steady state, both motors and cargo move es- sentially with the same velocity and therefore Vn = V0 ∀ n ≥ 1. Hence, from eqs.7, 8 and 10, we arrive at the following transcendental equation for ∆: N κℓ γ ∆ + f γ = ℓ(cid:2)W +(∆) − W −(∆)(cid:3) (20) Solution to eq.20 is to be reinserted in eq.19 to find the average cargo velocity. Let us first understand the nature of solution to the transcendental eq.20 for the simple case where the exter- nal force on the cargo is zero (f = 0). The left hand side is a straight line with slope equal to N κℓ/γ pass- ing through the origin. The right hand side is, in general, a smooth monotonic function of ∆ which varies from ∞ to −∞ (see fig.3 for a schematic picture). Therefore, as the slope N κℓ/γ becomes larger and larger, the solution to the equation ∆ becomes smaller in magnitude (sign is positive if w > v and negative if w < v). It suggests three different roles of κ, γ and N respectively in the transport mechanism: (a) When γ becomes larger, cargo experiences a large drag force while motors try to pull it away. This in- creases the motor-cargo separation as γ increases. (b) As κ is increased, the energy cost due to stretching increases, larger separation between motor and cargo is energetically unfavourable and this will enhance the backward hopping rate of the motor. As a result, we see a reduction in mean stretch ∆ with increase in κ. (c) Because all the motors are of same directionality, the number of springs which are connected in parallel increases (on an average) as N is increased. Due to additive nature of the spring constant in parallel springs, stretching of motors results in large energy cost and hence the mean separation between mo- tor and cargo decreases with N . We will refer to these points again in subsequent sections where we use the spe- cific rates given in eq.18 explicitly. In the presence of external force (f 6= 0), the solution to the transcendental eq.20 depends on the directionality of the motor bound to the cargo. If the net motion of the cargo is towards plus direction (i.e. w > v) on the micro- tubule, then f < 0 corresponds to a resisting force and decrease in f would increase the magnitude of ∆. On the other hand if the net motion of the cargo is towards minus direction (i.e. w < v), then f < 0 corresponds to assisting force and decrease in f would decrease the magnitude of ∆. (ii) Effective diffusion coefficient: Starting from eq.15, the long time behaviour of the variance hη2 i i can be cal- culated exactly for simple cases like N = 1, 2, results for which are given in Appendix B. Based on these results, we conjecture that for a cargo pulled by N motor proteins, the effective diffusion coefficient is α2 D0 + N ( κ γ )2 D1 . (21) γ(cid:17)2 2(cid:16)α + N κ   Deff(N ) = ℓ2  Interestingly, eq.21 implies that Deff ∝ N −1 as N → ∞. N -dependence of Vavr and Deff: The asymptotic re- sult obtained above may be understood using an intu- itive argument in the following way. For a free motor on the filament, w and v are forward and backward hopping rates respectively, with a fixed jump length ℓ. The aver- age velocity and diffusion coefficient for this are given by Vf = ℓ(w − v) and Df = ℓ2(w + v)/2 respectively. How- ever, when multiple motor proteins are coupled through cargo, one may naively visualise whole system of motors as an effectual random walker pulling a cargo with mod- ified hopping rates we ∼ N w, ve ∼ N v and jump length ℓe ∼ ℓ/N . Then, the effective velocity Ve = ℓe(we − ve) be- comes independent of N while the effective diffusion coef- e(we + ve)/2 ∼ N −1 as N → ∞. The scaling ficient De = ℓ2 holds for a large range of stiffness, particularly when N is large. However, as discussed already, the exact effective 6 Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins velocity and diffusion coefficient depend on the effective stiffness (κe = N κ) in a highly non-linear manner. This gives an additional weak N dependence to Vavr and Deff at small N . (iii) Average force on the motor and force-fluctuations: The instantaneous force fn on the n'th motor is de- termined by the separation between the motor and cargo at that instant, i.e., fn = −κℓ(xn − x0) = −κℓ∆n. Be- cause motor-cargo separation ∆n changes randomly with time, fn is also varies stochastically. The present formal- ism allows us to systematically determine the statistics of force experienced by the motor proteins and its load shar- ing properties. By the transformation of variables given in eq.3, we write the average force experienced by the motor as hfni = −κℓ∆ for identical motors. hfni has two con- tributions: one is due to the external force f and other is due to the viscous drag force. To quantify the contri- bution by the viscous drag, we evaluated the difference between force experienced by the motor from f /N , i.e., define δf = fn − (f /N ). From eq.19, we notice a relation between the average deviation in the force hδf i and the average cargo velocity Vavr, i.e., hδf i = −γ Vavr N , (22) which suggests that the contribution due to the viscous drag is negligible (hfni = f /N ) when Vavr = 0 i.e. when cargo is stalled or when the motors are large in number (N ≫ 1). It must be noted that equal [15] and stochastic [19,20] load-sharing models of motor-driven transport have been discussed in the literature. However, a meticulous study of force experienced by the motor and its deviation from mean field limit in stochastic load sharing model is still missing. The present formalism allows us to systemati- cally determine the statistics of force experienced by the motor proteins and its load sharing properties. To investi- gate the deviation of force experienced by the motor from the mean field approximation, we study the standard de- viation in the force experienced by the motor defined as ni − hfni2. In particular, note that if the force per motor is always the same, σf (N ) would vanish: this limiting behaviour is achieved when stiffness κ is suffi- ciently small, that is when motor-cargo interaction is neg- ligible. With increase in the stiffness, fluctuations in the instantaneous force fn relative to its mean value increases. As a result, σf (N ) is an increasing function of κ. It may be easily shown that σ2 0 i − 2hη0ηni] for n ≥ 1. From the expressions for hηpηpi in Appendix B, specific results for N = 1 and N = 2 are obtained: σf (N ) = phf 2 f = κ2ℓ2[hη2 n i + hη2 D0 + D1 σf (1) = κℓvuuut   ,  γ(cid:17) 2(cid:16)α + κ σf (2) = κℓvuut" αD0 + (α + κ γ )D1 2α(α + 2 κ γ ) #. (23) Because D1 is a function of κ, the nature of the depen- dence of σf on κ is not obvious from eq.23. However, as we will see later from numerical simulation that, σf (N ) is an increasing function of κ and N . (iv) Stall force: To determine the expression for stall force f N (force corresponding to vanishing cargo velocity, s i.e., Vavr = 0) within this formalism, we put the RHS of eq.20 to zero, the solution of which may be denoted ∆s(κ). Then, from the LHS, the stall force is given by fs(κ) = −N κℓ∆s(κ): after substitution of ∆s(κ) we find the stall force to be f N s = N(cid:20) 1 βℓ log(cid:16) v w(cid:17) − κℓ(cid:18) 1 2 − θ(cid:19)(cid:21) (24) Eq.24 completes the set of results we obtained from our approximate analytical treatment of the problem. We now proceed to discuss the results from numerical simulations. 2.4 Numerical simulations to determine the force on the cargo fc = κℓPN In numerical simulations, we used Brownian dynamics for cargo motion, along with fixed time-step kinetic Monte- Carlo scheme for motor dynamics. All the motors and the cargo are initialised at xn = 0 (0 ≤ n ≤ N ) at t = 0 and their locations are updated in each time step δt = 10−5s. From the noted locations of the cargo and all the mo- tors (x0, x1...xN ) in the present time t, the separation be- tween n'th motor and the cargo ∆n = xn − x0 is used n=1 ∆n and the energy cost for forward/backward jump ε± n = (κnℓ2/2)[2∆n±1]. In the next time step t+δt, the location of the cargo is estimated according to the over-damped Langevin equation (eq.1) and motor locations are updated using the rates given in eq.18. Position of the cargo and all the individual motors are recorded for a long time (this time scale depends on the number of motors N and stiff- ness κ; the time required to reach steady state is different in each cases, but typically varies from 5s to 25s), which enabled us to determine various averaged quantities of in- terest in the steady state. In particular, the average veloc- ity is estimated from the relation hδxi/δτ where, hδxi is the average distance covered in the observation time δτ . The averaging is carried out over several identical copies (typically 50000) of the system. Similarly the effective dif- fusion coefficient is evaluated as [hδx2i − hδxi2]/2δτ . 2.5 Comparing theory and simulations Velocity versus κ: We have chosen the set of parameters given in Table 1, and found the solution for ∆ from the transcendental equation 20 numerically. Then the average cargo velocity Vavr is evaluated from eq.19. As explained earlier, due to increase in energy cost for the movement of motors on the filament, average separation ∆ is reduced at larger κ. Therefore, the average velocity of the cargo decreases with κ. Increase in the number of motors results Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins 7 β γ pN−1nm−1 0.2433 pNs nm−1 9.42 × 10−4 w s−1 v s−1 125.0028 0.0028 ℓ nm 8 θ 0.1 Table 1. A list of parameters used in our calculations and numerical simulations are shown here. The value for γ is chosen from [22]. For kinesin, ℓ = 8nm is observed [37] and θ = 0.1 is the typical value used for the same motor in [24]. Rates w and v are chosen in such a way that the free motor velocity Vf = ℓ(w−v) and the stall force of the free motor f 1 s = ln(v/w)/(βℓ) are equal to 1µms−1 and −5.5pN respectively. 1000 800 600 ) 1 - s m n ( 400 r v a V 200 1000 900 800 700 κ=0.05 pN nm-1 κ=0.1 pN nm-1 4 8 12 N 16 20 N=1 N=2 N=3 N=4 N=5 0 0 0.4 0.8 1.2 κ (pN nm ) -1 1.6 2 Fig. 4. Average cargo velocity Vavr is shown as a function of stiffness at zero external force (f = 0), for various numbers of cargo-bound motors. In inset, Vavr is shown as a function of number of motors N , at two different values of the stiffness. The dashed line is the analytical result obtained from eq.19 and the symbols are computer simulation results. in a very small increase in the velocity at small κ, but the enhancement is negligibly small at large κ and large N . We can see these effects in fig.4 where analytical results (lines) show good agreement with simulations (symbols). Diffusion coefficient versus κ: Because both forward and backward hopping rates decrease with increase in κ, the overall movement itself is hindered. Fluctuations about the mean are suppressed by the increase in stretching en- ergy cost at large stiffness. Therefore, the effective dif- fusion coefficient Deff (see eq.21) of the motor-cargo as- sembly, as predicted by theory, decreases with κ and van- ishes asymptotically. Results for Deff as a function of κ are shown in fig.5(a). Simulation results [symbols in fig.5(a)] also show good agreement with these observations. We see in fig.5(b) that, Deff decreases with N and the depen- dence becomes proportional to 1/N asymptotically, which is confirmed by the slope in the inset. We have seen in fig.4 that the velocity remains constant at larger motor num- bers although it increases initially with N . Therefore, as N increases, multiple motor-mediated transport becomes more deterministic. Mean force experienced by a motor and its load shar- ing features: In fig.6, the average force experienced by each motor hfni = −κℓ∆n is shown as a function of κ. Even 3000 2500 2000 1500 f f e 1000 1 - 2 ) s m n ( D 500 0 0 10000 0.4 ) 1000 1 - 2 s m n ( 100 f f e D 1 N=1 N=2 N=3 N=4 N=5 1.6 2 0.8 κ (pN nm ) -1 1.2 (a) e p o l s -0.6 -0.7 -0.8 -0.9 -1 -1.1 0 10 20 30 κ=0.05 pNnm-1 κ=0.1 pNnm-1 κ=0.5 pNnm-1 10 number of motors N 100 (b) Fig. 5. The effective diffusion coefficient (Deff ) of the cargo as a function of stiffness is shown in (a) for various numbers of cargo-bound motors. Deff decreases with stiffness as well as number of motors on the cargo. In (b), Deff is shown on a logarithmic scale against number of motors in for three dif- ferent κ. The slope of this line, equal to [ln Deff (N + 1) − ln Deff (N )]/[ln(N + 1) − ln(N )], approaches −1 as N increases, indicating that Deff decreases as 1/N asymptotically (see in- set). In both (a) and (b), dashed lines correspond to eq.21 and symbols are simulation results. in the absence of external load on the cargo (f = 0), due to the competition between viscous drag and elas- tic stretching, a motor experiences a net opposing force, whose average is hfni = −γVavr/N . The decrease of hfni as a function of κ is related to the reduction in Vavr with increasing stiffness. With increase in number of motors N , Vavr becomes independent of N (see fig.4 insets) and therefore, the force experienced by each motor decreases as 1/N asymptotically. We have looked at the dependence of hδf i (the average force on the motor due to viscous drag) on the stiffness, in the presence of external force f = −3pN. The ana- lytical expression (eq.22), in comparison with simulation results, is shown in fig.7 insets. Notably, the average force on the motor due to viscous drag in the absence of ex- ternal force, shown in fig.6, is slightly larger than that in the presence of force f = −3pN , as shown in fig.6 insets. 8 Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins N=1 N=2 N=3 N=4 N=5 ) N p ( > f δ < - 0.5 0.4 0.3 0.2 0.1 0 0 1.2 1 0.8 0.6 f=-3 pN 0.3 0.6 0.9 1.2 1.5 6 5 4 3 ) N ( σ f n > f < = r 2.5 2 1.5 1 0.5 0 0 N=1 N=2 f= -3pN 0.3 0.6 0.9 1.2 1.5 ) N p ( ) N ( σ f 2 ) N p ( > f < - n 0.4 0.2 0 0.3 1 0 0 0.3 1.2 1.5 0.6 κ (pN nm ) 0.9 -1 1.2 1.5 0.6 κ (pN nm ) 0.9 -1 Fig. 6. The figure show the mean force experienced by each motor hfni = −κℓ∆ as a function of stiffness at f = 0. Inset shows the force experienced by the motor due to the viscous drag in the medium hδf i = hfni − (f /N ) at force f = −3pN . Symbols correspond to results obtained in computer simula- tions. Fig. 7. The standard deviation σf (N ) (eq.23) of force ex- perienced by a single motor is shown here as a function of stiffness at f = −3pN. Inset shows, the fluctuation to mean ratio r = σf (N )/hfni at f = −3pN. The symbols are results obtained in computer simulations while the line corresponds to analytical result. This is expected because the opposing force reduces the average velocity of the motor-cargo complex, thereby re- ducing the viscous drag force on it. Further, with increase in stiffness, the average velocity of the motor decreases, and hence the force experienced due to the drag force also reduces accordingly. The mean squared fluctuation σf (N ) which quantifies the load sharing properties is shown in fig.7 for N = 1 and 2. From both simulations (symbols) as well as the analytical expression (eq.23, lines), we see that σf (N ) is an increasing function of κ, i.e., when motors are weakly interacting, they share load equally. For typical κ and N , the deviation is not small- almost 1pN for N = 1 and 2pN for N = 2, when κ = 0.5pNnm−1; the numbers be- come 2.5pN for N = 1 and 4pN for N = 2 when the stiffness is 1.5pNnm−1. The inset of fig.7 shows the coeffi- cient of variation r = σf (N )/hfni for N = 1, 2. Note that while r is less than (but comparable to) 1 for N = 1 (for κ <1.5pN/nm), it is typically greater than one for N = 2. Force experienced by an individual motor in an assembly is subject to large relative fluctuations when N is large. The average force experienced by the motor in the presence of elastic motor-cargo coupling has been studied numerically by Kunwar et al.[20] and it has been shown that, the force experienced by the motor decreases with the stiffness, similar to what we have seen in fig.6. More- over, in their work, broadening of the distribution of force experienced by the motor with increases in the stiffness was also observed, which is captured by σf (N ) in fig.7 in our study. Velocity versus f : We studied the force-velocity curve for a cargo driven by a single motor (N = 1) at various stiffnesses, results of which are shown in fig.8. We see that, for the chosen set of parameters in Table 1, the force- velocity curve is convex-up. The comparison with com- 1000 800 600 400 ) 1 - s m n ( r v a 200 V κ=0.001 pNnm-1 κ=0.01 pNnm-1 κ=0.1 pNnm-1 κ=0.5 pNnm-1 κ=1 pNnm-1 0 -200 0 1 3 2 4 -f (pN) 5 6 Fig. 8. The force-velocity curve is shown for a cargo driven by a motor protein (N = 1) at different stiffnesses κ of the motor-cargo linker(dashed line, eq.19). The computer simula- tion results are shown as symbols. puter simulations (symbols) show nice agreement with the theoretical results at small opposing forces (f < 0), while significant deviation is observed at larger opposing forces, particularly close to the regime where velocity becomes zero. Interestingly, in simulations, the velocity appears to cross zero at the same force for all κ values, indicating that the stall-force is independent of κ; a more detailed discussion on the same is given in the next paragraph. Stall force versus κ: We looked at stall-force f N s (κ) as a function of κ for a cargo driven by varying numbers of motor proteins. For fast convergence in simulations, we replaced the constant force f with a harmonic trap force f = −κtx where κt is the trap stiffness, whose value was chosen as 0.5pNnm−1 (we have performed some simula- tions in the constant force ensemble also, and verified that the results are not affected). The results for stall force as a Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins 9 N=1 N=2 N=3 N=4 N=5 38.5 33 27.5 22 ) N p ( N f - 16.5 s 11 5.5 0 0 0.1 0.2 0.3 κ (pN nm ) -1 0.4 0.5 Fig. 9. The results for stall force as a function of stiffness κ is shown here different N . The computer simulation results are shown as symbols. The dashed line corresponds to the analyt- ical expression given in eq.24. function of stiffness is shown in fig.9 for different numbers of cargo bound motors. In numerical simulations (symbols in the figure), the stall force is found to be independent of κ. However, the expression for the stall force given in eq.24 disagree with simulations, except in the limit κ → 0. The underlying reason is presumably the neglect of higher or- der terms in the expansion of the master equation (eq.12), not captured in the first order perturbation approxima- tion. To justify this argument, in Appendix A, we have given the Taylor coefficients (eq.11) of different orders in the presence of the stall force of the motor proteins and show that higher order terms are not negligible when κ is large. It is also pertinent to point out that force-velocity be- haviour of two elastically coupled motor proteins has been studied in [27], where stall force is found to be dependent on the stiffness. However, there are two key differences between our model and that in [27]. (i) In [27], forward hopping rate of the individual motor is linearly dependent on the force (which is derived from linear force-velocity re- lation), while backward hopping is absent. In our model, more general, thermodynamically consistent rates (eq.18) are used. (ii) Detachment of motors from the filament with load-dependent detachment rates is included in [27], while it is ignored on our study. We are presently developing an extended version of our model, which also include motor detachment from the filament. 3 Comparison with experiments In order to check how well our model describes the ob- served features of real motors, we study the force-velocity behaviour of kinesins reported in [37]. In the experiment, the velocity of a motor protein attached to a silica bead was studied by exerting controlled loads using an optical trap, in the presence of different ATP concentrations. The force-velocity behaviour found in Visscher et al. (1999) at 5µM and 2mM ATP concentration is displayed as squares in fig.10(a) and (b) respectively. We managed to reproduce 70 60 50 40 30 20 ) 1 - s m n ( r v a V 1000 κ=0.2 pNnm-1 experiment (Visscher et al., 1999) 5µM ATP ) 1 - s 800 600 κ=0.2 pNnm-1 experiment (Visscher et al., 1999) 2mM ATP m n ( 400 r v a 10 0 0 1 2 3 -f (pN) (a) V 200 4 5 6 0 0 1 2 6 7 8 4 3 5 -f (pN) (b) Fig. 10. In (a) and (b), the force-velocity behaviour predicted from our theory and simulation is compared with the force- velocity behaviour of kinesin observed by Visscher et al. [37] at 5µM and 2 mM ATP concentrations respectively. In both (a) and (b), experimental data (used with the permission of the publishers) are given as squares, analytical results as dashed lines and numerical simulations in circles. these results by tuning a few parameters in our model. The values of ℓ, β and γ are same as that in Table 1. The forward and backward hopping rates of the motor (w and v) are determined in such a way that the sin- gle free motor velocity [Vf = ℓ(w − v)] and stall force [f 1 s = ln(v/w)/βℓ] are consistent with experimental obser- vations in [37]. Specifically, at 5µM ATP concentrations, velocity and stall force of the motor are close to 70nms−1 and −5.5pN respectively, from which we get w ≈ 8.745s−1 and v ≈ 1.96 × 10−4s−1. Similarly, at 2mM ATP concen- tration velocity and stall force are respectively 1000nms−1 and −7pN, using which we get w ≈ 125s−1 and v ≈ 1.52×10−4s−1. The value of θ is reduced slightly (from 0.1 to 0.05) in order to get more accurate behaviour for 2mM ATP concentration case. The force-velocity behaviour ob- tained mathematically (dashed line) and computationally (circles) at stiffness κ = 0.2pNnm−1 is shown in fig.10(a) and (b) at 5µM and 2mM ATP concentrations respec- tively. However, at larger values of κ, the results do not seem to show good agreement with experimental obser- vations. Further, as the external force on the motor ap- proaches the stall force, analytical results show a slight de- viation from computer simulation results, suggesting the relevance of higher order corrections. 4 Summary and Conclusions In this paper, we explored the effects of elastic coupling between a cargo and the attached molecular motors on the statistical properties of transport. In our model, we re- garded motor domains of these proteins as biased random walkers with fixed jump length, while the cargo is sub- jected to thermal noise and an external applied force, in addition to elastic forces from the motors. To capture the elastic energy dependence of motor hopping rates, asym- metric exponential forms given in eq.18 were used. The stochastic hopping dynamics for N motor proteins, along with the over-damped Langevin equation for cargo mo- tion (eq.1) leads to a (N + 1)-variable composite master equation for the dynamics of the motor-cargo assembly. 10 Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins Employing a transformation of variables and first order perturbation expansion in the master equation govern- ing the motor-cargo dynamics, the dynamics of the av- erage quantities was systematically separated from that of fluctuations. The former satisfies a set of determinis- tic "macroscopic" equations, while the latter are governed by a LFPE, which yields equations for the various mo- ments and correlation functions. Both sets of equations were solved numerically to determine quantities of inter- est such as the average velocity and mean force experi- enced by the motor, effective diffusion coefficient of the motor-cargo complex and information on load sharing be- tween the motors. Using our model, we also reproduced the force-velocity behaviour of kinesin observed in an ex- periment [37] by minimal tuning of parameters. We made the following important observations in course of our study. (i) The average velocity and the effective diffusion coefficient of the motor cargo assembly reduces with increase of the elastic coupling constant κ. Our re- sults are consistent with some of the observations made in earlier studies [25,26] where it was reported that, the aver- age cargo velocity reduces as a function of stiffness of the motor-cargo linker. (ii) Asymptotically, the average veloc- ity becomes independent of motor number N while the effective diffusion constant decreases as 1/N . (iii) Even in the absence of external force, all the motors experi- ence a load due to viscous drag of the cytoplasm. The average force experienced by a motor decreases with κ, and this observation is consistent with earlier study by Kunwar et al.[20]. In the presence of opposing external force on the cargo, the average velocity of the cargo de- creases and hence the force due to the viscous drag also reduces accordingly. (iv) When κ is very small, motors are almost non-interacting and the force on the cargo is shared equally among the motors. On the other hand, as κ becomes larger, the deviation from this "mean field" behaviour becomes significant. Kunwar et al.[20] too have reported large fluctuations in the force experienced by in- dividual motor in a team of two motors pulling a cargo. (iv) The stall force is found to be independent of the stiff- ness κ in simulations, but this behaviour is not captured by the analytical results due to the neglect of relevant higher order terms in the Kramers-Moyal expansion. The complete absence of κ-dependence in the stall force is an important observation in our study, something which our analytical treatment failed to reproduce. This has important implications; for instance, this result guar- antees that the stall force measured for a motor using a glass bead as cargo in an in vitro optical trap experiment may be expected to be valid for a more flexible intracellu- lar cargo also. Recent experiments on transport of DNA- scaffold by motors having controlled separation between them, have opened possibility of studying directly motor- motor interaction between like motors during transport [12]. In such experiments, it may be possible to verify some of our predictions. We hope that the formalism developed here and its possible future extensions will be found use- ful in quantitative modelling of cargo transport involving multiple motor proteins. The authors would like to thank P.G. Senapathy Centre for Computing Resources, IIT Madras for computational support. DB thanks Udo Seifert for stimulating discussions during the Non-equilibrium Statistical Physics Workshop (2015) held at ICTS, Bengaluru. DB also acknowledges Sumesh Thampi, Raghu- nath Chelakkot and Amitabha Nandi for useful conversations. A On the relevance of higher order terms in the Kramers-Moyal expansion nkl... and β(r) The Taylor coefficients α(r) nkl.. given in eq.11 are derivatives of Vn and Dn respectively, evaluated at x = x. As we see from eq.8, V0 is function of all the variables (x0, x1...xN ), but depends only linearly on them. So corresponding coeffi- cients of order larger than one in the Taylor expansion are zero identically, i.e. α(r) 0kl... = 0 for r ≥ 2. D0 = D/ℓ2 is a constant, so all β(r) 0kl... for r ≥ 1 are zero. However, from eqs.7 and 18, we can see that the higher order coefficients in Taylor expansion of Vn and Dn are non-zero for n ≥ 1. But, Vn and Dn are functions of only xn and x0 through ∆n = xn − x0. Therefore, those terms involving cross derivatives i.e. α(r) nkl.. with k, l.. 6= n and k, l.. 6= 0 are zero [only for k, l.... = n, 0, α(r) nkl... and β(r) nkl.. survive]. Further, the coefficients with k = n corresponding to derivative with respect to xn and those with k = 0 correspond- ing to derivative with respect to x0 (keeping rest of the suffices same), differ only by sign. Therefore, it is enough to study the coefficients corresponding to k = l = .. = n i.e. α(r) nnn... and β(r) nnn.. for n ≥ 1. From eqs.7, 11 and 18, these are given by nkl... and β(r) α(r) β(r) nnn.. = (cid:0)βκnℓ2(cid:1)r nnn.. = (cid:0)βκnℓ2(cid:1)r r! r! (cid:2)(−θn)rW + (cid:2)(−θn)rW + n (∆n)(cid:3) , n (∆n)(cid:3) .(25) n (∆n) − (1 − θn)rW − n (∆n) + (1 − θn)rW − nnn.. = α(r) Further, for identical motors, ∆n = ∆, θn = θ, wn = w, 111.. and β(r) vn = v and therefore, at fixed N , α(r) nnn.. = β(r) 111.. for all N ≥ n ≥ 2. In fig.11, Taylor coefficients α(r) 111.. and β(r) 111.. shown (up to r = 5) for N = 1 at different κ and θ values. For the chosen set of parameters given in Table 1, when θ = 0.1 the terms corresponding to r = 1 are predominant in the range of stiffness we considered in this study (0 − 2pNnm−1). There- fore, the formalism developed here shows overall agreement with computer simulations. On the other hand, at θ = 0.5 and θ = 0.9, higher order terms become important and so the for- malism developed here is not appropriate at these θ values. But, at very small κ values such that βκℓ2 ≪ 1, we see from eq.25 that the higher order Taylor coefficients are small, and so the formalism is valid in this regime. In fig.12 and fig.13, we have shown results for Vavr and Deff as a function of κ obtained in computer simulations at θ = 0.5 and θ = 0.9. The variation in the velocity and diffusion coefficient as a function of stiffness κ is qualitatively similar to that of θ = 0.1. However, now a large deviation from the analytical results is seen, which highlights the inaccuracy of our approximation for these parameter values. For the set of parameters in Table 1, the Taylor coefficients 11.. and β(r) α(r) 11.. evaluated at f = −5.5pN are also shown for N = 1 in the fig.14. We saw in fig.11 (for θ = 0.1 column) that, when f = 0, only the first order coefficients are important within the range of stiffness considered. However at f = −5.5pN, all the Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins 11 θ=0.1 θ=0.5 θ=0.9 ) 1 - s ( ) r ( . . 1 1 α 400 200 0 -200 -400 -600 ) 1 - s ( ) r ( . . 1 1 β 400 200 0 -200 -400 -600 0 r=1 r=2 r=3 r=4 r=5 0.5 1 f f e D 1.5 2 0 0.5 1 1.5 2 2 0 1.5 1 κ (pNnm-1) 0.5 800 400 0 0 0.4 2000 1600 1200 N=1 N=2 N=3 N=4 N=5 0.8 κ (pNnm-1) 1.2 1.6 2 (a) N=1 N=2 N=3 N=4 N=5 1 - 2 ) s m n ( 2000 1600 1200 1 - 2 ) s m n ( f f e D 800 400 0 0 0.4 0.8 1.2 κ (pNnm-1) 1.6 2 (b) Fig. 13. Effective diffusion coefficient of the motor-cargo as- sembly is shown as a function of stiffness here for θ = 0.5 in (a) and θ = 0.9 in (b). Here, rest of the parameters are cho- sen from Table 1. The deviation of analytical results (lines) from simulation (symbols) is significant here due to prominent higher order corrections. . . 1 1 ) r ( α 100 0 -100 -200 -300 0 . . 1 1 ) r ( β r=1 r=2 r=3 r=4 r=5 1 0.5 1.5 κ (pNnm-1) 2 (a) 300 200 100 0 0 2 1 0.5 1.5 κ (pNnm-1) (b) Fig. 14. The coefficients α(r) 11.. given in eq.25 are shown as function of κ at f = −5.5pN for N = 1. Rest of the parameters are chose from Table 1. 11.. and β(r) as: F[Π(η)] ≡ Φ(g) =Z exp[ig · η]Π(η)dη0dη1.. Using properties of the Fourier transform, 11.. and β(r) Fig. 11. Taylor coefficients α(r) 11.. given in eq.25 are shown here (up to r = 5) as function of κ at θ = 0.1, θ = 0.5 and θ = 0.9. For θ = 0.1 higher order terms are much smaller compare to first order terms. On the other hand for θ = 0.5 and θ = 0.9, higher order terms are significant. 1000 ) 1 - s m n ( 800 600 400 r v a V 200 0 0 0.4 0.8 1.2 κ (pN nm-1) 1000 (a) 800 600 400 r v a ) 1 - s m n ( V 200 0 0 0.4 0.8 1.2 κ (pN nm-1) N=1 N=2 N=3 N=4 N=5 1.6 2 N=1 N=2 N=3 N=4 N=5 1.6 2 (b) Fig. 12. Average velocity of the cargo as a function of stiffness is shown here for θ = 0.5 in (a) and θ = 0.9 in (b). All the other parameters are chosen from Table 1. The deviation of analyt- ical results (lines) from simulation (symbols) is significant in these cases, compared to that in θ = 0.1 case (fig.4). higher order terms are large compared to first order terms. This results in the deviation of analytical results for stall force from the computer simulations in fig.9. B Evaluation of effective diffusion coefficient Π(η) in eq.12 is function of fluctuations η ≡ {η0, η1, ...ηN }. We define the Fourier transform of Π with respect the variables η F(cid:20) ∂mΠ(η) ∂ηm n (cid:21) = (−ign)mΦ(g) ; F [ηnΠ(η)] = ∂Φ(g) ∂(ign) eq.12 may be rewritten as follows: 12 Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins ∂Φ ∂t = N (−ign)"(cid:0) xn − V n(cid:1) Φ −Xk Xn=0 − 1 2 Some useful properties of Φ here are: α(1) nk ∂Φ ∂(igk)# Xn=0 Dng2 N hηpi = and hηpηqi = ∂Φ ∂(igp)(cid:12)(cid:12)(cid:12)(cid:12)g=0 ∂2Φ ∂(igp)∂(igq)(cid:12)(cid:12)(cid:12)(cid:12)g=0 From eq.26 and the definition of hηpi in eq.27, we arrive at: d dt hηpi = N Xn=0 α(1) pn hηni − ( xp − Vp) (28) As {ηp} are defined as fluctuations about averages xp, we require that hηpi = 0, which is realized by putting the non- homogeneous term in eq.28 to zero, thereby arriving at eq.14. Further, using the condition hηpi = 0, from eq.26 and the definition of hηpηqi in eq.27, we obtain: N dhηpηqi dt = Xn=0hDnδpnδqn + α(1) pn hηqηni + α(1) qn hηpηnii . (29) For N = 1, second moments and correlation functions are found explicitly as follows: hη2 0 is = D0s2 + D0(3α + 2 κ s2(s + α + 2 κ γ )2 γ )s + 2α2D0 + 4D1( κ γ )(s + 2α + 4 κ γ ) γ + α)s2 γ )(s + 2α + 4 κ γ ) D1s3 + D1 2 (2 κ s2(s + 2α)(s + α + 2 κ + nΦ (26) hη2 1 is = hη2 2 is = 2[α2(D0 + D1) + 4( κ γ )2D1 + 3α( κ s2(s + 2α)(s + α + 2 κ (27) hη0η1is = hη0η2is = h(αD0 + D1 s2(s + α + 2 κ γ )2 γ )D1]s + 4α3D0 + 8αD1( κ γ )(s + 2α + 4 κ γ ) γ )s + 2α2D0 + 4( κ γ )(s + 2α + 4 κ γ ) γ )2D1i κ hη1η2is = 2αh(αD0 + D1 κ γ )s + 2α2D0 + 4( κ γ )2D1i . (33) s2(s + 2α)(s + α + 2 κ γ )(s + 2α + 4 κ γ ) Similar to the previous case, it is easily shown that hη2 0 i = hη2 1 i = hη2 2 i ≈ 2(Deff /ℓ2)t in the large t limit where Deff (2) = ℓ2" α2D0 + 2( κ 2(α + 2 κ γ )2D1 γ )2 # . (34) For both N = 1 and N = 2, the variance shows diffusive be- haviour in the long-time limit. For a general N motor system, we therefore expect hη2 i i ∼ 2(Deff /ℓ2)t, where Deff is the effec- tive diffusion coefficient of the motor-cargo assembly. Based on a simple extrapolation of the analytical results for N = 1 and N = 2, we conjecture eq.21 as the effective diffusion coefficient for arbitrary N . d dt d dt κ hη2 hη2 0 i = D0 + γ (cid:2)hη0η1i − hη2 0 i(cid:3) 1 i = D1 + α(cid:2)hη0η1i − hη2 1 i(cid:3) γ(cid:19) hη0η1i 1 i −(cid:18)α + hη2 κ γ 0 i + κ hη0η1i = αhη2 d dt References 1. J. Howard, Mechanics of Motor Proteins and Cytoskeleton, (Sinauer Press, Sunderland, MA, 2001). 2. A. B. Kolomeisky, Motor Proteins and Molecular Motors, (30) (CRC Press, Boca Raton, FL, 2015). To solve the above equations, we define the Laplace transforms 0 exp(−st)hηpηqidt. eq.30 is now expressed in ma- trix form and solved for the moments in Laplace space. For N = 1, the complete solution is hηpηqis =R ∞ hη2 s2(s + α + κ 0 is = hD0s2 + D0(3α + κ 1 is = hD1s2 + D1(α + 3 κ hη0η1is = h(D0α + D1 γ )s + 2α2D0 + 2( κ γ )(s + 2α + 2 κ γ ) γ )s + 2α2D0 + 2( κ γ )(s + 2α + 2 κ γ ) γ )s + 2α2D0 + 2( κ γ )(s + 2α + 2 κ γ ) s2(s + α + κ s2(s + α + κ hη2 γ )2D1i γ )2D1i γ )2D1i κ . (31) 1 is ≈ 2(Deff /ℓ2s2), and We observe that as s → 0, hη2 hence hη2 1 i ≈ 2(Deff /ℓ2)t where Deff is the effective one- dimensional diffusion coefficient of the cargo on the filament, given by 0 is = hη2 0 i = hη2 Deff (1) = ℓ2" α2D0 + ( κ 2(α + κ γ )2D1 γ )2 # . (32) 3. D. Chowdhury, Phys. Rep. 529, (2013) 1. 4. A. Desai and T. J. Mitchison, Ann. Rev. Cell. Dev. Biol. 13,(1997) 83. 5. S. M. Block and L. S. B. Goldstein and B. J. Schnapp, Nature 348 (1990),348. 6. S. P. Gross, Curr. Biol. 17, (2007) R478. 7. S. P. Gross, Phys. Biol. 1, (2004) R1. 8. M. A. Welte, Curr. Biol. 14, (2004) R525. 9. V. Soppina, A. K. Rai, A. J. Ramaiya, P. Barak and R. Mallik, Proc. Natl. Acad. Sci. USA 106, (2009) 19381. 10. A. G. Hendricks, E. Perlson, J. L. Ross, H. W. Schroeder III, M. Takito and E. L. F. Holzbaur, Curr. Biol. 20, (2010) 697. 11. A. R. Rogers, J. W. Driver, P. E. Constantinou, D. K. Jamisonb and M. R. Diehl, Phys. Chem. Chem. Phys. 11, (2009) 4882. 12. K. Furuta, A. Furuta, Y. Y. Toyoshima, M. Amino, K. Oiwa and H. Kojima, Proc. Natl. Acad. Sci. USA 110, (2013) 501. 13. C. M. Coppin, J. T. Finer, J. A. Spudich and R. D. Vale, Proc. Natl. Acad. Sci. USA 93, (1996) 1913. 14. C. M. Coppin, D. W. Pierce, L. O. Hsu and R. D. Vale, Proc. Natl. Acad. Sci. USA 94, (1997) 8539. 15. S. Klumpp and R. Lipowsky, Proc. Natl. Acad. Sci. USA For N = 2, a similar analysis gives the moment-transforms 102, (2005) 17284. Deepak Bhat, Manoj Gopalakrishnan: Transport of organelles by elastically coupled motor proteins 13 16. M. J. Muller, S. Klumpp and R. Lipowsky, Proc. Natl. Acad. Sci. USA 105, (2008) 4609. 17. M. J. I. Muller and S. Klumpp and R. Lipowsky, Biophys. J. 98, (2010) 2610. 18. M. J. I. Muller and S. Klumpp and R. Lipowsky, J. Stat. Phys. 133, (2008) 1059. 19. A. Kunwar, S. K. Tripathy, J. Xu, M. K. Mattson, P. Anand, R. Sigua, M. Vershinin, R. J. McKenney, C. C. Yu, A. Mogilner and S. P. Gross, Proc. Natl. Acad. Sci. USA 108, (2011) 18960. 20. A. Kunwar, M. Vershinin, J Xu and S P Gross, Curr. Biol. 18, (2008) 1173. 21. D. Materassi, S. Roychowdhury, T. Hays and M. Salapaka, BMC Biophys. 6, (2013) 14. 22. S. Bouzat and F. Falo, Phys. Biol. 7, (2010) 046009. 23. S. Bouzat and F. Falo, Phys. Biol. 8, (2011) 066010. 24. J. W. Driver, A. R. Rogers, D. K. Jaminson, R. K. Das, A. B. Kolomeisky and M. R. Diehl, Phys. Chem. Chem. Phys. 12, (2012) 10398. 25. F. Berger, C. Keller, S. Klumpp, and R. Lipowsky, Phys. Rev. Lett. 108, (2012) 208101. 26. F. Berger, C. Keller, R. Lipowsky and S. Klumpp, Cell. Molec. Bioeng. 6, (2013) 48. 27. F. Berger, C. Keller, S. Klumpp and R. Lipowsky, Phys. Rev. E 91, (2015) 022701. 28. E. Zimmermann and U. Seifert, Phys. Rev. E 91, (2015) 022709. 29. S. Bouzat, Phys. Rev. E 93, (2016) 012401. 30. R. Mallik, B. C. Carter, S. A. Lex, S. J. King and S. P. Gross, Nature 427, (2004) 649. 31. A. K. Rai and A. Rai and A. J. Ramaiya and R. Jha and R. Mallik, Cell 152, (2013) 1. 32. N. G. van Kampen, Stochastic Processes in Physics and Chemistry (Elsevier, Amsterdam, 2007). 33. M. E. Fisher and A. B. Kolomeisky, Proc. Natl. Acad. Sci. USA 96, (1999) 6597. 34. T. Schmiedl and U. Seifert, Eur. Phys. Lett. 83, (2008) 30005. 35. E. B. Stukalin, H. Phillips III and A. B. Kolomeisky, Phys. Rev. Let. 94, (2005) 238101. 36. E. B. Stukalin and A. B. Kolomeisky, Phys. Rev. E 73, (2006) 031922. 37. K. Visscher, M. J. Schnitzer and S. M. Block, Nature 400, (1999) 184.
1805.04971	3	1805	2018-12-29T01:33:06	Propagating wave in the flock of self-propelled particles	[ "physics.bio-ph", "physics.flu-dyn" ]	We investigate the linearized hydrodynamic equations of interacting self-propelled particles in two dimensional space. It is found that the small perturbations of density and polarization fields satisfy the hyperbolic partial differential equations---that admit analytical propagating wave solutions. These solutions uncover the questionable traveling band formation in the flocking state of self-propelled particles. Below the critical noise strength, an unstable disordered state (random motion) undergoes a transient vortex and evolves to an ordered state (flocking motion) as unidirectional traveling waves. There appear two possible longitudinal wave patterns depending on the noise strength, including single band in stable state and multiplebands in unstable state. A comparison of theoretical and experimental studies is presented.	physics.bio-ph	physics	Propagating wave in the flock of self-propelled particles Waipot Ngamsaad1, ∗ and Suthep Suantai2 1Division of Physics, School of Science, University of Phayao, Phayao 56000, Thailand 2Department of Mathematics, Faculty of Science, Chiang Mai University, Chiang Mai 50200, Thailand (Dated: January 1, 2019) We investigate the linearized hydrodynamic equations of interacting self-propelled particles in two dimensional space. It is found that the small perturbations of density and polarization fields satisfy the hyperbolic partial differential equations -- that admit analytical propagating wave solutions. These solutions uncover the questionable traveling band formation in the flocking state of self- propelled particles. Below the critical noise strength, an unstable disordered state (random motion) undergoes a transient vortex and evolves to an ordered state (flocking motion) as unidirectional traveling waves. There appear two possible longitudinal wave patterns depending on the noise strength, including single band in stable state and multiplebands in unstable state. A comparison of theoretical and experimental studies is presented. PACS numbers: 02.30.Jr, 47.35.-i, 87.10.Ed I. INTRODUCTION The onset of collective motion can be found in vari- ous systems of the self-propelled objects, ranging from macromolecules, microorganism, animal, human, and swarming robots (see Ref. [1] and references therein). The physics aspect of this phenomenon has been a cur- rent active research topic. The minimal paradigm that can be used to describe this dynamics successfully is acknowledged to the Vic- sek model [2]. In this model, the point-like self-propelled particles (SPP) move at constant velocity in the direction of their orientation unit vector [2]. The particles inter- act locally by trying to align their directions of motion in presence of noise. If the noise strength is greater than the critical value, the SPP move in random direction. If the noise strength is below the critical value, the particles transit from random motion (disordered state) to flocking (ordered state), where they form the coherence clusters so that the individual members tend to move together in the same direction. The Vicsek model can be viewed as the flying XY spin model where the phenomenological hy- drodynamic equations have been proposed for description at a continuum level by Toner and Tu [3, 4]. Recently, the hydrodynamic equations of SPP were derived from specified individual-based dynamics using several coarse- gaining frameworks, such as the Smoluchowski equation [5 -- 7] and the Boltzmann equation [7 -- 10]. Based on hydrodynamic theory, the long wavelength mode fluctuations of density and velocity fields propa- gating as sound waves in SPP had been predicted by Tu and Toner [11]. Later, the moving bands of the ordered state in the disordered state background were obviously found in the large-scale simulations of SPP by different researchers [9, 12 -- 14]. Two distinct robust pattern forms ∗ [email protected] of the traveling waves in SPP including solitary moving band [9, 15, 16] and moving multistripes [17] have been explored. Apart from wave patterns, fluctuating flock- ing states [17] and stationary radially symmetric asters [15] in SPP have also been presented. In the experi- ments, the existence of traveling bands in active matter has been demonstrated for the system of actin filaments [18] and colloidal rollers [19]. Recently, the linear sound wave in a system of colloidal rollers has been explored experimentally [20]. Analytical work has been carried out in order to gain deep insight into wave propagation dynamics in SPP. The standard method is the linear stability analysis of the hydrodynamic equations [9, 15, 17]. Several authors agree that the traveling waves in SPP owe their emer- gence to instability of the homogeneous states [9, 15, 17]. However, the linear stability analysis provides only the dispersion relation [9] that is inadequate to characterize the spatiotemporal wave patterns. In a more rigorous study, the propagative Ansatz, in which the wave pro- file travels along a direction with constant speed, has been postulated to be a solution for hydrodynamic equa- tions of SPP in one-dimensional space (1D) [9, 17, 21, 22]. This approach reduces the hydrodynamic equations from nonlinear partial differential equations (PDEs) to nonlin- ear ordinary differential equations (ODEs). The nonlin- ear ODEs can be recast further into equivalent Newton's equation of motion for single particle moving in a poten- tial field in the presence of friction by using a dynam- ical framework [21, 22]. This approach seems likely to classify the three different types of propagating patterns in SPP successfully, including solitary wave, multistripes wave and polar-liquid droplet. Nonetheless, solving the exact wave profiles explicitly by using this framework is challenging due to its nonlinearity. As classified in the textbook of Whitham [23], there are two classes of wave solutions for the linear or non- linear PDEs which consist of hyperbolic wave and dis- persive wave solutions. The difference is that the hyper- bolic wave propagates in two opposite directions along an arbitrary axis, a case in which the speeds need not be equal [24 -- 26]. Obviously, the previous analyses rely on the dispersive wave solution that propagates in one direction [9, 15, 17, 21, 22]. In contrast to this, we de- duced that the linearized hydrodynamic equations of SPP can formulate the hyperbolic-type PDEs. Therefore, the previous propagating wave assumption, which belongs to the dispersive wave solution [9, 15, 17, 21, 22], is still an incomplete wave feature of SPP. In this work, we investigate the linearized hydrody- namic equations of SPP, which can be combined into the linear wave PDEs [24 -- 26]. Instead of performing the con- ventional mode analysis [9, 15, 17], we solve for the exact space and time dependent solutions of these equations by using the Riemann method [24 -- 26]. These linear analyt- ical solutions are capable of capturing the dynamics of SPP in the vicinity of early and final states of the sys- tem. Especially, they can be used to classify the wave pattern formation in the flocking state of SPP clearly. 2 where l is the interaction range [27]. Although differ- ences in physical parameters, Eq. (1) and Eq. (2) have a form identical to the phenomenological model proposed by Toner and Tu [3, 4] and the coarse-grained equations obtained by using the Boltzmann theory [9]. The homogeneous states of Eq. (1) and Eq. (2) admit arbitrary constant density ρ0 with two possible values of polarization W0, given by ε ≥ ε0 ε < ε0 (3) W 0 = W0 =(0, p8ε (ε0 − ε)/γ, where ε0 = γρ0/2, which is defined as the critical noise strength value. Above the critical point (ε > ε0), the system is in disordered state with zero polarization where the SPP move in random direction. Below the critical point (ε < ε0), the system transitions into an ordered state where the SPP tend to move together in the same direction with nonzero polarization, called flocking. II. HYDRODYNAMIC EQUATIONS In this study, we consider a particular variant Vic- sek model that has been studied by Farrell et al. [27]. The advantage of this variant model is that it can explic- itly map the microscopic physical parameters to the hy- drodynamic equations through a coarse-grained process. Adapted from Ref. [27], the hydrodynamic equations that describe evolution of particle number density field ρ(r, t) and polarization field W (r, t) in two-dimensional space (2D), are given by III. LINEARIZED WAVE EQUATIONS Now we study the dynamics of SPP in the vicinity of homogeneous states. We suppose that the homoge- neous polarization aligns in x-direction. Thus, we de- fine the solutions as follows: ρ(r, t) = ρ0 + n(r, t) and W (r, t) = W0 x + u(r, t), where n(r, t) and u(r, t) are small perturbations in density and polarization fields, re- spectively, called perturbations for convenience. Substi- tuting these solutions into Eq. (1) and Eq. (2) by first retaining the first-order terms and then obtaining the linearized hydrodynamic equations of SPP, 8ǫ W2(cid:19) W 2 v0 γ2 ρ − ε − ρt = −v0∇ · W , 2 ∇ρ +(cid:18) γ W t = − v0 (W · ∇) W − v2 0 v0∇(cid:0)W2(cid:1) + 16ε∇2W , 3γ 16ε 5γ 32ε 5γ 16ε − + v0W (∇ · W ) (1) (2) where v0 is the moving speed of the particle, ǫ describes noise strength and γ describes the strength of align- ment. The polarization field is associated with the parti- cle velocity field V (r, t) in such a way that v0W (r, t) = ρ(r, t)V (r, t). These equations are coarse-gained dy- namics of an N point-like SPP system in which parti- cles move at constant speed v0 in the direction of their orientation unit vector and interact with the vicinity or neighborhood via noisy alignment rule [2]. The position ri(t) and the orientation angle θi(t) of the ith particle at time t evolve with the following equations of motion: ri = v0pi and θi = Pj6=i F (θi − θj, ri − rj) + √2εηi(t), where the unit vector pi(θi) = cos θix + sin θiy and ηi(t) is a white noise with zero mean and unit vari- ance. The local pairwise alignment interaction is given by F (θ, r) = γ sin(θ)/(πl2), if r ≤ l (otherwise F = 0), (4) nt = −v0∇ · u, ut = − γ γ 2 (5) W0h + v2 0 16ε∇2u, v0 2 ∇n + α0u + where α0 = (ε0 − ε − γ 2 0 ) and h = nx − 2ε W0 (x · u) x− v0 8ε [3 (x · ∇) u− 5x (∇ · u) + 5∇ (x · u)]. The vector field h tends to drive the polarization field to the mean direction and affects only the flocking state (W0 6= 0). Operating Eq. (4) with ∂t and using Eq. (5) (similarly, operating Eq. (5) with ∂t and using Eq. (4)), we arrive at the following equations. 8ε W 2 ntt − α0nt = c2∇2n − utt − α0ut = c2∇2u + c2∇ × (∇ × u) + W0v0∇ · h + O(κ∇2nt), (6) γ 2 γ 2 W0ht (7) +O(κ∇2ut), 0 and κ = v2 where c = v0√2 16ε . Noting that the third-order derivative terms in Eq. (6) and Eq. (7) can be ignored as we are interested only in the evolution of large flocking cluster or long wavelength (λ) mode, λ ≫ πv0 2√εα0 . Ob- viously, Eq. (6) and Eq. (7) belong to the wave equations or hyperbolic PDEs [24 -- 26]. A. Perturbation of disordered state From Eq. (6) and Eq. (7), the perturbations of the disordered state (W0 = 0) satisfy the telegraph equations [24 -- 26] ntt − α0nt = c2∇2n, utt − α0ut = c2∇2u + c2∇ × (∇ × u) . (8) (9) Specially, Eq. (9) reveals a vortex, defined by ω = ∇× u. The governing equation for the perturbed vorticity was obtained by taking the curl operator (∇×) to Eq. (4) with W0 = 0, 1 (10) ωt = (ε0 − ε) ω + κ∇2ω. 2 (ε0−ε)tn and By using the following solutions n = e ω = e(ε0−ε)t ω, we found that n satisfies a simple lin- 4 (ε0 − ε)2 n, which ear wave equation ntt = c2∇2n + 1 looks similar to the Klein-Gordon equation, while ω sat- isfies the diffusion equation ωt = κ∇2 ω [25]. So that, c is interpreted as the speed of sound in disordered phase that has the magnitude of about 0.707 of the individual particle velocity v0 [16]. And, κ is diffusion constant. It implies that the disordered state is unstable below the critical point (ε < ε0) and it evolves to the ordered state to form the flocking state. The transient vortex can be observed in the early stages of simulations of the Vicsek- type model [28]. However, the vortex in the experimental system such as colloidal rollers seems robust due to the effect of additional repulsive interaction and confinement [29], which have been excluded in our investigation. 3 Eq. (7), the wave equations for the ordered state in this case are given below. ntt = c2nyy + O(κntxx), vtt = c2vyy + O(κvtxx). (13) (14) w is assumed to relax to zero in the ordered state. By neglecting the third-order derivative term, Eq. (13) and Eq. (14) are the plane wave equations with the well- known d'Alembert solution -- where the initial condition splits into two waves that propagate in opposite direc- tions along the y-axis with the speed of sound c [24 -- 26]. Since the perturbations do not change shape from the ini- tial conditions for this sort of wave, we ignore this mode in the present study. In addition, the speed of sound in the experimental system of colloidal rollers is direction- dependent [20]. It results from the hydrodynamic and electrostatic interactions in the system of colloidal rollers [19, 20] that have been excluded in our model. We are now looking for the analytical space and time dependent solution of longitudinal waves. By dropping the third-order derivative term, we rewrite Eq. (11) or Eq. (12) φtt + 2νφtx − c2φxx + αφt + βφx = 0, (15) where φ can refer to either n or w, since all the equations are in identical form. The initial conditions for Eq. (15) are given by φ(x, 0) = f (x), Dtφ(x, 0) = φt(x, 0) + νφx(x, 0) ≡ g(x). Eq. (15) is a second-order PDE whose dt(cid:1)2 characteristic equation is given by(cid:0) dx dt(cid:1)− c2 = dt = ν ± √ν2 + c2 [24 -- 26]. From the characteristic 0 or dx equation, obviously, Eq. (15) is a hyperbolic-type PDE and it can be reduced to a canonical form by introducing the curvilinear coordinates given below. − 2ν(cid:0) dx B. Perturbation of ordered state η = x + c−t, ξ = x − c+t, (16) As shown in Eq. (6) and Eq. (7), the perturbations around the flocking or the ordered state (W0 6= 0) tend to be biased to the mean direction by vector field h. It is observed, at least in simulations, that the moving bands are unidirectional waves [9, 12 -- 14, 21, 22]. We write such a symmetry-broken field as u(r, t) = w(r, t)x + v(r, t)y, where w and v are x- and y-component of the small per- turbed polarization field, respectively. Now, we consider the longitudinal mode, where the wave profiles propa- gate in the same direction as that of mean polarization (ny = wy = vy = 0). From Eq. (6) and Eq. (7), the wave equations in this case are provided by ntt + αnt = c2nxx − 2νntx − βnx + O(κntxx), (11) wtt + αwt = c2wxx − 2νwtx − βwx + O(κwtxx), (12) where α = 2 (ε0 − ε), β = γ 32ε W0v0. Noting that v is decoupled and tends to eventually decay to small value by a bias-diffusion process. 2 W0v0, and ν = 3γ Next, we consider the transverse mode where the wave profiles propagate perpendicular to the direction of mean polarization (nx = wx = vx = 0). From Eq. (6) and where c± = √ν2 + c2 ± ν, so that ν is exactly the collec- tive speed of SPP induced by the alignment interaction. Applying the transformations in Eq. (16), we rewrite Eq. (15) in ηξ-plane 1 4Λ2 φηξ + k−φη + k+φξ = 0, , k+ = αc+−β 4Λ2 (17) , and Λ = √ν2 + c2 = where k− = − αc−+β 2 (c− + c+). Now the solutions of Eq. (17) depend on the two wave variables, φ(x, t) = φ(η, ξ). According to ν > 0 in the ordered state, the wave speeds c± are always positive so that η and ξ are left- and right-propagating wave variables, respectively. In the presence of collective motion, the wave speeds in the flocking state are larger than the speed of sound in the disordered phase. This supports the supersonic wave structure as pointed out by Ihle [16]. IV. RIEMANN METHOD Finding the solution of Eq. (17) subjected to the initial data is called a Cauchy problem, which can be solved by using the Riemann method [24 -- 26]. This approach can solve the general form of linear hyperbolic PDE in 1D, but case study in the presence of ν term is scarce [24 -- 26]. Therefore, we provide the procedure for solving Eq. (17) as follows. For convenience in further calculation, we rewrite Eq. (17) L[φ] = φηξ + k−φη + k+φξ = 0, (18) c−+c+ and x = c+η+c−ξ where L is linear operator. From Eq. (16), we have that t = η−ξ c−+c+ . When t = 0, we have η = ξ = x, which is the straight line in the ηξ-plane. Therefore, the initial conditions in ηξ-plane (Cauchy data) are transformed to φη=ξ = f (ξ), (19) (20) 1 Λ M[φ]η=ξ = (φt(ξ) + νφx(ξ)) ≡ 1 Λ where we define operator M[∗] = ∂η ∗ −∂ξ∗. The starting point of the Riemann method is to find a smooth function R(η, ξ), called the Riemann function, that satisfies the adjoin equation g(ξ), L∗[R] = Rηξ − k−Rη − k+Rξ = 0, (21) where L∗ is adjoin operator [24 -- 26]. This approach can reduce the second-order PDE to the first-order integral 4 equation. From Eq. (18) and Eq. (21) it evaluates that RL[φ] − φL∗[R] = Pη + Qξ = 0, (22) where P = 1 2 (Rφξ − φRξ) + k−Rφ and Q = 1 2 (Rφη − φRη) + k+Rφ. By using Green's theorem [24 -- 26, 30] in Eq. (22), we have RRD (Pη + Qξ) dηdξ = HC (P dξ − Qdη) = 0, where D is the region bounded by the positively oriented closed curve C. We integrate along the three edges of a triangle in ηξ-plane whose ver- tices with positive orientation are given by C0 = (η0, ξ0), C1 = (η0, η0) and C2 = (ξ0, ξ0). In this way, we choose a path that dη = 0 along C0-C1 line, dξ = 0 along C2-C0 line and dη = dξ along C1-C2 line, containing the initial data in Eq. (19) and Eq. (20). So that, the Riemann method turns our problem to a line integral equation Z C1 C0 P dξ +Z C2 C1 (P − Q)η=ξdξ −Z C0 C2 Qdη = 0. (23) To calculate the integral terms in Eq. (23), the Riemann function must satisfy following conditions: Rξ− k−R = 0 when ξ = ξ0, Rη−k+R = 0 when η = η0 and R = 1 when η = η0 and ξ = ξ0. After evaluating Eq. (23) with the properties of Riemann function and initial data Eq. (19) and Eq. (20), we have φ(η0, ξ0) = 1 2 [R(η0, η0)f (η0) + R(ξ0, ξ0)f (ξ0)] − 1 2Z ξ0 η0 {R(ξ, ξ)M[φ]η=ξ − M[R]η=ξf (ξ) + 2ΓR(ξ, ξ)f (ξ)} dξ, (24) where Γ = k+− k−. Eq. (24) is analytical solution of our main problem in ηξ-plan. The remain ingredient is the exact form of the Riemann function. A. Riemann function To find the Riemann function, we define R(η, ξ) = exp [k+ (η − η0) + k− (ξ − ξ0)]Ψ(η, ξ). Substituting it to Eq. (21), we obtain Ψηξ − k2 4 Ψ = 0, where k2 = 4k−k+. We assume that this equation has a particular solution in the form Ψ(q) where q = (η − η0) (ξ − ξ0) [24 -- 26]. Using this assumption, we have q Ψ(q)+ Ψ(q)− k2 4 Ψ(q) = 0. This equation can be transformed further with another new variable θ = k√q and then we have Ψ(θ)+ 1 Ψ(θ)−Ψ(θ) = 0. Finally, it found that Ψ(θ) exactly satisfies the zeroth- order modified Bessel equation whose solution has been known. After gathering all terms, the Riemann function is provided by θ R(η, ξ) = A(η, ξ)I0(cid:16)kp(η − η0) (ξ − ξ0)(cid:17) , (25) where A(η, ξ) = exp [k+ (η − η0) + k− (ξ − ξ0)] and I0 is the zeroth-order modified Bessel function [30]. It can calculate that Rη = k+R + k2 and Rξ = k−R + k2 , where I1(θ) = I0(θ) which is first-order modified Bessel function [30]. And it shows that this Riemann function satisfies all required conditions. 2 (η − η0) A(η, ξ) I1(θ) 2 (ξ − ξ0) A(η, ξ) I1(θ) θ θ As shown later, we transform the solution in Eq. (24) back to xt-plane by using the Riemann function Eq. (25) subjected to the initial data Eq. (19) and Eq. (20). Since ξ becomes dummy variable now, we let η0 = x + c−t and ξ0 = x − c+t. B. Stability analysis We now find stability of solution by considering the exponential the obtained analyti- factor cal line that A(ξ, ξ) = in Eq. exp [k+ (ξ − η0) + k− (ξ − ξ0)]. For ε < ε0, it found that k− < 0, due to c±, α and β are always positive, while k+ can be either negative or positive. Solving (25) along η = ξ the inequality, we found that k+ < 0 if ε > 7 11 ε0 and k+ > 0 if ε < 7 11 ε0. For ξ0 ≤ ξ ≤ η0 or equivalent to −c−t ≤ x−ξ ≤ c+t, therefore A(ξ, ξ) always decays when 0 < ε < 7 11 ε0 (stable regime). In contrast, A(ξ, ξ) can grow when 7 11 ε0 < ε < ε0 (unstable regime). The growth rate is highest at ξ = ξ0 and trends to decrease as ξ < ξ0. Consequently, k2 > 0 if 7 11 ε0 < ε < ε0 (unstable regime) and k2 < 0 if 0 < ε < 7 11 ε0 (stable regime). Using the relation Im(s) = i−mJm(is) where Jm(s) is the Bessel function of order m, the Riemann function for k2 < 0 changes to R(η, ξ) = A(η, ξ)J0(cid:16)kp(η − η0) (ξ − ξ0)(cid:17). Gartering all terms, the analytical wave solution of Eq. (17) in space and time variables is provided by 5 φ(x, t) = 1 2hea−tf (x + c−t) + ea+tf (x − c+t)i + 1 2 e−(µx−σt)Z x+c−t x−c+t eµξ [F (x − ξ, t)f (ξ) + G(x − ξ, t)g(ξ)] dξ, (26) where a− = 2Λk−, a+ = −2Λk+, µ = k+ + k−, and σ = k−c+− k+c−. For −c−t < x < c+t and 4k−k+ = k2 > 0, the propagators F and G are given by F (x, t) = −ΓJ0(ks(x, t)) + Λkt J1(ks(x, t)) s(x, t) , (27) 1 Λ G(x, t) = J0(ks(x, t)), (28) where s(x, t) = √c2t2 + 2νxt − x2 and Γ = k+−k−. The wave profile in Eq. (26) is nonzero for the position from x − c−t to x + c+t, called domain of dependence [25, 26], that supports the finite band formation and discontinu- ous front as found in simulations [9, 13, 16]. V. DISCUSSION From Eq. (26), the analytical solution indicates that initial profiles of the small perturbed density and polar- ization fields lose their configuration and propagate in both positive and negative directions of x-axis with un- equal speed. Due to c+ > c−, the propagation in the positive direction is faster than in the negative. Be- low the critical point, ε < ε0, we found that k+ > 0 when ε < 7 11 ε0 while k− is always negative. As t ≫ 0, according to Λ > 0, the left-propagating initial profile decays to zero whereas the right-propagating wave grows for ε > 7 11 ε0 (unstable regime) and decays for ε < 7 11 ε0 (stable regime). There- fore, the propagating waves in SPP trend move in the direction of mean polarization vector as found in simula- tions [9, 12 -- 14, 21, 22]. 11 ε0 and k+ < 0 when ε > 7 To this point, there exists another transition noise strength at 7 11 ε0 that separates the spatiotemporal pat- tern formation of the propagating wave in the flocking state of SPP into two regimes as mentioned by Chat´e [13]. Let us consider the unstable regime where et al. 4k−k+ = k2 > 0. The asymptotic Bessel function is 2 ) for s ≫ 1. With this character, it shows that the perturbations of the unstable ordered state propagate as waves with spa- tial oscillatory pattern or multiplebands in 2D, that has given by Jm(ks) = q 2 πks cos(ks − π 4 − mπ 9 k = π√k−k+ . Thus, we have ε′ (cid:17) + 1 11(cid:1) (1 − ε′) 64(cid:16) 1−ε′ q 11 2 (cid:0)ε′ − 7 λw = 2π v0 ε0 , (29) been observed in simulations [13, 17, 21, 22]. The wave profiles grow with the fastest rate in the position of the leading front and slower for the tandem position, at least in the early stage. Therefore, k is equivalent to wavenumber which relates to the wavelength λw as fol- lows: λw = 2π where ε′ = ε . The wavelength in Eq. (29) can be used ε0 to approximate the width of the stripes and it shows that the moving speed, v0, of the particle, plays a role in regu- lating the bands width. In the opposite situation, for the stable regime where 4k−k+ = −k2 < 0, the propagators change to F (x, t) = −ΓI0(ks(x, t)) − Λkt I1(ks(x, t)) s(x, t) , (30) G(x, t) = 1 Λ I0(ks(x, t)). (31) 1√2πks The asymptotic form of the modified Bessel functions, eks for s ≫ 1, shows the given by Im(ks) ∼ nonoscillatory wave patterns or a single band in 2D. In the stable ordered state, the perturbations eventually de- cay to smaller values. Thus, our linear approximation should be valid over long time scale. In long time scale x2 t → ∞, we approximate s ≃ ct + ν t . Thus, below the noise threshold, the perturbations converge to the homogeneous ordered state as biased Gaussian waves. However, the homogeneous ordered state is not observed in the simulation; instead, this state is replaced by the fluctuating flocking state [13, 17]. c x − 1 Λ2 c3 2 VI. CONCLUSION In summary, based on the linearized hydrodynamic theory of self-propelled particles, the small perturbed density and polarization fields are governed by the hy- perbolic partial differential equations. As opposed to the 6 previous analytical studies that rely on the dispersive wave solution, our analytical hyperbolic wave solutions reveal some different aspects of spatiotemporal pattern formations in self-propelled particles. Below the critical noise strength, the homogeneous disordered state is un- stable that is growing into the ordered state and gener- ates the vortex flow of perturbation polarization field. The perturbations in the homogeneous ordered state evolve as two possible unidirectional longitudinal prop- agating waves separated by a threshold noise strength. This includes single band in the stable state below the threshold value and multiplebands in the unstable state above the threshold value. We believe that these special solutions could provide the basic knowledge for studying the dynamics of more complex self-propelled particles, by including hydrodynamic and electrostatic interactions, in the future work. ACKNOWLEDGMENTS This research was supported by the Research Grant for New Scholar (Grant no. MRG5980258 ) funded by The Thailand Research Fund (TRF) and Office of the Higher Education Commission (OHEC). [1] T. Vicsek and A. Zafeiris, Phys. Rep. 517, 71 (2012). [2] T. Vicsek, A. Czir´ok, E. Ben-Jacob, I. Cohen, and [18] V. Schaller, C. Weber, C. Semmrich, E. Frey, and A. R. Bausch, Nature 467, 73 (2010). O. Shochet, Phys. Rev. Lett. 75, 1226 (1995). [19] A. Bricard, J.-B. Caussin, N. Desreumaux, O. Dauchot, [3] J. Toner and Y. Tu, Phys. Rev. Lett. 75, 4326 (1995). [4] J. Toner and Y. Tu, Phys. Rev. E 58, 4828 (1998). [5] A. Marchetti, and C. and D. Bartolo, Nature 503, 95 (2013). [20] D. Geyer, A. Morin, and D. Bartolo, Nat. Mater. 17, 789 (2018). Baskaran M. Phys. Rev. E 77, 011920 (2008). M. Baskaran and [6] A. Phys. Rev. Lett. 101, 268101 (2008). C. Marchetti, [21] J.-B. Caussin, A. Solon, A. Peshkov, H. Chat´e, and D. Bartolo, T. Dauxois, J. Tailleur, V. Vitelli, Phys. Rev. Lett. 112, 148102 (2014). [7] E. Bertin, A. Baskaran, H. Chat´e, and M. C. Marchetti, [22] A. P. Solon, J.-B. Caussin, D. Bartolo, H. Chat´e, and Phys. Rev. E 92, 042141 (2015). J. Tailleur, Phys. Rev. E 92, 062111 (2015). [8] E. Bertin, M. Droz, and G. Gr´egoire, [23] G. B. Whitham, Linear and nonlinear waves, Vol. 42 Phys. Rev. E 74, 022101 (2006). (John Wiley & Sons, 2011). [9] E. Bertin, M. Droz, and G. Grgoire, [24] A. G. Webster, Partial differential equations of mathe- J. Phys. A: Math. Theor. 42, 445001 (2009). matical physics (Dover Publications, 1955). [10] A. Peshkov, I. S. Aranson, E. Bertin, H. Chat´e, and F. Ginelli, Phys. Rev. Lett. 109, 268701 (2012). M. Toner, and Tu, J. [11] Y. Ulm, Phys. Rev. Lett. 80, 4819 (1998). [12] G. Gr´egoire and H. Chat´e, [25] T. Myint-U and L. Debnath, Partial differential equa- tions for scientists and engineers (North-Holland, 1987). [26] R. B. Guenther and J. W. Lee, Partial differential equations of mathematical physics and integral equations (Dover Publications, 1996). [27] F. D. C. Farrell, M. C. Marchetti, D. Marenduzzo, and Phys. Rev. Lett. 92, 025702 (2004). [13] H. Chat´e, F. Ginelli, G. Gr´egoire, Phys. Rev. E 77, 046113 (2008). [14] F. Ginelli, F. Peruani, M. Bar, Phys. Rev. Lett. 104, 184502 (2010). and F. Raynaud, J. Tailleur, Phys. Rev. Lett. 108, 248101 (2012). [28] A. Czir´ok, H. E. Stanley, and T. Vicsek, and H. Chat´e, J. Phys. A: Math. Theor. 30, 1375 (1997). [15] A. Gopinath, M. F. Hagan, M. C. Marchetti, and A. Baskaran, Phys. Rev. E 85, 061903 (2012). [16] T. Ihle, Phys. Rev. E 88, 040303 (2013). [17] S. Mishra, A. Baskaran, and M. C. Marchetti, [29] A. Bricard, J.-B. Caussin, D. Das, C. Savoie, V. Chikkadi, K. Shitara, O. Chepizhko, F. Peruani, D. Saintillan, and D. Bartolo, Nat. Commun. 6, 7470 (2015). [30] G. B. Arfken, Mathematical Methods for Physicists (Aca- Phys. Rev. E 81, 061916 (2010). demic Press, San Diego, 1985).
1102.4594	2	1102	2011-08-12T13:01:49	Growth of immobilized DNA by polymerase: bridging nanoelectrodes with individual dsDNA molecules	[ "physics.bio-ph", "cond-mat.other", "cond-mat.soft" ]	We present a method for controlled connection of gold electrodes with dsDNA molecules (locally on a chip) by utilizing polymerase to elongate single-stranded DNA primers attached to the electrodes. Thiol-modified oligonucleotides are directed and immobilized to nanoscale electrodes by means of dielectrophoretic trapping, and extended in a procedure mimicking PCR, finally forming a complete dsDNA molecule bridging the gap between the electrodes. The technique opens up opportunities for building from the bottom-up, for detection and sensing applications, and also for molecular electronics.	physics.bio-ph	physics	Growth of immobilized DNA by polymerase: bridging nanoelectrodes with individual dsDNA molecules Veikko Linko,*a Jenni Leppiniemi,b Boxuan Shen,a Einari Niskanen,c Vesa P. Hytönenb and J. Jussi Topparia 5 Submitted to Nanoscale We present a method for controlled connection of gold electrodes with dsDNA molecules (locally on a chip) by utilizing polymerase to elongate single-stranded DNA primers attached to the electrodes. Thiol- modified oligonucleotides are directed and immobilized to nanoscale electrodes by means of dielectro- phoretic trapping, and extended in a procedure mimicking PCR, finally forming a complete dsDNA molecule bridging the gap between the electrodes. The technique opens up opportunities for building from the bottom-up, for detection and sensing applications, and also for molecular electronics. 10 15 1. Introduction During recent years DNA has been proven to be a very flexible and versatile molecule within almost any areas of nanotechnology.1 Many DNA-based templates and constructs,2-4 even functional ones,5 have been introduced as well as numerous assays making use of the unique properties of DNA.6,7 Various approaches to implement molecular devices with a broad range of functions for nanoelectronics and bionanotechnology, are mostly based on superior self-assembly properties of DNA and well- developed tool kits, including many manipulation techniques. However, precise spatial control of diverse molecular components and assemblies is still a huge challenge. the One of the methods which has already enriched possibilities of DNA as a key player in bionanotechnology is Polymerase Chain Reaction (PCR). It is one of the basic tools in molecular biology and the principle of the reaction was already invented in 1970,8 but was not implemented to the laboratory work until during 1990’s – Nobel Prize was awarded to Kary B. Mullis in 1993. PCR allows exponential amplification of a target DNA sequence, a template, by using short synthetic DNA oligonucleotides as reaction primers. The method employs thermostable polymerase and temperature cycling, which enables repetition of the amplification step several times, thus allowing one to detect and characterize even a single copy of DNA. There exist countless applications for PCR in fields such as diagnostics, juridical research and personal medicine, just to mention a few. Typically, a PCR reaction is performed in solution. However, it has been shown that PCR can also be carried out when a DNA primer used for amplification is immobilized on a substrate. The first demonstration of “immobilized PCR” was carried out by Rasmussen et al. (1994),9 who successfully detected leukemia virus and Salmonella by utilizing PCR with one of the primers covalently immobilized via 5’-phosphate group by carbodiimide chemistry. Several reports have applied this principal 20 25 30 35 40 45 50 methodology to develop novel DNA-based methods, such as preparation of arrays of long DNA sequences by “on-chip” elongation.10 A method, where both of the primers are immobilized in large amounts to extensive surfaces, has also been demonstrated, named as “bridge amplification”.11 Here, we introduce a concept of growing individual dsDNA molecules locally on a chip from immobilized primers with nanoscale precision by mimicking a typical PCR procedure (however, we are not amplifying any target DNA sequence). The presented technique can serve as a bridge between novel DNA- nanotechnology and common molecular biology methods, while resolving also the problem of the precise spatial control. The method is based on directed concentration and immobilization of short thiol-modified single-stranded primers to the ends of fingertip-type gold nanoelectrodes by utilizing alternating-current dielectrophoresis (AC-DEP) (see Fig. 1(a)). Dielectrophoresis consists in the movement of any polarizable particle in a nonuniform electric field,12 and it has been widely exploited in trapping of diverse variety of objects,13,14 including controllable micron-15-17 and nanoscale17-22 manipulation and positioning of DNA. The DEP-immobilized primers are then extended by polymerase during repetitive cycles similar to the PCR. During the annealing steps the elongated strands can 1) form a complete dsDNA molecule having a template sequence and bridging the gap between the electrodes, or 2) pair with the complementary strands originated from a template (Figs. 1(b), 3(e)-(h)). Yet, by utilizing multielectrode geometries, it is possible to immobilize primers only to the desired electrodes (see Fig. 1(a)) enabling specific and sequence depended growing of dsDNA between the chosen electrodes. This provides new possibilities for building from the bottom-up with DNA, and the method can find applications in the fields of detecting and sensing as well as in molecular electronics. 55 60 65 70 75 80 1 Fig. 1 (a) Left: A schematic view of the electrode-specific dielectrophoretic (DEP) trapping of different thiol-modified primers with a multielectrode system. An AC voltage signal is applied one by one to the desired electrode gathering the primers to its end (also along the whole electrode), while keeping the other electrodes grounded. In the figure the electrode connected to an AC voltage source is collecting primers (blue tail), while the immobilization of other type of primers (green tail) to the opposite electrode has already been performed. Upper right: An AFM image of the multielectrode structure. The scale bar is 100 nm. Lower right: A confocal microscope image of 40 nt long primers labeled with different dye molecules [Cy3 (red) and Cy5 (blue)] separately trapped and immobilized to the opposite electrodes. The scale bar is 1 μm. The multielectrode geometry shown here was not utilized in the actual elongation experiments, but was used to show that the electrode-specific trapping is achievable enabling sophisticated wiring systems through the further optimization. (b) A sequence of schematic cross-sectional images presenting the polymerase growing method. A sample, containing primers (black and light blue) immobilized to the electrodes (yellow), is placed into a tube with required PCR reagents, such as dsDNA template (green strands), Taq DNA polymerase (violet) and nucleotides (red). Step 1. At the initialization / denaturation temperature the template melts. Step 2. In the annealing step the template strands pair with the complementary primers and the polymerase binds to the primer-template hybrid. Step 3. The polymerase elongates the immobilized primers. During later cycles of the procedure the extended primer can serve as a template strand for primers attached to the opposite electrode (in the steps 2 and 3). Step 4. After denaturation and annealing, the extended primers can form a bridge over the gap of the electrodes via hybridization (lower part) or pair again with the complementary template strands (upper part) (see also Figs. 3(e)-(h)). immobilize the Cy5-marked strands. The fluorescence spots of both types of the oligonucleotides could be simultaneously seen on the separate electrodes (see Fig. 1(a)) proving the reliable immobilization and practicability of the technique. However, the trapping procedure described above is not practical for minute examination of the kinetics of bridging the electrodes by polymerase for the following reasons: 1) it is a bit laborious method for extending the immobilized primers due to the observed relatively low yield in our (not fully optimized) elongation process (see below), 2) moreover, the used strands contained 3’-dye molecules and thus were not suitable for being elongated. Due to this reasoning, in order to show that our method of extending the immobilized primers by polymerase to form a dsDNA molecule across the gap is really exploitable, the elongation experiments were carried out using a slightly simplified scheme. As forward and reverse primers we used 5’- thiol-modified 38 nt ssDNA molecules consisting of the (CT)8- spacer followed by the 22 nt long part matching to the terminal sequences of the complementary strands of the 414 bp template.22,23 The added spacer facilitates the hybridization of the template and attaching of the polymerase to the active/matching part of the immobilized primer. The trapping was carried out similarly as stated above, but now we used an array of adjacent fingertip type electrode pairs (single electrode pairs are presented in Figs. 2 and 3) to gain more data in a single run. Also, both primers were immobilized simultaneously, i.e., ~10 μl of primer solution containing both the primers (~20 nM in the Hepes/NaOH buffer) was pipetted onto the chip and an AC-voltage of 4.5 Vpp (VDC = 0) was applied between the electrodes for 1-2 minutes. Finally the sample was gently rinsed with 40-50 μl of distilled water and dried with nitrogen flow. The efficiency of the DEP trapping of the primers was first 2. Results and discussion 2.1 DEP trapping and immobilization of primers The first step in the procedure is to prepare a gold nanoelectrode structure on a silicon oxide substrate by electron beam lithography, for trapping the primers to the certain locations on a substrate, i.e. to the selected electrodes. The dimensions of the fabricated electrodes are 20 nm × 20 nm (width × height) for the 8-electrode system (Fig. 1(a)) or 100-170 nm × 20 nm for the fingertip electrodes (Figs. 2 and 3), and the separation between the opposite electrodes is 100-140 nm, roughly corresponding to the length of the grown dsDNA molecule (template for growing: 414 bp partial complementary DNA of chicken avidin22,23). To demonstrate the feasibility of the selective trapping with the multielectrode geometry we used two dye-labelled 40 nt long 5’- hexanethiol-modified oligonucleotides with fluorescent dye molecules, Cy3 and Cy5, attached to the 3’-ends. First, 12 μl of 0.3 μM Cy3-labeled oligonucleotide solution (3 mM Hepes / 2 mM NaOH buffer) was pipetted onto the chip, and the trapping field was created by applying a sinusoidal 1 MHz AC-voltage of 5.0 Vpp with DC-offset of 1.3 V to the desired electrode, while keeping the other electrodes grounded. The gathering of the oligonucleotides to the trap, and in particular to the chosen electrode, was studied in situ under a confocal-microscope (Olympus FluoView 1000, 60× oil objective; lasers 543 nm and 633 nm were used to excite Cy3 and Cy5 dyes, respectively). After a successful immobilization of Cy3-modified oligonucleotides, the trapping voltage was switched off and the sample was gently rinsed with distilled water. Second, the same procedure was repeated but now using the opposite electrode to 60 50 55 35 40 45 5 10 15 20 25 30 2 Fig. 2 (a) A confocal microscope image presenting the trapping and immobilization of 22 nt Cy3-labelled primers in between the fingertip- type electrodes. The scale bar is 500 nm. (b) Simulated DEP trapping potential (color coded surface) in the vicinity of 100 nm wide fingertip- type electrodes (shown as an AFM image). Potential is calculated at the plane just above the electrodes, i.e., about 25 nm above the substrate. Trapping area is the region where the absolute value of the DEP potential is higher than the thermal energy of DNA, i.e., where the potential is below the flat horizontal surface, corresponding to the negative of thermal energy, -3/2 kBT (UDEP + Uthermal = 0). It can be seen that the strongest trapping spots lie at the ends of the electrodes but trapping also takes place along the electrodes. validated by comparing the finite element method (FEM) simulations of the time-averaged DEP trapping potential (UDEP = -½ α E2, where the estimated value for the polarizability α ≈ 2×10-33 Fm2 for a 40 nt ssDNA was used21) to the estimated thermal energy of the primers (Uthermal = 3/2 kBT) at the room temperature. The results showed that the absolute value of the DEP potential in the vicinity of the electrodes is well above the thermal energy resulting in a deep trapping well for ssDNA molecules as shown in Fig. 2(b). Yet, to ensure that the primers are trapped and immobilized in a desired way, similar ssDNA molecules labeled with dye molecules (22 nt ssDNA with hexanethiol at the 5’-end and Cy3-modification at the 3’-end) were trapped with the above-mentioned parameters followed by the confocal microscope imaging confirming the result (see Fig. 2(a)). Further, the fluorescence was still visible in the gap region after heating the sample in a PCR buffer to 95 °C, indicating covalent binding of the primers to the electrodes (covalent sulphur-gold bond). Thus, the majority of the trapped primers can be assumed to stay immobilized during the elongation. The results of the trapping of the primers without dye molecules, i.e. the primers feasible for a polymerase extension, were verified by atomic force microscope (AFM) imaging (Veeco, Dimension 3100) (see Figs. 3(a)-(d)). 2.2 Extension of primers and bridging the electrodes As the final step, the chip containing the electrode structure and the immobilized primers (this time two types of primers at each electrode) was placed into a 0.2 ml tube with the reagents and components required for the standard PCR procedure (see Experimental section). The following programme for a thermal cycler (Biometra T3 Thermoblock, Biotron, Germany) was used: 1) 94 °C, 5 min; 2) 94 °C, 40 s; 3) 50 °C, 3 min; 4) 72 °C, 4 min; 5 10 15 20 25 30 45 50 35 40 5) 4 °C, where the cycles 2-4 were repeated 25-50 times. Finally, the chip was removed from the tube and washed with distilled water similarly as after the trapping process. Since the very low electrical conductivity of a long dsDNA is known to be strongly dependent on the environment (types and concentrations of ions of the buffers),22,24-26 a reliable electrical verification of successful growth, i.e. bridging the gap, was not practicable. In addition, the utilized single strand spacers in the primers can be considered as insulators. Also, since the low amount of grown individual molecules does not produce strong enough fluorescent signal to employ optical detection either, the successful growth was detected by imaging the samples again with AFM and comparing the images taken before and after growing procedure. In Fig. 3, examples of the results are shown. In Figs. 3(e) and 3(f) a single ~150 nm long dsDNA molecule, grown during the procedure and comprised of the extended primers attached to the opposite electrodes, is bridging the gap between the electrodes. Figs. 3(g) and 3(h) present alternative results of the process, where elongated primers are paired with template strands via hybridization during the annealing step, forming ~150 nm long dsDNA molecules at the edges of the electrodes. 55 3. Experimental section 3.1 Nanoelectrode preparation The nanoelectrodes were fabricated on an oxidized silicon chip by using electron beam lithography and evaporation of metal (15- 18 nm gold on top of 1-2 nm titanium) in an ultrahigh vacuum (UHV) chamber. In addition, PMMA residues from the lift-off were cleaned off the electrode structure with an oxygen plasma flash in a reactive ion etcher. This procedure also made the SiO2 surface hydrophilic, which greatly enhances the DEP-trapping. 3.2 DNA strands The sequences of the thiol-modified (5’end) and either Cy3 or Cy5-dye-labeled (3’end) DNA oligonucleotides used in the optimization procedure of the DEP-trapping were the following: long oligos: 5’-(CT)16GATGGCTT-3’-Cy3 and 5’- 40 nt (CT)16GAAAAAGC-3’-Cy5; and 22 nt long ssDNA molecules: 5’-GCCAGAAAGTGCTCGCTGACTG-3’-Cy3 (purchased from Biomers as HPLC-purified). In the actual elongation experiments the sequences of the 38 nt forward and reverse primers (without dye-molecules) contained 22 nt long sequences complementary to spacer: 5’- the template and the additional 16 nt (CT)8GCCAGAAAGTGCTCGCTGACTG-3’ 5’- and (CT)8TTCTCGACAAGCTTTGCGGGGC-3’, where 5' Thiol Modifier C6 S-S (Disulfide) was attached to 5’ends (ordered from Integrated DNA Technologies as dual HPLC-purified). 3.3 Reagents for elongation of primers Reagents for elongation of the primers and their amounts are presented here in the order of mixing (reagent, amount, final concentration): H2O (distilled), 69.9 μl, -; 10× Taq buffer [750 85 70 75 80 60 65 3 Fig. 3 (a)-(d) “Before extension” AFM images of bunches of 38 nt primers immobilized to the electrodes. (e)-(h) “After extension” AFM images of the same samples shown in Figs. (a)-(d). Note that the height in all the 3D images is presented in a logarithmic scale. (e)-(f) A single grown dsDNA molecule is bridging the gap between electrodes. (g)-(h) A few dsDNA molecules grown on the edge of the electrode. The non-specifically bound primers have been detached during the elongation. The gap between the electrodes is ~100-120 nm in all samples. The graphs below are cross sections along the blue lines on (e)-(h), showing the characteristic height of the dsDNA molecule on a SiO2 substrate, i.e. 1-2 nm. nM Tris-HCl (pH 8.8), 200 nM (NH4)2SO4, 0.1% Tween 20], biolab facilities; A. Kuzyk and P. Törmä (Aalto University School of Science and Technology, Helsinki) for momentous 10.0 μl, 1×; MgCl2 (25 mM), 8.0 μl, 2.0 mM; dsDNA template discussions. Academy of Finland (projects 218182, 130900 and (51 ng/μl), 1.5 μl, 0.76 ng/μl; dNTP mix (2 nM), 10.0 μl, 0.2 mM; 115976) is acknowledged for financial support. V.L. thanks Taq polymerase (5 U/μl), 0.6 μl, 1.5 U / 50 μl. Taq DNA polymerase (recombinant) (with Taq buffer and MgCl2) and Finnish Academy of Science and Letters (Väisälä Foundation), Finnish Cultural Foundation (Central Finland Regional Fund), dNTP mix were purchased from Fermentas. Finnish Foundation for Technology Promotion (TES), and National Doctoral Programme in Nanoscience (NGS-NANO). J.L. thanks Tampere Graduate Program in Biomedicine and Biotechnology. B.S. thanks Centre of Expertise Programme, The Nanotechnology Cluster Programme 090002-501. 4. Conclusions 35 40 5 10 15 20 25 30 4 In summary, we have demonstrated a novel method to grow individual dsDNA molecules on a chip based on a controllable directing of the primers via dielectrophoresis, and elongation of the immobilized primers by polymerase. The developed technique can serve as a tool in exploiting the on-chip growing of DNA and also PCR on a new level. In this particular proof-of- principle experiment, the yield has been rather low, but there are still several aspects to optimize. By making further improvements on the procedure and designing an electrode pattern suitable for the application in question, it opens up opportunities for detecting single molecules or molecule combinations, and also for fabricating bottom-up based nanostructures from DNA at desired locations on a chip, e.g. sophisticated DNA wiring systems and networks27 or DNA-based multiswitching units programmable to react on planned targets. Finally, DNA-programming platform can be envisioned, where identical electrode geometries are tailored by using different combinations of primers and template sequences. Acknowledgements We would like to thank J. Ihalainen, J. Ylänne and M. Vihinen- Ranta (Nanoscience Center, University of Jyväskylä) for use of Notes and references a Nanoscience Center, Department of Physics, University of Jyväskylä, P.O. Box 35, FI-40014, Jyväskylä, Finland, Fax: +358 14 260 4756; Tel: +358 14 260 4722; E-mail: [email protected] b Institute of Biomedical Technology, University of Tampere and Tampere University Hospital, FI-33014, Tampere, Finland c Nanoscience Center, Department of Biological and Environmental Science, University of Jyväskylä, P.O. Box 35, FI-40014, Jyväskylä, Finland. Present address: Department of Biochemistry, Erasmus University Medical Center, Dr. Molewaterplein 50, 3015 GE Rotterdam, The Netherlands 1 T. H. LaBean and H. Li, Nano Today, 2007, 2, 26. 2 N. C. Seeman, Nature, 2003, 421, 427; N. C. Seeman, Nano Lett., 2010, 10, 1971. 3 S. M. Douglas, H. Dietz, T. Liedl, B. Högberg, F. Graf and W. M. Shih, Nature, 2009, 459, 414. 4 C. E. Castro, F. Kilchherr, D. N. Kim, E. L. Shiao, T. Wauer, P. Wortmann, M. Bathe and H. Dietz, Nature Methods, 2011, 8, 221. 5 B. Yurke, A. J. Turberfield, A. P. Mills Jr., F. C. Simmel and J. L. Neumann, Nature, 2000, 406, 605. 6 R. Chhabra, J. Sharma, Y. Ke, Y. Liu, S. Rinker, S. Lindsay and H. Yan, J. Am. Chem. Soc., 2007, 129, 10304. 45 50 55 60 65 7 N. V. Voigt, T. Tørring, A. Rotaru, M. F. Jacobsen, J. B. Ravnsbaek, R. Subramani, W. Mamdouh, J. Kjems, A. Mokhir, F. Besenbacher and K. V. Gothelf, Nature Nanotech., 2010, 5, 200. 8 K. Kleppe, E. Ohtsuka, R. Kleppe, I. Molineux and H. G. Khorana, J. Mol. Biol., 1971, 56, 341. 9 S. R. Rasmussen, H. B. Rasmussen, M. R. Larsen, R. Hoff-Jørgensen and R. J. Cano, Clin Chem., 1994, 40, 200. 10 M. von Nickisch-Rosenegk, X. Marschan, D. Andresen, A. Abraham, C. Heise and F. F. Bier, Biosens Bioelectron., 2005, 20, 1491. 11 D. H. Bing, C. Boles, F. N. Rehman, M. Audeh, M. Belmarsh, B. Kelley and C. P. Adams, Bridge amplification: a solid phase PCR system for the amplification and detection of allelic differences in single copy genes, Genetic Identity Conference Proceedings, Seventh International Symposium on Human Identification, 1996. 12 H. A. Pohl, Dielectrophoresis: the Behaviour of Neutral Matter in Nonuniform Electric Field, Cambridge University Press, Cambridge, UK, 1978. 13 P. J. Burke, Encycl. Nanosci. Nanotechnol., 2004, 6, 623. 14 M. P. Hughes, Nanotechnology, 2000, 11, 124. 15 M. Washizu and O. Kurosawa, IEEE Trans. Ind. Appl., 1990, 26, 1165. 16 M. Washizu, O. Kurosawa, I. Arai, S. Suzuki and N. Shimamoto, IEEE Trans. Ind. Appl., 1995, 31, 447. 17 R. Hölzel and F. F. Bier, IEEE Proc.-Nanobiotechnol., 2003, 150, 47. 18 L. Ying, S. S. White, A. Bruckbauer, L. Meadows, Y. E. Korchev and D. Klenerman, Biophys. J., 2004, 86, 1018. 19 A. Wolff, C. Leiterer, A. Csaki and W. Fritzsche, Front Biosci., 2008, 13, 6834. 20 A. Kuzyk, B. Yurke, J. J. Toppari, V. Linko and P. Törmä, Small, 2008, 4, 447. 21 S. Tuukkanen, A. Kuzyk, J. J. Toppari, H. Häkkinen, V. P. Hytönen, E. Niskanen, M. Rinkiö and P. Törmä, Nanotechnology, 2007, 18, 295204. 22 S. Tuukkanen, A. Kuzyk, J. J. Toppari, V. P. Hytönen, T. Ihalainen and P. Törmä, Appl. Phys. Lett., 2005, 87, 183102. 23 M. L. Gope, R. A. Keinänen, P. A. Kristo, O. M. Conneely, W. G. Beattie, T. Zarucki-Schulz, B. W. O’Malley and M. S. Kulomaa, Nucleic Acids Res., 1987, 15, 3595. 24 D. Porath, G. Cuniberti, R. Di Felice, Top. Curr. Chem., 2004, 237, 183. 25 V. Linko, S.-T. Paasonen, A. Kuzyk, P. Törmä and J. J. Toppari, Small, 2009, 5, 2382. 26 V. Linko, J. Leppiniemi, S.-T. Paasonen, V. P. Hytönen and J. J. Toppari, Nanotechnology, 2011, 22, 275610. 27 Y. Eichen, E. Braun, U. Sivan and G. Ben-Yoseph, Acta Polym. 1998, 49, 663. 5 10 15 20 25 30 35 40 45 5
1302.2798	1	1302	2013-02-12T14:14:49	Molecular motors robustly drive active gels to a critically connected state	[ "physics.bio-ph", "cond-mat.soft", "q-bio.SC" ]	Living systems often exhibit internal driving: active, molecular processes drive nonequilibrium phenomena such as metabolism or migration. Active gels constitute a fascinating class of internally driven matter, where molecular motors exert localized stresses inside polymer networks. There is evidence that network crosslinking is required to allow motors to induce macroscopic contraction. Yet a quantitative understanding of how network connectivity enables contraction is lacking. Here we show experimentally that myosin motors contract crosslinked actin polymer networks to clusters with a scale-free size distribution. This critical behavior occurs over an unexpectedly broad range of crosslink concentrations. To understand this robustness, we develop a quantitative model of contractile networks that takes into account network restructuring: motors reduce connectivity by forcing crosslinks to unbind. Paradoxically, to coordinate global contractions, motor activity should be low. Otherwise, motors drive initially well-connected networks to a critical state where ruptures form across the entire network.	physics.bio-ph	physics	Molecular motors robustly drive active gels to a critically connected state José Alvarado1, Michael Sheinman2, Abhinav Sharma2, Fred C. MacKintosh2, Gijsje H. Koenderink1 1 FOM Institute AMOLF, Science Park 104, 1098 XG Amsterdam, The Netherlands 2 Department of Physics and Astronomy, Vrije Universiteit, 1081 HV Amsterdam, The Netherlands * e-mail:[email protected], [email protected] Molecular motors robustly drive active gels to a critically connected state Alvarado et al. Abstract Living systems often exhibit internal driving: active, molecular processes drive nonequilibrium phenomena such as metabolism or migration. Active gels constitute a fascinating class of internally driven matter, where molecular motors exert localized stresses inside polymer networks. There is evidence that network crosslinking is required to allow motors to induce macroscopic contraction. Yet a quantitative understanding of how network connectivity enables contraction is lacking. Here we show experimentally that myosin motors contract crosslinked actin polymer networks to clusters with a scale-free size distribution. This critical behavior occurs over an unexpectedly broad range of crosslink concentrations. To understand this robustness, we develop a quantitative model of contractile networks that takes into account network restructuring: motors reduce connectivity by forcing crosslinks to unbind. Paradoxically, to coordinate global contractions, motor activity should be low. Otherwise, motors drive initially well-connected networks to a critical state where ruptures form across the entire network. Page 2 of 21 Molecular motors robustly drive active gels to a critically connected state Alvarado et al. Introduction One of the defining qualities of soft matter is that it is readily driven far from thermodynamic equilibrium by external stress. Driving forces such as those due to an electric field or shear can drive colloidal suspensions and polymer networks into fascinating non- equilibrium patterns, including banded1,2, jammed3, and randomized steady states4. Much progress has been made in understanding such externally driven systems5. By contrast, living soft matter systems such as cells and tissues naturally exhibit a unique form of internal driving in the form of mechanochemical activity6,7. A prominent example is the cytoskeleton, a meshwork of protein polymers and force-generating motor proteins that constitutes the scaffold of cells. In solutions of purified cytoskeletal filaments and motors, remarkable self-organized patterns have been observed8,9, inspiring theoretical work of these so-called active gels10. More recently, attention has shifted to the important role of network connectivity in active gels, which can be controlled by the number of crosslinks between filaments. In weakly connected systems, motors slide filaments to form static or dynamic clusters11-14. In the opposite limit of a well-connected, elastic network, motors generate contractile stresses as they pull against crosslinks, which can dramatically change the elastic properties of the network15,16 or lead to contraction17,18. The existence of a threshold connectivity that separates these two behaviors has been proposed, since macroscopic contractions are known to occur above certain minimum values of crosslink or actin concentration14,17,19,20. We should expect remarkable critical behavior at the threshold of contraction. Recent theoretical models predict diverging correlation length-scales and a strong response to external fields21-24 at the threshold of rigidity. In suspensions of self-propelled patches, critical slowing was predicted at the threshold of alignment25. Yet the threshold of contraction still remains poorly understood, and experimental evidence of criticality in active gels remains lacking. Page 3 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state Here, we experimentally study model cytoskeletal systems composed of actin filaments and myosin motors. We vary network connectivity over a broad range by adding controlled amounts of crosslink protein. We show that the motors can actively contract the networks into disjoint clusters that exhibit a power-law size distribution. This behavior is reminiscent of classical conductivity percolation26, for which a power-law size distribution of clusters occurs close to a critical point. However, in sharp contrast to this equilibrium phenomenon, we observe critical behavior over a wide range of initial network connectivities. To understand this robustness, we develop a general theoretical model of contractile gels that can quantitatively account for our observations. In this model, motors not only contract the network, but also reduce the connectivity of initially stable networks down to a marginal structure by promoting crosslink unbinding. Below this marginal connectivity, the network no longer supports stress and the system rapidly devolves to disjoint clusters which reflect the critical behavior of the marginal structure. Our model predicts cluster size distributions that agree well with experiment. Moreover, it predicts an inverse relationship between cluster size and motor activity, which we also confirm experimentally. Experiment: motors rupture networks into clusters In order to resolve the interplay between motor activity and network connectivity in active cytoskeletal networks, we develop a biomimetic model system with a well-controlled composition (Fig 1a). Networks are formed by initiating actin filament polymerization, which results in a semiflexible polymer meshwork with a pore size of ~0.3 µm. We control the motor activity by adding different amounts of myosin motors, expressed in terms of the myosin-to- actin molar ratio, RM = [myosin] / [actin]. We control the network connectivity by adding different amounts of the crosslink fascin, which can simultaneously bind to two neighboring Page 4 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state actin filaments (see Methods). We express the crosslink density in terms of the fascin-to-actin molar ratio, RC = [fascin] / [actin]. To ensure that we can observe motor-driven contraction on all scales, from microscopic to macroscopic, we prepare networks in customized flow-cells, which fit entirely in the field-of-view of the 4× objective of a confocal microscope (see Methods). To track the temporal evolution of the networks, we acquire time-lapse movies starting from 1 minute after the initiation of actin polymerization, where the solution is still homogeneous, until 2 hours afterwards. To resolve the influence of network connectivity, we first prepare a series of networks with constant myosin activity (RM = 0.01) and gradually increasing crosslink density (RC). Even at low RC, the motors can contract actin networks (Supplementary Movie 1). However, contraction occurs only on a small length scale, as seen in the time projection image in Fig. 1b. However, when we increase RC, contraction occurs on a larger length scale (Fig 1c, Supplementary Movie 2). The motors break the network up into multiple disjoint clusters. At still higher RC, motor activity contracts the entire network into a single dense cluster which often retains the square shape of the assay chamber (Fig 1d, Supplementary Movie 3). To quantify the effect of connectivity on the length scale of network contraction, we developed an image processing algorithm (Supplementary Movie 4) which identifies the clusters in the final image and traces their origin back in time. As shown in Fig. 1, the initial areas of each cluster are small in weakly crosslinked networks (panel d). The smallest clusters are ~30 µm in size, which corresponds to the typical distance between myosin motor clusters in the absence of cross-links (Supplementary Figure 1). However, the clusters increase in size when the crosslink density is increased (panel f). In strongly crosslinked networks, the entire network forms one cluster (panel g). Page 5 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state Qualitatively, the transition from local to macroscopic contraction is reminiscent of a classical conductivity percolation transition. Below this transition, a system is only locally correlated and cannot establish connections over long distances. Only above a certain critical connectivity can the system establish global correlations. In order to determine the extent of agreement between our experimental results and percolation theory, we investigate three key predictions26. First, conductivity percolation theory predicts how connectivity determines the size of the largest and second-largest connected clusters. Connectivity is quantified by the probability p of creating a connection. The largest cluster (of size ξ1) is predicted to increase monotonically with p, while the second-largest cluster (of size ξ2) should exhibit a peak right at the conductivity percolation threshold, where ξ1 and ξ2 both approach the system size, L (Fig 1h, inset). Our experiments agree with this prediction: the measured cluster sizes, ξ1 and ξ2, are both small at low crosslink density and increase monotonically with increasing crosslink concentration until they approach the system size, L ≈ 2.5 mm, around RC ~ 0.01 (Fig 1h). Above this threshold connectivity, ξ1 remains close to L whereas ξ2 decreases towards zero as the entire network contracts to one large cluster. Second, percolation theory predicts how cluster sizes are distributed: around the critical point, we should find a power law with an exponent of −2. To test this prediction, we begin by looking for networks which satisfy ξ1 ~ ξ2 ~ L. We replot all measurements separately in ξ1-ξ2- space (Fig. 2a). Because ξ2 < ξ1 by definition, all samples are located within a triangle in ξ1-ξ2- space. We can clearly identify the samples at the triangle’s peak, where ξ1 ~ ξ2 ~ L. We denote this peak as the critically connected regime. To the left of the peak are samples with low RC, which we denote the local contraction regime. To the right of the peak are samples with high RC, the global contraction regime. Page 6 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state Do the samples in the critically connected regime really exhibit critical behavior? To test this more rigorously, we plot the entire distribution of cluster sizes (Fig 2b). We represent the observed distribution as a histogram (open circles), where power-law distributions appear as straight lines on a log-log plot. We additionally plot complementary cumulative probability distributions (solid lines), whose visual form does not depend on bin size. We find that our experiments are again consistent with percolation theory: the critically-connected regime indeed exhibits a cluster-size distribution that is statistically consistent with a power-law across more than two orders of magnitude in measured area27. The power-law exponent is −1.9, close to the exponent of −2 predicted by percolation theory. The distributions of the other two regimes furthermore agree with percolation theory. The local contraction regime exhibits a short-tail distribution with a sharp cut-off. The global contraction regime exhibits a bimodal distribution with two well-separated length scales: the percolating cluster with size ξ1 ~ L and other small disjointed clusters with a typical size of ξ2 << L. Third, percolation theory predicts that only systems that are close to the critical point should exhibit a power law. But this prediction is difficult to reconcile with our data: the critically connected regime in ξ1-ξ2-space (Fig 2a) is populated by samples which span a wide range of cross-link densities (from RC = 0.01 to RC = 0.1). This is also reflected in Fig. 1h, which shows a broad ξ2-peak that is over half an order of magnitude wide in RC, in sharp contrast with the narrow ξ2-peak expected from percolation theory (inset of Fig 1h). We can therefore conclude that classical conductivity percolation theory cannot provide a complete description of the physics of active, contractile networks. Page 7 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state Simulation: network restructuring Percolation theory describes a network with a fixed connectivity. This can be appropriate for equilibrium fiber networks without internal driving. However, in motor-driven networks, the total connectivity can change significantly28-30. When we image our networks at high resolution, we see that motors actively pull on network strands and disconnect them, thereby reducing connectivity (Supplementary Movie 5). Crosslinks bind only transiently (~10 s in case of fascin31), and their binding kinetics are typically stress-dependent32. There is strong evidence that unbinding of fascin crosslinks is promoted under stress. For instance, in gliding assays where actin-fascin bundles move over immobilized myosin motors, the motors actively zipped open the bundles33. We hypothesize that such stress-dependent binding kinetics allow motor activity to drive initially well-connected networks down towards a critically connected state. To test this hypothesis, we develop a computational model of contractile actin-myosin networks using molecular dynamics. We model actin filaments with a planar triangular lattice of nodes connected by line segments of length l0 (Fig 3a). Filaments possess stretching modulus k and can strain-stiffen34 and buckle35. We set the average number z of line segments connected to a node (i.e. coordination number) to 4.0. Point-like crosslinks are randomly placed on nodes with probability p, which depends on crosslink concentration c. We assume first-order kinetics of crosslink (un)binding, which yields p = c/(1+c). We model the crosslinks by freely-hinged constraints, which prevent relative sliding of connected filaments. Motor activity results in contractile stresses13,36,37, which we model by pairs of forces f between nodes. Every node has mobility µ and experiences an effective, free-draining viscosity, η. The network evolves over time to achieve force balance at the nodes (Fig 3b). For fixed crosslinks, network connectivity remains unchanged and ξ1 and ξ2 remain constant. We now introduce into the model the important ingredient of network restructuring: connectivity can change via crosslink unbinding Page 8 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state and rebinding. The unbinding rate of a crosslink koff increases exponentially with the tension T according to Bell’s law32: koff = koff,0 exp(T / T0), where koff,0 denotes the off-rate in the absence of tension, and T0 a characteristic tension (Fig 3c). To account for rebinding, we consider the probability that an unbinding event is followed by a rebinding event at the same location before filaments are separated, which is given by exp(-c kon d / T µ), where d is an effective distance on the order of the mesh size over which filaments can move with velocity equal to T µ and kon is the binding rate of a crosslink. The effective unbinding rate is thus given by koff = koff,0 exp(T / T0) exp(-c kon d/ T µ). By varying c across many simulations (keeping f constant), we recover the three regimes found in experiment: the local contraction (Fig 3d,e; Supplementary Movie 6), critically connected (Fig 3f,g; Supplementary Movie 7), and global contraction regimes (Fig h,i; Supplementary Movie 8). The crosslink-dependence of ξ1 and ξ2 versus c (Fig 3j) as well as the cluster size distributions (Supplementary Figure 2) are fully consistent with experiment. The model clearly reveals that motor activity broadens the ξ2-peak: in the absence of active network restructuring (panel j, open symbols), only a narrow region (yellow stripes) around the critical point exhibits critical behavior. In the presence of network restructuring (panel j, closed symbols), this region broadens (solid yellow box). Motor-driven network restructuring can therefore account for the surprising robustness of critical behavior we found in experiment. Motors promote network restructuring So far we have investigated the effect of connectivity in experiment and simulation (RC and N), but kept motor activity constant (RM and f). Network restructuring breaks networks into clusters because motor stresses unbind crosslinks. Increased motor activity should therefore increase tension on crosslinks, enhance their unbinding, and lead to smaller clusters. To test this Page 9 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state hypothesis, we simulate well-connected networks with constant c but varying motor activity (modeled through changes in the force f). Increased force indeed leads to smaller clusters (Fig 4a-d). At low force (f / k < 0.1), the networks contract macroscopically (ξ1 >> ξ2), while at higher force levels ξ1 sharply decreases as f increases (Fig 4e). Increasing f allows networks that would otherwise contract globally to exhibit clusters with a power-law size distribution. In order to validate these predictions, we perform experiments in well-connected networks (RC = 0.02) where we change motor activity by varying the myosin-to-actin molar ratio RM. In agreement with the model’s prediction, the length scale of contraction strongly depends on myosin concentration. For low motor concentrations up to RM = 0.002, the networks appear stationary for the entire duration of the experiment. Large-scale collective breathing fluctuations are visible, indicative of a strongly connected network, but the motors exert insufficient force to contract the network (Supplementary Movie 9). Increasing the motor concentration to RM = 0.005 results in a drastic change: the entire network collapses into one large cluster mediated by a uniform global contraction (Fig. 4f,h; Supplementary Movie 10). However, a further increase of RM results in smaller clusters (Fig 4g,i; Supplementary Movie 11). At high motor densities, ξ1 decreases in a manner consistent with the model’s prediction (Fig 4j) and we again recover scale-free cluster size distributions (Supplementary Figure 3.) These results lead to a counterintuitive consequence: in order to coordinate contractions over macroscopic length scales, less motor activity is needed. Increasing motor activity only yields small clusters. Motors can nucleate many concurrent ruptures In order to better understand the effect of force on cluster size, we consider the opposing limits of local and global contraction in our simulations. These two regimes are clearly separated Page 10 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state by the critically connected regime, as evident in the schematic phase space diagram in Fig 5. The global contraction regime is located at the bottom-right corner, where motor forces are low and network connectivity is high. In this limit, networks are rigid, filaments remain straight, and the network deforms affinely22. On the opposite corner of the phase diagram, where connectivity is low and force is high, we find the local contraction regime. Such weakly connected, loose networks deform nonaffinely, and filaments are significantly bent. We can interpret these limits by considering two relevant timescales, τoff and τrelax. The first timescale is the characteristic crosslink unbinding time τoff = koff-1. The tension T experienced by a crosslink depends on both the motor force f and the network configuration, which can change over time. Although the full dependence of crosslink tension on motor force is complex, the qualitative behavior is clear: when filaments are straight, motor stress does not greatly induce crosslink tension; when filaments are bent, crosslinks experience tension (Supplementary Figure 4). The second timescale, τrelax, is the time it takes for filaments in the network to relax in response to a crosslink unbinding event. We estimate the values of τoff and τrelax from previous work31,38: τoff,0 ~ 1–10s τrelax ~ 0.1–1s. The above value for τrelax is set by the thermal equilibration of individual filaments. It acts as an upper bound: forces can cause faster relaxation. Therefore in the absence of tension, τoff > τrelax. We now consider how these timescales respond to the two limits of local and global contraction. In the global contraction limit, f and T are small, and τoff > τrelax holds: once a crosslink unbinds, the network fully relaxes before the next crosslink unbinds. This well-known limit corresponds to a quasistatic process39. Boundary conditions determine how the network evolves in this limit: networks fixed at rigid boundaries build up stress and rupture via the Page 11 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state nucleation of a large crack at a microscopic flaw, reminiscent of Griffith’s criterion40. Unanchored networks contract affinely, or drive shape changes when coupled to deformable boundaries41. In the opposite limit of local contraction, f and T are large, and the network satisfies τoff < τrelax: strong internal driving causes crosslinks to unbind quickly. Many cracks that rupture the network into clusters form across the whole network, rather than nucleating at a single flaw. The presence of a finite viscosity in our model is essential for this behavior. Neglecting viscosity leads to τrelax = 0, and networks fail only via quasistatic crack propagation39. In between the two limits of global and local contractions, we find critically connected networks with a scale-free distribution of clusters. For zero force, this regime is narrow and centered around the critical point. As forces increase, this regime broadens and shifts to higher connectivities. This rightward shift reflects an asymmetry where motor activity reduces connectivity, rather than increasing it. The broadening shows that increased motor activity drives networks more robustly to a critical state. Intriguingly, robust critical behavior has been demonstrated in many biological systems42-46. Internal driving could underlie robust criticality47, but so could other mechanisms, including natural selection48,49. Disentangling these mechanisms cannot be addressed by studying living systems alone. Here we report robust criticality in a minimal model system and show that internal driving is directly responsible. These results may help explain criticality in other biological contexts and may prove useful in designing the physical properties of synthetic active materials, which have recently become available50. Our framework offers a minimal microscopic mechanism that should help in modeling contractile systems in biology. Recent studies in live cells suggest that motor myosin-driven cytoskeletal ruptures play an important functional role in cell division,51, whereas they Page 12 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state contribute to developmental defects in developing embryos52. Consistent with our findings, decreased connectivity caused dramatic rupture of the ventral furrow into clusters of cells in developing fly embryos. We anticipate that our framework applies more generally to tissues of interconnected cells53,54, where a supracellular actomyosin network transmits forces over tissue length scales. Methods Protein Preparation. Actin and myosin were prepared from rabbit psoas skeletal muscle (Supplementary Information). Myosin II was labeled with Alexa Fluor 488 NHS ester (Invitrogen, Paisley, UK); actin was labeled with Alexa Fluor 594 carboxylic acid, succinimidyl ester13. Recombinant mouse fascin was prepared from T7 pGEX E. coli 55. Sample Preparation. Samples were mixed to yield a final buffer composition of 20 mM imidazole pH 7.4, 50 mM potassium chloride, 2 mM magnesium chloride, 1 mM dithiothreitol, and 0.1 mM adenosine triphosphate (ATP). Furthermore, 1 mM trolox, 2 mM protocatechuic acid, and 0.1 µM protocatechuate 3,4-dioxygenase were added to minimize photobleaching. The ATP level was held constant by addition of 10 mM creatine phosphate disodium and 0.1 mg mL-1 creatine kinase. The actin concentration was held constant at 12 µM (0.5 mg mL-1). Freshly mixed actoymyosin solutions were loaded onto polyethylene-glycol-passivated flowcells with a geometry of 2.5 x 2.5 x 0.1-mm3 (Supplementary Information) and sealed with either Baysilone silicone grease (Bayer, Leverkusen, Germany) or uncured PDMS (Dow Chemicals, Midland, MI, USA). The time evolution of the network structure was observed with a Nikon PlanFluor 4x objective (NA 0.13), which allows the network to fit entirely within the objective’s field of view. Page 13 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state Image Analysis. Cluster sizes were determined by a customized algorithm, implemented in MATLAB. Time-lapse images of contracting actomyosin networks were analyzed, starting from the final acquired frame. Cluster evolution, determined from Voronoi diagrams of myosin foci, was tracked by looping the algorithm backwards in time (Supplementary Information). Definition of ξ1 and ξ2. For experimental results, we measure the areas ai of the initial network that contract together, which we define as clusters. We define ξ1 as the weighted mean of cluster sizes li (square root of area), in analogy to the definition of the correlation length from percolation theory26: ξ1 := ∑i li ai2 / ∑i ai2 This length scale is dominated by the largest cluster. We furthermore define ξ2 in analogy to percolation theory: ξ2 := ∑’i li ai2 / ∑’i ai2 where ∑’i denotes summation over all clusters except for the largest cluster, as well as long edge clusters (Supplementary Information). This length scale is dominated by the second-largest cluster. For simulation results, ξ1 and ξ2 are given by the square root of the harmonic-averaged area of the largest and second largest clusters, respectively, over 10-100 disorder realizations for each set of parameters. Page 14 of 21 Molecular motors robustly drive active gels to a critically connected state Alvarado et al. References 1.&Fielding, S. M., Cates, M. E. & Sollich, P. Shear banding, aging and noise dynamics in soft glassy materials. Soft Matter 5, 2378–2382 (2009). 2.&Vissers, T., van Blaaderen, A. & Imhof, A. Band Formation in Mixtures of Oppositely Charged Colloids Driven by an ac Electric Field. Phys. Rev. Lett. 106, 228303 (2011). 3.&Weeks, E., Crocker, J., Levitt, A., Schofield, A. & Weitz, D. Three-dimensional direct imaging of structural relaxation near the colloidal glass transition. Science 287, 627–631 (2000). 4.&Corté, L., Chaikin, P. M., Gollub, J. P. & Pine, D. J. Random organization in periodically driven systems. Nat. Phys. 4, 420–424 (2008). 5.&van Hecke, M. Jamming of soft particles: geometry, mechanics, scaling and isostaticity. J. Phys.: Condens. Matter 22, 033101 (2009). 6.&Jülicher, F., Kruse, K., Prost, J. & Joanny, J.-F. Active behavior of the cytoskeleton. Phys. Rep. 449, 3–28 (2007). 7.&Zemel, A., De, R. & Safran, S. A. Mechanical consequences of cellular force generation. Curr. Opin. Solid St. M. 15, 169–176 (2011). 8.&Nédélec, F. J., Surrey, T., Maggs, A. C. & Leibler, S. Self-organization of microtubules and motors. Nature 389, 305–308 (1997). 9.&Sanchez, T., Chen, D. T. N., DeCamp, S. J., Heymann, M. & Dogic, Z. Spontaneous motion in hierarchically assembled active matter. Nature 491, 431–434 (2012). 10.&Marchetti, M. C. et al. Soft Active Matter. arXiv cond-mat.soft, (2012). 11.&Backouche, F., Haviv, L., Groswasser, D. & Bernheim-Groswasser, A. Active gels: dynamics of patterning and self-organization. Phys. Biol. 3, 264–273 (2006). 12.&Smith, D. et al. Molecular motor-induced instabilities and cross linkers determine biopolymer organization. Biophys. J. 93, 4445–4452 (2007). 13.&Soares e Silva, M. et al. Active multistage coarsening of actin networks driven by myosin motors. Proc. Natl Acad. Sci. USA 108, 9408–9413 (2011). 14.&Köhler, S., Schaller, V. & Bausch, A. R. Structure formation in active networks. Nat. Mater. 10, 462–468 (2011). 15.&Mizuno, D., Tardin, C., Schmidt, C. F. & MacKintosh, F. C. Nonequilibrium mechanics of active cytoskeletal networks. Science 315, 370–373 (2007). 16.&Koenderink, G. H. et al. An active biopolymer network controlled by molecular motors. Proc. Natl Acad. Sci. USA 106, 15192–15197 (2009). 17.&Bendix, P. M. et al. A quantitative analysis of contractility in active cytoskeletal protein networks. Biophys. J. 94, 3126–3136 (2008). 18.&Köhler, S., Bausch, A. R. & München, G. G. Contraction Mechanisms in Composite Active Actin Networks. PLOS ONE 7, 1–8 (2012). 19.&Wang, S. & Wolynes, P. G. Active contractility in actomyosin networks. Proc. Natl Acad. Sci. USA 109, 6446–6451 (2012). 20.&Wang, S. & Wolynes, P. G. Tensegrity and motor-driven effective interactions in a model cytoskeleton. J. Chem. Phys. 136, 145102 (2012). 21.&Wyart, M., Liang, H., Kabla, A. & Mahadevan, L. Elasticity of floppy and stiff random networks. Phys. Rev. Lett. 101, 215501 (2008). 22.&Broedersz, C. P., Mao, X., Lubensky, T. C. & MacKintosh, F. C. Criticality and isostaticity in fibre networks. Nat. Phys. 7, 983–988 (2011). Page 15 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state 23.&Sheinman, M., Broedersz, C. P. & MacKintosh, F. C. Nonlinear effective-medium theory of disordered spring networks. Phys. Rev. E 85, 021801 (2012). 24.&Sheinman, M., Broedersz, C. & MacKintosh, F. Actively Stressed Marginal Networks. Phys. Rev. Lett. 109, 238101 (2012). 25.&Weber, C., Schaller, V., Bausch, A. & Frey, E. Nucleation-induced transition to collective motion in active systems. Phys. Rev. E 86, 030901 (2012). 26.&Stauffer, D. & Aharony, A. Introduction To Percolation Theory. (Taylor & Francis, 1994). 27.&Clauset, A., Shalizi, C. R. & Newman, M. E. J. Power-Law Distributions in Empirical Data. SIAM Rev. 51, 661–703 (2009). 28.&Haviv, L., Gillo, D., Backouche, F. & Bernheim-Groswasser, A. A Cytoskeletal Demolition Worker: Myosin II Acts as an Actin Depolymerization Agent. J. Mol. Biol. 375, 325–330 (2008). 29.&Murrell, M. P. & Gardel, M. L. F-actin buckling coordinates contractility and severing in a biomimetic actomyosin cortex. Proc. Natl Acad. Sci. USA (2012).doi:10.1073/pnas. 1214753109 30.&Vogel, S. K., Petrasek, Z., Heinemann, F. & Schwille, P. Myosin motors fragment and compact membrane-bound actin filaments. eLife 2, (2013). 31.&Courson, D. S. & Rock, R. S. Actin Cross-link Assembly and Disassembly Mechanics for alpha-Actinin and Fascin. J. Biol. Chem. 285, 26350–26357 (2010). 32.&E Evans, K. R. Dynamic strength of molecular adhesion bonds. Biophys. J. 72, 1541 (1997). 33.&Ishikawa, R., Sakamoto, T., Ando, T., Higashi-Fujime, S. & Kohama, K. Polarized actin bundles formed by human fascin-1: their sliding and disassembly on myosin II and myosin V in vitro. J. Neurochem. 87, 676–685 (2003). 34.&Storm, C., Pastore, J. J., MacKintosh, F. C., Lubensky, T. C. & Janmey, P. A. Nonlinear elasticity in biological gels. Nature 435, 191–194 (2005). 35.&Chaudhuri, O., Parekh, S. H. & Fletcher, D. A. Reversible stress softening of actin networks. Nature 445, 295–298 (2007). 36.&Mizuno, D., Head, D., MacKintosh, F. & Schmidt, C. Active and Passive Microrheology in Equilibrium and Nonequilibrium Systems. Macromolecules (2008). 37.&Lenz, M., Thoresen, T., Gardel, M. & Dinner, A. Contractile Units in Disordered Actomyosin Bundles Arise from F-Actin Buckling. Phys. Rev. Lett. 108, 238107 (2012). 38.&Gisler, T. & Weitz, D. Scaling of the microrheology of semidilute F-actin solutions. Phys. Rev. Lett. 82, 1606–1609 (1999). 39.&Heussinger, C. Stress relaxation through crosslink unbinding in cytoskeletal networks. New J. Phys. 14, 095029 (2012). 40.&Griffith, A. A. The Phenomena of Rupture and Flow in Solids. Phil. Trans. R. Soc. Lond. A (1921). 41.&Martin, A. C., Kaschube, M. & Wieschaus, E. F. Pulsed contractions of an actin-myosin network drive apical constriction. Nature 457, 495–499 (2009). 42.&Camalet, S., Duke, T., Jülicher, F. & Prost, J. Auditory sensitivity provided by self-tuned critical oscillations of hair cells. Proc. Natl Acad. Sci. USA 97, 3183–3188 (2000). 43.&Veatch, S. L. et al. Critical Fluctuations in Plasma Membrane Vesicles. ACS Chem. Biol. 3, 287–293 (2008). 44.&Zhang, H. P., Be’er, A., Florin, E. L. & Swinney, H. L. Collective motion and density fluctuations in bacterial colonies. Proc. Natl Acad. Sci. USA 107, 13626–13630 (2010). Page 16 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state 45.&Mora, T. & Bialek, W. Are biological systems poised at criticality? J. Stat. Phys. 144, 1–35 (2011). 46.&Bialek, W. et al. Statistical mechanics for natural flocks of birds. Proc. Natl Acad. Sci. USA 109, 4786–4791 (2012). 47.&Furusawa, C. & Kaneko, K. Adaptation to Optimal Cell Growth through Self-Organized Criticality. Phys. Rev. Lett. 108, 208103 (2012). 48.&Halley, J. D. & Winkler, D. A. Critical-like self-organization and natural selection: Two facets of a single evolutionary process? BioSystems 92, 148–158 (2008). 49.&Torres-Sosa, C., Huang, S. & Aldana, M. Criticality Is an Emergent Property of Genetic Networks that Exhibit Evolvability. PLOS Comput. Biol. 8, e1002669 (2012). 50.&Bertrand, O. J. N., Fygenson, D. K. & Saleh, O. A. Active, motor-driven mechanics in a DNA gel. Proc. Natl Acad. Sci. USA 109, 17342–17347 (2012). 51.&Sedzinski, J. et al. Polar actomyosin contractility destabilizes the position of the cytokinetic furrow. Nature 476, 462–466 (2011). 52.&Martin, A. C. et al. Integration of contractile forces during tissue invagination. J. Cell Biol. 188, 735–749 (2010). 53.&Schwarz, U. & Safran, S. Elastic Interactions of Cells. Phys. Rev. Lett. 88, 048102 (2002). 54.&Schwarz, U. S. & Gardel, M. L. United we stand - integrating the actin cytoskeleton and cell- matrix adhesions in cellular mechanotransduction. J. Cell Sci. 125, 3051–3060 (2012). 55.&Gentry, B. S. et al. Multiple actin binding domains of Ena/VASP proteins determine actin network stiffening. Eur. Biophys. J. 41, 979–990 (2012). Page 17 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state Acknowledgements This work is part of the research programme of the Foundation for Fundamental Research on Matter (FOM), which is part of the Netherlands Organisation for Scientific Research (NWO). GK and JA were funded by a Vidi grant from the Netherlands Organization for Scientific Research (NWO). We thank M. Kuit-Vinkenoog, M. Preciado-López, and F.C. Tsai (AMOLF, Amsterdam, Netherlands) for help with purifications, S. Hansen and R.D. Mullins (UCSF, San Francisco, USA) for the fascin plasmid, K. Miura (EMBL, Heidelberg, Germany) for the Temporal Color Code ImageJ plugin, and C. Broedersz (Princeton University, NJ, USA) for insightful discussions. Author Contributions J.A. and G.K. designed experiments. J.A. performed experiments. M.S., A.S., and F.C.M. designed simulations. M.S. and A.S. performed simulations. All authors contributed to the writing of the paper. Additional Information The authors declare no competing financial interests. Page 18 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state Figure Captions Figure 1: Experiments with motor-driven networks show that initial connectivity controls the length scale of contraction. a: Schematic representation of the experiment. Actin filaments (black lines) are connected by crosslinks (purple circles), and myosin motors (green dumbbells) exert force dipoles (orange arrows) on actin filaments. b–d: Temporal evolution of three networks with varying amounts of fascin crosslinks (a. RC = 0.01; b. RC = 0.05; c. RC = 0.1). Actin and motor concentrations are constant: [actin] = 12 µM; RM = 0.01. Color corresponds to time according to calibration bar (b, left). Times (tstart, tend) in minutes after initiation of actin polymerization: b. (2, 20); c. (2,120); d. (1,5). Scale bar 1 mm. See Supplementary Movies 1–3. e–g: Decomposition into clusters, delimited by black lines. Color indicates the largest (blue) and the second-largest (pink) cluster, whose sizes correspond to ξ1 and ξ2 respectively. Note that (g) does not have a second-largest cluster because we exclude long edge domains from our analysis (Supplementary Figure 6). h: Dependence of ξ1 (blue circles) and ξ2 (pink triangles) on crosslink concentration (RC). Error bars denote standard errors of the mean for repeat experiments: 1, 6, 13, 14, 9, and 5 experiments for RC = 0.002, 0.005, 0.01, 0.02, 0.05, and 0.1, respectively. Inset: Predicted dependence of ξ1 and ξ2 on connection probability p according to percolation theory, given experimental parameters (Supplementary Information). Figure 2: Cluster size distributions depend on network connectivity, exhibiting power-law distributions when ξ1 ~ ξ2 ~ L. a. Scatter plot of 48 samples with different RC in ξ1-ξ2-space (see legend, top left). Boxes delimit different regimes: local contraction (ξ1 < 300 µm), critically connected (ξ1 ≥ 300 µm and ξ2 ≥ 300 µm), and global contraction (ξ1 ≥ 1500 µm and ξ2 < 300 Page 19 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state µm). Two data points with ξ2 = 0 are depicted here with ξ2 = 30 µm. b: Histogram (circles) and complementary cumulative probability distribution (solid lines) of cluster areas, a / µm2, for the three regimes. For the critically connected regime, data across more than two orders of magnitude (red circles) are statistically consistent with a power-law distribution (solid red lines) with an exponent of –1.91 ± 0.06, p = 0.52, where p > 0.1 indicates plausible agreement with a power law (Supplementary Information). Note that the slope of the complementary cumulative probability distribution is equal to one plus the slope of the histogram because the histogram is the absolute value of the derivative of the complementary cumulative probability distribution. Figure 3: Simulations show that motors can drive initially well-connected networks to a critical state. a: Schematic representation of the simulation. A triangular lattice of nodes, connected by line segments (black lines), contains an average of N crosslinks per node (purple circles). During the course of the simulation, pairs of nodes experience contractile forces (orange arrows) and move in response to these forces. b: Temporal evolution of a representative network in the absence of remodeling. c: Motors cause network restructuring by generating tension T on crosslinks that increases the off-rate koff. d-j: Simulated networks exhibit behavior consistent with experiment. See Supplementary Movies 4–6. d,f,g: Temporal evolution of three networks differing in initial connectivity: a. c = 0.025; b. c = 3; c. c = 10000. Force is constant: f / k = 50. Color corresponds to simulation time according to calibration bar (d, left). Box size L is 100 times longer than the initial lattice size l0. e,g,i: Decomposition into clusters, shaded by pastel colors. Bold color indicates the largest (blue) and the second-largest (pink) clusters, whose sizes correspond to ξ1 and ξ2 respectively. j: Dependence of ξ1 (blue circles) and ξ2 (pink triangles) on crosslink concentration c across repeat simulations. Open symbols indicate values at t = 0, which corresponds to passive networks described by classical percolation theory. Closed symbols Page 20 of 21 Alvarado et al. Molecular motors robustly drive active gels to a critically connected state indicate values at the end of the simulation, after the network has broken up into clusters. Yellow regions correspond to values of c for which ξ2 > L / 10 and the cluster size distribution exhibits a power-law. Note that this region is narrow for classical percolation theory (diagonal yellow stripes) but broadens substantially in response to active internal driving (solid yellow box). Figure 4: Simulation and experiment both show that increased motor force reduces cluster size. a,b: Temporal evolution of two simulated networks with constant network connectivity (c = 3) but with either (a) low force, f / k = 3; or (b) high force, f / k = 150. Color corresponds to simulation time. c,d: Decomposition into clusters. e: Dependence of ξ1 (blue circles) and ξ2 (pink triangles) on force f. f,g: Temporal evolution of two experimentally prepared networks with (f) low myosin concentration, RM = 0.005; or (g) high myosin concentration, RM = 0.02. Color corresponds to time. Times (tstart, tend) in minutes after initiation of actin polymerization: a. (2, 43); b. (2,14). The network connectivity is the same in both cases ([actin] = 12 µM, RC = 0.02). See Supplementary Movies 8–10. h,i: Decomposition into clusters. j: Dependence of ξ1 (blue circles) and ξ2 (pink triangles) on myosin concentration, given by RM. Scale bars 1 mm. Error bars denote standard errors of the mean for repeat experiments: 5, 14, and 5 experiments for RM = 0.005, 0.01, and 0.02, respectively. Dashed lines depict f-1 (panel e) and RM-1 (panel j). Figure 5: The critically connected regime broadens with increasing force. a: Schematic of proposed phase diagram in force-connectivity space, where the critically connected regime separates the local contraction and global contraction regimes. b, c: Dependence of ξ1 (b) and ξ2 (c) simulated over a broad range of force and connectivity. Page 21 of 21 Molecular motors robustly drive active gels to a critically connected state José Alvarado1, Misha Sheinman2, Abhinav Sharma2, Fred MacKintosh2, Gijsje Koenderink1 S U P P L E M E N T A R Y I N F O R M A T I O N Methods Protein preparation. Monomeric (G-) actin and myosin II were purified from rabbit psoas skeletal muscle1. G-actin was purified with a Superdex 200 column (GE Healthcare, Waukesha, WI, USA) and stored at −80 °C in G-buffer (2 mM tris-hydrochloride pH 8.0, 0.2 mM disodium adenosine triphosphate, 0.2 mM calcium chloride, 0.2 mM dithiothreitol). Myosin II was stored at −20 °C in a high-salt storage buffer with glycerol (25 mM monopotassium phosphate pH 6.5, 600 mM potassium chloride, 10 mM ethylenediaminetetraacetic acid, 1 mM dithiothreitol, 50% w⁄w glycerol). Creatine phosphate disodium and creatine kinase were purchased from Roche Diagnostics (Indianapolis, IN, USA), all other chemicals from Sigma Aldrich (St. Louis, MO, USA). Magnesium adenosine triphosphate was prepared as a 100 mM stock solution using equimolar amounts of disodium adenosine triphosphate and magnesium chloride in 10 mM imidazole pH 7.4. Sample Preparation. Fresh myosin solutions were prepared by overnight dialysis into myosin buffer (20 mM imidazole pH 7.4, 300 mM potassium chloride, 4 mM magnesium 1 FOM Institute AMOLF, Science Park 104, 1098 XG Amsterdam, The Netherlands 2 Department of Physics and Astronomy, Vrije Universiteit, 1081 HV Amsterdam, The Netherlands Alvarado et al. SUPPLEMENTARY INFORMATION chloride, 1 mM dithiothreitol) and used within four days. All frozen protein stocks (actin, myosin, fascin) were clarified of aggregated proteins upon thawing at 120,000 g for at least 5 min and used within four days. The proteins’ concentrations in the supernatant were determined by measuring the solution absorbance at 280 nm with a NanoDrop 2000 (ThermoScientific, Wilmington, DE, USA) and using extinction coefficients, in M-1 cm-1, of 26600 (actin2), 249000 (myosin3) and 66280 (fascin, computed from amino acid sequence4). Fluorescently labeled proteins were mixed with unlabeled proteins to yield a 10% molar ratio of dye to protein. During sample preparation, myosin and Alexa-488-myosin were mixed at high salt and then mixed into a tube containing fascin and buffer. This solution was mixed into a second tube containing actin and Alexa-594-actin to initiate polymerization and immediately inserted into glass flowcells pass iva ted by po ly-L- lys ine-po lye thy lene-g lyco l (Surface So lu t ions AG , Dübendorf , Switzerland). Preparation of flow cells. Glass flow cells were assembled by sandwiching strips of ParaFilm between a long cover slip (24 mm x 60 mm) and 2.5-mm-narrow glass strips which were manually cut from 40-mm-long cover slips. This yielded 2.5 x 2.5 x 0.1-mm3-large chambers (corresponding to ~0.6 µL). All glass was cleaned with piranha solution, rinsed in MilliQ water, and stored in isopropanol. Assembled flow cells were then passivated by applying 1M potassium hydroxide for 5 min, rinsing with MilliQ, drying with N2, applying 0.2 mg mL-1 poly-L-lysine-polyethylene-glycol (Surface Solutions AG, Dübendorf, Switzerland) for 30 min, rinsing with MilliQ, and drying with N2. Simulation. The values taken for the simulations are: the system size W=100, koff,0=10, T0=1, kon d/ µ=10, k=1. The buckling is implemented by vanishing force of a bond for a compression strain below 0.1. The stiffening is implemented by increase of the stretching constant by 100-fold for extension strain above 0.2. Alvarado et al. SUPPLEMENTARY INFORMATION Algorithm for determining cluster size We developed a MATLAB algorithm to determine sizes of contracting clusters from time- lapse images of contractile actomyosin networks. Actin filaments and myosin motors were fluorescently labeled to appear in separate channels. In short, this technique begins with the final frame of acquisition (Fig 5a), determines clusters, and tracks the expansion of these clusters back in time until the first frame of acquisition. The result is a decomposition of the initial network into clusters. Step 1: the final acquired image from the actin channel (Fig 5b) was median-filtered with a radius of 1px (Fig 5c). This step filters out noise. Step 2: the median-filtered image was thresholded using Otsu’s method5 (Fig 5d). The result of this step is a binary image of only black or white pixels. Contiguous groups of white pixels are called connected components. Each connected component corresponds to a cluster of actin and myosin. Step 3: the final acquired image from the myosin channel (Fig 5e) was also median filtered with a radius of 1px (Fig 5f). Step 4: the median-filtered image was thresholded using Otsu’s method (Fig 5g), again yielding a black-and-white image of connected components that correspond to clusters of actin and myosin. Step 5: the thresholded image was morphologically opened (successive dilation and erosion) using a 1-px-radius-disk as a structuring element (Fig 5h). This step serves as an additional filter, removing connected components smaller than the structuring element. Step 6: connected components from step 5 were assigned to connected components from step 2 (Fig 5i). Note that the connected components from step 2 (actin) are usually large, and Alvarado et al. SUPPLEMENTARY INFORMATION contain many smaller connected components from step 5 (myosin). Without this step, the disjoint connected components from step 5 could be erroneously treated as separate clusters. Step 7: domains were defined around each cluster, using the MATLAB function bwdist, which performs a distance transform (Fig 5j). This step decomposes the entire image into a Voronoi-like diagram, where domain boundaries occur halfway between connected components. Step 8: steps 3-7 were repeated for the myosin channel, starting with the final acquired frame (Fig 5k), looping through successive acquired images backwards in time (Fig 5l), and finally arriving at the first acquired frame (Fig 5m). The end result is the first acquired frame, where the actin and myosin signals are uniformly distributed, and decomposed into clusters. During the loop, steps 3-7 were unchanged, except for step 6: myosin connected components were joined not by using actin connected components, but by the domains from the previous iteration of step 7. Step 9: finally, the image of domains produced from step 7 of the final loop was cropped to the largest rectangle contained by the network. In two cases, adjustments to this routine were necessary. In one case, during step 2, Otsu’s threshold sometimes yielded large connected components that spanned the image (Fig 6a). This resulted in the network being erroneously represented as one large cluster (Fig 6b). This artifact was eliminated by choosing a more restrictive threshold (Fig 6c), which resulted in accurate domains (compare panels d and e). In another case, during step 5, morphological opening sometimes filtered out small, dim clusters (Fig 6f). This led to their corresponding domains to disappear, and small, neighboring clusters were reported as bigger clusters (Fig 6g). This artifact was eliminated by omitting step 5 (Fig 6h), yielding accurate domains (compare panels i and j). Alvarado et al. SUPPLEMENTARY INFORMATION Adjustment of domains We performed two types of adjustments to the cluster decompositions produced by our algorithm. First, we removed long edge domains from our analysis. These domains could be the result of enhanced interactions with the edge of the confining geometry, in addition to internal driving by myosin activity. We first search for domains that touch the border of the cropping rectangle (Fig 5m, dashed line). We next compute the major and minor axis of the ellipse that has the same normalized second central moment of each domain, as well as the orientation of the major axis. Finally, we consider edge domains which satisfy the following two conditions: (i) the major-axis-to-minor-axis ratio is greater than two; (ii) the major axis is oriented along the edge that the domain touches to within 45º. (Condition ii is dispensed for corner domains that touch two edges.) Edge domains that satisfy these two conditions are then omitted from the sums in the definition of ξ1 and ξ2, as well as when plotting distributions. Second, we compensated for fast clusters. Sometimes the displacement of a cluster between two successive frames was greater than the half-way distance to a neighboring cluster. Clusters would then leave their own domain and erroneously enter neighboring domains. This artifact mostly affected networks in the global contraction regime, where the global build-up of stress led to fast relaxation events. This artifact cannot be addressed by modifying the algorithm. We therefore manually corrected contraction domains to accurately reflect network evolution. A total of five corrections were performed, all of which are reported in Fig 7. These two adjustments to domain size affect our results for the global contraction regime. This is evident by inspecting the effect of the two adjustments on cluster size distributions (Fig 8a). However, the local contraction and critically connected regimes are largely unaffected. The power-law exponents determined from experiment are robust to the two adjustments described above (Fig 8b). Alvarado et al. SUPPLEMENTARY INFORMATION Statistical analysis of domain sizes In order to determine whether cluster size distributions were consistent with a power law, we employed a recently developed, rigorous statistical analysis6. This technique first fits observed data to a power law, determining both the best exponent and best lower cut-off. The lower cut-off is the minimum value above which the power law is fitted. It then compares the observed dataset with multiple synthetic datasets (generated from a power-law distribution using the best fit parameters) by computing the Kolmogorov-Smirnov statistic, which quantifies the “distance” between a dataset and the true power-law distribution. Finally, it computes a p-value, which is defined as the fraction of synthetic datasets whose distance is greater than the observed dataset. Therefore, larger p-values correspond to an increased likelihood that the observed dataset is consistent with a power law. A power law can be ruled out for p < 0.1. Percolation model Our model is based on three-dimensional network of N straight filaments of length L placed in a W x W x W box. The filaments are placed such that their position and orientation is uniformly distributed. Two filaments are considered to be intersecting if the shortest distance between them is less than a certain value which is taken to be of the order of size of the cross link. At this intersection these two filaments can be connected by a freely hinging crosslink. The probability that such a crosslink exists is denoted by p. Periodic boundary conditions are assumed in all directions. The line density NL/W3 is obtained from the experiments, and estimated to be ~20 µm-2. Our simulations show that the connectivity percolation occurs in the vicinity of p=0.33. Alvarado et al. References SUPPLEMENTARY INFORMATION 1.#Soares e Silva, M. et al. Active multistage coarsening of actin networks driven by myosin motors. Proc. Natl Acad. Sci. USA 108, 9408–9413 (2011). 2.#Pardee, J. D. & Spudich, J. A. Purification of muscle actin. Method. Enzymol. 85 Pt B, 164– 181 (1982). 3.#Margossian, S. S. & Lowey, S. Preparation of myosin and its subfragments from rabbit skeletal muscle. Method. Enzymol. 85 Pt B, 55–71 (1982). 4.#Artimo, P. et al. ExPASy: SIB bioinformatics resource portal. Nucleic Acids Res. 40, W597– W603 (2012). 5.#Otsu, N. A threshold selection method from gray-level histograms. IEEE Trans. Syst., Man, Cybern., Syst. 9, 62–66 (1975). 6.#Clauset, A., Shalizi, C. R. & Newman, M. E. J. Power-Law Distributions in Empirical Data. SIAM Rev. 51, 661–703 (2009). Alvarado et al. Figure Captions SUPPLEMENTARY INFORMATION Figure 1: Confocal image of actin (red) and myosin (cyan) in the absence of crosslinks ([actin] = 12 µM, RM = 0.01, t ~ 2 h after the initiation of actin polymerization). Myosin motors form small foci, which are separated approximately 30 µm apart. Figure 2: Simulated histogram (open circles) and corresponding cumulative probability distribution (closed circles) of cluster areas, a / l02, for the three conditions shown in the main text, Fig 3d-f. Solid red lines denote a power law with exponent −2.0. Figure 3: Increased myosin activity results in smaller clusters. a: Scatter plot of samples with different RM in ξ1-ξ2-space (see legend, top left). Crosslink concentration is constant (RF = 0.02). Boxes delimit different regimes: local contraction (ξ1 < 300 µm), critically connected (ξ1 ≥ 300 µm and ξ2 ≥ 300 µm), and global contraction (ξ1 ≥ 1500 µm and ξ2 < 300 µm). b: Histogram (circles) and corresponding cumulative probability distribution (solid lines) of cluster areas, a / µm2, for the three regimes. For the critically connected regime, data across more than two orders of magnitude are consistent with a power-law distribution with an exponent of –1.90 ± 0.06 (p = 0.12, amin = (20 ± 8) 103 µm2). Figure 4: Bent filaments induce tension on crosslinks. a: Two straight filaments (black lines) are crosslinked (purple circle) at their intersection. If forces (orange arrows) are balanced, the crosslink experiences zero tension. This is evident because if the crosslink unbinds (right), no relaxation occurs. b: A straight filament and a bent filament are crosslinked. Although forces are balanced, the crosslink here experiences tension. This is evident because if the crosslink unbinds (right), the bent filament relaxes to a straight conformation. Alvarado et al. SUPPLEMENTARY INFORMATION Figure 5. Cluster-size algorithm. See Supplementary Movie 11. a: Final image of a time-lapse acquisition of a contractile actomyosin network (t = 80 min). Alexa-594-actin is shown in red, DyLight-488-myosin in green. [actin] = 12 µM, RF = 0.02, RM = 0.02. b: Close-up of the actin channel, corresponding to dashed box, a. c: Median filter of b. d: Otsu threshold of c. e: Close- up of the myosin channel, corresponding to dashed box, a. f: Median threshold of e. g: Otsu threshold of f. h: Morphological opening of g. i: Superposition of d (red) and h (green). Note that some disjoint green clusters are contained within one large red cluster. In this case, they are treated as one large cluster. j: Superposition of original myosin signal (white) with domains (shades of beige), which result from a distance transform, implemented in MATLAB as the bwdist function. k: Myosin signal (white) and domains (shades of beige) for the entire sample before looping the algorithm, beginning with the final frame (t = 80 min). l: Close-up of k (dashed box) at three representative stages of loop progression as clusters expand in -t (from left to right: t = 20 min, 8 min, 4 min). m: Myosin signal and resulting clusters for the first acquired frame (t = 2 min). Note that myosin is distributed homogeneously across the entire network at the beginning of acquisition, and is fully decomposed into clusters. n: Clusters (outlined in black) after cropping to dashed box, k, with largest and second-largest clusters denoted in blue and pink, respectively. Scale bars: a: 1 mm, b-j: 200 µm, k: 1 mm, l: 500 µm, m: 1 mm. Figure 6. Two modifications of the cluster-size algorithm. a-e: First modification: skipping step 2 when it erroneously yields large, system-spanning connected clusters. a: Actin channel of original image (RF = 0.01, RM = 0.01, local contraction regime). b: Result of step 2. Note that the thresholded image does not resemble the individual clusters visible in the original image. c: Result of continuing the algorithm, which erroneously represents the sample as one large cluster. d: Result of the algorithm, skipping steps 1 and 2. Note that this image correctly represents Alvarado et al. SUPPLEMENTARY INFORMATION individual clusters. e: Overlay of acquired data, where color corresponds to time (calibration bar, top right). tstart = 1 min; tend = 20 min. Note that this image qualitatively captures cluster evolution, and is obtained independently of the cluster-size algorithm. Comparing to panels c and d shows that panel d more accurately represents true cluster size. f-l: Second modification: skipping step 5 when it removes small, dim myosin clusters. f: Result of step 4 (RF = 0.01, RM = 0.01, local contraction regime). g: Close-up of f, green dashed box. h: Result of step 5. i: Close- up of h, green dashed box. Note that morphological opening removes very small clusters. j: Result of continuing the algorithm, which erroneously joins together many small clusters in one large cluster. k: Result of the algorithm, skipping step 5. Note that this image correctly represents individual clusters. l: Time overlay, as in e. tstart = 2 min; tend = 30 min. Comparing to panels j and k shows that panel k more accurately represents true cluster size. Figure 7: Manual correction of five experiments. True sample dynamics is depicted in the time overlay (first column). For these five experiments, the algorithm produces excessively large domains (second column). Upon careful visual inspection of the original data, erroneous domains were manually corrected to their apparent true size (third column). Scale bar 1 mm. Figure 8: Results from modifying algorithm output. a: Distributions of cluster sizes (RM = 0.01) that result when either removing long edge domains (rows) or manually correcting domains (columns). Distributions are divided according to global contraction (top), critically connected (center), and local contraction (bottom) (see main text, Figure 2). Movie Captions Movie 1: Experiment depicting the time-lapse of a weakly connected network (RF = 0.01, RM = 0.01). The initially homogeneous actomyosin network breaks up into small clusters. Actin is Alvarado et al. SUPPLEMENTARY INFORMATION shown in red, and myosin is shown in cyan (white denotes overlap). Time after initiation of actin polymerization is shown at the top-right corner in hours:minutes:seconds. Movie 2: Experiment depicting the time-lapse of a medially connected network (RF = 0.02, RM = 0.01). The initially homogeneous actomyosin network breaks up into large and small clusters. Actin is shown in red, and myosin is shown in cyan (white denotes overlap). Time after initiation of actin polymerization is shown at the top-right corner in hours:minutes:seconds. Movie 3: Experiment depicting the time-lapse of a strongly connected network (RF = 0.01, RM = 0.01). The initially homogeneous actomyosin network contracts into one large cluster. A slower, secondary contraction subsequently occurs. Actin is shown in red, and myosin is shown in cyan (white denotes overlap). Time after initiation of actin polymerization is shown at the top-right corner in hours:minutes:seconds. Movie 4: Custom MATLAB algorithm for determining cluster size. Sample depicted corresponds to Movie 11. White pixels correspond to myosin clusters. Beige regions correspond to domains around myosin clusters. The algorithm tracks cluster evolution backwards in time, yielding the original network, decomposed into clusters. Movie 5: Experiment depicting a typical rupture event (RF = 0.02, RM = 0.01). Myosins exert forces that rupture the actin network, decreasing connectivity. Actin is shown in red, and myosin is shown in cyan (white denotes overlap). Time after initiation of actin polymerization is shown at the top-right corner in hours:minutes:seconds. Alvarado et al. SUPPLEMENTARY INFORMATION Movie 6: Simulation depicting the time-lapse of a locally contracting network (c = 0.025). The initially well-connected network breaks up into many small clusters. Nodes of a triangular lattice representing crosslink points (black dots) are connected by line segments representing actin filaments (red lines). Movie 7: Simulation depicting the time-lapse of a critically connected network (c = 3). The initially well-connected network breaks up into small and large clusters. Nodes of a triangular lattice representing crosslink points (black dots) are connected by line segments representing actin filaments (red lines) Movie 8: Simulation depicting the time-lapse of a locally contracting network (c = 10000). The initially well-connected network breaks primarily due to large crack. Nodes of a triangular lattice representing crosslink points (black dots) are connected by line segments representing actin filaments (red lines) Movie 9: Experiment depicting the time-lapse of a network with very few motors (RF = 0.02, RM = 0.002). The actomyosin network does not contract. Rather, it exhibits slow rearrangements on macroscopic length scales, reminiscent of breathing, and indicative of a well-connected network. Actin is shown in red, and myosin is shown in cyan (white denotes overlap). Time after initiation of actin polymerization is shown at the top-right corner in hours:minutes:seconds. Movie 10: Experiment depicting the time-lapse of a network with few motors (RF = 0.02, RM = 0.005). The actomyosin network contracts into one large cluster. Actin is shown in red, and myosin is shown in cyan (white denotes overlap). Time after initiation of actin polymerization is shown at the top-right corner in hours:minutes:seconds. Alvarado et al. SUPPLEMENTARY INFORMATION Movie 11: Experiment depicting the time-lapse of a network with many motors (RF = 0.02, RM = 0.02). The initially homogeneous actomyosin network breaks apart into clusters. Actin is shown in red, and myosin is shown in cyan (white denotes overlap). Time after initiation of actin polymerization is shown at the top-right corner in hours:minutes:seconds.
1211.4366	1	1211	2012-11-19T11:37:58	The Physics of Life: one molecule at a time	[ "physics.bio-ph", "q-bio.BM", "q-bio.SC" ]	The esteemed physicist Erwin Schroedinger, whose name is associated with the most notorious equation of quantum mechanics, also wrote a brief essay entitled 'What is Life?', asking: 'How can the events in space and time which take place within the spatial boundary of a living organism be accounted for by physics and chemistry?' The 60+ years following this seminal work have seen enormous developments in our understanding of biology on the molecular scale, physics playing a key role in solving many central problems through the development and application of new physical science techniques, biophysical analysis and rigorous intellectual insight. The early days of single molecule biophysics research was centred around molecular motors and biopolymers, largely divorced from a real physiological context. The new generation of single molecule bioscience investigations has much greater scope, involving robust methods for understanding molecular level details of the most fundamental biological processes in far more realistic, and technically challenging, physiological contexts, emerging into a new field of 'single molecule cellular biophysics'. Here, I outline how this new field has evolved, discuss the key active areas of current research, and speculate on where this may all lead in the near future.	physics.bio-ph	physics	Philosophical Transactions B. The Physics of Life: one molecule at a time Mark C. Leake1, 2 * 1Clarendon Laboratory, Dept of Physics, Parks Road, Oxford University, Oxford OX1 3PU, UK, 2Dept of Biochemistry South Parks Road Oxford, OX1 3QU, UK. *Correspondence: [email protected] Received Philtrans B 28 August, 2012. Accepted Philtrans B 13 September, 2012. Contributed review to theme issue of PhiltransB “Single molecule cellular biophysics ” for print publication spring 2013. ABSTRACT The esteemed physicist Erwin Schr ödinger, whose nam e is associated with the most notorious equation of quantum mechanics, also wrote a brief essay entitled “What is Life?”, asking: “How can the even ts in space and time which take place within the spatial boundary of a living organism be accounted for by physics and chemistry?” The 60+ years following this seminal work have seen enormous developments in our understanding of biology on the molecular scale, physics playing a key role in solving many central problems through the development and application of new physical science techniques, biophysical analysis and rigorous intellectual insight. The early days of single molecule biophysics research was centred around molecular motors and biopolymers, largely divorced from a real physiological context. The new generation of single molecule bioscience investigations has much greater scope, involving robust methods for understanding molecular level details of the most fundamental biological processes in far more realistic, and technically challenging, physiological contexts, emerging into a new field of “single molecule cellular biophysics ”. Here, I outl ine how this new field has evolved, discuss the key active areas of current research, and speculate on where this may all lead in the near future. 1. INTRODUCTION Richard Feynman, celebrated physicist and bongo-drum enthusiast, gave a lecture in 1959 viewed by nanotechnologists of the future as a prophecy imagining perfectly their own field. The title was “There ’s plenty of room at the bottom ”, and it discussed a potential future to con trol and manipulate machines and store information and on a length scale tens of thousands times smaller than that of the everyday “macroscopic ” wor ld [1]. It was a clarion call to engineers and scientists to establish a new discipline, later coined nanotechnology [2]. Feynmann alluded to this small scale as relevant to that of biological systems, and how cells could function at this scale to perform “all kinds of marvelous things ”. We now know that this f undamental minimal unit is the single biological molecule. It’s not to say tha t atoms comprising these molecules do not matter, nor sub-atomic particles that make up the individual atoms, nor smaller still the quarks of which the sub-atomic particles are composed. The point is, in general, we do not need to refer to length scales smaller than single molecules to understand most biological processes. Technological developments in experimental biological physics have been the primary driving force in establishing the field of single molecule biophysics, and even though the discipline in its modern form is only a human generation in age it is clear that at the often prickly interfaces between the physical and the life sciences, and at scale of the single biological molecule, many of the most fundamental questions concerning cellular systems are being addressed. This field is evolving into a new discipline of single molecule cellular biophysics [3]. It is manifest not only in investigations at the single molecule level using live cells as the test system, i.e. in vivo single molecule studies, but also by some highly ingenious single molecule studies in vitro that, although divorced from the native physiological context, have a very high level of complexity either in the make up of the experimental components studied or in the combinatorial single molecule biophysics methods used, which greatly enhance the physiological relevance of the data obtained. 2. THE ESTABLISHMENT OF SINGLE MOLECULE BIOPHYSICS (a) Why bother with single molecules? An experimental method which utilises single molecule biophysics gives us information on the position of a biomolecule in space at a given time or will allow the control and/or measurement of forces exerted by/on that molecule [4], or sometimes both. However, these approaches, despite being established for over two decades in dedicated scientific research laboratories around the world, are still technically challenging since they operate in a regime dominated by stochastic thermal fluctuations of water solvent molecules whose characteristic energy scale, that of kBT where kB is Boltzmann ’s constant and T the absolute temperature measured in Kelvin, is comparable to energy transitions involved in molecular processes in biology. Forces are characterized by the piconewton (pN) scale, and the length scale of molecules and complexes is of the order of a few nanometres (nm), two orders of magnitude smaller than the wavelength of visible light (figure 1). Why should we wish to perform such experiments which, as a rule, require measurements of tiny signals in environments of significant noise, in all but rare cases suffering from poor yields and, traditionally, being not remotely “high-throughput”? There already exist man y robust bulk ensemble average biophysical methods which illuminate several aspects of structure and function of cellular systems using well-characterized experimental apparatus [5, 6], with an effect of averaging over copious molecular events, typically resulting in low measurement noise. The principal reason for using novel physical methods and analyses for studying biological processes at the level of single molecules is the prevalence of molecular heterogeneity. One might suppose that the mean average property of ~1019 molecules (roughly the number of molecules in 1 ml of water, equivalent to 1/(18x1000)th of a mole), as is the case for most bulk ensemble average techniques, is an adequate representation of the properties of any given single molecule. In some exceptional biological systems this is true, however, in general this is not the case. This is because single biological molecules usually exist in multiple states, intrinsically related to their biological functions. A state here is a measure of the energy locked in to that molecule. For example, there are many molecules which exist in multiple spatial conformations, such as molecular motors, with each conformation having a characteristic energy state. Although there may be a single conformation which is more stable than the others for these tiny molecular machines, several shorter-lived conformations still exist which are utilized in different stages of motion and force generation. The mean conformation would look something close to the most stable of these many different conformations, but this single average parameter does not tell us a great deal about the behaviour of the other shorter-lived but essential states. Bulk ensemble average analysis, irrespective of what experimental property is measured, can not probe multiple states in a heterogeneous molecular system. Also, temporal fluctuations in the molecules from a population result in broadening the distribution of a measured parameter from a bulk ensemble experiment which can be difficult to interpret physiologically. These thermal fluctuations are driven by collisions from the surrounding water molecules (~109 per second - biological molecules are often described as existing in a thermal bath) which can drive biological molecules into different states. In an ensemble experiment this may broaden the measured value, making reliable inference difficult. In single molecule measurements these states can often be probed individually. Furthermore, there is a danger of lack of synchronicity in ensemble experiments. The issue here is that different molecules within a large population may be doing different things at different times, molecules may for example be in different conformations at a given time, so the average snapshot from the large population encapsulates all such temporal fluctuations resulting in a broadening of the distribution of any molecular parameter being investigated. The root cause of molecular asynchrony is that in most ensemble experiments the population is in steady-state, that is the rate of change between forward and reverse molecular states is identical. If the system is briefly taken out of equilibrium then transient molecular synchrony can be obtained, such as by forcing all molecules into just a single state, however this by definition is a transient effect so practical measurements are likely to be short-lived and technically challenging. These molecular-synchronizing methods include chemical and temperature jumps such as in stopped-flow reactions, electric and light field methods to align molecules, as well as freezing a population or causing it to form regular crystals. A danger with such approaches is that the normal physiological function may be different. Some biological tissues, for example cell membranes and muscles, are naturally ordered on a bulk scale and so these have historically generated the most physiologically relevant ensemble data. The real strength of single molecule biophysics experiments is that these sub-populations of molecular states can be investigated. The importance to biology is that this multiple-state heterogeneity is actually an essential characteristic of the normal functioning of molecular machines; there is a fundamental instability in these molecules which allows them to switch between multiple states as part of their underlying physiological function. A final point to note is that, although there is a wide range in concentration of biological molecules inside living cells, the actual number of molecules that are directly involved in any given biological process at any one time is generally low. Biological processes at this level can therefore be said to occur under minimal stoichiometry conditions in which just a few stochastic molecular events become important. In fact, it can often be these rarer, single molecule events that may be the most significant to cellular processes, and so it becomes all the more important to investigate life at the level of single molecules, and many approaches developed from the physical sciences have now been established focussed upon using single molecule biophysics techniques to address fundamental biological questions [7]. (b) The first generation of single molecule biophysics investigations Single molecule biophysics is still a youthful field, in the context of the traditional “core ” sciences. The first definitive s ingle biological molecule investigations used pioneering electron microscopy techniques to produce metallic shadow replicas of large, filamentous molecules including DNA and a variety proteins [8], using fixed samples in a vacuum. Single particle detection began in non-biological samples, involving trapping single elementary particles in a gaseous-phase in the form of a single electron [9], and later as a single atomic ion [10]. The first single molecule biophysics investigation in which the surrounding medium included that one compound essential to all known forms of life, namely water, came with the fluorescence imaging in the lab of Boris Rotman in 1961 with the detection of single molecules of the enzyme b- galactosidase by chemically modifying one of its substrates to make it fluorescent, and observing the emergence of these molecules during the enzyme-catalysed reaction inside microscopic droplets [11] - although the sensitivity of detection at that time was not sufficiently high to monitor single fluorescent molecules directly, this particular assay utilised the fact that a single molecule of the b-galactosidase enzyme could generate several thousand substrate molecules which could be detected and thereby indicate the presence of a single enzyme. Comparable observations were made lab of Thomas Hirshfeld over a decade later in aqueous solution without the need for microdroplets using the organic dye fluorescein, similar in structure to the fluorogenic component in the 1961 Rotman study, attached via antibodies to single globulin protein molecules, each with 80-100 individual fluorescein molecules bound [12]. The decade that followed involved marked developments in measurement sensitivity, including fluorescence detection of single molecules of a liquid-phase solution of the protein phycoerythrin labelled with ~25 molecules of the orange organic dye rhodamine [13], as well as parallel developments in the detection of single molecules in solids using optical absorption of a non-biological sample [14]. The seminal single molecule biophysics work that came in the subsequent decade involved in vitro studies, experiments done, in effect, in the test tube. In the first instance, these investigations were driven by developments in a newly established technique of optical trapping, also known as laser or optical tweezers. The ability to trap particles using laser radiation pressure was reported by Arthur Ashkin, forefather of optical trapping, as early as 1970 [15], though the modern form which results in a net optical force on refractile/dielectric particles of higher refractive index than the surrounding medium roughly towards the intensity maximum of a focussed laser (figure 2a-c), was developed in the early 1980s by Ashkin and co- workers [16], and these optical force-transduction devices have since been applied with great diversity to study single molecule biophysics [17, 18]. Arguably, the key pioneering biophysical investigation involving optical trapping used only a relatively weak optical trap in combination with a very sensitive sub-nm-precise detection technique called back focal plane interferometry [19], with micron-sized beads conjugated to molecules of the motor protein kinesin to monitor the displacement of single kinesin motors on a microtubule filament track, which indicated quantized stepping of each motor of a few nm consistent with the structural periodicity of kinesin binding sites on the microtubule [20]. This was followed by a study on another molecular motor of a type of myosin protein which was implicated in the generation of force during muscle contraction in its interaction with F-actin filaments [21].This investigation utilised two independent optical traps to tether a single filament and lower it onto a third, surface-immobilized, bead which had been functionalized with the “motor-active” par t of the myosin molecule. This was the first study to clearly measure both the quantized nature of displacement and force of a single molecular motor to nm/pN precision. Biopolymer molecules were also the source of seminal single molecule biophysics investigations, using optical trapping to measure the mechanical molecular properties by stretching molecules and observing how the forces that developed changed with end-to-end displacement. These were applied to both single and double-stranded DNA [22] and RNA [23] nucleic acids (the latter study also investigating folding/unfolding transitions in the model RNA hairpin structural motif), as well as large modular proteins made up of repeating motifs of either the immunoglobulin or fibronectin family including many proteins related to the class of giant muscle proteins known as titins [24, 25, 26, 27]. A complementary technique of AFM force spectroscopy also emerged at around the same time. Surface probe techniques originated through the seminal work of Gerd Binning using the scanning tunnelling microscope (STM) [28] that measured electron tunnelling between a sample surface and micron-sized probe tip (a quantum mechanical effect whose probability depended exponentially on the tunnelling distance involved) as a measure of the surface topography. This developed into atomic force microscopy (AFM) [29], in which a similar probe tip, typically composed of silicon nitride, detects primarily Van der Waals forces from a sample surface, allowing imaging of surface topography to sub-nm precision. AFM force spectroscopy instead of imaging the surface uses a probe tip as a fishing-rod to clasp ends of molecules bound to gold-coated surface, and subsequently stretch them in retracting the tip away from the surface. This approach was used on modular protein constructs of titin to demonstrate forced unfolding of individual immunoglobulin modules. In doing so, this seminal paper showed evidence for a single molecule “signature ” - a physical meas urement indicating that there really is a single molecule under investigation, as opposed to multiples or noise, and in the case of AFM force spectroscopy this signature was a characteristic “sawtooth” pattern of the molecular force-extension trace that indicated dramatic changes in molecular extension of ~20-30 nm whenever one of the immunoglobulin modules made a forced transition from folded to unfolded conformations [30]. Developments in optical imaging, most importantly fluorescence microscopy, had an enormous impact on pushing single molecule biophysics forward. These have included molecular interaction methods using single molecule F örster resonance energy transfer (smFRET) in which energy can be transferred non-radiatively between differently coloured donor and acceptor dye molecules, each designed to be attached to biological structures which transiently interact as part of their biological function. FRET occurs provided there is suitable spectral overlap between the emission and absorption spectra, and the two molecules are both oriented appropriately and within less than ~10 nm of each other. The first clear report of smFRET measurements involved monitoring single molecule assembly of the DNA double helix [31]. Fluorescence imaging was also applied to monitor rotation of single molecules of the rotary motor F1-ATPase by attachment of a rhodamine- tagged fluorescent filament of F-actin conjugated to the F1-ATPase rotor subunit, which demonstrated clear rotation of this vital biological machine responsible for the generation of the universal cellular fuel ATP, but also showed the motion occurs in quantized angular units mirroring the symmetry of the enzyme’s atomic structure [32]. In another pioneering study, single molecule fluorescent dye imaging was used to monitor the movement of tagged myosin molecules to show that they travelled along F-actin tracks in a hand-over-hand mechanism. This was the first study to show unconstrained walking of a single molecular motor, using nm-precise localization in the form of Gaussian fitting of the “point spread function” image of each single fluorescent d ye molecule, which the investigators denoted as fluorescence imaging with one nanometre accuracy, or FIONA [33]. A seminal in vitro study which links to several key in vivo investigations involved the application of high-speed millisecond fluorescence imaging to monitor real-time diffusion of single lipid molecules labelled with an organic dye, expressed in an artificial lipid bilayer [34], thus acting as a mimic for real cell membranes. Here, investigators could track single molecules with an accuracy better than the optical resolution limit (~200-300 nm) using a method which estimated the centre of the fuzzy diffraction-limited intensity image of single dye molecules to within a few tens of nm precision by using Gaussian fitting to the raw images (a method that was originally applied almost a decade earlier to determine the centre position of 190 nm diameter kinesin- coated beads conjugated to microtubules from non-fluorescence brightfield differential interference contrast (DIC) images to within 1-2 nm precision [35]). 3. THE “GOLDEN AGE ” – THE EMERGENCE OF SINGLE MOLEC “CELLULAR ” BIOPHYSICS (a) Approaches that investigate living, functional cells W ith so much exemplary single molecule biophysics research performed in the test tube, a question which should be addressed is: why do we care about studying molecular details in live-cell, or near live-cell, environments? Test ULE tube environments are significantly more controllable, less contaminated and come associated with less measurement noise. The best answer is that cells are not test tubes. A test tube experiment is a much reduced version of the native biology containing only components which we think/hope are important. We now know definitively that even the simplest cells are not just bags of chemicals, but rather have localized processes in both space and time. Also, the effective numbers of molecules involved in many cellular processes are often low, sometimes just a few per cell, and these minimal stoichiometry conditions are not easy to reproduce in the test tube without incurring a significant reduction in physiological efficiency. Single molecule biophysics investigations in vivo are, however, technically very difficult. Here, fluorescence microscopy is an invaluable biophysical tool. It results in exceptionally high signal-to-noise ratios for determining the localization of molecules tagged with a fluorescent dye but does so in a way that is relatively non-invasive compared to other single molecule biophysics methods. This minimal perturbation to native physiology makes it a probe of choice in single molecule biophysics studies in the living cell. Many of the improvements in our ability to detect single molecules have been driven by developments in the technology that allows photons to be efficiently collected from molecular report probes, several of which are fluorescent, including both “point ” detectors such as the photomultiplier tube (PMT) to pixel arrays of the next generation high quantum-efficiency cameras called electron multiply charge-coupled devices (EMCCDs), and these comparative technologies are reviewed in this Theme Issue [36]. It was only as recently as the year 2000 that the first definitive single molecule biophysics investigation involving a living sample was performed - by Sako and others [37] in which the investigators performed single molecule live-cell imaging on the cell membrane, here the high-contrast imaging technique of total internal reflection fluorescence microscopy (figure 3a), or TIRF [38], monitoring fluorescently-labelled EGF ligands binding to membrane receptor, and by Byassee and others [39] in which the researchers performed single molecule live-cell imaging inside the centre of a cell using confocal microscopy to monitor fluorescently-labelled transferrin molecules undergoing endocytosis. Significant developments have been made over the past decade in the field of live-cell super-resolution imaging [40],the ability to perform optical imaging in vivo at a spatial resolution better than that predicted from the Abbe optical resolution limit of ~0.61l/NA, where l is the detected wavelength for imaging and NA is the numerical aperture of the imaging system (typically set by the objective lens of the optical microscope of ~1.2-1.5), in particular an ability to monitor functional molecular complexes with such precision [41, 42]. There are several reviews that the reader can seek to discover the state-of- the-art in regards to various super-resolution technologies, however in this Theme Issue, super-resolution methods are reviewed in the context of a relatively new and highly promising technique called optical lock-in detection (OLID) which permits dramatic improvements to imaging contrast in native cellular imaging, far in excess of other competing super-resolution methods [43]. Recent developments in cellular single molecule fluorescence imaging have include the ability to definitively count molecules that are involved in functional biological processes integrated in the cell membranes of live cells, for example to quantify multiple protein subunit components in relatively large molecular machines such as the bacterial flagellar motor [44, 45] or single ion channels [46], and to combine counting with tracking of relatively mobile components around different spatial locations in the cell, such as molecular machines involves in protein translocation [47] and ATP fuel generation via oxidative phosphorylation, or OXPHOS [48, 49]. The state-of-art of our ability to image molecular components in cell membranes has led to substantial improvements to our understanding of their complex architecture, reviewed in two articles in this Theme Issue for model bacterial systems [50] as well focussing on putative zones of molecular confinement in the membrane, commonly referred to as lipid rafts [51]. By modifying the modes of fluorescence illumination, for example using narrow-field [34] or slimfield imaging [52], it has been possible to increase the excitation intensity in the vicinity of single cells to allow millisecond single molecule imaging. This has permitted visualization of native components normally expressed in the cytoplasm of cells whose viscosity is 100-1,000 times smaller than that of the cell membrane and so would be expected to diffuse at a faster rate by this same factor, allowing observation of gene expression bursts [53], regulation of transcription factors [54] and quantification of functional replisome components used in bacterial DNA replication machines [55]. Despite the central importance of fluorescence methods for single molecule cellular imaging there are also non-fluorescence detection techniques which can generate highly precise. For example, scanning probe microscopy (SPM) techniques. These cover a range of experimental approaches allowing topographical detail from the surface of a sample to be obtained by laterally scanning a probe across the surface. There are more than 20 different types of SPM methods currently developed which measure a variety of physical parameters as the probe is placed in proximity to a sample surface, and the most popular to date has been AFM (figure 3b). In this Theme Issue, Klenerman et al [56] reviews SPM techniques in the context of singe molecule precise imaging on the topographical details of live cells, namely probe-accessible features present on the cell membrane, and discusses in depth a relatively novel SPM approach of scanning ion conductance microscopy, or SICM (figure 3c). Another non-fluorescence technique which shows significant future potential for single molecule cellular biophysics is surface enhanced Raman scattering (SERS). Raman scattering is an inelastic process such that scattered photons from a sample have a marginally different frequency to those of the incident photons due primarily to vibrational energy transfer from the molecular orbitals in the sample, either resulting in a loss of energy from the photons (Stokes scattering) or, less commonly, a gain (anti-Stokes scattering). However, to detect the presence of a single molecule in a sample using Raman spectroscopy requires significant enhancement to the standard method used to acquire a scattering spectrum from a bulk, homogeneous sample. The most effective method utilises surface enhancement, which is reviewed in this Theme Issue [57], involving placing the sample in a colloidal substrate of gold or silver nanoparticles tens of nm in diameter. Photons from a laser will induce surface plasmons in the metallic particles, and in the vicinity of the surface the local electric field E associated with the photons is enhanced by a factor E4. The enhancement depends critically on the size/shape of the nanoparticles, but typically generates a better measurement sensitivity by a factor ~1014, particularly effective if the molecule itself is conjugated to the nanoparticle surface. This enhancement can be sufficient to detect single biomolecules. (b) In vitro methods of high complexity This is not to say that in vitro experiments are intrinsically bad and in vivo experiments are definitively good. Rather, they each provide complementary information. In vitro experiments are detached from a true physiological setting, but the level of environmental control is high. In vivo experiments are more demanding technically and are subject both to greater experimental noise and intrinsic biological variation - being in a native physiological environment is appealing at one level but offers difficulty in interpretation since there is a potential lack of control over other biological processes not directly under study but which may influence the experimental results. Next generation in vitro single molecule biophysics approaches are characterized by a much greater complexity than those involved in the early days of the field. In this Theme Issue, some of these often highly involved novel test tube approaches are discussed in Duzdevich and Greene [58], with a particular emphasis on a high-throughput single molecule biophysics method to investigate the binding of proteins to DNA, called DNA curtains. One particular focus of recent in vitro single molecule experiments has been the FoF1-ATPase enzyme, a highly complex machine composed of two rotary molecular motors of the membrane-integrated Fo motor and the hydrophilic F1 motor, which are ultimately responsible for the generation of cellular ATP. In this Theme Issue, recent single molecule biophysics approaches to investigate this vital, ubiquitous enzyme are reviewed in Sielaff and B örsch [59], with novel confirmation that the m echanism of nanoscale stepping of the F1 component elucidated in a thermophilic enzyme at room temperature, in which molecular rotation has been fuelled by the hydrolysis of ATP in the opposite direction to that involved during ATP manufacture, is shared by the mesophilic E. coli F1 enzyme, suggesting that even in markedly different environments there are common modes of action to this ubiquitous, essential molecular machine (Bilyard et al [60]). (c) Novel automated and bio-computational techniques Single molecule biophysics experiments are often plagued with noise, with the effective signal-to-noise ratio being sometimes barely in excess of 1 and generally less than 10. This constitutes an enormous analytical challenge to reliably detect a true signal and not erroneously measure noise. Molecular events are often manifest as some form of transient step signal in a noisy time-series, for example a motor protein might move via stepping along a molecular track. Thus, the challenge becomes one of reliable step-detection from noisy data. The aim is to assemble quantitative statistics of such step events in a fully objective, automated way. Edge-preserving filtration of the raw, noisy data is often the first tool employed, which preserves distinct edge event in time-series, such as the simple median filter, or better still the Chung-Kennedy filter which consists of two adjacent running windows whose output is the mean from the window possessing the smallest variance [26, 27] - a step event may then be classed as “true” on the basis of the change in the mean an d variance between the two windows being above some pre-agreed threshold. A significant issue with step-detection from a data time-series is that detection is sensitive to the level of threshold set. An alternative approach where all steps in a series are expected to be of the same size is to convert the time-series into a frequency-domain using a Fast Fourier transform, and then detect the periodicity in the original trace by looking for a fundamental peak in the associated power spectrum, which has been used to good effect for the estimation of molecular stoichiometry using step-wise photobleaching of fluorescent proteins [44]. A recent improvement to objectifying single molecule biophysics data is in how the distributions of single molecule properties are rendered. Traditional approaches used histograms, however these are highly sensitive to histogram bin size and position. A more general, objective approach uses kernel density estimation (KDE) - data are convolved with a Gaussian whose width is the measurement error for that property in that particular experiment, and whose height is normalized so that the area under the Gaussian is precisely one (i.e. one detected event), used to good effect in studying single molecule architectures of the bacterial replisome [55]. Spatial dynamics of single molecules and complexes inside living cells is a feature of biological processes. However, due to the low signal-to-noise ratio involved in cellular imaging experiments, the analysis of the motions of molecular complexes is non-trivial. In this Theme Issue, Robson et al [61] describe a novel method implementing a well-known weapon in the statistician’s armoury called Bayesian inference to robustly determine the underlying different modes of molecular diffusion relevant to live-cell imaging in both an objective and automated manner. One of the biggest challenges to single molecule biophysics is the traditionally low-throughput nature of experiments. In this Theme Issue, Ullman et al [62] describe methods combining automated microfluidics and novel imaging/analysis to dramatically improve the high-throughput nature. 4. THE CONTRIBUTIONS IN THIS THEME ISSUE This Theme Issue presents a series of articles from leaders in the field offering new insight into some of the latest developments of single molecule biophysics research which has now moved towards a far greater physiological relevance into the regime of addressing real, cellular questions. In summary, these articles include: i. A comprehensive review of new approaches in photon detection technology essential to modern single molecule cellular biophysics research [36]. Novel insights into super-resolution fluorescence imaging using the exceptionally high-contrast method of optical lock-in detection, OLID [43]. An appraisal of the increasing use of model bacteria as fundamental biological for addressing testbeds experimental questions using single molecule techniques [50]. A robust comparison of the single molecule biophysics methods which probe the nanoscale architectures of lipid microdomains in cell membranes [51]. A description of new, exciting single molecule surface probe technologies for living cells, including surface ion conductance microscopy, SICM [56]. A discussion of promising new single molecule cellular biophysics probing techniques using surface enhanced Raman spectroscopy, SERS [57]. A review of elegant, in vitro approaches to comb out single DNA ii. iii. iv. v. vi. vii. tethers for investigating single molecule protein translocation [58]. viii. An exploration of the state-of-the-art in single molecule biophysics ix. x. methodologies for experimentally probing the molecular means of ATP generation in cells [59]. Novel, cutting-edge single molecule biophysics research showing how the rotary molecular motors used in ATP generation in cell species which experience markedly different physical environments share fundamental mechanistic features [60]. New illustrating powerful new bio-computational research approaches to characterize the underlying modes of molecular diffusion from live-cell single molecule imaging [61]. xi. investigation demonstrating how single molecule A novel experiments on live cells can be made substantially more high- throughput by utilising ingenious engineering developments in microfluidics and computational improvements to optical microscope automation [62]. 5. THE OUTLOOK - BEYOND THE SINGLE MOLECULE AND THE SINGLE CELL The development of single molecule cellular biophysics represents a coming- of-age of methods using physics to understand life at the molecular level. There is great potential to now apply these novel technologies into areas that may have a large future impact on society, including those of bionanotechnology, systems and synthetic biology, fuel production for commerical use and single molecule biomedicine. As a scientific field, single molecule cellular biophysics is undergoing enormous expansion and is likely to be a key discipline in revealing underlying mechanistic features of biological processes in cells, with significant implications for the shape of both biophysical and biomedical research in the future. The industrial motivation to miniaturize synthetic bio-inspired devices is already starting to feedback into academic research laboratories in catalysing a general down-sizing approach for measurement apparatus. There is a compelling need to push this area of physiologically relevant interfacial science forward significantly, and this can only be truly facilitated by future generations of life and physical scientists talking to each other. Folk from each side of the bioscience fence traditionally blend like oil and water, such immiscibility often stemming from unfortunately early academic choices that schoolchildren make. However, what is needed now is an appreciation that some of the most fundamental concepts in each discipline can be shared by both camps, once elements of unwieldy language and overly complex maths have been put aside. The outlook for single molecule cellular biophysics is highly promising, but it is fundamentally driven by the enthusiasm of the talented researchers willing to take a punt and cross bridges into areas of science unknown. Preparation of this article was supported by a Royal Society University Research Fellowship and EPSRC research grant (EP/G061009) to M.C.L. The help of the anonymous referees who commented on contributions to this Theme Issue is gratefully acknowledged, as is the diligent assistance and patience of Helen Eaton of the Royal Society during the publication process. Figure 1. A schematic representation of the length scale of biological molecules and complexes in the context of larger macroscopic length scale entities. Figure 2. Optical trapping. (a) Ray-optic depiction of the trapping force for an optically trapped particle - a parallel Gaussian-profile laser beam is focussed and refracted by the trapped particle, such that equal and opposite changes in momentum on either side of the particle cancel out resulting in zero net force when the particle is roughly at the laser focus. But, (b) when the particle is laterally displaced from the focus the net momentum change experienced due to the reaction forces when refracted beams of light emerge from the particle are directed back towards the laser focus, illustrated by the momentum vector plots. (c) Displacement of a micron sized bead in an optical trap, the lateral trapping force is proportional to the lateral displacement x(t) where time is time, also the forwarded radiation pressure pushes the bead a little away from the precise laser focus. (d) Single optical trap stretch of a titin molecule tethered to a microscope coverslip via antibodies Ab1 and Ab2 binding to opposite termini of the titin molecule. (e) Similar titin stretches using a suction micropipette combined with an optical trap and (f) dual optical traps. Figure 3. Schematics of (a) TIRF, (b) AFM and (c) SICM. REFERENCES 1. Feynman, R. P. 1959. There’s Plenty of Room at t he Bottom. Lecture transcript deposited at Caltech Engineering and Science, 23:5, 22-36, and at www.its.caltech.edu/~feynman/plenty.html 2. Taniguchi, N. 1974. On the basic concept of 'nano-technology'. Proc. Intl. Conf. Prod. Eng. Tokyo, Part II, Japan Society of Precision Engineering. 3. Leake, M.C. 2012. Singe-molecule cellular biophysics, 1st edn. Cambridge University Press. 4. Harriman, O. L. J. and Leake, M. C. 2011. Single molecule experimentation in biological physics: exploring the living component of soft condensed matter one molecule at a time. J. Phys.: Condens. Matter. 23, 503101. 5. Van Holde, K E., Johnson, C. and Shing Ho, Pui. 2005. Principles of Physical Biochemistry, 1st edn. Pearson Education. 6. N ölting, B. 2009. Methods in Modern Biophysics, 2nd edn. Springer. 7. Lenn, T. and Leake, M. C. 2012. Experimental approaches for addressing fundamental biological questions in living, functioning cells with single molecule precision. Open Biol, 2, 120090. 8. Hall, C. E. 1956. Method for the Observation of Macromolecules with the Electron Microscope Illustrated with Micrographs of DNA. Biophys Biochem Cytol. 2, 625-628. 9. W ineland, D., Ekstrom, P. and Dehmelt, H. 1973. Monoelectron Oscillator. Phys. Rev. Lett. 31, 1279. 10. Nagourney, W ., Sandberg, J. and Dehmelt, H.. 1986. Shelved optical electron amplifier: Observation of quantum lumps. Phys Rev Lett. 56, 2797. 11. Rotman, B. 1961. Measurement of activity of single molecules of beta-D- galactosidase. Proc. Natl. Acad. Sci. USA 47, 1981-1991. 12. Hirschfeld, T. 1976. Optical microscopic observation of single small molecules. Appl. Opt. 15, 2965-2966. 13. Nguyen, D. C., Keller, R. A., Jett, J. H. and Martin, J. C. 1987. Detection of single molecules of phycoerythrin in hydrodynamically focused flows by laser-induced fluorescence. Anal Chem 59, 2158-2161. 14. Moerner, W . E. and Kador, L. 1989. Optical detection and spectroscopy in a solid. Phys. Rev. Lett. 62, 2535-2538. 15. Ashkin, A. 1970. Acceleration and Trapping of Particles by Radiation Pressure. Phys. Rev. Lett. 24, 156-159. 16. Ashkin, A., Dziedzic, J. M., Bjorkholm, J. E. and Chu, S. 1986. Observation of a single-beam gradient force optical trap for dielectric particles. Opt. Lett. 11, 288-290. 17 . Svoboda, K. and Block, S. M. 1994. Biological applications of optical forces. Annu. Rev. Biophys. Biomol. Stuct. 23, 247-85. 18. Moffitt, J. R., Chemla, Y. R., Smith, S. B. and Bustamante, C. 2008. Recent advances in optical tweezers. Annu. Rev. Biochem. 77, 205-28. 19. Gittes, F. and Schmidt, C. F. 1998. Interference model for back-focal- plane displacement detection in optical tweezers. Opt. Lett. 23, 7-9. 20. Svoboda, K., Schmidt, C. F., Schnapp, B. J. and Block, S. M. 1993. Direct observation of kinesin stepping by optical trapping interferometry. Nature 365: 721-727. 21. Finer, J. T., Simmons, R. M. and Spudich, J. A. 1994. Single myosin molecule mechanics: piconewton forces and nanometre steps. Nature 368, 113-119. 22. Smith, S. B., Cui, Y. and Bustamante, C. 1996. Overstretching B-DNA: the elastic response of individual double-stranded and single-stranded DNA molecules. Science 271, 795-799. 23. Liphardt J., Onoa, B., Smith, B. S., Tinoco Jr., I. and Bustamante, C. 2001. Reversible unfolding of single RNA molecules by mechanical force. Science 292, 733-7. 24 . Kellermayer, M. S., Smith, S. B., Granzier, H. L. and Bustamante, C. 1997. Folding-unfolding transitions in single titin molecules characterized with laser-tweezers. Science 276, 1112-1116. 25. Tskhovrebova, L., Trinick, J., Sleep, J. A. and Simmons, R. M. 1997. Elasticity and unfolding of single molecules of the giant muscle protein titin. Nature 387, 308-312. 26. Leake, M. C., W ilson, D., Bullard, B. & Simmons R. M. 2003. The elasticity of single kettin molecules using a two-bead laser-tweezers assay. FEBS Lett. 535, 55-60. 27. Leake, M. C., W ilson, D., Gautel, M. & Simmons, R. M. 2004. The elasticity of single titin molecules using a two-bead optical tweezers assay. Biophys. J. 87, 1112-1135. 28. Binnig, G., Rohrer, H., Gerber, Ch. and Weibel, E. (1982). Tunneling through a controllable vacuum gap. Applied Phys. Lett. 40, 178-180. 29. Binnig, G., Quate, C. F. and Gerber, C. 1986. Atomic Force Microscope. Phys. Rev. Lett. 56, 930-933. 30. Rief, M., Gautel, M., Oesterhelt, F., Fernandez, J. M. and Gaub, H. E. 1997. Reversible unfolding of individual titin immunoglobulin domains by AFM Science 276, 1109-1112. 31. Ha, T., Enderle, Th., Ogletree, D. F., Chemla, D. S., Selvin, P. R. and Weiss, S. 1996. Probing the interaction between two single molecules: Fluorescence resonance energy transfer between a single donor and a single acceptor. Proc. Natl. Acad. Sci. U.S.A. 93, 6264-6268. 32. Noji, H., Yasuda, R., Yoshida, M. and Kinosita, K. J. 1997. Direct observation of the rotation of F1-ATPase. Nature 386, 299-302. 33. Yildiz, A., Forkey, J. N., McKinney, S. A., Ha, T., Goldman, Y. E. and Selvin, P. R. 2003. Myosin V walks hand-over-hand: Single fluorophore imaging with 1.5-nm localization. Science 300, 2061-2065. 34. Schmidt, T., Schütz, G. J., Baumgartner, W ., Gr uber, H. J. and Schindler, H. 1996. Imaging of single molecule diffusion. Proc. Natl. Acad. Sci. USA 93, 2926-2829. 35. Gelles, J. Schnapp, B. J. and Sheetz, M. P. 1988. Tracking kinesin-driven movements with nanometre-scale precision. Nature. 331, 450-453. 36. Michalet X et al. 2013 Development of new photoncounting detectors for single-molecule fluorescence microscopy. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0035). 37. Sako, Y., Minoguchi, S. and Yanagida, T. 2000. Single-molecule imaging of EGFR signalling on the surface of living cells. Nature Cell Biol 2, 168-172. 38. Axelrod, D., Burghardt, T. P. and Thompson, N. L. 1984. Total internal reflection fluorescence. Ann. Rev. Biophys. Bioeng. 13, 247-268. 39. Byassee, T. A., Chan, W . C. and Nie, S. 2000. Probing single molecules in single living cells. Anal Chem. 72, 5606-11. 40. Chiu, S.-W . and Leake, M. C. 2011. Functioning nanomachines seen in real-time in living bacteria using single-molecule and super-resolution fluorescence imaging. Int. J. Mol. Sci. 12, 2518-2542. 41. Leake, M. C. 2010. Shining the spotlight on functional molecular complexes: the new science of single-molecule cell biology. Commun Integr Biol. 3, 415-418. 42. Dobbie, I. M., Robson, A., Delalez, N. and Leake, M. C. 2009. Visualizing Single Molecular Complexes In Vivo Using Advanced Fluorescence Microscopy. J Vis Exp 31, 1508. 43. Yan Y, Petchprayoon C, Mao S, Marriott G. 2013 Reversible optical control of cyanine fluorescence in fixed and living cells: optical lock-in detection immunofluorescence imaging microscopy. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0031). 44. Leake, M. C., Chandler, J. H., Wadhams, G. H., Bai, F., Berry, R. M. and Armitage, J. P. 2006. Stoichiometry and turnover in single, functioning membrane protein complexes. Nature. 443, 355-8. 45. Delalez, N. J., Wadhams, G. H., Rosser, G., Xue, Q., Brown, M. T., Dobbie, I. M., Berry, R. M., Leake, M. C. and Armitage, J. P. 2010. Signal- dependent turnover of the bacterial flagellar switch protein FliM. Proc Natl Acad Sci U S A 107, 11347-11351. 46. Ulbrich, M. H. and Isacoff, E. Y. 2007. Subunit counting in membrane- bound proteins. Nat Meth. 4, 319-21. 47. Leake M. C., Greene, N. P., Godun, R. M., Granjon,,T., Buchanan, G., Chen, S., Berry, R. M., Palmer, T. and Berks, B. C. 2008. Variable stoichiometry of the TatA component of the twin-arginine protein transport system observed by in vivo single-molecule imaging. Proc Natl Acad Sci U S A. 105, 15376-15381. 48. Lenn, T., Leake, M. C. and Mullineaux, C. W.2008. Are Escherichia coli OXPHOS complexes concentrated in specialised zones within the plasma membrane? Biochem. Soc. Trans., 36, 1032-1036. 49. Lenn T., Leake, M. C. and Mullineaux, C. W. 2008 In vivo clustering and dynamics of cytochrome bd complexes in the Escherichia coli plasma membrane. Mol. Microbiol. 70, 1397-1407. 50. Ritchie K, Lill Y, Sood C, Lee H, Zhang S. 2013 Single molecule imaging in live bacteria cells. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0355). 51. Klotzsch E, Schutz G. 2013 A critical survey of methods to detect plasma membrane rafts. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0033) 52. Plank, M., Wadhams, G. H. and Leake, M. C. 2009. Millisecond timescale slimfield imaging and automated quantification of single fluorescent protein molecules for use in probing complex biological processes. Integr. Biol. 1, 602612. 53. Cai, L., Friedman, N. and Xie, X. S. 2006. Stochastic protein expression in individual cells at the single molecule level. Nature 440, 358-62. 54. Elf, .J, Li, G. W . and Xie X. S. 2007. Probing transcription factor dynamics at the single-molecule level in a living cell. Science 316, 1191-4. 55. Reyes-Lamothe R., Sherratt D. J. and Leake, M. C. 2010. Stoichiometry and architecture of active DNA replication machinery in Escherichia coli. Science, 328, 498-501. 56. Klenerman D, Shevchuk A, Novak P, Korchev Y, Davis S. 2013 Imaging the cell surface and its organization down to the level of single molecules. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0027). 57. Wang Y, Irudayaraj J. 2013 Surface enhanced Raman spectroscopy (SERS) at single molecular scale and its implications in biology. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0026). 58. Greene E, Duzdevich D. 2013 Towards physiological complexity with in vitro single-molecule biophysics. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0271). 59. Boersch M, Sielaff H. 2013 Twisting and subunit rotation in single FOF1- ATP synthase. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0024). 60. Berry R, Bilyard T, Nakanishi-Matsui M, Steel B, Pilizota T, Nord A, Hosokawa H, Futai M. 2013 Highresolution single-molecule characterization of the enzymatic states in Escherichia coli F1-ATPase. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0023). 61. Robson, A, Burrage, K, Leake, MC. 2013. Inferring diuffusion in single cells at the single molecule level. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0029). 62. Ullman G, Walldén M, Marklund E, Mahmutovic A, Razinkov I, Elf J. 2013 Hi-throughput gene expression analysis at the level of single proteins using a microfluidic turbidostat and automated cell tracking. Phil. Trans. R. Soc. B 368. (doi:10.1098/rstb.2012.0025).
1908.08373	1	1908	2019-08-14T15:12:20	Structure and efficiency in bacterial photosynthetic light-harvesting	[ "physics.bio-ph", "quant-ph" ]	Photosynthetic organisms use networks of chromophores to absorb sunlight and deliver the energy to reaction centres, where charge separation triggers a cascade of chemical steps to store the energy. We present a detailed model of the light-harvesting complexes in purple bacteria, including explicit interaction with sunlight; energy loss through radiative and non-radiative processes; and dephasing and thermalizing effects of coupling to a vibrational bath. An important feature of the model is that we capture the effect of slow vibrational modes by introducing time-dependent disorder. Our model describes the experimentally observed high efficiency of light harvesting, despite the absence of long-range quantum coherence. The one-exciton part of the quantum state fluctuates due to slow vibrational changes, but remains highly mixed at all times. This lack of long-range coherence suggests a relatively minor role for structure in determining the efficiency of bacterial light harvesting. To investigate this we built hypothetical models with randomly arranged chromophores, but still observed high efficiency when typical nearest-neighbour distances are comparable with those found in nature. This helps to explain the efficiency of energy transport in organisms whose chromophore networks differ widely in structure, while also suggesting new design criteria for efficient artificial light-harvesting devices.	physics.bio-ph	physics	Structure and efficiency in bacterial photosynthetic light-harvesting Susannah Bourne Worstera, Clement Strossa,b, Felix M. W. C. Vaughana,b,c, Noah Lindenb, and Frederick R. Manbya,1 a: Centre for Computational Chemistry, School of Chemistry, University of Bristol, Bristol BS8 1TS, UK b: School of Mathematics, University of Bristol, Bristol BS8 1TW, UK and c: Bristol Centre for Complexity Sciences, University of Bristol, Bristol, BS2 8BB, UK∗ Photosynthetic organisms use networks of chromophores to absorb sunlight and deliver the energy to reaction centres, where charge separation triggers a cascade of chemical steps to store the energy. We present a detailed model of the light-harvesting complexes in purple bacteria, including explicit interaction with sunlight; energy loss through radiative and non- radiative processes; and dephasing and thermalizing effects of coupling to a vibrational bath. An important feature of the model is that we capture the effect of slow vibrational modes by introducing time-dependent disorder. Our model describes the experimentally observed high efficiency of light harvesting, despite the absence of long-range quantum coherence. The one- exciton part of the quantum state fluctuates due to slow vibrational changes, but remains highly mixed at all times. This lack of long-range coherence suggests a relatively minor role for structure in determining the efficiency of bacterial light harvesting. To investigate this we built hypothetical models with randomly arranged chromophores, but still observed high efficiency when typical nearest-neighbour distances are comparable with those found in nature. This helps to explain the efficiency of energy transport in organisms whose chromophore networks differ widely in structure, while also suggesting new design criteria for efficient artificial light-harvesting devices. 9 1 0 2 g u A 4 1 ] h p - o i b . s c i s y h p [ 1 v 3 7 3 8 0 . 8 0 9 1 : v i X r a ∗ To whom correspondence should be addressed. E-mail: [email protected] 2 Photosynthesis is the process by which organisms absorb energy in the form of sunlight and convert it into chemical energy, and it is the engine that produces almost all global biomass. Since the molecular architecture needed to carry out photosynthesis is costly to produce and maintain, sunlight is initially captured by an extensive antenna network of chlorophyll molecules (chromophores), through which absorbed energy must then be transported to specialised reaction centres to begin the photosynthetic reaction cycle. Experiments show that transport through the antenna can be highly efficient: absorption of a sufficiently energetic photon almost always leads to the charge-separation process that is a prerequisite for chemical storage of the energy [1 -- 3]. Two-dimensional spectroscopic investigations present the possibility of a substantial role for quantum coherence in achieving high efficiency [4, 5], an idea that was initially supported by a variety of theoretical and computational studies [6 -- 11]. More recently this has been contested on the grounds that the oscillatory signature observed in the spectroscopy measurements might have been incorrectly interpreted [3, 12 -- 14], or might be a result of the (unnatural) laser pulses used to excite the sample [15, 16]. Furthermore, even if the oscillatory pattern truly revealed quan- tum coherence, it might not necessarily have an important functional role [17, 18]. The research nevertheless sparked considerable interest in characterising the quantum state of the antenna over time and identifying criteria for efficiency. This has proved a challenging undertaking due to the complexity of the system and the number of factors influencing its behaviour. In particular, it has been shown that the nature of the light source [15, 16], the method used to model the vibrational environment [19], the inclusion (or not) of energetic and spatial disorder [12, 20], and the consid- eration of short-time versus steady-state dynamics [16, 17, 21 -- 23] all influence the way the system evolves. We aim to bring all these ideas together in a comprehensive and realistic model that includes detailed descriptions of each step of the process; from the continuous absorption of weak, incoherent sunlight to the final consumption of excitons at a reaction centre. We apply this model to a larger system than has yet been considered, including both of the major purple-bacteria light harvesting systems (LHI and LHII) and the branched chromophore chains of the reaction centre (RC). In addition to a drive to understand the functioning of natural photosynthetic antennae, the suggestion that coherence may play a role in exciton transport inspired several attempts to design artificial systems that exploited the same principles. These systems, both hypothetical and real, require exquisite control over the placement and orientation of pigment chromophores in order to create and preserve the necessary coherence [20, 24 -- 26]. This makes them difficult to realise and those that have been created are small and delicately constructed. In the second part of this paper we consider an artificial antenna system of much simpler construction: a box of chro- mophores placed randomly (as if in solution) around a central reaction centre. Remarkably, the high efficiencies we observe demonstrate that structure (in the sense of a detailed arrangement of chromophores) does not play a significant role in determining the efficiency of an antenna. This could offer a pathway to designing cheaper, more scalable synthetic light harvesting devices. 3 THEORY The flow of energy through a photosynthetic antenna system is governed by a complex interplay of several physical processes: coherent evolution of non-stationary states; relaxation due to coupling with vibrational modes in the environment; absorption and emission of light; non-radiative decay, which results in exciton loss, and, finally, consumption of excitons at the reaction centre, triggering the biochemical cascade that ultimately leads to production of adenosine triphosphate. For reasons justified below, we model the system using a Lindblad master equation with a time-dependent Hamiltonian. Thus the evolution of the system density matrix is described by the Liouville-von Neumann equation (cid:105) (cid:104) H(t), ρ(t) dρ(t) dt = − i + (Lrad + Lbath + Lnr + Lsink) ρ(t). (1) Although each chromophore is a large, complex molecule, its participation in exciton transport is dominated by its electronic ground state ψ0(cid:105) and first excited state ψ1(cid:105) under biologically relevant conditions. Consequently, we model each chromophore site as a two-level system, with the states of a network of chromophores modelled as Hartree products of the single-site states. We consider only the n + 1 states with either zero (0(cid:105) = 01, 02,··· , 0n(cid:105)) or one (i(cid:105) = 01 ··· 1i ··· 0n(cid:105)) exciton [18]. Rather than focusing on a single component, as has been done previously, we consider transport across the whole antenna using a model system consisting of one LHII and one LHI/RC complex (SI Appendix, Fig. S1). The system contains a total of n = 65 chromophores. Unitary dynamics The unitary dynamics of the exciton are governed by a Hamiltonian of the form (cid:88) i H = Ei i(cid:105)(cid:104)i + Vij j(cid:105)(cid:104)i (2) (cid:88) i(cid:54)=j 4 where Ei are the excitation energies at each site and Vij is the strength of the interaction between the transition dipole moments µi on sites i and j, calculated using the point dipole approximation. The point dipole approximation overestimates the nearest neighbour coupling in the LHI and RC complexes so these values were corrected to match the results of more detailed calculations (SI Appendix). Realistic disorder is introduced into the Hamiltonian by randomly drawing both Ei and µi from distributions computed using time-dependent density functional theory (TDDFT) on snapshots of a molecular dynamics (MD) simulation [12]. The unit vectors µi are taken separately from randomly selected MD snapshots. Interaction with light Different light sources induce very different dynamics [15, 16, 22]. Coherent pulses, localised in time and space, excite coherent superpositions of energy eigenstates, which then evolve coherently. Continuous incoherent illumination excites stationary eigenstates with well-defined energies. To understand the dynamics of photosynthesis it is therefore crucial to use a realistic model of the interaction between the system and incident radiation. We capture this interaction using a Lindblad operator (Lrad) derived from the Born-Markov quantum optical master equation, including terms for both absorption (Labs rad) and spontaneous and stimulated emission (Lemit rad ). The rates of each process are parameterised to correctly account for the initial blackbody temperature of the sun, the attenuation of the radiation as it passes through the atmosphere, and the anisotropy of absorption into different eigenstates (SI Appendix). At 10−4 ps−1 the rate of absorption is orders of magnitude slower than any of the other processes governing the dynamics of the antenna. As a result, the system remains almost entirely in the zero-exciton ground state for all time. We find that the amount of excited state present at any given time is on the order p = 10−9.[27] To accelerate equilibration to the correct excited-state population, simulations are started from an initial state ρinit = (1 − p)0(cid:105)(cid:104)0 + pψ(cid:105)(cid:104)ψ with the √ excited state manifold in a completely delocalised pure state ψ(cid:105) = 1/ j j(cid:105). As will be seen, the dynamics soon washes out any specific signatures of this initial state. n(cid:80) Capture of excitons at the reaction centre At the reaction centre, excitons are used to trigger a series of electron transfers. As this process is effectively irreversible [28], we treat the reaction centre as a sink through which excitons are continuously lost. It is described by the Lindblad operator Lsinkρ = Γsink 0(cid:105)(cid:104)jρj(cid:105)(cid:104)0 − 1 2 (cid:19) {j(cid:105)(cid:104)j , ρ} , 5 (3) (cid:18) (cid:88) j∈SP where the sum runs over the chromophores in the 'special pair', the rate of exciton loss Γsink is taken as the rate of the first step in the electron transfer chain [29, 30], and {·,·} denotes anticommutator. Coupling to the environment The vibrational modes of the light-harvesting complexes and the surrounding medium have a huge impact on how excitons are transported through the antennae. They can act both to reduce the amount of excited state in the system by offering a pathway for non-radiative decay back to the ground state, and to redistribute remaining excitations by acting as a driving force for relaxation within the excited state manifold. Vibrations that are slow on the timescale of dynamics relevant for photosynthetic energy transport disrupt the symmetrical ring structure of the individual light- harvesting complexes, introducing variations in the excitation energies of individual chromophores, and in the interactions between them. We include each of these effects separately in our model. The symmetry-breaking effects of slow vibrations are captured by introducing static disorder into the Hamiltonian, as described above. The exact form of the disorder will vary slowly with the motion of the vibration over the lifetime of an exciton. We approximate this time-dependence by generating a new, random Hamiltonian at regular intervals of 0.1 ps, corresponding approximately to a vibrational frequency below which higher harmonic levels become significantly occupied at 300 K. To our knowledge, this is the first time any attempt has been made to include the dynam- ical effects of slow vibrations in a master equation description of photosynthetic exciton transfer, although a similar technique has been used successfully to describe the photochemistry of small organic molecules [37]. Non-radiative loss of excitons competes with their biochemical utilisation at the reaction centre. We model it using a Lindblad operator Lnr that has an identical form to Lsink (equation 3), with TABLE I. Rate constants and timescales for Lindblad terms Γsink Γnr Γdeph Value / ps−1 0.125 0.001 References Timescale 29, 30 31 -- 35 7, 36 8 ps 1 ns 90 fs 11 6 j summed over every site in the system and a non-radiative decay constant Γnr. We consider two different models for the Lindblad term Lbath describing the relaxation processes. The first is a local dephasing model Ldephρ = Γdeph (cid:18) (cid:88) j j(cid:105)(cid:104)j ρj(cid:105)(cid:104)j − 1 2 {j(cid:105)(cid:104)j , ρ} (cid:19) , (4) in which the coherences between the states j(cid:105) localised on each site decay exponentially with a decay constant Γdeph. Our second model is a global thermalizing model [38] (cid:18) (cid:88) a,b,a(cid:54)=b (cid:19) Lthermρ = kb→a 2π a(cid:105)(cid:104)bρb(cid:105)(cid:104)a − 1 2 {b(cid:105)(cid:104)b , ρ} , (5) in which exciton population is transferred between eigenstates b(cid:105) and a(cid:105) at a rate kb→a. The local model provides an intuitive picture of the effect of random environmental fluctuations but forces the system to relax to the maximally mixed state ρmm = I/n (SI Appendix). By contrast, the global approach relaxes the system to the correct instantaneous thermal equilibrium state but does not explicitly account for the proximity and strength of coupling between chromophores. In a regime where inter-chromophore couplings are small on the scale of the system-bath coupling, this can result in an unphysical flow of energy between spatially well-separated and uncoupled chromophores [39]. A further concern is the possible role of non-markovianity in the system-bath coupling [40, 41]. However we find that the global Lindblad approach gives good qualitative agreement with preliminary calculations on small model systems using the exact hierarchical-equations-of-motion (HEOM) approach [42]. The rate constants in equation 5 are determined using the Redfield theory expression [43] (cid:18) ωab (cid:19) kBT ka→b = exp kb→a =2πω2 ab ((1 + n(ωab))J(ωab) + n(ωba)J(ωba)) , (6) where the relationship between ka→b and kb→a ensures that detailed balance is obeyed. The angular frequency ωab is that of the energy gap between states a(cid:105) and b(cid:105) and n(ω) = 1/(exp(ω/kBT ) − 1) is the Bose-Einstein distribution. The spectral density J(ω) describes the coupling of the excitonic system to the vibrational modes of the environment, under the assumption that all modes can be modelled as harmonic 7 oscillators. This assumption breaks down for flat-bottomed, low-frequency modes, whose effect, however, we capture separately by periodically changing the Hamiltonian, as described above. In the absence of a full spectral density for the antenna system, we use the form proposed by Renger and Marcus [44]: JRM(ω) = 0.8 e−√ ω3 7! 2ω4 1 ω/ω1 + 0.5 e−√ ω/ω2, ω3 7! 2ω4 2 (7) which is fitted to the experimental spectra of a monomer pigment-protein complex (B777) with characteristic frequencies ω1 = 0.10483 rad ps−1 and ω2 = 0.364625 rad ps−1. Using this spec- tral density, the global model in equation 5 accurately reproduces the experimentally measured relaxation time constant for an LHII ring [45 -- 47] (SI Appendix, Fig. S2). Efficiency The efficiency of a photosynthetic apparatus can be defined in a number of different ways, not all of which are equivalent [21]. In this work, we will define the efficiency as E[ρ] = (cid:104)0Lsinkρ0(cid:105) (cid:104)0(Lsink + Lnr + Lemit rad )ρ0(cid:105) . (8) For each unit of energy that is absorbed by the antenna, the efficiency tells us what fraction is used productively by the sink. This measure reflects how well the system is adapted to transport and use the absorbed energy. Note that it does not tell us what fraction of incident light reaches the reaction centre as this would require additional consideration of how efficiently the system absorbs light. While this is also an important question, we will not consider it here. RESULTS Bacterial antenna system Under the influence of Lbath, the excited-state manifold relaxes rapidly towards a steady-state. When Lbath = Ldeph, the dynamics for every realization of H equilibrates to ρmm. When Lbath = Ltherm the steady-state is the thermal state of the current Hamiltonian, in which eigenstates are populated according to the Boltzmann distribution. The thermal state changes when the Hamiltonian changes, with an average trace distance of 0.4 between instantaneous thermal states. Following a change in the Hamiltonian, the exciton manifold reequilibrates to the new 8 FIG. 1. Trace distance between the time-evolving state and the thermal state ρtherm =(cid:80) α exp(Eα/kBT )/Z of the current Hamiltonian as a function of time for the LHI-LHII model system. The Hamiltonian changes at 1 ps, as does the corresponding ρtherm. It takes several hundred femtoseconds for the system to requilibrate, under the influence of the Hamiltonian and the thermalizing Lindblad operator Ltherm. thermal state within a few hundreds of femtoseconds (see figure 1), in keeping with experimentally measured equilibration times in individual LHCs [45]. The Hamiltonian changes on a similar timescale (every 100 fs), such that the quantum state of the antenna is constantly evolving (see figure 2a). However, since individual, instantaneous thermal states are not too dissimilar, the system spends much of its time in states that are close to thermal and the time-averaged state ρav, shown in figure 2b, is close (trace distance ≈ 0.15) to the average of the thermal states for all realisations of the Hamiltonian (cid:104)ρtherm(cid:105)av. The longer the period between changes of the Hamiltonian, the more closely ρav resembles (cid:104)ρtherm(cid:105)av. Thermal states are, by nature, mixed states. The extent to which the state of a system is mixed is quantified by tr ρ2, which is 1 for a pure state and 1/N for the maximally mixed state. It may be shown analytically (SI Appendix) that the full average over the thermal states of all possible Hamiltonians is exactly the maximally mixed state. Correspondingly, the values of tr ρ2, given in table II, for the average states of both the full system and each component subsystem are small and close to their values for the maximally mixed state. On average then, the system is not only in a highly mixed state but in one that equally favours equivalent sites within a ring. 9 FIG. 2. Evolution of the state of the exciton manifold in the full LHI-LHII model system. The Hamiltonian changes every 0.1 ps. Dephasing is implemented using the global model Ltherm (an equivalent figure for the local model can be found in SI Appendix, Fig. S5). Rates of other processes are given in table I. (a) Trace distance of the instantaneous evolving state ρ(t) from the current thermal state ρtherm, which changes with the Hamiltonian (only a short time sample is shown so that the details of the evolution are more clearly visible). The coloured bar along the top illustrates the amount of time spent close to a thermal state, with dark blue marking stretches of time where the trace distance is ≤ 0.2, and light blue where the trace distance is > 0.2. (b) Plot of log10 Re(ρav). The renormalized average density matrix ρav is calculated from 5 ps onward, to avoid any influence from the specific choice of initial state. (c) Degree of purity quantified by tr ρ(t)2. (d) The inverse participation ratio of ρ(t). For (c) -- (d), the mean value of each quantity starting from 5 ps is displayed on the graph. TABLE II. Properties of subsystem states calculated for the renormalized substates of the average state tr ρ2 Min. value of tr ρ2 IPR Max. value of IPR LHI-LHII B875 B850 B800 0.083 0.095 0.141 0.114 0.015 0.03 0.06 0.11 3.60 8.86 6.73 1.57 9 65 32 18 The average state, furthermore, contains very little coherence, as evidenced by small values of the inverse participation ration (IPR). The IPR, [48] (cid:104)(cid:80) jk ρjk(cid:105)2 (cid:80) jk ρjk2 , nsite IPR = (9) 10 provides a measure of how coherently delocalised a state is. It is bounded between 1 (incoherent and fully localised) and nsite (fully delocalised coherent pure state). In B800 particularly, there is essentially no coherence (IPR = 1.57), reflecting the fact that the chromophores are well-spaced and weakly coupled. Stronger coupling in the B850 and B875 rings introduces coherences between neighbouring sites, raising the value of the IPR. However, when ρB850 and ρB875 are expressed in the reduced basis of the BChl dimer subsystem, to more accurately reflect the structure of these rings, they are also found to be essentially maximally mixed (SI Appendix, Fig. S3). Overall, the excitonic subsystem is, on average, in a highly mixed, incoherent state that never- theless places population equally on all sites. Instantaneously, the system is never in its average state but rather fluctuates continuously between instantaneous thermal states. However, since the thermal states all have similar forms (though favouring different chromophore sites), the properties of the system do not fluctuate a great deal. Plots of tr ρ(t)2 and IPR in figure 2 (panels c and d) indicate that, at all times, the state of the system remains highly mixed with no long-range coherence. The efficiency of the antenna, plotted in figure 3, remains fairly constant over time, reflecting the consistent form of the quantum state throughout the evolution. The efficiencies obtained with the two variants of Lbath span the experimentally measured efficiencies for photosystem II in higher plants (84% -- 91%) [1, 3]. Using Lbath = Ltherm, the time-averaged efficiency (cid:104)E[ρ(t)](cid:105)av is 96%. This is likely an overesti- mate, since the global model does not account for the weaker coupling between B875 and the RC, which significantly retards the rate of exciton transfer between them [49]. Hence, this model tends to overestimate the population of the RC (SI Appendix, Fig. S4), on which E is heavily dependent (SI Appendix). By contrast, using Lbath = Ldeph underestimates the efficiency by failing to capture the energy gradient between LHII and LHI and consequently underestimating the population of the B875 ring and the rate of exciton transfer onto the RC. Using the local dephasing model (cid:104)E[ρ(t)](cid:105)av is 62%. This value is highly sensitive to the number of antenna chromophores, since exciton density is distributed equally between all chromophores under the action of Ldeph. Therefore, increasing the number of chromophores reduces the population on any single site. For example, if only LHI is modelled, (cid:104)E[ρ(t)](cid:105)av = 75%. In addition to observing minimal fluctuations in the efficiency, we find that (cid:104)E[ρ(t)](cid:105)av is closely reproduced (97% rather than 96% for the global model) by E[ρav], the efficiency of the time- averaged state. Together, these results indicate that, despite short-time fluctuations in the state, 11 FIG. 3. Efficiency of the LHI-LHII antenna over the same period of evolution plotted in figure 2. From top to bottom, the traces correspond to efficiency of the full antenna (one LHI and one LHII) with Lbath = Ltherm; efficiency of the LHI/RC complex with Lbath = Ldeph, and efficiency of the full antenna with Lbath = Ldeph. The mean efficiency, displayed to the right hand side of each trace, is calculated from 5 ps. the system evolves continuously through states that are similar in character both to each other and to their time average. An artificial light-harvesting system The highly symmetric ring structures of natural photosynthetic antenna systems are often though to imply a special physical significance for the particular spatial arrangement of chro- mophores (such as maintaining long-lived coherence). We investigated this hypothesis by calcu- lating the efficiency of a box of 32 chromophores (for comparison with LHI) placed and oriented randomly around a central reaction centre. The 'antenna' chromophores were not allowed within 3 A of each other or within 45 A of the middle of the reaction centre (SI Appendix). The dimensions of the box were adjusted to alter the concentration of antenna chromophores without changing the ratio of chromophores to reaction centre. Relaxation was implemented using the local dephas- ing model (Ldeph), which is sensitive to chromophore spacing and particularly appropriate to the lower concentrations investigated. Each concentration was sampled 32 times with different random configurations of the chromophores. The results are shown in figure 4. 12 FIG. 4. Efficiency as a function of chromophore concentration in a (a) cubic (d× d× d) box or (b) pseudo-2D (d × d × 3A) slab of 32 chromophores arranged randomly around a centrally placed reaction centre. Each concentration was sampled 32 times. Range bars indicate one standard deviation above and below the mean efficiency. Horizontal dotted lines indicate, for comparison, the efficiency of the LHI-RC complex in purple bacteria, using the local dephasing model. In each sample, the efficiency was averaged over the evolution from 10 ps to 100 ps. (c) Efficiency as a function of dephasing rate for one arrangement of chromophores in a cubic box with d = 125 A. (d). Example structure of a chromophore box with d = 95 A (corresponding to 106/(d/A)3 = 1.17). Two clear trends are visible. Firstly, as the concentration of chromophores is reduced, the mean efficiency decreases. More weakly coupled chromophores do not transfer the exciton as quickly to the reaction centre (if at all) so more excited state is lost through non-radiative decay and emission processes. Secondly, the standard deviation in the efficiency increases with decreased concentration. This can by understood by considering that in a small box, at high concentration, the chromophores will always be close to each other and to the reaction centre. At lower concen- trations, the random placement of chromophores will sometimes result in clusters of closely-spaced, strongly coupled chromophores and other times in all the chromophores being far apart and only weakly coupled. Crucially, when the chromophores were closely spaced, there was very little vari- ation in the efficiency over the 32 sampled configurations, indicating that the orientation of the chromophores was not significant. The same behaviour was observed whether the chromophores were placed freely in 3D (figure 13 4a), as they might be in a liquid phase, or confined to a pseudo-2D slab (figure 4b), to mimic the layered structure of natural antenna networks. Whilst chromophore spacing is an important factor for determining efficiency, it is equally important to balance the rates of competing processes in the system. Figure 4c illustrates the impact of varying the dephasing rate. There is a sharp drop in efficiency when the dephasing rate exceeds 100 ps−1, where continual measurement by the environment prevents the system evolving. Typically, simulations have shown that efficiency drops off at low dephasing rates too, as the exciton is trapped by Anderson localisation. However, in our calculations this effect is largely mitigated by the coupling to slow vibrations. DISCUSSION We have examined the nature and evolution of the quantum state of a photosynthetic antenna system under constant illumination by a natural, incoherent light source. When talking about the state of light harvesting systems it is common to refer only to the part of the state that contains exactly one exciton. However, it is worth noting that the overall quantum state is, to a very good approximation, the zero-exciton ground state. This contrasts with the often-employed picture of photon absorption triggering a large change in the excitonic state of a single chromophore. While this picture is rele- vant for understanding ultrafast spectroscopic experiments, it has little bearing on the phenomenon of photosynthesis in sunlight. Within the detailed model we have constructed of the many processes governing the state and dynamics of the excitonic subsystem, accurately capturing the effect of interacting with a vibrational environment is particularly challenging, since the similar strengths of inter-system and system-bath coupling places the problem uncomfortably between the ranges of validity of commonly used approximations [3, 50]. The two approaches we have employed embody the two opposing regimes of strongly coupled chromophores interacting collectively with the bath (Ltherm) and weakly coupled chromophores interacting independently with a local bath (Ldeph). Both methods make the markovian approximation so will not capture the small non-markovian effects that may arise from strongly coupled discrete vibrational modes. Although neither approach can, on its own, reliably describe the effect of the vibrational environment, we obtain valuabe insights by observing features that arise from both descriptions. Firstly, we conclude that the state of the excitonic subsystem is at all times highly mixed 14 with little or no coherence. Indeed, both descriptions indicate that, on the timescale of the sink (around 10 ps), the average state is close to the maximally mixed within each ring subsystem. In other words, exciton density is distributed equally between equivalent sites in an incoherent manner. Over the whole antenna, the distribution of excited state population is skewed towards LHI because of the energy gradient between successive ring components and the large number of chromophores [25, 49]. The importance of this effect is exemplified by the low efficiency obtained using the local dephasing model, which spreads exciton density uniformly over the whole antenna system. The coupling to a vibrational environment, and the consequent relaxation, is an important mechanism for transporting excitations through the antenna and overcoming disorder-driven lo- calization. The role of dephasing in exciton transport has been noted before and is sometimes referred to as environment-assisted transport [21, 51, 52]. What we additionally see here is that slow vibrational modes keep the exciton density moving continuously around the antenna. This is in good agreement with atomistic simulations of LHII, which also show excitons moving per- petually around the system, with a bias towards lower energy ring components [53]. By allowing exciton density to explore the whole antenna, the vibrational environment performs the role that was previously postulated to be the purpose of long-lived coherences. It is therefore unsurprising that, along with others [17, 21, 54], we observe that excitonic coherence is not a necessary condition for high efficiency. Without the requirement for coherence, the rigid structural requirements that have been put forward to maintian coherence can be relaxed [24 -- 26]. Indeed, the results of figure 4 demonstrate that the orientation of chromophores has little bearing on the antenna efficiency. Instead, the typical nearest neighbour distance appears to be the key determinant of efficiency. However, effects such as concentration quenching and the energetic cost of synthesising chromophores (which we did not include in our simulation) would make high chromophore concentrations undesirable. We suggest that perhaps the organisation of chromophores into circular structures may simply be a means of avoiding the extremely high pigment concentration that would be required to achieve the right separation in a completely disordered solution. It is possible that the same end could be achieved in an artificial system without the need for complex protein architecture, for example, by anchoring chromophores to a bead. However, to design an appropriate artificial matrix to support the chromophores, futher investigation would need to be carried out into the role of environment in determining the rates of dephasing and non-radiative decay, which need to be carefully balanced to maintain efficiency. ACKNOWLEDGMENTS 15 We gratefully acknowledge the funding agencies that supported this work: CS was supported through the doctoral training grant of the Engineering and Physical Sciences Research Council (EPSRC); FMWCV was supported through the EPSRC Bristol Centre for Complexity Sciences; SBW is supported by a research fellowship from the Royal Commission for the Exhibition of 1851. [1] E. Wientjes, H. Van Amerongen, and R. Croce, Journal of Physical Chemistry B 117, 11200 (2013). [2] C. P. Rijgersberg, R. Van Grondelle, and J. Amesz, Biochimica et Biophysica Acta - Bioenergetics 592, 53 (1980). [3] F. Fassioli, R. Dinshaw, P. C. Arpin, and G. D. Scholes, Journal of The Royal Society Interface 11, 20130901 (2013). [4] G. S. Engel, T. R. Calhoun, E. L. Read, T. K. Ahn, T. Mancal, Y. C. Cheng, R. E. Blankenship, and G. R. Fleming, Nature 446, 782 (2007). [5] H. Lee, Y.-C. Cheng, and G. R. Fleming, Science 316, 1462 (2007). [6] T. R. Calhoun and G. R. Fleming, Physica Status Solidi (B) Basic Research 248, 833 (2011). [7] E. Harel and G. S. Engel, Proceedings of the National Academy of Sciences 109, 706 (2012). [8] G. Panitchayangkoon, D. Hayes, K. A. Fransted, J. R. Caram, E. Harel, J. Wen, R. E. Blankenship, and G. S. Engel, Proceedings of the National Academy of Sciences 107, 12766 (2010). [9] J. Strumpfer, M. S¸ener, and K. Schulten, Journal of Physical Chemistry Letters 3, 536 (2012). [10] E. Collini, C. Y. Wong, K. E. Wilk, P. M. Curmi, P. Brumer, and G. D. Scholes, Nature 463, 644 (2010). [11] N. F. van Hulst, R. J. Cogdell, D. Brinks, R. Hildner, and J. B. Nieder, Science 340, 1448 (2013). [12] C. Stross, M. W. Van der Kamp, T. A. A. Oliver, J. N. Harvey, N. Linden, and F. R. Manby, The Journal of Physical Chemistry B 120, 11449 (2016). [13] N. Christensson, H. F. Kauffmann, T. Pullerits, and T. Mancal, Journal of Physical Chemistry B 116, 7449 (2012). [14] H.-G. Duan, V. I. Prokhorenko, R. J. Cogdell, K. Ashraf, A. L. Stevens, M. Thorwart, and R. J. D. Miller, Proceedings of the National Academy of Sciences 114, 8493 (2017). [15] T. Mancal and L. Valkunas, New Journal of Physics 12 (2010), 10.1088/1367-2630/12/6/065044. [16] P. Brumer and M. Shapiro, Proceedings of the National Academy of Sciences 109, 19575 (2012). [17] I. Kassal, J. Yuen-Zhou, and S. Rahimi-Keshari, Journal of Physical Chemistry Letters 4, 362 (2013). [18] M. Tiersch, S. Popescu, and H. J. Briegel, Philosophical Transactions of the Royal Society A: Mathe- matical, Physical and Engineering Sciences 370, 3771 (2012). [19] F. Fassioli, A. Olaya-Castro, S. Scheuring, J. N. Sturgis, and N. F. Johnson, Biophysical Journal 97, 16 2464 (2009). [20] M. Sarovar and K. Birgitta Whaley, New Journal of Physics 15, 013030 (2013). [21] D. Manzano, PLoS ONE 8, 1 (2013). [22] P. Brumer, The Journal of Physical Chemistry Letters 9, 2946 (2018). [23] T. Mancal and L. Valkunas, New Journal of Physics 12 (2010), 10.1088/1367-2630/12/6/065044. [24] S. Buckhout-White, C. M. Spillmann, W. R. Algar, A. Khachatrian, J. S. Melinger, E. R. Goldman, M. G. Ancona, and I. L. Medintz, Nature Communications 5, 5615 (2014). [25] E. A. Hemmig, C. Creatore, B. Wunsch, L. Hecker, P. Mair, M. A. Parker, S. Emmott, P. Tinnefeld, U. F. Keyser, and A. W. Chin, Nano Letters 16, 2369 (2016). [26] L. Olejko and I. Bald, RSC Adv. 7, 23924 (2017). [27] A simple kinetics calculation gives the steady-state probability of being in the excited state as p = kabs/(Γsink + Γnr + kspont + kstim + kabs), where kspont, kstim and kabs are, respectively, the rates of spontaneous emission, stimulated emission and absortion. An order of magnitude estimate gives p = 10−9. [28] R. E. Blankenship, D. M. Tiede, J. Barber, G. W. Brudvig, G. Fleming, M. Ghirardi, M. R. Gunner, W. Junge, D. M. Kramer, A. Melis, T. A. Moore, C. C. Moser, D. G. Nocera, A. J. Nozik, D. R. Ort, W. W. Parson, R. C. Prince, and R. T. Sayre, Science 332, 805 (2011). [29] S. R. Greenfield, M. Seibert, Govindjee, and M. R. Wasielewski, The Journal of Physical Chemistry B 101, 2251 (2002). [30] B. Demmig-Adams, G. Garab, and W. Adams Iii, Advances in Photosynthesis and Respiration Includ- ing Bioenergy and Related Processes Non-Photochemical Quenching and Energy Dissipation in Plants, Algae and Cyanobacteria Non-Photochemical Quenching and Energy Dissipation in Plants, Algae and Cyanobacteri (2014). [31] M. A. Palacios, F. L. De Weerd, J. A. Ihalainen, R. Van Grondelle, and H. Van Amerongen, Journal of Physical Chemistry B 106, 5782 (2002). [32] K. L. Zankel, D. W. Reed, and R. K. Clayton, Proceedings of the National Academy of Sciences 61, 1243 (2006). [33] A. Pandit, N. Shirzad-Wasei, L. M. Wlodarczyk, H. Van Roon, E. J. Boekema, J. P. Dekker, and W. J. De Grip, Biophysical Journal 101, 2507 (2011). [34] R. Monshouwer, M. Abrahamsson, F. van Mourik, and R. van Grondelle, The Journal of Physical Chemistry B 101, 7241 (2002). [35] M. a. Bopp, Y. Jia, L. Li, R. J. Cogdell, and R. M. Hochstrasser, Proceedings of the National Academy of Sciences of the United States of America 94, 10630 (1997). [36] A. F. Fidler, V. P. Singh, P. D. Long, P. D. Dahlberg, and G. S. Engel, Journal of Physical Chemistry Letters 4, 1404 (2013). [37] A. J. Schile and D. T. Limmer, arXiv:1905.0029v [cond-mat.stat-mech] (30 Apr 2019). [38] M. Ostilli and C. Presilla, Physical Review A 95, 062112 (2017). 17 [39] P. P. Hofer, M. Perarnau-Llobet, L. D. M. Miranda, G. Haack, R. Silva, J. B. Brask, and N. Brunner, New Journal of Physics 19 (2017), 10.1088/1367-2630/aa964f. [40] F. Caruso, A. W. Chin, A. Datta, S. F. Huelga, and M. B. Plenio, Physical Review A - Atomic, Molecular, and Optical Physics 81, 1 (2010). [41] F. M. W. C. Vaughan, N. Linden, and F. R. Manby, Journal of Chemical Physics 146 (2017), 10.1063/1.4978568. [42] F. M. W. C. Vaughan, Exciton Energy Transfer in Photosynthetic Light Harvesting Complexes, Ph.D. thesis, University of Bristol (2017). [43] V. May and O. Kuhn, Charge and Energy Transfer Dynamics in Molecular Systems (Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, Germany, 2011). [44] T. Renger and R. A. Marcus, Journal of Chemical Physics 116, 9997 (2002). [45] T. A. Stuart, M. Vengris, V. I. Novoderezhkin, R. J. Cogdell, C. Neil Hunter, and R. Van Grondelle, Biophysical Journal 100, 2226 (2011). [46] T. Pullerits, M. Chachisvilis, M. R. Jones, C. N. Hunter, and V. Sundstrom, Chemical Physics Letters 224, 355 (1994). [47] A. Freiberg, K. Timpmann, S. Lin, and N. W. Woodbury, Journal of Physical Chemistry B 102, 10974 (1998). [48] T. Meier, V. Chernyak, and S. Mukamel, The Journal of Physical Chemistry B 101, 7332 (1997). [49] R. E. Blankenship, Molecular Mechanisms of Photosynthesis, 1st ed. (Blackwell Science Ltd, Oxford, UK, 2002). [50] A. Olaya-Castro and F. Fassioli, Procedia Chemistry 3, 176 (2011). [51] K. M. Pelzer, T. Can, S. K. Gray, D. K. Morr, and G. S. Engel, Journal of Physical Chemistry B 118, 2693 (2014). [52] F. Caruso, A. W. Chin, A. Datta, S. F. Huelga, and M. B. Plenio, The Journal of Chemical Physics 131, 105106 (2009). [53] A. Sisto, C. Stross, M. van der Kamp, M. O'Connor, G. T. Johnson, E. G. Hohenstein, F. R. Manby, D. R. Glowacki, and T. J. Martinez, Phys. Chem. Chem. Phy. 19, 5 (2017). [54] C. C. Jumper, S. Rafiq, S. Wang, and G. D. Scholes, Current Opinion in Chemical Biology 47, 39 (2018).
1410.2468	2	1410	2016-07-12T06:31:32	Revealing the properties of the radical-pair magnetoreceptor using pulsed photo-excitation timed with pulsed rf	[ "physics.bio-ph", "quant-ph" ]	The radical-pair mechanism is understood to underlie the magnetic navigation capability of birds and possibly other species. Experiments with birds have provided indirect and in cases conflicting evidence on the actual existence of this mechanism. We here propose a new experiment that can unambiguously identify the presence of the radical-pair magnetoreceptor in birds and unravel some of its basic properties. The proposed experiment is based on modulated light excitation with a pulsed laser, combined with delayed radio-frequency magnetic field pulses. We predict a resonance effect in the birds' magnetic orientation versus the rf-pulse delay time. The resonance's position reflects the singlet-triplet mixing time of the magnetoreceptor.	physics.bio-ph	physics	Revealing the properties of the radical-pair magnetoreceptor using pulsed photo-excitation timed with pulsed rf K. Mouloudakis and I. K. Kominis Department of Physics, University of Crete, Heraklion 71003, Greece The radical-pair mechanism is understood to underlie the magnetic navigation capability of birds and possibly other species. Experiments with birds have provided indirect and in cases conflicting evidence on the actual existence of this mechanism. We here propose a new experiment that can unambiguously identify the presence of the radical-pair magnetoreceptor in birds and unravel some of its basic properties. The proposed experiment is based on modulated light excitation with a pulsed laser, combined with delayed radio-frequency magnetic field pulses. We predict a resonance effect in the birds' magnetic orientation versus the rf-pulse delay time. The resonance's position reflects the singlet-triplet mixing time of the magnetoreceptor. I. INTRODUCTION II. RADICAL-PAIR MODEL USED FOR THE SIMULATIONS Animal magnetoreception [1 -- 4] and specifically avian magnetoreception [5 -- 9] is a long-standing and still unresolved scientific puzzle. A wealth of data [10 -- 18] has made the mag- netic navigation capabilities of birds unquestionable. How- ever, the particular mechanism underlying this capability re- mains elusive. Magnetite crystals in the bird's upper beak [19 -- 23] and the photo-initiated radical-pair mechanism [24] in the avian retina are the two prevalent hypotheses behind the biophysical realization of avian magnetoreceptors. Regard- ing the latter, the specific radical-pair (RP) magnetoreceptor is still unknown, even though cryptochrome has been a major protein candidate supporting magnetic sensitive RP reactions [25 -- 28]. A significant experimental signature of the RP mechanism was the radio-frequency resonance effect [29], where radio- frequency (rf) magnetic fields transverse to the static field and of particular frequencies were shown to disorient the birds. This directly pointed to the RP mechanism since the molecule- specific electron spin resonances are expected to be excited by resonant rf fields. However, a recent experiment studying rf disorientation could not reproduce this resonance effect [30]. Moreover, the magnitude of the disorienting rf fields used in [29, 30] is far smaller than theoretically required by the RP mechanism [31]. To our understanding, experiments with cw light excitation and cw magnetic noise have reached their lim- its in how much more information they can extract. It thus appears that further progress in making a convincing case for the RP compass requires new experimental signatures. We here propose a new experiment using pulsed photo- excitation combined with pulsed rf magnetic fields, in a way that can unambiguously identify the presence of the radical- pair compass and extract its basic parameters. In Section 2 we discuss the RP model used for the analysis. In Section 3 we proceed to examine pulsed photoexcitation pulses fol- lowed by pulsed rf magnetic fields, the rf pulses following the laser pulses by a variable delay time. Singlet RPs are insen- sitive to magnetic fields, while triplet RPs are randomized by the rf magnetic fields. Hence only when the rf pulse is de- layed with respect to the laser pulse by the S-T mixing time will one observe the disorientation of the compass. In Section 4 we discuss the experimental implementation. We use a simple RP model to produce the simulations con- veying the idea behind the proposed experiment. In partic- ular, we consider an RP with one nuclear spin in the donor molecule, having an anisotropic hyperfine coupling with the donor's electron. The hyperfine tensor is considered to have Axx = A and all other elements zero, thus the magnetic Hamil- tonian is H = w (cid:0)cosf (s1x + s2x) + sinf (s1y + s2y)(cid:1) + As1xIx (1) Here w is the electron Larmor frequency in the applied static magnetic field, taken to be on the x-y plane, s1i and s2 j refer to the i-th and j-th spin component of the donor's and acceptor's electron, respectively, and Ix is the x-component of the donor's single nuclear spin. The other pertinent rates are seen in Fig.1. The singlet and triplet recombination rates are taken equal and denoted by k. To close the reaction we also consider an inter- system crossing rate kisc transforming triplet neutral products into the singlet precursors. Light excites the ground state DA molecules to ∗DA at a rate G , and charge transfer leads to the creation of singlet RPs. Since the rate of the latter process is [28] much larger than G and all other rates of the problem, the rate of RP creation is G . For the same reason, i.e. the fact that the population of ∗DA is drained practically instantaneously, there is no need to consider stimulated emission of the excit- ing light. The population of the singlet precursors DA is taken to be the signaling state carrying the magnetic field information to- wards further neural processing leading to the bird's orienta- tion. In many spin-chemistry calculations the RPs are consid- ered to be all initialized in the singlet state at time t = 0 and one then calculates the reaction yields resulting at the end of a single reaction cycle. For this work, however, we need to con- tinuously create RPs at a rate G and calculate the steady-state population of the neutral DA molecules, Sg, in the scheme of a continuously running and closed reaction of Fig.1. To do so, we add a source term to the Haberkorn master equation for the RP density matrix r : dr dt = G Sgr 0 − i[H ,r ] + R(r ), (2) where r 0 = QS/Tr{QS} is the initial density matrix of singlet RPs having zero nuclear spin polarization, and R(r ) = − kS 2 (QSr + r QS) − kT 2 (QTr + r QT) (3) is the reaction super-operator describing singlet and triplet RP recombination. We used the traditional (Haberkorn) master equation, since any quantum effects [32] beyond this approach are not relevant to this work. Nevertheless, we checked the results of our master equation, involving singlet-triplet deco- herence, and they are qualitatively the same. The first term in Eq. (2) creates G Sg RPs per unit time in the state r 0. To close the reaction we also consider the following two rate equations for Sg and the corresponding triplet ground state population, Tg: dSg dt dTg dt = −G Sg + kSTr{QSr } + kiscTg = kTTr{QTr } − kiscTg (4) (5) The first of the above equations describes the depopulation of Sg by photoexcitation at the rate G and the population of Sg by (i) the singlet RP recombination and (ii) the intersystem crossing from TDA at the rate kisc. The second describes the depopulation of TDA at the rate kisc and its population by the triplet RP recombination. Finally, when solving the system of equations (2), (4) and (5), the initial condition is Sg(t = 0) = 1. Before moving to the main part of this work, i.e. the pulsed photoexcitation for which the excitation rate G is time- dependent, we first discuss the continuous illumination case G = const in order to get some insight into the quantities of interest. We first note that in our numerical work (except for the Hamiltonian evolution of Fig.3) all rates will be given rel- ative to the recombination rate k = kS = kT = 1. Accordingly, time will have units 1/k = 1. In Fig. 2 we plot the steady-state population Sg, evaluated numerically from (2), (4) and (5), as a function of f for two values of constant G is the angle between the mag- netic field (lying on the x-y plance) and the x-axis defining the hyperfine anisotropy. The avian compass is based on the f -modulation of the population Sg. We define maxf {Sg} − minf {Sg} maxf {Sg} + minf {Sg} , where f D S ≡ (6) and call it f -visibility. The measured heading error in exper- iments with birds is inversely proportional to D S. It is seen that the higher G , the faster is drained the ground state DA, hence the smaller its steady state population. For the pulsed photo-excitation we use an average excitation rate G = 0.25. What is of interest for the time-delay resonance effect to be presented in the following is the time evolution of the RP state resulting just from the Hamiltonian term in the master equa- tion (2). Using this Hamiltonian time evolution, we plot in Fig.3 the triplet state probability hQTi as a function of time for three different angles f . It is seen that the first instance in time when the triplet state is reached, i.e. when hQTi ≈ 1, is largely independent of f and, as expected, scales as 1/A. 2 *DA charge transfer Radical-pair with density matrix ρ Γ(t) SD + A H TD + A kS kT DA kisc TDA Singlet ground state with population Sg Triplet ground state with population Tg FIG. 1: Radical-pair reaction dynamics. The population of the sin- glet donor-acceptor precursor DA is considered to be the signal car- rying the magnetic field information into deeper stages of neural processing. This population is drained by photoexcitation at the rate G (t), which in this particular work is time-dependent. It is in- creased by the radical-pair singlet recombination and by the inter- system crossing from the triplet ground state, introduced in order to close the reaction. The singlet and triplet recombination rates are kS and kT, respectively, and H is the magnetic Hamiltonian induc- ing singlet-triplet oscillations between the singlet and triplet radical- pairs, SD•+A•− and TD•+A•−. The charge transfer from the photo- excited molecule ∗DA is much faster than all other rates, hence the rate of creation of radical-pairs is effectively G . Sg 0.42 0.41 0.40 0.39 0.38 0.37 0.36 Γ=0.25 Γ=1.0 0 π/4 π/2 3π/4 φ Sg 0.150 0.145 0.140 0.135 0.130 0.125 0.120 π FIG. 2: Angular modulation of the singlet ground state (DA) popu- lation for two different values of a constant excitation rate, G = 0.25 (dashed blue line) and G = 1.0 (solid red line). The parameters of the RP model are w = kS = kT = 1, A = 10 and kisc = 0.1. For the higher excitation rate the state DA is depleted faster and hence both the population Sg and the difference maxf {Sg} − minf {Sg} become smaller. III. PHOTOEXCITATION PULSES FOLLOWED BY RF PULSES We will here provide a detailed analysis of the idea of the proposed experiment. There are three main ingredients to this idea. First, as well known, the singlet state is not sensitive T Q 1.0 0.8 0.6 0.4 0.2 0.0 φ=0 φ=π/4 φ=π/2 (a) Γ(t) τ Γ0 Tr τST ~~ 6/A 0 5 10 15 20 time (1/A) (b) ΩRF(t) τd Tr Ω0 τRF FIG. 3: Singlet-triplet mixing driven by the Hamiltonian H of Eq. (1) with w = 1. We plot the triplet expectation value hQTi as a func- tion of time (in units of 1/A) for three different angles f . It is seen that the first instance of S-T conversion is independent of f and takes place at a time t ST ≈ 6/A for this particular Hamiltonian. (c) 3 t t to any magnetic field, constant or alternating. The mecha- nism through which the avian RP compass is disoriented by rf fields necessarily starts with the induced spin randomization of the triplet state. Second, if the photo-excitation is pulsed, the transformation of singlet RPs to triplet RPs takes place in well defined times, given the S-T mixing frequency W ST. Third, if the radio frequency pulses are delayed with respect to the light pulses, as shown in Fig.4, it is expected that by varying the delay time t d, the birds' magnetic orientation, as measured by D S, will exhibit a resonance, as an increasing de- lay will correspond to an increasingly triplet character of the RP's spin state. The resonance dip will happen at a particu- lar delay t d such that the RPs that were photo-excited to the singlet state will have oscillated into a predominantly triplet spin character. Observing this resonance dip will thus (i) un- ambiguously reveal the presence of the radical-pair magne- toreception mechanism and (ii) unravel the mixing frequency W ST of the particular magneto receptor molecule. The above picture is exemplified in the following. The photo-excitation rate G (t) is shown in Fig.4a. It consists of pulses of amplitude G 0, pulse width t and repetition time Tr. The amplitude of the photo-excitation pulses, G 0, is given a value such that the time average G of G (t) is the same as the G = 0.25 case of continuous excitation shown in Fig.2. We choose t = 0.005 for the pulse width and Tr = 2 for the pulse repetition time, hence G 0 = G Tr/t = 100. To include the presence of the pulsed rf we add to the Hamiltonian (1) the term Hrf = W rf(t)cos(w rft + y )(cid:0)s1z + s2z(cid:1) (7) We took the rf magnetic field to be polarized along the z- axis, perpendicular to the static magnetic field lying on the x-y plane. W rf(t) is the pulse train envelope shown in Fig.4b. The pulse amplitude and width are W 0 and t rf, respectively. The FIG. 4: (a) Photo-excitation rate G (t), consisting of a pulse train with pulse amplitude G 0, pulse width t and pulse repetition time Tr. (b) Envelope of the rf field W rf(t), consisting of a pulse train with pulse amplitude W 0, pulse width t rf and pulse repetition time Tr. This pulse train is delayed from the photo-excitation pulse train by t d. (c) Rf carrier wave modulated by the envelope shown in (b). In order for the rf frequency spectrum to be continuous and simulate noise we insert a random pulse-to-pulse phase difference y . 0 = 15w 0, is taken W pulse delay time with respect to the photoexciation pulses is t d, which is variable. The pulse repetition time is the same as for G (t), i.e. Tr. The amplitude of the rf magnetic field, given in terms of its Rabi frequency W , i.e. the rf-field amplitude is 15 times earth's field. We note that this is way higher than the rf-field amplitudes experimentally found to disorient the birds. As mentioned in the introduction and clearly stated in [30], it is still an unresolved puzzle why the theoretically required rf-field amplitude is so much higher than what is experimentally observed to disorient the birds. We further elaborate on this point in the following Section on the experimental implementation. Finally, we take t rf = 0.1. The rf carrier we use, shown in Fig.4c, is a cosine wave of frequency w rf = 20. In the experiment one must use pulsed noise of a bandwidth similar to [30]. To simulate that theo- retically we include a pulse-to-pulse random phase y in the cosine wave. Without this phase the rf pulse train would have a discrete Fourier spectrum. With the inclusion of these ran- dom phases we theoretically simulate the pulsed rf noise since now the Fourier spectrum of W rf(t)cos(w rft + y ) is continu- ous and has a bandwidth of about 1/t rf. In Fig.5 we depict the time-delay resonance effect. The change of D S from the off- resonant to the on-resonant time delay is significant enough (about a factor of 3) that the compass should disorient on res- onance. We see that by varying the hyperfine coupling A the A=10 A=5 following numerical estimates we take 1/k = 1 m s. In any case, an educated guess of k must be made in order to set the timescale of the experiment. 4 1.0 0.5 τd (units of 1/k) 1.5 3.0 2.5 2.0 1.5 ) % ( S Δ 1.0 0.0 FIG. 5: Time-delay resonance effect predicted in this work. Shown is the f -visibility as a function of the time delay t d of the rf-pulses with respect to the laser pulses, for two different values of the hyperfine coupling A. For the radical-pair we took k ≡ kS = kT = 1, ksic = 0.1 and w = 1. For the pulse trains we took G 0 = 100, t = 0.005, Tr = 2, t rf = 0.1, W rf = 15 and w rf = 20. It is seen that for higher A, singlet- triplet mixing is taking place faster, hence the time-delay required to hit the triplet state is smaller. For zero time delay the f -visibility for this model is about 3%, and at the resonance dip it falls by a factor of 3 for the chosen value of W rf. resonance's position is shifted in accordance with Fig.3. That is, according to Fig.3, the S-T mixing time is about 6/A, and for the two values used for the hyperfine coupling, A = 5 and A = 10, the position of the time-delay resonance is t d ≈ 1 and t d ≈ 0.5, respectively. The different resonance width ob- served in Fig.5 is due to the different interplay of the S-T mix- ing (dependent on A) with the pulse repetition time Tr. We finally note that Fig.5 was produced by a moving average of the actual result in order to remove a (still visible) modulation artifact stemming from the numerical scanning of the delay time t d. We have checked that the resonance phenomenon persists for a multi-nuclear spin radical pair. In particular, we run the same simulation for a radical-pair containing up to 4 nu- clear spins. We note that by choosing the relevant hyperfine couplings so that the angular modulation of Fig. 2 is signif- icant, we also obtain a significant resonance dip like in Fig. 5. In other words, it appears that if the compass has evolved to reach an optimum angular yield dependence, it will ex- hibit the resonance effect we presented. On the other hand, by no means do we claim that the effect will be experimen- tally detected no matter what. What we claim is that this is a viable measurement to do with live birds, and if the reso- nance phenomenon is realized, it will provide for a clean and information-rich signature of the radical-pair magnetorecep- tor. IV. EXPERIMENTAL IMPLEMENTATION All rates of the problem have been expressed relative to the recombination rate k, which was given the value 1. For the A. Laser Pulses Pulsed lasers with pulse duration on the order of 1-10 ns, a repetition rate on the order of 200-500 kHz and a wave- length within the sensitivity window of the avian magnetore- ceptor are commercially available. The pulsed laser can be fed into a diffuser and illuminate the birds' cage just like the regular illumination with lamps or diodes. For ns lasers, any pulse broadening by the diffuser is negligible given the much slower reaction and magnetic dynamics. In other words, since we took t = 0.005 (in units of 1/k) for the laser pulse width, any pulse broadening will leave the pulse width still much smaller than the magnetic and recombination dynamics tak- ing place at the timescale 1/k = 1. Regarding the laser pulse peak intensity, in the case of continuous illumination a flux of about 1016 photons/s/m2 is known [34, 35] to be enough for the compass to function. Assuming a total illumination area on the order of 1 m2, the light source's average power should then be about 5 mW (at 500 nm). We took the pulse width to be 400 times smaller than the pulse repetition time, so to get the same average photoexcitation rate the pulse peak power should be 2 W. For a 1 ns pulse this translates into a pulse energy of 2 nJ, which is well within the capabilities of commercially available and simple table-top lasers. B. Radio-frequency pulses In our calculations we took the rf pulse width to be t rf = 0.1, which is small enough compared to a typical mixing fre- quency W ST ≈ 1 (see Fig.3). This pulse width translates to 100 ns. In producing Fig.5 we scanned the delay time in steps of 0.02, translating to 20 ns. To summarize, we need 50-100 ns wide rf pulses modulating noise of bandwidth of about 10 MHz, the delay of the pulses being scanned in steps of about 20-50 ns. Such rf pulse generators are commercially avail- able. Similarly, the power of the rf magnetic field should be the one used in [30] scaled up by the ratio Tr/t rf ≈ 20 since now we have pulsed and not continuous rf. Again, this is read- ily achievable. V. CONCLUSIONS We have proposed an experiment using pulsed photo ex- citation in conjunction with properly delayed pulses of radio frequency magnetic fields to study the response of avian mag- netic orientation. If the radical-pair mechanism is indeed re- sponsible for the avian compass, a robust resonance will ap- pear in the measured birds' orientation versus delay time be- tween laser and rf pulses. Further, the particular delay time at the resonance's dip is the inverse of the singlet-triplet mix- ing frequency of the magneto receptor molecule. We analyzed this experiment using a generic radical-pair model, but the re- alization of the experiment as well as the result we obtained for the time-delay resonance effect is robust and independent of the particular radical-pair model. For example, one could consider an RP with just one non-zero recombination rate, e.g. the singlet, and no intersystem-crossing. The singlet ground state population would again be the signaling state, depend- ing on f through the different time spent by the RP in the triplet state. Similar results would be obtained in this case. The same experiment could also be used for other magneto receptive species [36 -- 38] in which the RP mechanism is pre- sumed to exist. 5 Acknowledgments We acknowledge support from the European Union's Sev- enth Framework Program FP7-REGPOT-2012-2013-1 under grant agreement 316165. [1] T. P. Quinn and E. L. Brannon, J. Comp. Physiol. 147, 547 [21] A. F. Davila, M. Winklhofer, V. P. Shcherbakov and N. Peter- (1982). [2] J. B. Phillips and S. C. Borland, Nature 359, 143 (1992). [3] S. Johnsen and K. J. Lohmann, Phys. Today 61 (3), 29 (2008). [4] S. Qin et al., Nature Mat. 15, 217 (2016). [5] R. Wiltschko and W. Wiltschko, Animal Behaviour 65, 257 (2003). [6] W. Wiltschko and R. Wiltschko, Science 176, 62 (1972). [7] H. Mouritsen, Nature 484, 320 (2012). [8] D. T. Rodgers and P. J. Hore, Proc. Natl. Acad. Sci. USA 106, 353 (2009). [9] T. Ritz, S. Adem and K. Schulten, Biophys. J. 78, 707 (2000). [10] C. Walcott and R. P. Green, Science 184, 180 (1974). [11] R. Wiltschko and W. Wiltschko, Naturwissenschaften 65, 112 son, Biophys. J. 89, 56 (2005). . [22] I. A. Solovyov and W. Greiner, Biophys. J. 93, 1493 (2007). [23] M. M. Walker, J. Theor. Biol. 250, 852008 (1999). [24] K. Schulten, C. E. Swenberg and A. Weller, Z. Phys. Chem. 111, 1 (1978). [25] A. Cashmore, J. Jarillo, Y. J. Wu and D. Liu, Science 284, 760 (1999). [26] H. Mouritsen et al., Proc. Natl. Acad. Sci. USA 101, 14294 (2004). [27] I. A. Solovyov, D. Chandler and K. Schulten, Biophys. J. 92, 2711 (2007). [28] I. A. Solovyov and K. Schulten, J. Phys. Chem. B 116, 1089 (2012). (1978). [29] T. Ritz, P. Thalau, J. B. Phillips, R. Wiltschko and W. [12] U. Munro, J. A. Munro, J. B. Phillips, R. Wiltschko and W. Wiltschko, Nature 429, 177 (2004). Wiltschko, Naturwissenschaften 84, 26 (1997). [13] M. E. Deutschlander, J. B. Phillips and S. C. Borland, J. Exp. Biol. 202, 891 (1999). [14] G. A. Gudmundsson and R. Sandberg, J. Exp. Biol. 203, 3137 (2000). [15] M. N. Williams and J. M. Wild, Brain Res. 889, 243 (2001). [16] M. Zapka et al., Nature 461, 1274 (2009). [17] D. Heyers, M. Zapka, M. Hoffmeister, J. M. Wild and H. Mouritsen, Proc. Natl. Acad. Sci. USA 107, 9394 (2010). [30] S. Engels et al., Nature 509, 353 (2014). [31] K. V. Kavokin, Bioelectromagnetics 30, 402 (2009). [32] I. K. Kominis, Mod. Phys. Lett. B 29, 1530013 (2015). [33] R. Wiltschko, K. Stapput, H.-J. Bischof and W. Wiltschko, Frontiers in Zoology 4 (2007) 5. [34] W. Wiltschko, R. Wiltschko and U. Munro, Naturwis- senschaften 87, 36 (2000). [35] W. Wiltschko and R. Wiltschko, Naturwissenschaften 89, 445 (2002). [18] H. Mouritsen and P. J. Hore, Current Opinion in Neurobiology [36] R. J. Gegear, L. E. Foley, A. Casselman and S. M. Reppert, 22, 343 (2012). Nature 463, 804 (2010). [19] J. L. Kirschvink and J. L. Gould, Biosystems 13, 181 (1981). [20] V. P. Shcherbakov and M. Winklhofer, Eur. Biophys. J. 28, 380 [37] B. Paulus et al., Febs 282, 3175 (2015). [38] E. P. Malkemper et al., Sci. Rep. 4, 9917 (2015). (1999).
1905.06377	1	1905	2019-05-15T18:33:45	Limitations in Predicting Radiation-Induced Pharmaceutical Instability during Long-Duration Spaceflight	[ "physics.bio-ph", "q-bio.BM" ]	As human spaceflight seeks to expand beyond low-Earth orbit, NASA and its international partners face numerous challenges related to ensuring the safety of their astronauts, including the need to provide a safe and effective pharmacy for long-duration spaceflight. Historical missions have relied upon frequent resupply of onboard pharmaceuticals; as a result, there has been little study into the effects of long-term exposure of pharmaceuticals to the space environment. Of particular concern are the long-term effects of space radiation on drug stability, especially as missions venture away from the protective proximity of the Earth. Here we highlight the risk of space radiation to pharmaceuticals during exploration spaceflight, identifying the limitations of current understanding. We further seek to identify ways in which these limitations could be addressed through dedicated research efforts aimed towards the rapid development of an effective pharmacy for future spaceflight endeavors.	physics.bio-ph	physics	Limitations in Predicting Radiation-Induced Pharmaceutical Instability during Long-Duration Spaceflight Rebecca S. Blue1,2,, Jeffery C. Chancellor3, Erik L. Antonsen4,5, Tina M. Bayuse6, Vernie R. Daniels6, and Virginia E. Wotring7 1Aerospace Medicine and Vestibular Research Laboratory, The Mayo Clinic Arizona, Scottsdale, AZ 85054, USA 2GeoControl Systems, Inc., Houston, TX 77058, USA 3Department of Physics & Astronomy, Texas A& M University, College Station, TX 77843, USA 4Department of Emergency Medicine and Center for Space Medicine, Baylor College of Medicine, Houston, TX 77030, USA 5National Aeronautics and Space Administration (NASA), Johnson Space Center, Houston, TX 77058, USA 6KBRwyle, Houston, TX 77058, USA 7Department of Pharmacology and Chemical Biology and Center for Space Medicine, Baylor College of Medicine, Houston, TX 77030, USA Author to whom correspondence should be addressed: Rebecca Blue, [email protected] ABSTRACT As human spaceflight seeks to expand beyond low-Earth orbit, NASA and its international partners face numerous challenges related to ensuring the safety of their astronauts, including the need to provide a safe and effective pharmacy for long-duration spaceflight. Historical missions have relied upon frequent resupply of onboard pharmaceuticals; as a result, there has been little study into the effects of long-term exposure of pharmaceuticals to the space environment. Of particular concern are the long-term effects of space radiation on drug stability, especially as missions venture away from the protective proximity of the Earth. Here we highlight the risk of space radiation to pharmaceuticals during exploration spaceflight, identifying the limitations of current understanding. We further seek to identify ways in which these limitations could be addressed through dedicated research efforts aimed towards the rapid development of an effective pharmacy for future spaceflight endeavors. Introduction With the expansion of human spaceflight outside of low-Earth orbit (LEO), NASA and its international partners face numer- ous challenges related to ensuring the safety of their astro- nauts. Among these challenges is the ability to provide a safe and effective pharmacy with sufficient capability to manage both planned and unforeseen medical conditions that may arise during flight. The ability to provide a safe and effec- tive pharmacy to crews is contingent upon multiple factors, such as the stability of any medication for the duration of a given mission, the effectiveness of that medication in the unique space environment, and the provision of appropriate and sufficient medications to meet the unique physiological and psychological challenges the crew may face. There is a paucity of evidence regarding pharmaceutical stability in the space environment, largely because this issue has not historically been a pressing concern for human space- flight. Short-duration flights of the Mercury, Gemini, Apollo, and Space Shuttle eras minimized the need for prolonged medication shelf life, and the selection of healthy crewmem- bers minimized the need for ongoing medication provision for chronic disease. Careful maintenance of crew health and stringent flight rules regarding the more dangerous activities during spaceflight, such as extravehicular activity (EVA), have largely obviated the need for emergency medication provision. Even now, with missions to the International Space Station (ISS) lasting 6 months or longer, crews have been able to rely on medication availability through retirement of expired medications and frequent resupply rather than contending with questions of degradation, storage, and the impact of the space environment (with environmental concerns related to a myriad of factors such as vibration, humidity, and space radiation exposure). As a result, investments in the systematic collection of data for the characterization of medication use, efficacy, side effects, pharmacokinetics, pharmacodynamics, and long-term stability have been a lower priority than other health and human performance investments. With the push for exploration missions to the moon and Mars, these questions have become a more pressing concern. One potential risk to pharmaceutical stability arises from long-term exposure to the space radiation environment. While gamma radiation exposure has been used terrestrially for ster- ilization procedures in select pharmaceuticals, space radiation differs considerably from such practices because of differ- ences in type of radiation, dose, dose-rate, and length of exposure. It is unclear whether long-term exposure to space radiation may affect stability, alter drug ingredients, or pro- duce potentially toxic byproducts, particularly in drugs that have undergone degradation reactions. 1 Here, we seek to present the current understanding of phar- maceutical stability in the space radiation environment. In particular, we have attempted to highlight the gaps in current knowledge and the difficulties in translating terrestrial-based radiation studies to a meaningful interpretation of drug re- sponse to space radiation. We hope to identify high-yield opportunities for future research that might better define and mitigate the space radiation risk to a future formulary for exploration spaceflight. The Interplanetary Space Radiation Envi- ronment The effects of radiation are due to the transfer of energy from a charged particle to the medium it travels through. The amount of energy that can be transferred is a function of the particle's kinetic energy, charge, and mass.1, 2 The effects of indirect ionizing radiation (e.g. gamma, x-ray) are negliglible compared to effects caused by direct ionizing charged particle risk. The more charge a particle has the greater ability it has to ionize the medium it traverses, depositing more energy per unit path length (defined as increased linear energy transfer, or LET) in a traversed material. Future space exploration endeavors will include manned ex- peditions beyond the protection of the Earth's magnetic field. These long-duration missions, which may span months to years, will require additional protection for the human crews on board. The space radiation environment is a complex mix of charged particles originating from several sources. Within an exploration vehicle (one intended to travel outside of the Earth's geomagnetic field), the the intravehicular radiation environment primarily consists of relativistic heavy-charged particles attributed to galactic cosmic rays (GCR, chronic, isotropic background radiation). The GCR spectrum, and thus the intravehicular radiation environment, primarily consists of ionized hydrogen (protons, approximately 85%) and he- lium (alpha, approximately 14%) nuclei, but also includes less abundant ionized particles of higher atomic weight.3, 4 Despite their rarity, heavier particles contribute a dispropor- tionately high amount of overall radiation dose-exposure due to their relatively high LET. In addition, as GCR ions pass through vehicular structures, interaction with vehicle materi- als can cause fragmentation (or "spallation") of heavier ions into more numerous particles of lower atomic weight. This process can produce cascades of ions resulting in a destructive capability in addition to that of the primary ions. Accounting for all such interactions increases the complexity of predicting the intravehicular radiation environment. It is particularly difficult to shield from GCR exposures given the isotropic, highly penetrating nature, and the relative energies of the GCR spectra.5 An additional, off-nominal source of charged particle radia- tion can be attributed to solar particle events (SPEs), where particles are ejected from the sun in prompt and short-lived bursts of energy. SPEs consist primarily of protons and elec- trons with a relatively small contribution from heavy nuclei. Unlike GCR, SPEs are anisotropic.6 SPE radiation is pri- marily composed of protons with kinetic energies ranging from 10 MeV up to several GeV (determined by the rela- tivistic speed of particles).3 SPEs are capable of accelerating an abundance of protons that can occasionally result in high dose-rates in the interplanetary environment. For example, a particularly large event in October 1989 is predicted to have delivered dose-rates as high as 1,454mGy/hour for a short pe- riod of time to an exposed astronaut in a vehicle with 5g/cm2 of aluminum-equivalent shielding traveling in interplanetary space.3, 7 While rare, SPE exposure would be in addition to the nominal intravehicular dose, attributed to GCR nuclei, expected to be approximately 0.028mGy/hour during travel in interplanetary space.4 Interplanetary intravehicular doses would be altered by the peak flux, energy spectrum, and du- ration of any given SPE as well as shielding thickness and material makeup of the vehicle. Similarly, the contribution of radiation exposure from SPEs and resultant effects on phar- maceuticals would depend upon intravehicular dose and any additional shielding. As missions to the moon or Mars will expand human pres- ence from LEO to interplanetary space, intravehicular radia- tion exposure will increase. Vehicles, and the pharmaceuticals on board, will be exposed to higher cumulative GCR exposure and increased risk for transient SPE exposures. As a result, the risk of radiation-induced alterations of pharmaceutical sta- bility, structure, potency, and potential toxicity will increase with future missions (Figure 1). Mechanisms of Radiation Impact A majority of pharmaceutical radiation risk research is derived from terrestrial analogs rather than the full particle and energy spectrum of the space radiation environment. Accordingly, differences in the relative abilities of terrestrial and space radiation to induce damage in a target have yet to be elucidated. This property of different types of radiation to induce different levels and kinds of damage is known as radiation quality. Radiation quality is thought to be dependent on LET, which can be characterized by energy deposition pattern. Charged particles traverse a material in an approximately straight line, transferring energy through interactions with the medium's nuclei and electrons. Imparted energy may be enough to knock an electron out of an atom, ionizing the atom or leaving it in an excited, non-ionized state. The ejected electron can have enough energy to leave the immediate vicinity of the charged particle's path and produce a notable track of its own. This results in a densely ionizing core along the charged particle's path (where energy continues to be deposited in an approximately straight line), as well as a sparsely ionizing penumbra generated by expelled electrons (where energy is deposited throughout the material randomly). In addition to differences in radiation quality, substrate com- position is an important factor in radiation-induced damage. To date, most radiation research has been conducted in biolog- ical models, where a majority of the substrate is water. In this 2/11 Figure 1. Factors limiting understanding of pharmaceutical stability in the space radiation environment. Radiation from galactic cosmic rays (GCR) is not graphically depicted but should be considered ubiquitous in the space environment. PK/PD: Pharmacokinetics/Pharmacodynamics. scenario, radiation is more likely to hit water than a biologi- cally relevant target (e.g. DNA). However, even if radiation impacts water rather than a target, it can still induce damage to a target via generation of free radicals, which can diffuse to interact and ionize a target within range. This form of damage is known as indirect ionization, while damage caused from radiation hitting a target is known as direct ionization. In pharmaceuticals, where a greater percentage of substrate composition consists of target molecules, direct ionization is far more likely than in biological substrates. The difference between percentage of target interactions that are direct ver- sus indirect is highly dependent on substrate and projectile energy. It has been observed that direct ionization can cause in- creased damage compared to indirect ionization, particularly in the decomposition of chemical bonds, creation of radiolysis products, and damage to polymer structure.8 -- 10 Studies using biological substrates therefore do not provide good analogues for pharmaceutical research. Furthermore, clustered damage imparted by densely ionizing space radiation combined with the higher target concentration in pharmaceutical formularies could potentially interact, resulting in outcomes that have not yet been characterized. Challenges in Reproducing Radiation Dose, Dose-Rate, and Formulation Sensi- tivity Another area of uncertainty is the effect of low doses on pharmaceuticals. Terrestrial pharmaceutical radiosterilization techniques generally use doses of 25-50kGy, which far exceed those expected for even cumulative Mars mission doses (ap- proximately 0.5Gy). Delivery of radiosterilization doses over a matter of minutes or hours considerably exceeds dose-rates anticipated in interplanetary space, where such doses would be accrued over 2-3 years by current estimates. It has been suggested that, if a pharmaceutical is found to be stable at higher doses or dose-rates (such as those provided by radiosterilization techniques), then the pharmaceutical should be stable at more limited exposures (such as those delivered in space).11 Some evidence for this argument has been provided by the expected level of damage from indirect ionization. De- spite the relatively higher concentration of target molecules in pharmaceutical than biological substrates, damage to pharma- 3/11 ceuticals from indirect ionization does still occur, particularly in liquid pharmaceuticals where target molecules are less concentrated than in solid formulations. Indirect ionization- induced damage stems from the formation of free radical species; in water-based formulations, this includes the gener- ation of radical oxygen species (oxygen ions and hydrogen peroxide, H2O2) from the breakdown of water.12 -- 14 Studies have indicated that the concentration of these radicals from exposures to radiosterilization doses (25-50kGy) is generally well below toxic levels.15, 16 A recent NASA technical paper indicated that nanomolar concentrations of radiation byprod- ucts could be produced in exposed pharmaceuticals, but cited low anticipated radiolytic yield in liquid-based pharmaceuti- cals (based on modeled calculations) as sufficient evidence that irradiated pharmaceuticals should be stable in the space environment.11 However, even the low nanomolar concentrations of ions predicted by that technical report could be enough to suf- ficiently alter local pH in drug products, which could alter chemistry or drive degradation reactions.18 In addition, stud- ies using electron spin resonance, a sensitive method for the detection of free radicals,19, 20 have demonstrated that alter- ation of radiation dose changes the concentration and type of free radicals produced, often with unpredicted complexity or type of resultant radical species.20 These complex reactions could alter subsequent radical-induced damage.20 Dose-rate may be an important factor in the activity of radical species, and many pharmaceuticals are demonstrated to be more stable at higher dose-rates. It has been theorized that high dose- rate increases oxygen consumption, resulting in decreased presence of oxygen radicals (or the rapid consumption of any radical species generated) and associated damage.16, 20 Shorter-duration exposures may produce fewer long-lived oxy- gen radical species and, as a result, less prolonged opportunity for delayed damage than protracted exposures. In evidence to these arguments, one historical study of spaceflight-approved pharmaceuticals compared drug stability at variable radiation dose ranging from 0.1-50Gy and found drug degradation asso- ciated with moderate radiation exposure where no instability was noted at higher doses.17 It is worth reiterating that these dose ranges include exposures that are substantially greater than even cumulative anticipated doses in long-duration, ex- ploration spaceflight. Liquid pharmaceuticals are often considered less stable than solid or powdered drugs given the greater potential for free radical formation in water-based formulations, the pos- sibility of interactions between substrate and excipients (the pharmacologically inert compounds in a given dosage for- mulation), incomplete dissolution of substrate, crystallization of dissolved compounds, and other alterations of drug sus- pensions over time. Water-based drugs will undergo more frequent hydrolysis reactions, driving more prevalent and more rapid degradation reactions. As a result, the rare discus- sions of pharmaceutical stability in the context of space radia- tion have focused on liquid formulations and postulated that, should liquid formulations be determined to be stable in the space radiation environment, solid or semi-solid formulations would be of no additional concern.11 However, some studies demonstrate radiation-induced instability in solid or powder formulations, with reports of radical trapping in excipient lat- tices leading to a longer presence of free radicals in powder or solid drugs than in liquid formulations.21 -- 24 In addition, the interaction of particles with solid or powder substrates may produce increased types and complexity of ion species due to spallation. Spallation ions impacting stored pharmaceuti- cals may cause increased direct and indirect ionizations or induce additional chemical reactivity in the substrate. In short, it is unclear how drugs of any formulation may respond to the unique qualities of space radiation, simply because such responses have not been studied to any degree of fidelity. Given the uncertainty of drug response to alterations of dose, dose-rate, or exposure time, radiosterilization is only approved for well-documented procedures of declared dosage (most commonly 25-50kGy) and dose-rate, and deviation from the designated dose or dose-rate is assumed to be capable of al- tering the final drug product.16, 20 In evidence to this concern, the United States Pharmacopeia (USP) regards radiosterilized pharmaceuticals as entirely new products, and pharmaceuti- cal companies are required to submit new drug applications and demonstrate safety, potency, and lack of toxic breakdown products for approval of radiosterilization in any marketed drug.16, 25 The use of terrestrial analog radiation to predict the response of pharmaceuticals in the space environment di- rectly contradicts the standard approach to safety and stability review of irradiated pharmaceuticals. Finally, it should be noted that many of the pharmaceuticals currently included in a spaceflight formulary are not approved by the U.S. Food and Drug Administration (FDA) for terrestrial radiosterilization procedures. Challenges in Emulating the Space Envi- ronment Accurate simulation of the complex space radiation environ- ment for pharmaceutical testing via terrestrial analog is cur- rently not possible, given limitations in radiation type and dose-rate of exposure. Space radiation studies, pharmaceuti- cal or otherwise, often make use of a recently updated GCR simulator at the NASA Space Radiation Laboratory (NSRL) at Brookhaven National Laboratory in Brookhaven, New York. To date, the NSRL is the only U.S. government facility ca- pable of generating heavy-charged particles at energies and spectra that approximate the space environment.26, 27 Recent improvements now allow for rapid switching between ion species, providing rapid and consecutive exposures to dif- ferent mono-energetic ion beams.26 Rapid switching of ion beams may be sufficient in simulation of the complex space environment, particularly as previous modeling has suggested that the likelihood of multiple ion species traversing a small volume at the same time (i.e. traversing a single drug tablet) is exceedingly low.28 This rapid switching technique is a poten- 4/11 Figure 2. NASA data from an NSRL study performed by L. Putcha demonstrating variable drug sensitivity to radiation exposure for clavulanate (as a combination medication, amoxicillin-clavulanate) and promethazine.17 All drug products were measured at time-zero, control and irradiated products were analyzed at the same time following exposures. The solid green line indicates USP-accepted lower limits of percent API content compared to label claims. Note the variable sensitivity both by radiation beam exposure (proton, in red, or iron, in blue) and by dose received (0.1-50Gy). In this study, drugs demonstrated increased degradation to 10Gy exposures compared to 50Gy exposures, suggesting that pharmaceutical stability at higher-dose exposure may not necessarily translate to stability at lower-dose exposures. However, dose and dose-rate of high exposures were significantly greater than even cumulative anticipated doses in long-duration, exploration spaceflight. Further, there is only limited documentation regarding research design or even the full results of this study, limiting our ability to interpret findings. tial improvement when compared to use of photon or single ion exposures and offers more insight than studies with con- siderably higher doses than those expected during spaceflight, such as the doses used in radiosterilization literature.3, 26 However, even this simulator utilizes exposures that are appreciably different from those anticipated in the interplane- tary space environment, generally delivering cumulative an- ticipated mission doses over short periods of time.26 While the NSRL simulator is capable of providing more protracted doses, limitations of funding for long-term experiments gen- erally limit exposure times, causing deviation of the analog from GCR. The simulator cannot generate the full spectrum of ions or spallation ions that make up the GCR spectrum; instead, exposures are limited to only a sampling of some of the heavy ions that contribute to GCR, and these ions are delivered sequentially rather than simultaneously. Further, the simulator lacks the capacity to generate the pions (sub- atomic particles) or neutrons that would follow spallation reactions in the intravehicular environment,3, 27 though these would be expected to account for 15-20% of an intravehicular exposure.27, 29 These factors may limit ability to translate terrestrial analog studies to an understanding of the true risk of space radiation pharmaceutical exposure. Even so, the NSRL simulator is one of the few simula- tors available to study space-like radiation in the terrestrial environment. To date, there are remarkably few studies of pharmaceuticals at this facility. In 2011, Chuong et al. studied the stability of solid formulations of vitamin B during space- flight and in terrestrial radiation analogs, making use of an older radiation simulator at the NSRL for some exposures.30 The authors studied vitamins that had flown onboard the Space Shuttle and ISS for 2-4 weeks or 12-19 months, comparing them to terrestrial controls and vitamins exposed to the terres- trial radiation beam. While the NSRL exposures were used as a terrestrial radiation study arm to examine radiation effects on the vitamin, it is noteworthy that the NSRL exposures were mono-energetic exposures of 0.1-50Gy, using either hydrogen or iron radiation sources.30 The upper limits of exposures in this study greatly exceed those expected during even long duration and exploration spaceflight. The USP allows variation of the active pharmaceutical ingredient (API, the ingredient imparting the desired physi- ological effect of a medication) within 90-150% of package label content for vitamin B. The Chuong study did identify statistically significant, though acceptable, variation in con- tent of ground (unflown, non-irradiated) and NSRL irradi- ated samples, and even identified one flown sample with API concentration well below acceptable ranges.30 However, the authors suggest that instability was most likely related to API formulation, excipient interaction, or even packaging, and stated that, since NSRL samples were found to be stable (at notably higher dose-rate than flown samples), radiation was not the cause of instability.30 As discussed above, this rea- soning is questionable given the numerous factors that limit translation of simulated radiation exposures to the true space environment. Additional research conducted at the NSRL by NASA on numerous pharmaceuticals in an effort to delineate the effects 5/11 H+H+FeFeof various radiation doses on drug stability similarly utilized 0.1-50Gy doses of proton or iron mono-energetic beams.17 Data released suggests dose-variable alterations of API, with increased degradation noted at 10Gy exposures compared with those at 50Gy exposures delivered over equivalent time intervals (Figure 2).17 Study exposures are still notably higher than those expected during spaceflight but suggest that sta- bility at high-dose exposures may not necessarily translate to stability at low-dose exposures, and that dose and dose-rate alterations may significantly impact stability. Unfortunately, there is only limited documentation regarding research design or even the full results of this study, limiting our ability to fully interpret findings. Mechanisms of Pharmaceutical Instability Pharmaceuticals can become unstable through alteration of either their physical or chemical properties. Alteration of physical properties includes changes in appearance or con- sistency; alteration of chemical properties includes loss of potency, alteration of excipients, excipient-active ingredient interactions, or toxic degradation.31, 32 In order to determine that a pharmaceutical is unchanged by exposure to the radi- ation environment, a drug must be demonstrated following exposure to have no significant alteration of its API(s) while at the same time have no significant development of degradation products that are either toxic themselves or in some way alter the pharmaceutical properties of the original medication.18 The USP provides guidelines for acceptable API content in medications approved by the FDA, commonly within 10% of label-specified content (though this can vary considerably by drug type or API).25 A medication would be considered radiosensitive if API concentration fails to meet USP require- ments following radiation exposure. Alterations of API can affect drug potency, efficacy, and safety rendering the drug less effective, ineffective, or potentially dangerous. There are numerous documented cases of pharmaceuti- cals being altered by radiation exposure at sterilization doses (25-50kGy). For example, irradiation of metoclopramide hydrochloride produced a number of degradation products fol- lowing radiation exposure,33 and gamma sterilization of cer- tain beta-blockers has been demonstrated to alter the pharma- ceuticals' color and appearance and affect the melting point of the drug preparations.34 Even compounds that are molecularly similar may have vastly different responses to irradiation.35 Cephradine and cefotaxime, both solid-form cephalosporin antibiotics of similar molecular structure, demonstrate signif- icantly different radiosensitivity when exposed to identical sterilization doses of gamma radiation. Cephradine degrades significantly and has been determined to be unstable under irradiation36 where cefotaxime demonstrates high resistance and stability.37, 38 As molecular alterations can change the saturation of the compound or the presence or absence of re- active groups such as alcohols, acids, or ketones, even minor differences in API structure can affect radiosensitivity. In addition to altering API, radiation exposure can result in the generation of degradation products and may alter the medication, whether or not the API is affected, by damaging the structure or action of excipients. For example, radiation is known to alter the chemical structure of various polymer drug delivery systems, causing increased cross-linking of poly- mers in some cases and inducing polymer chain breakage in others.35, 39 Cross-linked polymers, with higher molecular weight, may cause issues with insolubility,40 and chain break- age of some polymeric microspheres used for drug delivery have been associated with high production of free radicals and instability of the resultant compounds.35, 41 In some stud- ies, alteration of excipients has been demonstrated to affect dissolution rates and controlled release of API.42, 43 Radiosensitivity is highly specific to dose, dose-rate, radi- ation type (photon, electron, proton, heavy ion, etc), chem- ical composition, excipient content, and drug formulation. However, it must also be stated that any radiation-induced pharmaceutical risk must be weighed in the context of the multitude of other factors that may render a flown drug un- stable in the space environment. Mission duration will soon extend beyond approved shelf life for many medications cur- rently included in onboard medical kits. Current medications aboard the ISS are replenished through regular resupply and removal of older drugs; this may not be possible with future missions to the moon or Mars.18 Older drugs may be at higher risk of degradation from chronic exposure to the radiation environment. NASA currently repackages some of the flown pharmaceu- ticals to manage mass and volume constraints and to limit packaging waste in the closed environment of a space ve- hicle. However, repackaging itself may affect shelf life or stability of stored medications or alter their response to ra- diation exposure.18, 44 For example, nuclei interacting with packaging material could produce additional progeny ions that alter the chemical composition of pharmaceuticals within.31 Previous studies have suggested various packaging materials that may be intrinsically better for radiation shielding, such as polyethylene;31, 45 -- 47 however, there are insufficient data regarding ideal packaging technique or long-term shelf life of pharmaceuticals packaged in such materials, and any novel packaging approach intended for use onboard future missions would be subject to USP review and guidelines.25 Ultimately, choice of packaging materials should address radiation sensi- tivity as well as additional shelf life concerns, particularly as it remains unclear how the factors of drug age, repackaging, shelf life, and radiation exposure will interact to determine pharmaceutical response. There have been very few examinations of pharmaceuti- cals actually exposed to the space environment, including a ground-controlled study of API in flown pharmaceuticals conducted by Du et. al.49 and a convenience sampling of pharmaceuticals returned to Earth after ¿550 days aboard the ISS by Wotring (see Figure 3).48 More recently, Cory et. al.51 and Wu et. al.52 sought to analyze potency, pu- rity, and drug degradation in certain pharmaceuticals flown 6/11 Figure 3. To date, there have been few studies of pharmaceuticals flown in the space environment. The studies presented in the figure included various evaluations of active pharmaceutical ingredient (API), physical characteristics, impurity products, and degradation, as indicated.30, 48 -- 52 Only one study by Chuong et. al.30 included "radiation arm," a subset of ground controls that were irradiated with either hydrogen or iron ions at high dose and dose-rate dissimilar to the space environment. Drugs in red text were found to have alterations of API, physical characteristics, or contain significant concentrations of degradants or impurities after flight in one or more preparation of the indicated pharmaceutical. Multivitamin preparations were analyzed only for B-complex API stability. Drugs contained API concentrations within acceptable limits at time of study analysis, but would fail API analysis according to current standards. †Drugs contained unspecified or unidentified impurity products of unknown significance. •Multivitamin content demonstrated time-related instability but showed no alteration specifically related to spaceflight exposure. aboard the ISS, though these results have yet to be published. Two additional papers addressed multivitamin stability after spaceflight exposure, including Chuong et al.30 and Zwart et al.50 In general, most pharmaceuticals tested after flight have been found to meet USP requirements for API concen- tration, though notable exceptions occurred. For example, Du et. al. found that amoxicillin-clavulanate, levofloxacin, trimethoprim, sulfamethoxaxole, furosemide, and levothyrox- ine degraded before their expiration dates.18, 49 The study additionally identified alterations of physical appearance of some medications. Wotring identified degradation and impu- rity products in aspirin, ibuprofen, loratadine, modafinil, and zolpidem.48 The two multivitamin studies identified alteration of multivitamins over time in both ground and flown samples when compared to time-zero controls, but neither found con- vincing evidence of degradation specific to spaceflight-flown formulations.30, 50 While these studies have provided at least some much-needed pilot data, they are limited by the ability to provide adequate ground-control, control of confounders, or appropriate reproducibility given limited sample size, and can provide only an initial awareness that flown pharmaceuti- cals may not be stable in the space environment. It is worth emphasizing that pilot data do suggest that expected radia- tion exposures may be sufficient to affect medication stability. While yet unpublished, reported results from more recent ex- periments performed by Cory et. al. and Wu et. al. were similarly limited by exposure-time variables, limited ground controls, and drug lot variability, but again suggest instability despite theoretical expectations to the contrary.51, 52 Finally, despite decades of pharmaceutical use in space- flight, there is limited knowledge regarding alterations of pharmacokinetics (absorption, metabolism, and excretion of a medication) and pharmacodynamics (drug effects on the body) in the space environment. As the human body under- goes significant physiological and metabolic changes during spaceflight, it stands to reason that the effects of pharmaceu- ticals on an astronaut may change during flight.53 However, research on this issue has largely been limited to observational 7/11 API and Physical CharacteristicsAcyclovirAmoxicillin/ClavulanateAtorvastatinAzithromycinCefadroxilCipro(cid:31)oxacinClotrimazoleDextroamphetamineEpinephrineFluconazoleFurosemideIbuprofenImipenem/CilastinLevo(cid:31)oxacinLevothyroxineLidocaineMetoprololMetronidazoleMupirocinNasal CobolaminePhenytoinProgestin/EstrogenPromethazineRisedronateSertralineSilver SulfadiazineSulfamethoxazole/ TrimethoprimTemazepamTriamcinoloneCentrum Silver® MultivitaminWomen's Once-A-Day® MultivitaminAspirinAcetaminophenIbuprofenLoratadine* †Loperamide †PseudoephedrineMelatonin Moda(cid:30)nil †Zolpidem †13 Days354 Days597 Days881 DaysVariable Exposure550 DaysVariable ExposureAPI (B Vitamin only)API, Degradants UnpublishedResultsGround Control AvailableUnpublishedResultsVariable ExposureIbuprofenLevo(cid:31)oxacinPhenytoinSertralineValacyclovirIbuprofenPromethazineAzithromycin13 Days353 Days596 Days880 DaysCentrum Silver® Multivitamin •Vitamin D Supplement •APIRadiation Armreports and analog studies.18, 53 Without directed studies to examine the multifactorial impact of the space environment on pharmaceutical response, it is difficult to fully understand how the additional risks from space radiation may further alter drug response, if at all, during exploration missions. Discussion Numerous confounders, limited spaceflight studies, and chal- lenges in translation of terrestrial analog evidence to space- flight have all hindered our ability to draw meaningful con- clusions regarding the stability of pharmaceuticals during exploration spaceflight. As NASA looks towards the chal- lenges associated with missions involving increased distance from Earth, the current inability to provide a safe and effective pharmacy for exploration spaceflight has been identified as a major research gap.54 To address this issue, NASA recently developed a Pharmacy Research Plan in which pharmaceuti- cal stability and radiation risk are highlighted as unknowns that should be addressed in dedicated research efforts prior to lunar or Mars missions.55 However, this research plan faces challenges including approaching mission design-freeze deadlines and a need to declare a planned formulary for fast- approaching exploration missions, expected to occur within the next decade of spaceflight. As an adjunct to NASA's research plan, recent literature has provided potential solutions for storage- and radiation-related stability concerns. For example, there has been some sugges- tion that cryogenic storage conditions may be protective to pharmaceuticals during spaceflight.31, 56 Such methods have been demonstrated to be successful during radiosterilization processes, providing increased stability of medications during gamma or x-ray exposure.31, 57, 58 Even so, some formulations may demonstrate decreased stability with freezing; for many drugs, effects are unknown or unstudied. There have been no studies of cryogenically stored pharmaceuticals exposed to space-like radiation doses, dose-rates, or spectral complexity. It is difficult to predict the response of cryogenic pharma- ceuticals to the space environment, given the multitude of confounding factors and the paucity of data available. Similarly, previous literature has discussed the potential inclusion of "space-hardy" formulations, such as use of excipi- ents believed to be more stable in a radiation environment.31, 48 For example, formulations including starch, stearate, cellu- lose, and dextrose may be more likely to be stable than alterna- tives, based on results from the 2016 Wotring study31, 48 Other options include preparations including excipients such as man- nitol, nicotinamide, and pyridoxine, which have demonstrated radioprotective properties in terrestrial sterilization process- ing.31 A more thorough discussion of potential excipients for improved stability, radioprotective qualities, and antioxidant effects can be found in Mehta et al.31 However, it should be reiterated that much of the literature supporting inclusion or exclusion of excipients for protective or stability properties is again based on incomplete data, convenience sampling, or radiation exposures dissimilar to the space environment, lim- iting the translation of findings particularly for long-duration, exploration missions. Further, altering or adding excipients would change drug formulation; the resultant product would be considered a new drug and, per USP regulations, would require a new application for drug approval and demonstration of safety, potency, and lack of toxic breakdown products. Finally, there has been discussion of limiting the impact of pharmaceutical irradiation through the inclusion of onboard shielding.56 In collaborative efforts to protect human crew from radiation exposure, there has been much discussion re- garding the inclusion of a heavily-shielded compartment of thick aluminum or other radioprotective material, or alter- natively by the use of "multi-purpose shielding solutions," such as barriers composed of water or food supplies, on ex- ploration vehicles.56, 59, 60 Pharmaceuticals could be stored within a shielded compartment to reduce radiation exposures. While these innovative efforts show promise, it is important to remember that shielding designs may be limited by mass and volume constraints and lift-mass capabilities of the launching vehicle. It is premature to assume that idealized shielding will be successfully implemented in early exploration vehicles. Many of the shield designs are intended for "just-in-time" deployment for protection of crew;60 in such circumstances, crew would have to retrieve onboard pharmacy stores and transfer them into the shielded space to protect drugs from SPEs. Even if a high degree of shielding were to be imple- mented, such a compartment would only mitigate transient exposures associated with large SPEs, and protracted expo- sure to GCR would continue to pose a threat to drug stability in long-duration spaceflight. Ultimately, successful mitigation of radiation risk relies upon a more thorough understanding of the potential effects of radiation upon pharmaceuticals, insight regarding which pharmaceuticals are at highest risk for radiation-induced dam- age, and an awareness of how the myriad of spaceflight-related factors (e.g. altered pharmacokinetics and pharmacodynam- ics, radiation dose, radiation dose-rate, packaging, shelf life, etc.) affect an exposed drug. Careful and controlled study of pharmaceutical stability, with ground controls and appropriate sample size, would greatly improve our understanding of the multifactorial risks to pharmaceuticals in space. Additional ground-based studies comparing the effects of gamma, x-ray, or electron beam to proton or heavy ion exposure may improve understanding of how to better translate terrestrial literature to the context of space radiation. Utilization of the ISS as a research platform, with long-duration storage of pharmaceu- ticals and well-designed and controlled studies of shelf life and radiation exposure, could provide much-needed under- standing of stability in actual spaceflight conditions. However, such studies would need to be initiated rapidly, as the ISS is intended for decommissioning within the next decade. With rapidly approaching exploration mission dates, NASA and its international partners seek a mature pharmaceutical formulary that can be realized before vehicle and mission design freezes occur. Inclusion of pharmaceuticals (particularly novel or 8/11 complex pharmaceuticals not currently included in the ISS formulary) onboard future manned or unmanned missions outside of LEO could provide additional data of drug stability in the actual deep-space environment; however, most of these missions do not return payloads to the Earth, limiting analysis options. Maximizing analysis opportunities requires program- matic commitments to sample return; this in turn depends upon an increased understanding of the importance of these data. As a comprehensive research plan onboard the ISS may not be feasible given time and financial constraints, study of comparative effects of single- and multi-energetic exposures may improve our understanding of the complexity of the space radiation environment and its impact on pharmaceuticals. As discussed above, use of dose and dose-rates that more closely emulate the space environment may provide more useful or accurate results than reliance upon high dose and dose-rate ex- posures. Further, inclusion of shielding materials in terrestrial analog design, to both mitigate dose and simulate potential spallation reactions, may better mimic the ionic composition of the intravehicular environment.29 Comparative studies of drugs both including and excluding shielding or packaging ma- terials may provide insight regarding the relative contribution of such materials to pharmaceutical degradation reactions. Careful and thorough evaluation of pharmaceuticals ex- posed to the space environment is overdue, and the paucity of data limits appropriate translation of terrestrial studies for understanding of space radiation exposures. It is critical to address these knowledge gaps before missions to the moon or Mars are underway. There are significant advances that can be achieved by a well-planned research effort that provides both actual flight data from flown pharmaceuticals aboard available research platforms, such as the ISS, as well as translational studies of comparative effects of space-like radiation in ter- restrial analogs, allowing for better interpretation of historical terrestrial radiation understanding in the context of space- flight. Use of improved and robust modeling techniques that better emulate the space environment, careful study of various formulations, alternate drug choices, and packaging materi- als, and consideration of novel techniques, such as cryogenic storage, could provide much-needed advances towards the development of a pharmaceutical capability for interplanetary flight. Acknowledgments The authors acknowledge Kerry Lee, Caitlin Milder, and S. Robin Elgart, in association with NASA's Space Radiation Analysis Group, for their assistance. We further acknowledge the support of the Exploration Medical Capability Element and the Clinical Pharmacy at NASA Johnson Space Center. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily repre- senting the official policies or endorsements, either expressed or implied, of the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Gov- ernmental purpose notwithstanding any copyright annotation therein. JC acknowledges the Texas Advanced Computing Center (TACC) at The University of Texas at Austin for pro- viding HPC resources that have contributed to the research results reported within this paper. Competing interests The authors declare that they have no competing interests. Author Contributions RB developed the concept of the review. JC contributed to the discussion on space physics. All authors contributed to the review of the literature, discussion on the interpretation of research outcomes to spaceflight operations, and drafting of the manuscript. References 1. Attix, F. H. Introduction to radiological physics and radiation dosimetry (Wiley, New York, 1986). 2. Hall, E. J. & Giaccia, A. J. Radiobiology for the radiol- ogist (Wolters Kluwer Health : Lippincott Williams and Wilkins, Philadelphia, PA, 2012). 3. Chancellor, J. et al. Limitations in predicting the space radiation health risk for exploration astronauts. NPJ Mi- crogravity 4 (2018). 4. Zeitlin, C. et al. Measurements of energetic particle radiation in transit to mars on the mars science laboratory. science 340, 1080 -- 1084 (2013). 5. Cucinotta, F., Kim, M. & Chappell, L. Space radiation cancer risk projections and uncertainties - 2012. Tech. Rep. NASA/TP-2013-217375, National Aeronautics and Space Administration (2013). 6. Gombosi, T. I. Physics of the space environment (Cam- bridge University Press, 1998). 7. Hu, S., Kim, M.-H. Y., McClellan, G. E. & Cucinotta, F. A. Modeling the acute health effects of astronauts from exposure to large solar particle events. Health Phys. 96, 465 -- 476 (2009). 8. Kempner, E. Effects of high-energy electrons and gamma ray directly on protein molecules. J Pharm Sci 90, 1637 -- 46 (2001). 9. Kempner, E. Direct effects of ionizing radiation on macro- molecules. J Polym Sci Part B Polym Phys 49, 827 -- 31 (2011). 10. Silindir, M. & Ozer, A. Sterilization methods and the comparison of e-beam sterilization with gamma radiation sterilization. J Pharm Sci 34, 43 (2009). 11. Kim, M. & Plante, I. An assessment of how radiation incurred during a mars mission could affect food and 9/11 pharmaceuticals. Tech. Rep. NAS9-02078, Wyle Science, Technology, and Engineering Group (2015). 12. Jacobs, G., Donbrow, M., Eisenberg, E. & Lapidot, M. The use of gamma-irradiation for the sterilization of water for injections and normal saline solution for injection. Acta Pharm Suec 14, 287 -- 92 (1977). 13. Jacobs, G. & Eisenberg, E. The reconstruction of powders for injection with gamma-irradiated water. Int J Appl Radiat Isot 32, 180 -- 1 (1981). 14. Plessis, T. The radiation sterilisation of pyrogen-free water in polyethylene sachets. Radiat Phys Chem 14, 289 -- 98 (1979). 15. Eisenberg, E. & Jacobs, G. The development of a formula- tion for radiation-sterilizable urea broth. J Appl Bacteriol 58, 21 -- 5 (1985). 16. Jacobs, G. Radiation sterilization of parenterals. Pharm Technol 2, 1 -- 6 (2007). 17. Daniels, V., Bayuse, T., McGuire, K., Antonsen, E. & Putcha, L. Radiation impact on pharmaceutical stability: Retrospective data review. Tech. Rep. 09940, Proceed- ings of the NASA Human Research Program Investigator Workshop (2018). 18. Wotring, V. Space Pharmacology (Springer: International Space University, New York, NY, 2012). 19. Crucq, A. & Tilquin, B. Method to identify products induced by radiosterilization: A study of cefotaxime sodium. J Pharm Belg 51, 285 -- 8 (1996). 20. Gibella, M. et al. Electron spin resonance studies of some irradiated pharmaceuticals. Radiat Phys Chem 58, 69 -- 76 (2000). 21. Ambro, H., Kornacka, E., Marciniec, B., Orgrodowczyk, M. & Przybytniak, G. EPR study of free radicals in some drugs gamma-irradiated in the solid state. Radiat Phys Chem 58, 357 -- 66 (2000). 22. Dicle, I. EPR Study of Radiation-Induced Radicals in Glutaric and Amino Acid Derivatives in Solid State. Ra- diat Eff Defects Solids 170, 393 -- 8 (2015). 23. Koseoglu, R., Koseoglu, E. & Koksal, F. Electron param- agnetic resonance of some gamma-irradiated drugs. Appl Radiat Isot Data Instrum Methods Use Agric Ind Med 58, 63 -- 8 (2003). 24. Taiwo, F., Patterson, L., Jaroszkiewicz, E., Marciniec, B. & Ogrodowczyk, M. Free radicals in irradiated drugs: an epr study. Free Radic Res 31, 231 -- 5 (1999). 25. United States Pharmacopeia and National Formulary (USP40-NF35) (United States Pharmacopeial Convention, Rockville, MD, 2016). 26. Norbury, J. W. et al. Galactic cosmic ray simulation at the NASA Space Radiation Laboratory. Life Sci. Space Res. 8, 38 -- 51 (2016). 27. Slaba, T. et al. Simulator reference field and a spectral ap- proach for laboratory simulation. Tech. Rep. NASA/TP- 2015-218698 (2015). 28. Curtis, S. Fluence rates, delta rays and cell nucleus hit rates from galactic cosmic rays. Tech. Rep., National Aeronautics and Space Administration (2013). 29. Chancellor, J. C., Guetersloh, S., Cengel, K., Ford, J. & Katzgraber, H. Emulation of the Space Radiation Environment for Materials Testing and Radiobiological Experiments. Tech. Rep. 1706.02727 (2017). 30. Chuong, M., Prasad, D., Leduc, B., Du, B. & Putcha, L. Stability of Vitamin B Complex in Multivitamin and Multimineral Supplement Tablets after Space Flight. J Pharm Biomed Anal 55, 1197 -- 200 (2011). 31. Mehta, P. & Bhayani, D. Impact of space environment on stability of medicines: Challenges and prospects. Journal of Pharmaceutical and Biomedical Analysis 136, 111 -- 19 (2017). 32. Carstensen, J. & Rhodes, C. Drug Stability: Principles and Practices, Third Edition (CRC Press, New York, NY, 2000). 33. Maquille, A., Jiwan, J.-L. & Tilquin, B. Radiosterilization of drugs in aqueous solutions may be achieved by the use of radioprotective excipients. Int J Pharm 249, 74 -- 82 (2008). 34. Marciniec, B., Ogrodowczyk, M., Czajka, B. & Hofman, M. The influence of radiation sterilisation on some beta- blockers in the solid state. Thermochim Acta 514, 10 -- 15 (2011). 35. Hasanain, F., Guenther, K., Mullett, W. & Craven, E. Gamma sterilization of pharmaceuticals -- a review of the irradiation of excipients, active pharmaceutical ingredi- ents, and final drug formulations. PDA J Pharm Sci Technol 68, 113 -- 37 (2014). 36. Signoretti, E. et al. Ionizing radiation induced effects on cephradine: Influence of sample moisture content, irradiation dose, and storage conditions. Drug Dev Ind Pharm 19, 1693 -- 708 (1993). 37. Barbarin, N., Crucq, A. & Tilquin, B. Study of volatile compounds from the radiosterilization of solid cephalosporins. Radiat Phys Chem 48, 787 -- 94 (1996). 38. Barbarin, N., Tilquin, B. & de Hoffmann, E. Radiosteril- ization of cefotaxime: investigation of potential degrada- tion compounds by liquid chromatography - electrospray mass spectrometry. J Chromatogr 929, 41 -- 61 (2001). 39. Chapiro, A. Physical and chemical effects of ionizing radiations on polymeric systems. In Technical Develop- ments and Prospects of Sterilization by Ionizing Radiation (1974). 40. Martini, L., Collett, J. & Attwood, D. The influence of gamma irradiation on the physicochemical properties 10/11 of a novel triblock-copolymer of beta-caprolactone and ethylene oxide. J Pharm Pharmacol 49, 601 -- 5 (1997). 41. Montanari, L. et al. Gamma irradiation effects on poly (dl- lactictide-co-glycolide) microspheres. J Control Release Soc 56, 219 -- 29 (1998). 42. Maggi, L. et al. Chemical and physical stability of hydrox- ypropylmethylcellulose matrices containing diltiazem hy- drochloride after gamma irradiation. J Pharm Sci 92, 131 -- 41 (2003). 43. Maggi, L. et al. Polymers-gamma ray interaction: Effects of gamma irradiation on modified release drug delivery systems for oral administration. Int J Pharm 269, 343 -- 51 (2004). 44. Wotring, V. Risk of therapeutic failure due to ineffective- ness of medication. Tech. Rep. NASA/JSC-CN-32122, National Aeronautics and Space Administration (2011). 45. Guetersloh, S. et al. Polyethylene as a radiation shielding standard in simulated cosmic-ray environments. Nucl. Instrum. Methods Phys. Res 252, 319 -- 32 (2006). 46. Wook, J. et al. Polyethylene/boron-containing compos- ites for radiation shielding. Thermochim Acta 585, 5 -- 9 (2014). 47. Harrison, C. et al. Polyethylene/boron nitride composites for space radiation shielding. Polym. Sci. 109, 2529 -- 38 (2008). 48. Wotring, V. Chemical Potency and Degradation Prod- ucts of Medications Stored over 550 Earth Days at the International Space Station. AAPS J 18, 210 -- 6 (2016). 49. Du, B. et al. Evaluation of physical and chemical changes in pharmaceuticals flown on space missions. The AAPS journal 13, 299 -- 308 (2011). 50. Zwart, S., Kloeris, V., Perchonok, M., Braby, L. & Smith, S. Assessment of nutrient stability in foods from the space food system after long-duration spaceflight on the ISS. J. Food Sci. 74, H209 -- 217 (2009). 51. Cory, W., James, V., Lamas, A., Mangiaracina, K. & Moon, J. Analysis of Degradation of Pharmaceuticals Stored on the International Space Station. Tech. Rep. 17091, Proceedings of the NASA Human Research Pro- gram Investigator Workshop (2017). 52. Wu, L. & Chow, D. Degradation Analysis of Medications from ISS Using LC-MS/MS Assays: NSBRI RFA 15- 01 First Award Fellowship, Final Report. Tech. Rep., National Space Biomedical Research Institute (2016). 53. Gandia, P., Saivin, S. & Houin, G. The influence of weightlessness on pharmacokinetics. Fundam Clin Phar- macol. 19, 625 -- 36 (2005). 54. Antonsen, E. et al. The risk of adverse health outcomes and decrements in performance due to in-flight medical conditions. Tech. Rep. NASA/JSC-20170004604, Na- tional Aeronautics and Space Administration (2017). 55. Daniels, V., Bayuse, T., Mulcahy, R., McGuire, K. & An- tonsen, E. The pathway to a safe and effective medication formulary for exploration spaceflight. Tech. Rep. 13658, Proceedings of the NASA Human Research Program In- vestigator Workshop (2017). 56. Jaworske, D. & Myers, J. Pharmaceuticals exposed to the space environment: Problems and prospects. Tech. Rep. NASA/TM-2016-218949, National Aeronautics and Space Administration (2016). 57. Meents, A., Gutmann, S., Wagner, A. & Schulze-Briese, C. Origin and temperature dependence of radiation dam- age in biological samples at cryogenic temperatures. Proc. Natl. Acad. Sci. 107, 1094 -- 9 (2010). 58. Moyne, P., Botella, A. & Rey, L. Sterilization of in- jectable drugs solutions by irradiation. Radiat. Phys. Chem. 63, 703 -- 4 (2002). 59. Huff, J. et al. Evidence report: Risk of radiation carcino- genesis (2016). 60. Simon, M., Cerro, J. & Clowdsley, M. Radworks storm shelter design for solar particle event shielding. Tech. Rep. NF1676L-15945, NASA Langley Research Center (2011). 11/11
1109.0743	1	1109	2011-09-04T19:29:19	Stochastic modeling of p53-regulated apoptosis upon radiation damage	[ "physics.bio-ph", "q-bio.MN" ]	We develop and study the evolution of a model of radiation induced apoptosis in cells using stochastic simulations, and identified key protein targets for effective mitigation of radiation damage. We identified several key proteins associated with cellular apoptosis using an extensive literature survey. In particular, we focus on the p53 transcription dependent and p53 transcription independent pathways for mitochondrial apoptosis. Our model reproduces known p53 oscillations following radiation damage. The key, experimentally testable hypotheses that we generate are - inhibition of PUMA is an effective strategy for mitigation of radiation damage if the treatment is administered immediately, at later stages following radiation damage, inhibition of tBid is more effective.	physics.bio-ph	physics	Stochastic modeling of p53-regulated apoptosis upon radiation damage Divesh Bhatt, Zoltan Oltvai, and Ivet Bahar. 1 Introduction Studying cellular response to radiation damage is important from the perspectives of both radiotherapy and the mitigation of radiation damage. In the case of the former, the goal is the induction of cell death, whereas in the latter case a cellular response leading to organismal survival is required. The fate of the organism after radiation damage is linked in a complex manner to cellular fate. In the short term, an extensive cell death due to radiation damage will lead to the death of the organism. In the long term (relevant only if the organism survives the short term effects), it the survival of healthy and robust cells that dic- tate the organism's survival (for example, absence of long term cancerous cells). In this manuscript, we focus on the immediate survival of the organism after radiation damage. The mechanism of cell death after radiation damage is purported to be apoptosis via caspase activation that occurs several hours after radiation injury. Thus, it is natural to investigate cellular apoptotic machinery. Cellular response to radiation is very complex and involves, possibly, proteins that respond to DNA damage and the formation of free radicals that modify cellular biochemistry. Although the exposure to radiation may be momentary, the effect of the exposure on cellular biochemistry may be long lived. Further, several proteins that are expressed transiently after radiation damage, may trigger downstream response such as caspase activity that occurs long after their expression. In view of the complex cellular response, protein -- protein interactions that lead to that response, and possible importance of temporal evolution of the system, it is important to study systemically the time-dependent cellular response to stress. 1 There have been various efforts to model cellular response to radiation, as well as to model apoptosis. Efforts to model cellular response to radiation were largely pre- cipitated by the observation, at the level of a single cell, of oscillatory p53 response to radiation damage. 1 In particular, the oscillatory response of p53 to radiation damage has been modeled mostly using deterministic simulations with kinetic rate laws, 1 -- 4 and, to a significantly lesser extent, via stochastic simulations. 5 Cellular apoptosis has been modeled, independent of p53 radiation response, via deterministic simula- tions by several research groups, 6 -- 9 including efforts to include cell -- cell variability in a probabilistic manner. 10,11 Recently, there have been a few efforts to link the p53 response to cell sur- vival/death using deterministic simulations, 12 -- 14 or with inclusion of limited stochas- ticity. 15 Due to increasingly larger pool of experimental observation on the proteins involved, and their cellular compartmentalization, in the apoptotic network, model- ing of transient system behavior remains an active area of research and serves as a platform for evaluating pharmacological strategies. Spatio-temporal modeling using stochastic simulation methods allows for detailed examination of the kinetics of biochemical reaction networks within a cell. From such simulations, one can address several different issues: spatial localization of reac- tions (e.g., on the mitochondrial membrane), as well as studying the time evolution of molecular concentrations in response to perturbations to the system. These per- turbations can be radiation stress on the cell, or the response to a given drug dose. Further, stochasticity allows for the inclusion of cell variability in a natural way, as well as for studying conditions where the number of certain types of molecules become very low and their presence/absence dictates system behavior. In this study, we perform stochastic simulations to study the time evolution of pro- teins, associated with apoptotic pathways, in response to radiation damage. Further, we check the efficacy of various treatment strategies, including polypharmacological effects. In particular, we use different inhibitors targeting different proteins in the reaction network. The manuscript is organized as follows. We first present the bio- 2 chemical reaction network, gathered from experimental observations, associated with radiation induced apoptosis. Then, we describe briefly the stochastic simulations that we perform to study cell response to radiation damage, followed by results and discussion. 2 Biochemical reaction network We present here the chain of biochemical events that we use to model apoptotic mechanisms that are activated after radiation damage. Instead of adopting a com- prehensive model that includes all the cellular processes that are activated, or affected, by radiation damage, we focus on a number of key proteins/genes as a first approxi- mation. An overview of the key interactions is given in Figure 1. Table 1 gives details about the proteins involved in the system. Several other proteins besides those in Figure 1 are included in the model and described in Table 1, since an interaction map, such as that in Figure 1, does not contain details of the biochemical events that lead to the individual interactions. Before we describe the radiation response of the network, we present the generic underlying network. 2.1 p53 module p53 is an integral component of cell death due to genotoxic stress (including ra- diation damage), and it activates apoptosis via both transcription -- dependent and transcription -- independent pathways. 16 -- 18 The exact effect of p53 in its transcription- independent role in apoptosis is still in-debate (see a recent review by Lindenboim et al. 19), we attempt to incorporate some of the key experimental observations. The response of p53 to radiation exposure has been studied in great detail at the individual cell level. The response of p53 to radiation determined experimentally forms the starting point of our modeling effort: we first focus on the p53 "module" (as highlighted in Figure 1 with the dashed rectangle) and calibrate our model to 3 reproduce the known p53 response. p53 exists in several forms in a cell based on its post-translational modifications (e.g., phosphorylation, ubiquitination at different sites). In this manuscript, we utilize a model that captures the known biological features: such as p53 oscillatory behavior upon radiation damage, 1,2,20 and a stress -- induced increase in nuclear p53 21 and its transcriptional activity. 22,23 p53 is translated in the cytoplasm (p53C) and this formation is represented by Φ κ1−→ p53C (1) where Φ represents a null state, and the above reaction represents the formation of p53C under normal operating conditions given by the rate κ1. p53C can translocate to the nucleus, especially when a cell undergoes stress. 21 This translocation is modeled by p53C κ2−→ p53n (2) where p53n is the nuclear p53. In the nucleus, p53n transcriptionally activates Mdm2 mRNA and this transcription activation is modeled as cooperative: 3 p53n tetramers on the DNA leads to this transcription activation. The oligomerization of p53n is depicted by 4p53n k+ 3−⇀↽− k− 3 p53no (3) where the reaction order of the forward reaction is 4. The oligomerized p53 leads to a direct transcription activation of Mdm2: p53no κ4−→ p53no + Mdm2m (4) The Mdm2 messenger RNA translocates to the cytoplasm where it is translated. The cytoplasmic Mdm2 (Mdm2C) can be imported into the nucleus. This series of events is described by the following reactions: Mdm2m κ5−→ Mdm2mC (5) 4 Mdm2mC κ6−→ Mdm2C Mdm2C κ7−→ Mdm2 (6) (7) The monoubiquitinated, nuclear p53 (p53NU) is formed via the interaction of p53n with Mdm2 (or similar ligases), 24 and we represent this by p53n + Mmd2 k+ 8−⇀↽− k− 8 p53.Mdm2 κ8−→ p53NU + Mdm2 A similar reaction is modeled to occur for the cytoplasmic p53 and Mdm2: p53C + Mmd2C k+ 9−⇀↽− k− 9 p53C.Mdm2 κ9−→ p53CU + Mdm2C Further, the ubiquitinated nuclear p53 can translocate to the cytoplasm: p53NU κ10−→ p53CU (8) (9) (10) The ubiquitinated cytoplasmic p53 can translocate to the mitochondria 25 as shown below: p53CU κ11−→ p53m (11) Upon severe genotoxic stress, p53n become transcriptionally active for pro-apoptotic proteins 23 such as PUMA and Bax, and such a transcriptionally active form is rep- resented by p53T: p53n κ12−→ p53T (12) Although the major role of p53T is transcriptional activation of pro-apoptotic pro- teins, we let it retain the capacity for oligomerizing and activating Mdm2 (this, though, is a minor effect), 4p53T k+ 13−⇀↽− k− 13 p53To p53To κ14−→ p53To + Mdm2 (13) (14) The increase in p53 upon radiation damage is not expected to be a step function. Similarly, return of p53 to baseline levels after removal of radiation source is not expected to be instantaneous. Indeed, oscillatory levels of p53 persist for an extended 5 period of time after radiation damage. 2 Further, p53 response to radiation is mediated by upstream proteins such as ATM. We do not model such upstream agents in specific detail, instead we represent the upstream events by the following set of reactions. Φ k+ 15−⇀↽− k− 15 A A κ16−→ A + B B κ17−→ Φ B κ18−→ B + p53C (15) (16) (17) (18) Thus, A activates B, which in turn activates p53C. The effect of these upstream reactions is a smooth increase in p53C concentration after radiation damage. In addition to the above equations, we model the independent degradation of p53C, p53n, p53CU, p53NU, p53T, and p53m to highlight the fact that external agents (i.e., those not modeled explicitly) have a role in modulating protein concen- trations. For example, several E3-ligases (other than Mdm2) ubiquitinate and label p53 for proteasomal degradation downstream. These reactions are given below: p53C κ19−→ Φ p53n κ20−→ Φ p53CU κ21−→ Φ p53NU κ22−→ Φ p53T κ23−→ Φ p53m κ24−→ Φ (19) (20) (21) (22) (23) (24) 2.2 p53 modulated apoptotic pathways Now, we discuss the mitochondrial apoptotic pathways downstream of p53, and the role of p53 module to the activation of mitochondrial apoptosis (see Figure 1). 6 p53T leads to transcription activation of proapoptotic PUMA and Bax, as modeled by and, p53T κ25−→ p53T + PUMA p53T κ26−→ p53T + Bax (25) (26) The roles of PUMA and Bax are discussed below. p53m binds to the anti-apoptotic protein Bcl-2 to inhibit the anti-apoptotic ac- tion of Bcl-2 on the mitochondria. 26 The biochemical equation associated with this reaction is given by p53m + Bcl-2 k+ 27−⇀↽− k− 27 p53m.Bcl-2 (27) with binding affinity given by k+ 27/k− 27. PUMA also binds to the antiapoptotic Bcl-2 member on the mitochondrial membrane. 27 PUMA + Bcl-2 k+ 28−⇀↽− k− 28 PUMA.Bcl-2 (28) Further, PUMA binds stronger to Bcl-2 than does p53, and can displace p53 from its complex with Bcl-2 18 - as can be seen by a combination of reactions 27 and 28, thus, freeing up p53 for further proapoptotic activity on the mitochondria, such as by activating Bax/Bak as shown below. Pro-apoptotic Bax translocates back and forth from the cytoplasm to the mito- chondria 28 to establish an equilibrium between the cytosolic and mitochondrial Bax concentrations: Bax k+ 29−⇀↽− k− 29 Baxmito (29) At the mitochondria, Bax interacts with Bcl-2 to form a complex 28 that prevents Bax oligomerization: Baxmito + Bcl-2 k+ 30−⇀↽− k− 30 Baxmito.Bcl-2 (30) Thus, the available Bcl-2 (that also forms complexes with p53 and PUMA as shown above) regulates the amount of Bax available for oligomerization. 7 Chipuk et al. 27 reported that p53m binds stronger to Bcl-2 than does Bax, and can displace Bax from its complex with Bcl-2. This reaction can be modeled via the following reaction: p53m + Baxmito.Bcl-2 k+ 31−⇀↽− k− 31 p53m.Bcl-2 + Baxmito (31) However, eq 31 is a straightforward linear combination of eqs 27 and 30, and, thus, the obvious requirement of consistent equilibrium constants (k+ 31/k− 31 = k+ 27k− 30/k− 37k+ 30) must be observed (or, eq 31 should not be considered explicitly). Active form of Bax on the mitochondrial membrane is needed for oligomeriza- tion, a process that requires further interaction of Bax with activators such as tBid. tBid localization to mitochondria 29,30 is an important event for Bax activation and is modeled by the following reaction: tBid k+ 32−⇀↽− k− 32 tBidmito (32) Subsequently, tBid activates mitochondrial Bax into an active (for oligomerization) form: tBidmito + Baxmito κ33−→ tBidmito + Bax∗ (33) It has also been suggested that tBid activates Bax in the cytosol which then localizes to the mitochondria. 31 However, the order of consecutive reactions is not expected to have a significant effect further downstream (such as caspase activation). A similar Bax (or Bak) activating effect can be achieved by p53m: 18 p53m + Baxmito κ34−→ p53m + Bax∗ and by PUMA, 32 PUMA + Baxmito κ35−→ PUMA + Bax∗ The activated Bax then oligomerizes on the mitochondrial membrane: Bax∗ + Bax∗ k+ 36−⇀↽− k− 36 Bax2 8 (34) (35) (36) Further, higher order oligomerization is modeled following Albeck et al. 9 Bax2 + Bax2 k+ 37−⇀↽− k− 37 Bax4 (37) Mitochondrial outer membrane permeabilization (MOMP) induced by Bax oligomer- ization results in the release of cytochrome-c from the mitochondria into the cyto- plasm. We model the release of cytochrome-c via the following process: 7 Bax4 + cytc mito κ38−→ Bax4 + cytc (38) where the above reaction serves to model the fact that the presence of a Bax pore is essential for release of cytochrome-c into the cytoplasm. Similarly, MOMP also releases Smac/Diablo into the cytoplasm: these proteins bind to anti-apoptotic IAP's and, thus, facilitate apoptosis. 33 Bax4 + Smacmito κ39−→ Bax4 + Smac (39) Cytoplasmic cytochrome-c interacts with Apaf in an ATP-dependent manner, and the heterodimer forms a heptameric complex called the "apoptosome". 34 Following Bagci et al. 7, we model these events via the following two reactions: cytC + Apaf k+ 40−⇀↽− k− 40 cytC.Apaf 7cytC.Apaf k+ 41−⇀↽− k− 41 apop (40) (41) Further, we use cooperativity in the formation of the apoptosome: the order of the forward reaction 41 is 4. The apoptosome then incorporates caspase-9, that in its autocatalyzed, activated form cleaves procaspase-3 (C3) to form active caspase-3 (C3∗). Following Albeck et al., 9 we use the following reactions for these processes: apop + C9 k+ 42−⇀↽− k− 42 apop.C9 apop.C9 + C3 k+ 43−⇀↽− k− 43 apop.C3.C9 κ43−→ apop.c9 + C3∗ 9 (42) (43) As mentioned above, tBid is an important factor for activation of the caspase cascade. Caspase-8, activated due to external death signals, is an important molecule responsible for Bid cleavage. 35 Additionally, active caspase-3 also results in Bid cleav- age, resulting in a positive feedback loop activating apoptosis. 36 We model the asso- ciated reactions following Bagci et. al. 7: C3∗ + Bid C8 + Bid k+ 44−⇀↽− k− 44 k+ 45−⇀↽− k− 45 C3 ∗ .Bid κ44−→ C3∗ + tBid C8.Bid κ45−→ C8 + tBid (44) (45) The concentration of active caspase-8 depends on external death factors. XIAP inhibits the activity of apoptosome 37,38 and promotes proteasomal degra- dation of caspase-3 39 preventing cell death. On the other hand, Smac inhibits the activity of XIAP. 33 These processes are modeled as, 9 apop.C9 + XIAP k+ 46−⇀↽− k− 46 apop.C9.XIAP C3∗ + XIAP C9 + XIAP k+ 47−⇀↽− k− 47 k+ 48−⇀↽− k− 48 C3∗.XIAP κ47−→ C3∗ Ub + XIAP C9.XIAP κ48−→ C9Ub + XIAP Smac + XIAP k+ 49−⇀↽− k− 49 Smac.XIAP (46) (47) (48) (49) In addition to above reactions, there are formation reactions for several species -- Bcl2, Bax, Apaf, C9, C3, XIAP, Smacm, cytcm, C8, and Bid. Further, these species, along with C3, C3U, C9U, Bax, and p53 forms mentioned above also undergo degradation reactions to help establish steady state. 3 Drug interactions In this section, we integrate the possible drugs to regulate apoptosis in normal cells into the biochemical network discussed above. In particular, we consider four specific 10 proteins targets -- Mdm2, PUMA, Bid, and C3∗ (Mdm2-I, PUMA-I, Bid-I, and C3-I, respectively). The actions of the inhibitors are modeled by the following reactions: Mdm2 + Mdm2 − I PUMA + PUMA − I Bid + Bid − I C3∗ + C3 − I k+ 50−⇀↽− k− 50 k+ 51−⇀↽− k− 51 k+ 52−⇀↽− k− 52 k+ 53−⇀↽− k− 53 Mdm2.Mdm2 − I, PUMA.PUMA − I, Bid.Bid − I, C3∗.C3 − I. (50) (51) (52) (53) and Known inhibitors of p53/Mdm2 interaction includes nutlins, BEB55, BEB59, and BEB69. Mustata et al. 40 have identified several PUMA inhibitors using phar- macophore modeling, and several of these compounds have been shown to inhibit PUMA -- induced apoptosis in vitro. 4 Stochastic simulation method The number of proteins in a typical cell (about picoliter) are significantly less than Avogadro's number: a nanomolar concentration of a protein corresponds to ∼ 600 molecules of that protein in the cell. Accordingly, for smaller cells and/or lower con- centrations, deterministic rate equations are not applicable. For this reason, we use the well -- known Gillespie algorithm for stochastic simulations of a system of chemical reactions. Details of the Gillespie Algorithm are given in the original papers. 41,42 Here we briefly describe some of the ideas behind the algorithm. For illustration, consider the following schematic reaction, A + B → AB, (54) and denote the stochastic rate constant by c. This stochastic rate is directly related to the macroscopic kinetic rate, k, by a simple relation (which, for the reaction in eq 1 11 is c = kV , where V is the reaction volume). At any instant of time, the propensity for the reaction in eq 52 is aα = NANBc (55) where NA and NB are the number of molecules of A and B in the reaction volume, and the subscript α denotes that the reaction index is α in the system of M chemical reactions (1 ≤ α ≤ M). In the Gillespie algorithm, one such reaction is chosen to be proportional to its propensity (we use the so-called Direct Reaction version), and the time is advanced based on the overall reaction propensity at that time. 41,42 The underlying assumptions are that the system is in thermal equilibrium (the number of reactive collisions are much less than the number of non-reactive collisions), and that it is well -- mixed. 4.1 Heterogeneity Real cellular systems are heterogeneous - e.g., mitochondria and cytoplasm offer vastly different environments (and, there may be heterogeneities within these compartments themselves). Thus, at a first glance, the well-mixed assumption of the Gillespie algo- rithm seems to preclude an application to real cellular systems. However, Bernstein 43 developed a protocol to extend the Gillespie algorithm to heterogeneous environments. In that protocol, 43 the cell (or, any simulation volume) is subdivided into elements and a reaction is reproduced and treated as an individual reaction in each subvolume. Accordingly, each reactant is treated as a different molecule in each subvolume (e.g., Ai 6= Aj, where i and j are different subvolumes), and the reaction propensities are altered accordingly. Further, each subvolume is well -- mixed -- thus allowing for formulating the system using the Gillespie algorithm. In context of this work, the nucleus, the cytoplasm, and the mitochondrial membrane are well-mixed, distinct subvolumes. This protocol results in an increase in the number of reactions both due to the fact that each reaction is treated independently in different subvolumes, as well as addi- 12 tional "reactions" (due to diffusion) relating to the conversion of Ai to Aj, as described in detail by Bernstein. 43 Based on the identity of a compartment/subvolume, a type of protein may not be present in that compartment (for example, the apoptosome does not permeate through the mitochondrial membrane). 5 Model parameters We obtain several parameters from experimental results, and obtain others in the range of previous computational studies. 7,9 Here we discuss some known experimental rate/equilibrium constants, and the initial concentrations of the species involved in the biochemical reaction network discussed above. The dissociation constant, k− 27/k+ 27, for p53 and pro-apoptotic Bcl-2 family mem- bers have been given variously as 160 nM 27 and 535 nM 26 -- both obtained via surface plasmon resonance. We choose an intermediate value of the dissociation constant (333 nM). PUMA binds significantly more strongly, and Chipuk et al. 27 report the dissociation constant k− 28/k+ 28 = 10 nM, a value that we use. Edlich et al. 28 report the on and off rate of Bax from cytosol and mitochondria as approximately the same, and establish a first order reaction (as represented by reaction 3) with k+ 29 = 4.9 × 10−3 s−1 and k− 29 = 4.7 × 10−3 s−1. For reaction 31, it is reported that p53 displaces Bax from its complex with Bcl-2 at equimolar concentration, whereas Bax displaces p53 from its complex with Bcl-2 at a 50 times higher concentration. 18 This implies that k+ 31/k− 31 = 50. Since eqs 27, 30, and 31 are not independent, the above stated values imply that k− 30/k+ 30 ≈ 17 µM. This value of dissociation constant is significantly higher than values reported in the literature (0.1 µM 44 and 20 nM 45). We use k− 30/k+ 30 nM, except where noted explicitly -- and, as shown below, the results obtained are robust to a wide range of values for this dissociation constant. Half lifes of a few proteins have also been reported in vivo. However, the direct use of such an information (for example, to compute the decay rates of the type A→ Φ) 13 is often not appropriate because the decay rate of protein A in vivo is convoluted with, for example, the decay of protein B that upregulates A. The parameters for the p53 module that differ for these three levels of radiation are given in Table 2, and the parameters that stay the same irrespective of the radiation level are given in Table 3. These parameters for the p53 module are chosen to reflect p53 oscillations observed experimentally. κ11 and κ12 do not affect the dose-dependent oscillatory behavior of p53, however, they are important for coupling of the p53 module to the apoptotic network, and the corresponding values are given in the text and figures below. 5.1 Initial concentrations When a normally unstressed cell is exposed to radiation, it undergoes a sequence of biochemical events that may, depending upon the radiation dose, lead to apoptosis. Thus, the appropriate initial conditions of cellular protein levels for such an apoptotic study are the steady state conditions in an unstressed cell. Irrespective of what initial number of molecules we start with, we first allow the system to establish a normal steady state (without radiation -- induced apoptosis) by selecting N-IR kinetic parameters from Table 2 (in addition to radiation -- independent kinetic equations/parameters). Subsequently, we perturb the system with radiation damage (modeled here as genotoxic effect), and allow the system to evolve. With enough stress, the system will eventually display apoptosis, with a high concentra- tion of pro-apoptotic proteins (such as active caspase-3). In essence, the initial con- centrations that are used to study radiation induced damage and subsequent drug treatments are the steady state conditions without any radiation/drug. For reference, the typical concentration a typical protein present in the cell under normal conditions is 10 -- 100 nm. 9 In a picoliter cell, the number of molecules of a typical protein is, thus, in the range 6000 -- 60000. On the other hand, proteins such as active caspase-3 (C3∗) are expected to be low under normal homeostatic conditions. 14 5.2 Modeling radiation damage Mitochondrial apoptosis is regulated by transcription -- dependent and independent role of p53. Thus, we focus on the p53 module (Figure 1) to model the effects of radiation damage. The downstream action of the apoptotic pathways (for example, represented by Bax oligomerization, cleavage of caspase-3) occur due to the response of the p53 module to radiation damage. Specifically, we model the response of p53 to three different radiation levels: no radiation (N-IR), low radiation (L-IR), and high radiation dose (H-IR). The procedure for the simulation is as follows. (1) Establish steady state for an unstressed cell (N-IR). (2) induce radiation damage (L-IR or H-IR), as modeled by use of the appropriate kinetic parameters from Table 2 for a certain time ∆T . (3) Return to unstressed parameters after ∆T . We note here that ∆T does not refer to the time duration of radiation exposure, rather it refers to the amount of time the radiation modifies the kinetic parameters. ∆T only indirectly refers to the exposure time: longer exposure to a given radiation dose will affect cellular biochemical kinetics for a longer duration. 6 Results 6.1 Radiation induced p53 oscillations Single -- cell experiments to monitor cellular p53 levels after radiation damage showed that individual cells show dramatic oscillations in p53 levels. In particular, these oscillations are sustained and show dose-dependent oscillation frequencies. 2 These experimentally observed results form the basis of simulating the p53 module using stochastic simulations: we reproduce the radiation dependent behavior of p53 re- sponse using stochastic simulations. Figure 2 shows the oscillations in p53n (the major p53 component upon radiation damage) as a function of time for two different radiation doses and exposures. The 15 chosen model for radiation damage reproduces the known p53 oscillations, that are dependent upon the radiation dose. At low levels, this frequency is ≈ 11 hr, whereas it decreases to ≈ 6 hr upon increasing the radiation dose. The differences in L-IR and H-IR parameters for the p53 module suggests one possible mechanisms for the observed radiation-dose dependent frequencies. Impor- tant factors in the model leading to this difference are the enhanced translocation of Mdm2 protein from the cytoplasm to the nucleus, and the enhanced translocation of Mdm2 mRNA from the nucleus to the cytoplasm. This was deliberately chosen to represent the time delay in Mdm2 response that is frequently utilized in deterministic models. 3,46,47 Thus, this study shows that such time delay can be modeled mechanistically in stochastic simulations (and allow for the modulation of oscillation frequency) by the realization that proteins/genes act in specific cellular contexts. Further, the increase in the translocation rate of the Mdm protein (along with increased translocation of p53) to the nucleus upon DNA damage has been suggested. 48 Further, as Figure 2 suggests, at low exposure time (modeled by ∆T = 2 × 104 s), there is a single peak in the p53 profile for both radiation levels. On the other hand, sustained p53 oscillations are obtained for high exposure times (∆T = 2×105s). Even for cells exposed to radiation for the same amount of time, it is possible that cell -- cell variability can lead to different amounts of observed oscillatory peaks (∆T is, after all, not the radiation exposure time, but is the time for which the kinetic parameters are altered from unstressed levels). 6.2 Role of p53 oscillations in apoptosis At higher radiation dosage, the oscillations in p53 occur at a higher frequency. How- ever, it is unclear if cellular apoptotic response is mainly determined by the oscillation frequency: is increased apoptosis at higher dose of radiation a direct result of the higher p53 oscillation frequency? To address this question, we compare the caspase 3 activity obtained via the use 16 of H-IR and L-IR parameters in Table 2, along with the kinetic parameters associated with the rest of the apoptotic machinery. Further, we use the same rates for apoptosis- related transcriptional activation of p53 (κ12 = 10−5 s−1), and for p53 translocation to mitochondrial membrane (κ11 = 10−5 s−1). The obtained caspase-3 concentrations for a cell are shown in Figure 3 for both H-IR and L-IR sets. Upon using a radiation-dose independent coupling of the p53 module with the apoptotic network, L-IR leads to an increase in caspase-3 activity upon radiation damage (radiation damage occurs at t = 0). On the other hand, H-IR leads to low caspase-3 concentration. Clearly, this result is at odds with experimental observation -- increased radiation dose leads to an increase in apoptosis. Correspondingly, if κ11 and/or κ12 are increased for H-IR, caspase-3 activity is increased (shown in Figure 3). A crucial insight that emerges from Figure 3 is that the dose-dependent radiation damage and cell death depend upon the coupling of the regulatory p53 module to the apoptotic machinery (defined via an increase in the p53 pro-apoptotic transcriptional activity, κ12, and an increase in p53 translocation to the mitochondria, κ11), and not merely on the p53-oscillation frequencies. We use the term "coupling" to describe how the p53-regulatory module leads to apoptosis via transcription-dependent (κ12) and transcription-independent (κ11) pathways. 6.3 Downstream apoptosis cascade The response of the apoptotic network downstream of the p53 module sheds further light on the role of pro- and anti-apoptotic proteins in the cell. Here, we discuss (i) oscillations downstream of p53, (ii) role of Bax -- Bcl-2 interactions, and (iii) the requirement for sufficient Bid cleavage for sustained apoptosis. Figure 4 shows the concentrations of PUMA, Bax∗, tBid, and caspase-3 after radiation damage with high dose (the qualitative response of low radiation dose is similar). PUMA mirrors the p53 oscillations, irrespective of the coupling strength. This phenomenon is due to direct activation of PUMA by p53T. Unlike oscillations in 17 PUMA level, oscillations in the concentration of activated Bax show a distinct de- pendence on the coupling strength: at low coupling strength (no apoptosis), Bax oscillations are sustained, but decay along with PUMA oscillations. On the other hand, a high coupling that results in apoptosis leads to a sustained increase in ac- tivated Bax at the expense of oscillations. Bax∗ increases initially due to the direct action of PUMA on Bax. This leads to the release of caspase-3 that cleaves tBid resulting in further Bax activation and a positive feedback loop. A sustained caspase-3 activity even after the decay of PUMA depends upon whether sufficient tBid is activated by the time PUMA decays. A sustained release of tBid and the positive feedback is very important for sustained caspase activity: for low coupling of p53 module and apoptotic network, insufficient PUMA activity results in an insufficient amount of caspase 3 release and tBid activation to maintain a sustained caspase activity. 6.3.1 Strength of interaction between Bax and Bcl-2 In Section 5, we discussed the strength of Baxm and Bcl-2 interactions (k− 30/k+ 30): dif- ferent values of this dissociation constant (either directly or indirectly) are suggested by different groups spanning several orders of magnitude. 18,44,45 Here, we discuss the effect of a range of this dissociation constant on the apoptosis model we use in this manuscript. Figure 5 shows caspase-3 and Baxm.Bcl-2 concentrations for two different values of the dissociation constants. A comparison of panels (a) and (c) shows that the downstream apoptotic activity is unaffected by dissociation constants that differ by two orders of magnitude (all other model parameters remain unchanged between these two panels): caspase-3 concentration is unaffected, although vastly differing amounts of Baxm.Bcl-2 complex are formed. Similarly, the different dissociation constants had an insignificant effect on systems showing a lack of apoptosis (panels (b) and (d)). The model is, thus, robust with respect to the strength of Baxm and Bcl-2 in- teraction. This suggests that cells can exhibit significantly differing strengths of this 18 interaction without affecting the cell fate given a genotoxic insult, and several alter- nate mechanisms can lead to the same cell fate. 6.4 p53 transcription-independent apoptosis So far, we have discussed p53 transcription dependent apoptosis via transcriptional upregulation of PUMA and Bax upon radiation damage. On the other hand, the role of p53 in a transcriptionally independent manner is also of importance. In its tran- scription independent role, it has been suggested that p53 in the cytoplasm can either directly activate Bax, or can indirectly activate Bax by binding to anti-apoptotic Bcl- 2 and, thus, preventing the latter's anti-apoptotic action. In this section, we explore this issue. Figure 6 shows the evolution of p53m (top panel) and caspase-3 (bottom panel) with time upon exposure to H-IR for different ratios of transcription dependent (κ12) and independent (κ11) apoptosis pathway strengths. In the figure, the black line represents the case where the strength of transcription dependent pathways alone is not sufficient to cause apoptosis. Upon an increase in κ11 (black, red, and blue lines -- in that order), p53m shows an increasingly upregulated behavior. Correspondingly, caspase-3 shows a sustained activity with an increase in p53 transcription independent pathway strength (for the same κ12). Strikingly, the removal of direct Bax activation by mitochondrial p53 (green line) results in complete absence of apoptosis for the p53 transcription-independent apop- tosis, even for a high strength of transcription independent pathway. This suggests that the indirect role of p53m in apoptosis by binding Bcl-2 is not sufficient for sus- tained apoptosis by the p53-transcription independent mode -- direct Bax activation by the mitochondrial p53 appears to be critical, too. 19 6.5 Role of inhibitors of specific proteins From a pharmacological viewpoint, it is important to identify protein targets that are good candidates for drug treatment to mitigate radiation damage. In this section, we focus on this issue with the aim of identifying potential targets suitable for treatments that are effective even if not administered immediately. Figure 7 illustrates the efficacy of treatments administered after two different de- lays: treatments administered immediately after radiation damage (15 minute delay, top panel), and treatments administered after a longer delay (12 hours, bottom panel). The radiation dose in Figure 7 leads to sustained caspase-3 activity in absence of drug treatment (bottom panel of Figure 4, blue curve). If treatment is available immediately after radiation damage, inhibition of PUMA is very effective in mitigating radiation damage. Similarly, inhibition of Bid also helps in mitigating radiation damage (although to a lesser extent than PUMA inhibition). In contrast, treatment via inhibition of Mdm2 and caspase-3 are not effective. On the other hand, if there is a longer delay before treatment is administered, PUMA inhibi- tion is an ineffective treatment and inhibition of Bid is the most effective treatment. The ineffectiveness of PUMA inhibition in this case results because PUMA activity till the administration of PUMA-I leads to sufficient activation of positive feedback loop involving caspase-3 and Bid: activation of Bax via tBid dominates the apoptotic response at this late stage. The main cause of apoptosis for the system in Figure 7 is the p53-transcription dependent pathway. Thus, Mdm2 inhibition, that decreases the ubiquitination of p53n and, subsequently, leads to an increase in p53T (and PUMA and Bax), increases the propensity of the system to undergo apoptosis. A comparison of the two panels of Figure 7 shows that administering Mdm2-I sooner after radiation damage leads to a quicker increase in caspase-3 activity. 20 7 Discussion 7.1 Testable hypotheses Several key, testable features of cellular response to apoptosis emerges from the model and are listed as following. (i) oscillatory response of PUMA, at the level of a single cell, to radiation damage, (ii) elevated tBid activity even after PUMA decays when an increase in apoptosis occurs, and (iii) inhibition of Bid is more effective in mitigating radiation damage than PUMA inhibition when the treatment is administered after a substantial delay. The last point is especially relevant with respect to developing a treatment strategy for mitigating radiation damage. Even a very potent inhibitor of PUMA may not prevent apoptosis when it is administered after a substantial delay. Indeed, the inhibitor used in the current model acts in nM concentrations, and despite being effective when administered immediately after radiation damage, is ineffective after a longer delay because of enough tBid activity to maintain a sustained caspase activity. Even if potent inhibitors of Bid are not currently known, it is possible to test the last hypothesis above using the following two approaches. Firstly, a si-RNA knockout of PUMA performed several hours after radiation damage should be ineffective in mitigating apoptosis due to radiation damage. Secondly, knocking out Bid several hours after radiation damage should be more effective in mitigating apoptosis due to radiation damage. 21 References [1] Bar-Or, R. L., R. Maya, L. A. Segel, U. Alon, and M. Oren. 2000. Generation of oscillations by the p53-mdm2 feedback loop: a theoretical and experimental study. Proc. Natl. Acad. Sci. 97:11250 -- 11255. [2] Geva-Zatorsky, N., N. Rosenfeld, S. I. andR. Milo, A. Sigal, E. Dekel, T. Yamitzsky, Y. Liron, P. Polak, G. Lahav, and U. Aron. 2006. Oscillations and variability in the p53 system. Mol. Sys. Biol. 10:1 -- 13. [3] Ma, L., J. Wagner, J. J. Rice, W. Hu, A. J. Levine, and G. A. Stolovitzky. 2005. A plausible model for the digital response of p53 to dna damage. Proc. Natl. Acad. Sci. 102:14266 -- 14271. [4] Loewer, A., E. Batchelor, G. Gaglia, and G. Lahav. 2010. Basal dynamics of p53 reveal transcriptionally attenuated pulses in cycling cells. Cell. 142:89 -- 100. [5] Proctor, C. J., and D. A. Gray. 2008. Explaining oscillations and variability in the p53-mdm2 systems. BMC Syst. Biol. 2:xx. [6] Eissing, T., H. Conzelmann, E. D. Gilles, F. Allgower, and E. Bullinger. 2004. Bistability analyses of a caspase activation model for ceceptor-induced apoptosis. J. Biol. Chem. 279:36892 -- 36897. [7] Bagci, E. Z., Y. Vodovotz, T. R. Billiar, G. B. Ermentrout, and I. Bahar. 2006. Bistability in apoptosis: Roles of bax, bcl-2, and mitochondrial permeability transition pores. Biophys. J. 90:1546 -- 1559. [8] Legewie, S., N. Bluthgen, and H. Herzel. 2006. Mathematical modeling identifies inhibitors of apoptosis as mediators of positive feedback and bistability. Plos Comput. Biol. 2:e120. [9] Albeck, J. G., J. M. Burke, S. L. Spencer, D. A. Lauffenburger, and P. K. 22 Sorger. 2008. Modeling a snap-action, variable-delay switch controlling extrinsic cell death. PLoS Biol. 6:e299. [10] Spencer, S. L., S. Gaudet, J. G. Albeck, J. M. Burke, and P. K. Sorger. 2009. Non-genetic origins of cell-to-cell variability in trail-induced apoptosis. Nature. 459:428 -- U144. [11] Skommer, J., T. Brittain, and S. Raychaudhuri. 2010. Bcl-2 inhibits apoptosis by increasing the time-to-death and intrinsic cell-to-cell variations in the mito- chondrial pathway of cell death. Apoptosis. 15:1223 -- 1233. [12] Wee, K. B., U. Surana, and B. D. Aguda. 2009. Oscillations of the p53-akt network: Implications on cell survival and death. Plos One. 4:e4407. [13] Pu, T., X.-P. Zhang, F. Liu, and W. Wang. 2010. Coordination of the nuclear and cytoplasmic activities of p53 in response to dna damage. Biophys. J. 99:1696 -- 1705. [14] Li, Z. Y., M. Ni, J. K. Li, Y. P. Zhang, Q. Ouyang, and C. Tang. 2011. Decision making of the p53 network: Death by integration. J. Theor. Biol. 271:205 -- 211. [15] Zhang, X. P., F. Liu, Z. Cheng, and W. Wang. 2009. Cell fate decision mediated by p53 pulses. Proc. Natl. Acad. Sci. 106:12245 -- 12250. [16] Haupt, Y., S. Rowan, E. Shaulian, K. Vousden, and M. Oren. 1995. Induction of apoptosis in hela cells by trans-activation-deficient p53. Genes Dev. 9:2170 -- 2183. [17] Caelles, C., A. Helmberg, and M. Karin. 1994. p53-dependent apoptosis in the absence of transcriptional activation of p53-target genes. Nature. 370:220 -- 223. [18] Chipuk, J. E., T. Kuwana, L. Bouchier-Hayes, N. M. Droin, D. D. Newmeyer, M. Schuler, and D. R. Green. 2004. Direct activation of bax by p53 mediates mitochondrial membrane permeabilization and apoptosis. Science. 303:1010 -- 1014. 23 [19] Lindenboim, L., C. Borner, and R. Stein. 2011. Nuclear proteins acting on mitochondria. Biochem. Biophys. Acta. 1813:584 -- 596. [20] Lahav, G., N. Rosenfeld, A. Sigal, N. Geva-Zatorsky, M. B. Elowitz, and U. Alon. 2004. Dynamics of the p53-mdm2 feedback loop in individual cells. Nat. Genet. 36:147 -- 150. [21] Marchenko, N. D., W. Hanel, D. Li, K. Becker, N. Reich, and U. M. Moll. 2010. Stress-mediated nuclear stabilization of p53 is regulated by ubiquitination and importin-a3 binding. Cell Death Differ. 17:255 -- 267. [22] Joers, A., V. Jaks, J. Kase, and T. Maimets. 2004. p53-dependent transcription can exhibit both on/off and graded response after genotoxic stress. Oncogene. 23:6175 -- 6185. [23] Lee, C. W., J. C. Ferreon, A. C. M. Ferreon, M. Arai, and P. E. Wright. 2010. Graded enhancement of p53 binding to creb-binding protein (cbp) by multisite phosphorylation. Proc. Natl. Acad. Sci. 107:19290 -- 19295. [24] Lee, J. T., and W. Gu. 2010. The multiple levels of regulation by p53 ubiquiti- nation. Cell Death Diff. 17:86 -- 92. [25] Marchenko, N., S. Wolff, S. Erster, K. Becker, and U. M. Moll. 2007. Monoubiq- uitylation promotes mitochondrial p53 translocation. EMBO J. 26:923 -- 934. [26] Tomita, Y., N. Marchenko, S. Erster, A. Nemajerova, A. Dehner, C. Klein, H. Pan, H. Kessler, P. Pancoska, and U. M. Moll. 2006. Wt p53, but not tumor-derived mutants, bind to bcl2 via the dna binding domain and induce mitochondrial permeabilization. J. Biol. Chem. 281:8600 -- 8606. [27] Chipuk, J. E., L. Bouchier-Hayes, T. Kuwana, D. D. Newmeyer, and D. R. Green. 2005. Puma couples the nuclear and cytoplasmic proapoptotic function of p53. Science. 309:1732 -- 1735. 24 [28] Edlich, F., S. Banerjee, M. Suzuki, M. M. Cleland, D. Amoult, C. Wang, A. Neutzner, N. Tjandra, and R. J. Youle. 2011. Bcl-xl retrotranslocates bax from the mitochondria into the cytosol. Cell. 145:104 -- 116. [29] Lutter, M., G. A. Perkins, and X. Wang. 2001. The pro-apoptotic bcl-2 family member tbid localizes to mitochondrial contact sites. BMC Cell Biol. 2:22. [30] Gonzalvez, F., F. Pariselli, P. Dupaigne, I. Budihardjo, M. Lutter, B. Antons- son, P. Diolez, S. Manon, J.-C. Martinou, M. Goubern, X. Wang, S. Bernard, and P. X. Petit. 2005. tbid interaction with cardiolipin primarily orchestrates mitochondrial dysfunctions and subsequently activates bax and bak. Cell Death and Diff. 12:614 -- 626. [31] Eskes, R., S. Desagher, B. Antonsson, and J.-C. Martinou. 2000. Bid induces the oligomerization and insertion of bax into the outer mitochondrial membrane. Mol. Cell Biol. 20:929 -- 935. [32] Ren, D., H.-C. Tu, H. Kim, G. X. Wang, G. R. Bean, O. Takeuchi, J. R. Jeffers, G. P. Zambetti, J. J.-D. Hsieh, and E. H.-Y. Cheng. 2010. Bid, bim, and puma are essential for activation of the bax- and bak-dependent cell death program. Sci. 330:1390 -- 1393. [33] Du, C., M. Fang, Y. Li, and X. Wang. 2000. Smac, a mitochondrial protein that promotes cytochrome c-dependent caspase activation by eliminating iap inhibition. Cell. 102:33 -- 42. [34] Acehan, D., X. J. Jiang, D. G. Morgan, J. E. Heuser, X. Wang, and C. W. Akey. 2002. Three -- dimensional structure of the apoptosome: implications for assembly, procaspase-9 binding, and activation. Mol. Cell. 9:423 -- 432. [35] Li, H., H. Zhu, C. J. Xu, and J. Yuan. 1998. Cleavage of bid by caspase 8 mediates the mitochondrial damage in the fas pathway of apoptosis. Cell. 94:491 -- 501. 25 [36] Slee, E. A., S. A. Keogh, and S. J. Martin. 2000. Cleavage of bid during cytotoxic drug and uv radiation-induced apoptosis occurs downstream of the point of bcl-2 action and is catalysed by caspase-3: a potential feedback loop for amplification of apoptosis-associated mitochondrial cytochrome c release. Cell Death Differ. 7:556 -- 565. [37] Hill, M. M., C. Adrian, P. J. Duriez, E. M. Creagh, and S. J. Martin. 2004. Anal- ysis of the composition, assembly kinetics and activity of native apaf-1 apopto- somes. EMBO J. 23:2134 -- 2145. [38] Twiddy, D., D. G. Brown, C. Adrain, R. Jukes, S. J. Martin, G. M. Cohen, M. MacFarlane, and K. Cain. 2004. Pro-apoptotic proteins released from the mitochondria regulate the protein composition and caspase-processing activity of the native apaf-1/caspase-9 apoptosome complex. J. Biol. Chem. 279:19665 -- 19682. [39] Suzuki, Y., Y. Nakabayashi, and R. Takahashi. 2001. Ubiquitin-protein ligase activity of x-linked inhibitor of apoptosis protein promotes proteasomal degrada- tion of caspase-3 and enhances its anti-apoptotic effect in fas-induced cell death. Proc. Natl. Acad. Sci. 98:8662 -- 8667. [40] Mustata, G., M. Li, N. Zevola, A. Bakan, L. Zhang, M. Epperly, J. S. Green- berger, J. Yu, and I. Bahar. 2011. Development of small-molecule puma inhibitors for mitigating radiation-induced cell death. Curr. Top. Med. Chem. 11:281 -- 290. [41] Gillespie, D. T. 1976. A general method for numerically simulating the stochastic time evolution of coupled chemical reactions. J. Comput. Phys. 22:404 -- 434. [42] Gillespie, D. T. 1977. Exact stochastic simulation of coupled chemical reactions. J. Phys. Chem. 81:2340 -- 2361. [43] Bernstein, D. 2005. Simulating mesoscopic reaction-diffusion systems using the gillespie algorithm. Phys. Rev. E. 71:041103. 26 [44] Fletcher, J. I., S. Meusburger, C. J. Hawkins, D. T. Riglar, E. F. Lee, W. D. Fairlie, D. C. S. Huang, and J. M. Adams. 2008. Apoptosis is triggered when prosurvival bcl-2 proteins cannot restrain bax. Proc. Natl. Acad. Sci. 105:18081 -- 18087. [45] Ku, B., C. Liang, J. U. Jung, and B.-H. Oh. 2011. Evidence that inhibition of bax activation by bcl-2 involves its tight and preferential interaction with the bh3 domain of bax. Cell Res. 21:627 -- 641. [46] Wagner, J., and G. Stolovitzky. 2008. Stability and time-delay modeling of negative feedback loops. Proc. IEEE. 96:1398 -- 1410. [47] Bottani, S., and B. Grammaticos. 2007. Analysis of a minimal model for p53 oscillations. J. Theor. Biol. 249:235 -- 245. [48] Li, C., L. Chen, and J. Chen. 2002. Dna damage induces mdmx nuclear transloca- tion by p53-dependent and -independent mechanisms. Mol. Cell. Biol. 22:7562 -- 7571. 27 Figure 1: A map of interactions in the apoptotic pathway. This map only shows a schematic of the interactions highlighting the main features of the model. Detailed biochemical reactions (that may involve proteins not specifically mentioned in this map) are given the text. 28 ] m n [ , n 3 5 p PSfrag replacements 12 10 8 6 4 2 0 ] m n [ , n 3 5 p PSfrag replacements 12 10 8 6 4 2 0 (a) H−IR, low duration ] m n [ , n 3 5 p PSfrag replacements −40 −20 0 20 40 time, hr 60 80 100 (c) L−IR, low duration ] m n [ , n 3 5 p PSfrag replacements −40 −20 0 20 40 time, hr 60 80 100 12 10 8 6 4 2 0 12 10 8 6 4 2 0 (b) H−IR, high duration −40 −20 0 20 40 time, hr 60 80 100 (d) L−IR, high duration −40 −20 0 20 40 time, hr 60 80 100 Figure 2: Sustained oscillations in p53 concentration as a result of the duration of alteration, via radiation, of cellular biochemical kinetics. Radiation damage at t = 0 leads to radiation -- dose dependent p53 oscillations. Single spikes in p53 concentrations (the left two panels) are observed when radiation induces a momentary change in cellular biochemical kinetics, and sustained oscillations (the two panels on the right), occur when the cellular biochemical kinetics are altered for a longer duration. Further, the frequencies of oscillations at the two radiation dosage (top right and bottom right panel) are similar to the experimentally observed values. 29 ) M n ( ] 3 PSfrag replacements e s a p s a C [ 0.8 0.6 0.4 0.2 0 L-IR (κ12 = 10−5) H-IR (κ12 = 10−5) H-IR (κ12 = 3 × 10−5) −40 −20 0 20 40 60 80 100 time (hr) Figure 3: An increase in pro-apoptotic transcriptional activity of p53 associates with higher radiation dose to induce apoptosis. Caspase activities induced by the higher radiation dose for two different pro-apoptotic transcriptional activity of p53 (κ12). A lower dose of radiation leads to a substantial caspase activity at a smaller value of κ12. κ11 = 10−5 for the three cases. 30 0.3 0.2 0.1 0 2 1.6 1.2 0.8 0.4 0 ) M n ( c n o c PSfrag replacements ) M n ( c n o c PSfrag replacements PUMA caspase−3 tBid 60 80 100 60 80 100 20 40 time (hr) −40 −20 0 PUMA caspase−3 tBid −40 −20 0 20 40 time (hr) Figure 4: Role of tBid activation for sustained caspase-3 activity. A comparison of the concentrations of PUMA, caspase-3, and tBid obtained after irradiation with a high dose at t = 0 with two different couplings (κ12) -- low value of κ12 in the top panel does not lead to enough tBid activity to sustain apoptotic activity. κ11 = 10−5 for both panels. 31 PSfrag replacements M n conc, nM , c n o c k− 30 k+ 30 = 2 µM PSfrag replacements conc, nM = 20 nM k− 30 k+ 30 M n , c n o c 0.02 (a) Baxm.Bcl−2 5 4 3 2 0.08 0.06 0.04 k− 30 k+ 30 = 20 nM −40 −20 0 20 40 60 time (hr) 0.2 k− 30 k+ 30 80 = 2 µM 0 100 (c) C3* 0.8 PSfrag replacements 0.6 Baxm.Bcl2 k− 30 k+ 30 = 2 µM k− 30 k+ 30 0.4 conc, nM = 20 nM 0.2 0 −40 −20 0 20 60 time (hr) 40 80 0 100 0.8 (b) C3* PSfrag replacements conc, nM 0.6 0.4 5 4 3 2 0.1 0.08 0.06 0.04 0.02 0 Baxm.Bcl2 k− 30 k+ 30 = 20 nM C3* −40 −20 0 40 20 60 time (hr) 0.08 0.04 M n , c n o c 80 0 100 k− 30 k+ 30 = 2 µM (d) M n , c n o c Baxm.Bcl2 C3* −40 −20 0 40 20 60 time (hr) 80 0 100 Figure 5: Robustness of the model (for caspase-3 activity) with respect to the strength of Baxm and Bcl-2 interactions. Panels (a) and (c) show the results for κ12 = 2 × 10−5 s−1 showing distinct apoptosis despite vastly different strengths of Baxm and Bcl-2 interaction. Panels (b) and (d) show an absence of apoptosis for different strengths of Baxm -- Bcl-2 interactions (and both are for κ12 = 10−5 s−1. 32 M PSfrag replacements n , ] 3 [C3∗], nM 5 p [ PSfrag replacements [p53], nM M n , ] ∗ 3 C [ 4 3 2 1 0 0.8 0.6 0.4 0.2 0 (10−5 (10−4 (10−3 (10−3 , 10−5) , 10−5) , 10−5) , 10−5)∗ −40 −20 0 20 40 time (hr) 60 80 100 (10−5 (10−4 (10−3 (10−3 , 10−5) , 10−5) , 10−5) , 10−5)∗ −40 −20 0 20 40 60 80 100 time (hr) Figure 6: Important role of direct Bax activation by mitochondrial p53 in inducing apoptosis via the p53 transcription-independent pathway. The top panel shows the mitochondrial p53 concentrations for different sets of (κ11,κ12) values, and the asterix for the green curve indicates that direct Bax activation by the mitochondrial p53 in this case is abrogated (κ34 = 0). The bottom panel shows the corresponding C3∗ concentrations. 33 ) M n ( ] 3 p s a c [ PSfrag replacements ) M n ( ] 3 p s a c [ PSfrag replacements 0.8 0.6 0.4 0.2 0 0.8 0.6 0.4 0.2 0 Mdm2−I PUMA−I C3−I Bid−I −40 −20 0 Mdm2−I PUMA−I C3−I Bid−I −40 −20 0 20 40 time (hr) 20 40 time (hr) 60 80 100 60 80 100 Figure 7: Effective mitigation of radiation damage via Bid inhibition at longer times. The top panel shows caspase-3 activity upon treatment via Mdm2, PUMA, caspase- 3, and Bid inhibition after treatment by the respective inhibitors 15 minutes after radiation damage, and the bottom panel shows caspase-3 activity after treatment administered 12 hours after radiation damage. In the top panel, caspase-3 activity after treatment with PUMA inhibitors (PUMA-I) is indistinguishable from zero. In the bottom panel, resulting caspase-3 activity after treatment by PUMA-I (red line) is similar to the activity after treatment by procaspase inhibitors (blue line). 34 Table 1: List of proteins in Figure 1 and their descriptions. symbol description p53n p53T p53C p53CU p53NU p53U p53m p53no p53To nuclear p53 stress -- induced p53 for apoptotic activation p53 in cytoplasm ubiquitinated p53 in cytoplasm ubiquitinated p53 in nucleus polyubiquitinated p53 for proteasomal decay mitochondrial p53 p53n oligomer for transcription activation of Mdm2 p53T oligomer for transcription activation pro-apoptotic PUMA/Bax Mdm2-mRNA mRNA of Mdm2 Mdm2n Mdm2/E3 ligase proteins in nucleus Mdm2C Mdm2/E3 ligase proteins in cytoplasm PUMA p53-upregulated modulator of apoptosis Bcl-2 Bax Baxn anti-apoptotic Bcl-2 family proteins proapoptotic Bcl-2 family members oligomerized pro-apoptotic Bcl-2 family proteins Baxm proapoptotic Bcl-2 members on the mitochondria Bid tBid cytc Smac Apaf XIAP apop C9 C3 C3∗ C3U C8 BH3-only pro-apoptotic proteins truncated Bid cytoplasmic cytochrome c cytoplasmic Smac protein Apaf-1 pro-apoptotic protein IAP-family proteins apoptosome caspase 9 procaspase 3 active caspase 3 ubiquitinated caspase 3 35 caspase 8 Table 2: Radiation dose dependent kinetic parameters for the p53 module. κ11 and κ12 are also dose dependent, but their values do not affect the observed, dose -- dependent p53 oscillatory behavior and are noted in the text and the figures. No radiation (N-IR) Low radiation (L-IR) High radiation (H-IR) κ2 (s−1) κ5 (s−1) κ7 (s−1) k+ 15 (µM/s) κ20 (s−1) κ23 (s−1) 10−4 2 × 10−4 10−4 0 10−4 10−4 10−4 2 × 10−4 10−4 2 × 10−7 10−4 10−4 10−3 5 × 10−3 5 × 10−4 5 × 10−7 10−5 10−5 36 Table 3: Kinetic parameters of the p53 module that are unaffected by radiation. Formation reactions: µMs−1 κ0 = 10−6 Unimolecular reactions: s−1 3 = 10−4 k− k− 8 = 0 κ9 = 10−3 13 = 10−4 k− κ16 = 10−2 κ19 = 10−4 κ24 = 10−3 κ4 = 10−1 κ8 = 10−3 κ6 = 5 × 10−4 k− 9 = 0 κ10 = 10−4 κ11 = 10−5 κ14 = 0.1 κ17 = 0.1 15 = 5 × 10−5 k− κ18 = 10−2 κ21 = 10−3 κ22 = 10−3 Bimolecular reactions: µM−1s−1 k+ 8 = 0.1 k+ 9 = 0.1 Special (cooperative) reactions 3 = 10 µM−3s−1 k+ 13 = 10 µM3s−1 k+ 37 Table 4: Kinetic parameters downstream of the p53 module. Formation reactions: µMs−1 10−5 for Bax, Bcl-2, Bid 10−4 for Apaf, C9, C3, XIAP, Smacm, cytc Unimolecular reactions: units s−1 κ25 = 10−3 28 = 10−3 k− 30 = 10−3 k− 36 = 10−3 k− 41 = 10−5 k− κ43 = 1 45 = 10−3 k− 47 = 10−3 k− κ48 = 0.1 κ26 = 10−3 27 = 10−3 k− k+ 29 = 5 × 10−3 k+ 32 = 5 × 10−3 29 = 5 × 10−3 k− 32 = 5 × 10−3 k− 37 = 10−3 k− 42 = 10−4 k− 44 = 10−3 k− κ45 = 1 κ47 = 0.1 49 = 10−3 k− 40 = 10−3 k− 43 = 10−4 k− κ44 = 1 46 = 10−4 k− 48 = 10−3 k− 10−3: PUMA, tBid, Bid, C3, C9, XIAP, Apaf 10−3: cytcm, Smacm, Smac, C3∗, C3U, C9U 10−3: Bax, Bax∗, Bcl2 5 × 10−4: Mdm2, Mdm2C 10−2: cytC Bimolecular reactions: µM−1s−1 k+ 27 = 3 × 10−3 k+ 28 = 0.1 k+ 31 = 1.0 31 = 2 × 10−2 k− κ34 = 5 × 10−2 k+ 37 = 0.1 k+ 40 = 0.3 k+ 44 = 0.3 k+ 47 = 1 κ35 = 0.5 κ38 = 10 k+ 42 = 3 × 10−2 k+ 45 = 0.3 k+ 48 = 1 Special (cooperative) reactions 41 = 105 µM−33s−1 k+ 38 k+ 30 = 0.6 κ33 = 0.5 k+ 36 = 0.6 κ39 = 10 k+ 43 = 3 × 10−2 k+ 46 = 3 × 10−2 k+ 49 = 0.1
1912.07899	1	1912	2019-12-17T09:44:52	QuanTI-FRET: a framework for quantitative FRET measurements in living cells	[ "physics.bio-ph", "physics.optics" ]	Foerster Resonance Energy Transfer (FRET) allows for the visualization of nanometer-scale distances and distance changes. This sensitivity is regularly achieved in single-molecule experiments in vitro but is still challenging in biological materials. Despite many efforts, quantitative FRET in living samples is either restricted to specific instruments or limited by the complexity of the required analysis. With the recent development and expanding utilization of FRET-based biosensors, it becomes essential to allow biologists to produce quantitative results that can directly be compared. Here, we present a new calibration and analysis method allowing for quantitative FRET imaging in living cells with a simple fluorescence microscope. Aside from the spectral crosstalk corrections, two additional correction factors were defined from photophysical equations, describing the relative differences in excitation and detection efficiencies. The calibration is achieved in a single step, which renders the Quantitative Three-Image FRET (QuanTI-FRET) method extremely robust. The only requirement is a sample of known stoichiometry donor:acceptor, which is naturally the case for intramolecular FRET constructs. We show that QuanTI-FRET gives absolute FRET values, independent of the instrument or the expression level. Through the calculation of the stoichiometry, we assess the quality of the data thus making QuanTI-FRET usable confidently by non-specialists.	physics.bio-ph	physics	QuanTI-FRET: a framework for quantitative FRET measurements in living cells Alexis Coullomb1, C´ecile M. Bidan1, Chen Qian2, Fabian Wehnekamp2, Christiane Oddou3, Corinne Albig`es-Rizo3, Don. C. Lamb2, and Aur´elie Dupont1,* 1Univ. Grenoble Alpes, CNRS, LIPhy, Grenoble, F-38000, France 2Department of Chemistry, Center for Nano Science (CENS), Center for Integrated Protein Science (CIPSM) and Nanosystems Initiative Munchen (NIM), Ludwig Maximilians-Universitat Munchen, Germany 3Institute for Advanced Biosciences, Universit´e Grenoble Alpes, INSERM U1209, UMR5309, F38700 La Tronche, France *[email protected] ABSTRACT Förster Resonance Energy Transfer (FRET) allows for the visualization of nanometer-scale distances and distance changes. This sensitivity is regularly achieved in single-molecule experiments in vitro but is still challenging in biological materials. Despite many efforts, quantitative FRET in living samples is either restricted to specific instruments or limited by the complexity of the required analysis. With the recent development and expanding utilization of FRET-based biosensors, it becomes essential to allow biologists to produce quantitative results that can directly be compared. Here, we present a new calibration and analysis method allowing for quantitative FRET imaging in living cells with a simple fluorescence microscope. Aside from the spectral crosstalk corrections, two additional correction factors were defined from photophysical equations, describing the relative differences in excitation and detection efficiencies. The calibration is achieved in a single step, which renders the Quantitative Three-Image FRET (QuanTI-FRET) method extremely robust. The only requirement is a sample of known stoichiometry donor:acceptor, which is naturally the case for intramolecular FRET constructs. We show that QuanTI-FRET gives absolute FRET values, independent of the instrument or the expression level. Through the calculation of the stoichiometry, we assess the quality of the data thus making QuanTI-FRET usable confidently by non-specialists. Introduction The theory behind Förster Resonance Energy Transfer (FRET) was first successfully described in 1946 but its application to biological systems, particularly in living cells, has only become popular in the late 1990s with the cloning of fluorescent proteins. Since the first cloning of the Green Fluorescent Protein (GFP), fluorescence microscopy has rapidly become a standard tool in cell biology. Fluorescence labelling allows the localization of a protein of interest in space and time in a biological specimen, from cells to animals. The labelling of several proteins in the same sample has been used to address protein-protein interactions in terms of colocalization. However, using standard fluorescence microscopy, determination of protein-protein distance is limited by the diffraction of light i.e., hundreds of nanometers. Förster Resonance Energy Transfer (FRET) methods circumvent this barrier by allowing the detection of distances below 10 nanometers between a donor fluorophore and an acceptor through non-radiative energy transfer mediated by dipole-dipole interactions. FRET measurements can distinguish between two proteins being in the same compartment or in direct contact. Moreover, the ability to measure nanometric variations allows for the detection of protein conformational changes1 -- 3. A large class of fluorescent biosensors have been engineered based on FRET to monitor protein function (kinase4, 5, GTPase6), calcium signals,7 or more recently, forces on the molecular scale8 -- 10. The most common design relies on a molecular recognition element coupled with two fluorescent proteins (FPs) expressed in the same amino-acid sequence (intramolecular FRET sensor). An intermolecular FRET design is also possible where the FPs are inserted on two independent moieties. In this case, the apparent stoichiometry can strongly vary, which makes a quantitative analysis much more difficult. There are two main approaches for measuring FRET in living cells: one is based on the change in fluorescence intensity and the other on the change in the donor fluorescence lifetime11. Fluorescence LIfetime Microscopy (FLIM) requires sophisticated instrumentation and analysis, and is often recognized as a quantitative method for live-cell measurements. Different strategies have been developed to measure FRET efficiency via the fluorescence intensity of the donor and/or of the acceptor, some involving the total photobleaching of one fluorophore or specific instruments for spectral imaging12 -- 14. The most compatible method with dynamic quantitative FRET imaging and live-cell imaging is based on the sensitized-acceptor emission. Because the collected fluorescence intensity depends strongly on numerous instrumental factors (excitation, filter set, camera sensitivity Figure 1. A. A schematic of an experimental setup used for the validation of the framework is shown. Three images are acquired in two snapshots by automatically alternating the laser excitation and splitting the camera in two detection channels corresponding the donor and acceptor channels. B. Framework for quantitative FRET analysis. The analysis requires three images combining the detection in the donor and the acceptor channels with the excitation of the donor and the acceptor. A calibration step allows the determination of four factors correcting for the crosstalks and the relative excitation and detection efficiencies of the donor and acceptor fluorophores. As a result, instrument-independent FRET probabilities and stoichiometries are calculated. Scale bar: 20µm. etc), this approach requires several corrections to calculate an instrument-independent FRET efficiency. The literature is rich of different correction factors and mathematical expressions of FRET indices15, 16. The idea of correcting for spectral crosstalks and at least for the difference in detection efficiency between donor and acceptor channels emerged concomitantly in the single-molecule17, 18 and in the live-cell imaging fields19 -- 21. It is now generally accepted that bleedthrough of the donor emission in the acceptor channel and direct excitation of the acceptor by donor excitation channel must be corrected by substracting their contributions. This requires the acquisition of three different signals, also called 3-cube strategy in live-cell imaging20. As such, the apparent FRET index varies with the fluorophore concentration and, even with additional normalization, the direct comparison of FRET values obtained independently is not possible22. To account for photophysical artifacts, we need to go back to physical equations and determine the origin of the signal in each channel. The next obstacle is the experimental determination of the correction factors. Existing methods require samples with known FRET efficiency23 or known concentration24 or even an additional experiment using acceptor photobleaching21. In this work, we clarify the theory coming from single-molecule studies25 and adapt it to live-cell imaging. We present a new method to determine all the correction factors in a robust manner without any additional photobleaching experiment or external calibration of the FRET efficiency. The only requirement for calibration is knowledge of the donor:acceptor stoichiometry, which is in general known by construction. The calibration can thus be achieved directly on the sample of interest or with FRET standards26. While the stoichiometry can be accurately measured in the last case, this information can always be used as a quality factor to discard aberrant pixels. No specialized microscope is required as the QuanTi-FRET (Quantitative Three-Image) method can be applied to any epifluorescence triplet of images acquired with commercial instruments. Here, we demonstrate that QuanTI-FRET allows for absolute FRET measurements that are independent of the instrumental setup and of the fluorophore concentration. Being robust and including an inherent data quality check, the method can be used confidently by non-specialists, especially for FRET-based biosensors applications. Theory To obtain as much information as possible from the sample, we follow a multiple excitation scheme (Fig. 1) as introduced by Kapanidis and colleagues for single molecule spectroscopy18 and close to the three-cube method in live-cell imaging19. By switching rapidly between both excitation sources, and splitting the emission into two channels on the camera, we acquire in two successive snapshots four images: 2/13 IDD : the detected signal in the donor channel after excitation at the donor wavelength, IDA : the detected signal in the acceptor channel after excitation at the donor wavelength, IAA : the detected signal in the acceptor channel after excitation at the acceptor wavelength. The fourth image IAD contains no information, only noise, and is discarded. In principle, only IDD and IDA are sufficient to calculate the transfer efficiency. That would be the case if the photons coming from the donor and the acceptor had the same detection efficiency. In practice, it is not possible to have such an instrument and, several corrections must be considered to get unbiased quantitative FRET efficiencies. The third image, IAA, is independent from the FRET efficiency but is required to calculate all the necessary corrections. One can write the intensity of the three types of signals as a function of the photophysical and instrumental parameters and the population of the different fluorophores, nA (acceptor) and nD (donor), and FRET probability, E: IAA = nALAσ A IDD = nDLDσ D IDA = nDLDσ D AexφAηAem Adet Dex(1− E)φDηDem DexEφAηAem Ddet Adet + nDLDσ D Dex(1− E)φDηDem Adet + nALDσ A DexφAηAem Adet (1) (2) (3) where Li is the excitation intensity at the wavelength chosen for excitation of fluorophore i, σ j i is the absorption cross section of j at the excitation wavelength of i, φi is the quantum yield of i and η j i is the detection efficiency of photons emitted by j in the detection channel i. The expression of IAA is the simplest as it only depends on species A (acceptor). For IDD, one has to take into account the probability to transfer energy to the acceptor, E, as acceptor photons are not detected in this channel. Finally, to express IDA, the FRET image, not only the signal coming from FRET events must be taken into account but also the two crosstalk terms: (i) the bleedthrough of photons emitted by the donor into the acceptor channel and (ii) the direct excitation of acceptor molecules with the donor specific wavelength. Some of the parameters in the above equations are difficult to measure. We follow a pragmatical approach and avoid the systematic determination of all twelve unknowns. First, we can simplify the expressions by defining a bleedthrough correction factor as αBT and a direct excitation correction factor as δ DE. Additionally, a correction factor for the different detection efficiencies in both channels is defined as γM, and similarly a correction factor for the different excitation efficiencies in both channels is defined as β X αBT = ηDem Adet ηDem Ddet δ DE = LDσ A Dex LAσ A Aex γM = φAηAem Adet φDηDem Ddet and β X = LAσ A Aex LDσ D Dex Hence, the notation is simplified and by inverting the previous set of equations, we obtain the FRET probability: E = IDA − αBT IDD − δ DEIAA IDA − αBT IDD − δ DEIAA + γMIDD (4) (5) In addition, as in Lee et al.25, we define the stoichiometry as the relative amount of donor molecules with respect to the total number of fluorophores in each pixel: S = nD nD + nA (6) From equations (1) and (2), we derive expressions for nD and nA and insert them into equation (6). By simplifying with the excitation correction factor β X defined in equations (4), equation (6) reduces to: S = 1 + IAA IDD 1 1 β XγM (1− E) (7) To decouple stoichiometry and FRET probability, we replace E by the expression given in equation (5). Finally the stoichiometry reads: S = IDA − αBT IDD − δ DEIAA + γMIDD IDA − αBT IDD − δ DEIAA + γMIDD + IAA/β X (8) 3/13 By including the crosstalk corrections into a corrected FRET image, Icorr equations defining the FRET probability and the stoichiometry in each pixel: DA = IDA − αBT IDD − δ DEIAA, we obtain two master E = S = Icorr DA Icorr DA + γMIDD Icorr DA + γMIDD Icorr DA + γMIDD + IAA/β X (9) (10) Both E and S can be calculated from the three experimental images, IDD,IDA and IAA, and four parameters, αBT ,δ DE ,γM and β X. All four correction factors are derived from the detailed expressions of the collected fluorescence intensities in the three different channels. The notations were chosen according to the consensus in the single-molecule field27 with a supplemental exponent for a direct understanding of the role of each correction factor. The crosstalk correction factors are already widely used in the 3-cube approaches21 and are straightforward to calibrate. Imaging a donor-only sample, in vitro or in cellulo, provides αBT ; similarly, imaging an acceptor-only sample provides δ DE. αBT depends only on the donor emission spectrum, the filter set and the spectral response of the camera. δ DE depends on the acceptor excitation spectrum but also on the ratio of the illumination power in the two channels. Under the same experimental conditions (same fluorophores, same filter set and illumination intensities), the crosstalk corrections to be brought to IDA depend only on the quantity of both fluorophores, given by IDD and IAA while αBT and δ DE are unchanged. The two other correction factors, γM ("M" for Emission) and β X ("X" for Excitation), are more difficult to determine. γM accounts for the difference in the measured fluorescence emission when the same number of donor or acceptor molecules are excited. Hence, it is related to the quantum yield and to the detection efficiency of the setup in each channel. β X accounts for the difference in energy absorption for each channel. Hence, it is related to the illumination intensity and the absorption cross-section of each fluorophore. γM has already been described, in single molecule17 and in live-cell imaging21. Several indirect strategies have been developed to determine the value of γM: from acceptor photobleaching21, 28, the use of a FRET sample with known FRET efficiency23, an interpolation from two constructions with very different FRET values29 or a fit of the relation between 1/S and E25. β X has been introduced by Lee et al.25 for single molecule experiments and a similar parameter has also been empirically introduced by Chen et al. for cells experiments29. If β X and γM are determined independently, β X has no effect on the FRET efficiency but just on the stoichiometry (see equations (9) and (10)) . Since the stoichiometry in single molecule studies is often limited to donor only, acceptor only and donor:acceptor complexes, S does not need to be accurate and β X is not necessary. On the contrary, we will show that S can be very useful in live-cell experiments even when the FRET construction has a well-defined stoichiometry. Calibration of the correction factors Having described the theory directly from the physical parameters of the fluorophores and of the experimental setup, the difficult part to achieve the calculation of quantitative FRET is to determine the four correction factors. As mentioned previously, the crosstalk correction factors are measured from donor-only and acceptor-only cells, and calculated as the ratios αBT = DA Idonor−only Idonor−only DD and δ DE = DA Iacceptor−only Iacceptor−only AA (11) These ratios are calculated in each pixel of all the imaged cells and the median value is kept. The correction factors γM and β X cannot be determined from the donor-only or acceptor-only samples where the FRET probability is equal to zero (or not defined). Another piece of information is necessary and is found in the stoichiometry. Equation (10) can be rewritten as β XγMIDD + β X Icorr DA = S 1− S IAA, (12) which is the equation of a plane in the 3D space defined by {IDD,Icorr DA ,IAA}. If the stoichiometry is known, the strategy is to fit the experimental data {IDD,Icorr DA ,IAA} to a plane and thereby determine β XγM and β X. If the FRET sample of interest has an unknown stoichiometry, another calibration experiment has to be made with a defined stoichiometry FRET probe. Practically, the pixel values of a whole dataset (N cells) are gathered in vectors X = [IDD,Icorr DA ] and Y = [IAA] and the matrix A = [γMβ X ,β X ] is determined such as X.A = Y by a least-square fitting. If the sample shows only one FRET value E with different fluorescence intensities (i.e. fluorophore concentrations), the pixel values will form a straight line in the 3D space {IDD,Icorr DA ,IAA} (Fig.2). As a result, an infinite number of planes can fit the dataset. For a good determination of β X and γM, it is therefore necessary that the FRET values of the dataset are sufficiently spread. The visualization and the calculation of the correction factors in the 3D space {IDD,Icorr DA ,IAA} is the originality of this work. We compare our approach with the two other related methods in the last section. 4/13 Figure 2. FRET measurements on the three FRET standards, C5V, C17V and C32V. (A) Triplet fluorescence images are shown for exemplary cells transfected with the three FRET standards: C5V (short linker), C17V (medium linker) and C32V (long linker). The calculated FRET maps for the individual cells are shown on the right plotted using the same color scale. The highest FRET is observed for the shortest linker construct C5V and decreases to the lowest FRET construct C32V. Scale bar: 20µm. Color bar: FRET efficiency in % (B) Scattered plot of all pixels values from all cells imaged in the {IDD,Icorr DA ,IAA} 3D-space and the fitted plane, side view as inset. The three FRET standard populations forming three distinct clouds are all lying on the plane defined by β X and γM. (C) Boxplot gathering cellwise FRET values of C5V, C17V and C32V measured independently in two different labs ([A] and [B]). After calibration, the same FRET median values were obtained. Results Validation of QuanTI-FRET using FRET standards To test the proposed method in live-cell experiments, we utilized the FRET standards developed by Thaler et al.24 and Koushik et al.26. The FRET standards consist of a pair of fluoroscent proteins, a donor (Cerulean) and an acceptor (Venus), separated by an amino-acid sequence of variable length. Three standards were used in the present work to calibrate the experimental setup: C5V, C17V and C32V, where the linker between donor and acceptor consisted of 5, 17 and 32 amino-acids respectively. The construct with the shortest linker, C5V, was expected to exhibit the largest FRET efficiency and the FRET efficiency to decrease as the linker length increases26. The FRET standards were expressed in Hela cells and imaged on the setup described in Figure1. As a first step for calibration, the crosstalk corrections corresponding to the donor, Cerulean, and the acceptor, Venus, must be determined. Hence, Cerulean-only cells and Venus-only cells were imaged. Using equation (11), the bleedthrough for Cerulean was calculated as αBT = 0.421± 0.002 (10 cells) and the direct excitation of Venus as δ DE = 0.1100± 0.0008 (12 cells). The pixelwise distributions of αBT and δ DE are shown in supplementary information (Fig.S1). The second step consists in the determination of γM and β X, the factors correcting for the difference in detection and excitation efficiencies in the different channels. The three necessary fluorescence images, IDD,IDA and IAA, of three exemplary cells transfected with C5V, C17V and C32V are shown in Figure 2A. All the pixel values {IDD,IDA,IAA} of all cells expressing the three constructs were gathered as one dataset and fitted with the plane equation (12) (Fig.2B). A mask of each cell was obtained and only the pixels coming from within the cells were kept. This equation has an additional unknown, S. An assumption on S is necessary at this step. By design, the CxV constructs should have on average one donor for one acceptor, i.e. S = 0.5. This assumes the maturation efficiency of the donor and the acceptor are close to 1. We will discuss the influence of maturation in the next section. For S=0.5, the plane equation reduces to: β XγMIDD + β X Icorr DA = IAA. (13) A given set of experimental conditions (laser power, filter set, fluorophores, stoichiometry) corresponds to one plane, and in this plane, a given FRET efficiency corresponds to a line. As seen in Fig.2B, the scattered data from the three standards appear as linear clouds lying on the same plane defined by β X and γM and the assumed stoichiometry S (S = 0.5, 1 donor:1 acceptor). 5/13 Here, the least square fitting of the plane yielded β X = 1.167± 0.008 and γM = 2.10± 0.02 with a coefficient of determination R2 = 0.995. Once all the correction factors are determined, the FRET probability can be measured. Since this dataset was used for calibration with the hypothesis of S = 0.5, the stoichiometry cannot be an output for this calibration dataset. Nevertheless, no assumption was made concerning E, and therefore, the FRET probability can be calculated on the same dataset as the one used for calibration. If the experiment of interest presents a sufficiently broad distribution of FRET probabilities to determine the plane in 3D, there is no need for a different experiment with FRET standards for calibration. Hence, calibration can be achieved on-the-fly on samples with known stoichiometry. More than 25 Hela cells expressing one CxV construct were measured. The median FRET probability was EC5V = 51.1 (s.d. = 12.2, 26 cells) for C5V, EC17V = 43.1 (s.d. = 11.8, 25 cells) for C17V and EC32V = 35.1 (s.d. = 11.5, 27 cells) for C32V, calculated over more than 3· 106 pixels. The uncertainty comes rather from the cell to cell variability than from the pixel statistics. Hence, the median FRET value per cell was taken (Fig.2C, dataset [A]) and the uncertainty calculated as the standard error of the mean yielding: EC5V = 50.3± 0.4, EC17V = 41.7± 0.8 and EC32V = 35.1± 0.8. To verify that the FRET probability was independent of the fluorescence intensity, the Spearman's rank correlation coefficient was calculated between E and IAA, the only channel not affected by FRET and just related to the fluorophore concentration. Gathering the data from all three standards, the resulting Spearman's coefficient was ρ = 0.04, confirming the absence of correlation between the fluorescence level and the calculated FRET probability. This is also true pixelwise on a single cell basis (see Supplementary Fig.S2) and cellwise comparing all cells expressing one FRET standard (see Supplementary Fig.S3). Similarly, we questioned the effect of the correction factor γM by calculating the Spearman's coefficient between E and the total donor fluorescence (γMIDD + Icorr DA ) without the correction, ρ = 0.111 (γM = 1), and with the correction ρ = 0.045 (γM = 2.10). Hence, correcting for the different detection efficiencies decreases the correlation by a factor 2.5 between the donor fluorescence and the calculated FRET probability. The goal of the QuanTI-FRET method is to enable the comparison of FRET-based experiments from different studies i.e., obtained independently in different laboratories in the world. To test this, we performed the same experiments a second time in a completely independent way: with a different instrument, in a different country, by a different team on another cell culture with fresh constructs ordered directly from Addgene. The experimental data was analyzed with the exact same procedure. The calibration gave the following correction factors αBT = 0.467± 0.001 (12 cells) and δ DE = 0.101± 0.003 (12 cells) for the crosstalks and β X = 2.03± 0.07 and γM = 1.35± 0.07 (R2 = 0.82) for the excitation and emission corrections. The FRET probability was measured for the three FRET standards giving EC5V = 50± 1.7 (10 cells), EC17V = 43± 1.6 (12 cells) and EC32V = 33± 1.5 (12 cells) (Fig.2C, dataset [B]). For an easier comparison, correction factors and FRET probabilities from lab [A] and [B] are gathered in Supplementary Materials (Table S1). The variability in this second dataset was larger as seen by the smaller coefficient of determination (R2 = 0.82) of the 3D fitting and the standard deviation of the FRET probability for each construct. Nevertheless, the FRET values obtained were in agreement with the first dataset ([A]). Hence, we show that measuring FRET with the QuanTI-FRET method is quantitative: the absolute FRET values are meaningful and can be compared from one lab to another. Taking advantage of S So far, the stoichiometry was used only to calibrate the system. However, once the experimental system has been calibrated, the QuanTI-FRET analysis can determine both E and S independently. In this case, additional information can be extracted from S. First of all, S can be used to evaluate the quality of the calibration and of the dataset. As in single molecule studies, the 2D histogram combining the stoichiometry and FRET probability histograms (Fig.3A) is a useful tool . In theory, the standard constructs with 1 donor for 1 acceptor should appear as a cloud corresponding to their average FRET efficiency, E0, and S = 0.5. A known stoichiometry of 1:1 donor:acceptor is also reasonable for a biosensor construct that contains both donor and acceptor fluorescent proteins that fold and mature with high efficiency. However, when looking for interactions between different proteins, a fraction of donor only and/or acceptor only constructs are expected. If free acceptors are also present in the image, the apparent FRET probability stays constant but the stoichiometry drops (Fig.3A). On the contrary, if free donor is present with the 1:1 construct, both S and E are affected. This variation can be described theoretically. If a solution containing a donor-acceptor construct, n0 D , the apparent FRET probability and the apparent stoichiometry are (see Supplementary Information): D, with an average FRET efficiency of E0 is mixed with free donor, n f ree Eapp = 1 + 1 1− E0 + nD E0 f ree/nD 0 and Sapp = 1 + nD 1/S0 + nD f ree/nD 0 f ree/nD 0 (14) 6/13 Figure 3. (A) Influence of free donor or free acceptor in the sample. Theoretical S-E histogram with trajectories corresponding to the addition of free donor or free acceptor to a construct with 1:1 donor to acceptor ratio. (B) Experimental histogram of S versus E for constructs showing different FRET values (C32V and C5V) or different stoichiometries (CVC and VCV) as well as pure donor (Cerulean) and pure acceptor (Venus). This histogram was calculated using only the crosstalk correction. (C). The same experimental E-S histogram with the complete calibration including γM and β X. In the completely corrected 2D histogram, the stoichiometry and FRET probability are uncorrelated (ρ = 0.02). (D) Exemplary triplet of images showing a cell expressing C32V with a low signal-to-noise ratio, Scale bar 15µm. (E) The corresponding RAW E and S maps and the FRET map for the images in panel D after filtering with the weigthed gaussian filter. (F) The corresponding stoichiometry histogram and the weights (W) as a function of the stoichiometry (line). The weights are given, for each pixel, by a gaussian function of the deviation from the expected stoichiometry (S = 0.5) with a variance σS = 0.1. The corresponding map of weights W is shown in (G). (H) Line profiles corresponding to the three maps shown in panel E. Due to high intensity background in an endosome, the FRET efficiency drops (thin grey line). This anomaly is also observable in the stoichiometry (blue). By weighting the image with the measured stoichiometry, such artifacts can be avoided (magenta). . 7/13 We can then write the analytical formula describing this mix in the E-S histogram: Sapp = E0/Eapp 1/S0 + E0/Eapp − 1 , (15) which is sketched in Fig.3A. In equations (2) and (3), we assumed that all the donors were able to FRET i.e., had an acceptor partner. If this is not the case and free donors exist, then E becomes an apparent FRET probability Eapp as in equation (14). If the experimental E-S histogram can be fitted to equation (15), the FRET probability, E0 of the 1:1 construct can be extracted. The presence of free donors can result from the poor efficiency of the acceptor fluorophore to fold. As demonstrated above, this case can easily be seen and treated with the QuanTI-FRET method. The presence of free acceptors does not affect the FRET efficiency once the system calibrated. If free acceptors are present in the calibration samples, one should at least evaluate and take into account the effective stoichiometry in order to obtain a reliable calibration and avoid the propagation of biases to the measurements of interest. If both free donors and free acceptors are present, the situation is more complicated due the ensemble measurement made in each pixel. But fortunately, most of FRET-based biosensors are formed with variants of GFP, in particular of the pair CFP/YFP, which fold well30, 31. The observation of the E-S 2D histogram gives a hint about the quality of the calibration. In theory, for a sample with a fixed stoichiometry, the FRET probability and the stoichiometry should be uncorrelated resulting in horizontal clouds in the 2D histogram. Figure 3B shows the experimental data from this work with crosstalk correction but with β X and γM both set equal to 1. The constructs C5V and C32V do not lie on a horizontal line whereas they should have the same stoichiometry. On the contrary, with the complete calibration of β X and γM (Fig. 3C), the two clouds lie on a horizontal line corresponding to S = 0.5 (Spearman's correlation coefficient between E and S: ρ = 0.02 for C5V-C17V-C32V). Once the system has been calibrated with FRET probes with a known stoichiometry, the stoichiometry becomes an output of the QuanTi-FRET analysis. Two additional FRET standards were imaged under the same conditions as before, CVC (2 donors:1 acceptor) and VCV (1 donor:2 acceptors)24. As these two constructs were not used to determinate γM and β X, no assumption was made with respect to their stoichiometry. Both constructs were built with the same fluorophore pair and imaged using the same conditions (filter set, laser power, camera), hence, the calibration was still valid. Practically, the experimental results gave S = 68.9± 0.2 for CVC (24 cells), with an expected value of 66%, and S = 35.1± 0.2 for VCV (9 cells), with an expected value of 33% (Supplementary Fig.S4). Importantly, CVC and VCV experiments are well calibrated, appearing as horizontal clouds in the E-S histogram (Fig.3C). On the same histogram, the acceptor (Venus) population was found at very low stoichiometry (S = 12± 4, 12 cells) as expected and the donor population is also found where expected at stoichiometry close to 1 (S = 98.7± 0.4, 10 cells). In the case of a fixed stoichiometry sample, as is the case for most FRET-based biosensors, S can still bring an important piece of information about the confidence. The usual way to determine the uncertainty about a pixel is to rely on the photon statistics: if the fluorescence signal is high, then a high confidence is assumed. This is certainly true for pure fluorescence imaging but, in the case of FRET, there are cases where a high fluorescence intensity occurs in pixels where the FRET is biased. For instance, FRET can be affected by the local chemical environment (pH), the local crowding or by any unequal effect on the fluorescence of the donor and acceptor. An example is shown on Figure 3D where lower-than-expected FRET efficiency was observed in certain bright intracellular vesicles. The corresponding raw results of the pixel-based analysis is shown in Fig.3E (SRAW and ERAW ) and line profiles are plotted (Fig.3H). For this example, the spot pointed to by the arrow has a high fluorescence intensity in the three channels but the stoichiometry differs from the expected 50% (close to 65%). Similarly, dark, out-of-cell regions of the image also show deviation from the expected stoichiometry. We define a confidence index W as: − (S− S0)2 2σ 2 S , W = e (16) where S0 is the expected S and σS is a parameter to tune the sensitivity. W renders the deviation from an expected stoichiometry as a score between 0 and 1 (S = S0) with a gaussian shape (Fig.3F). This confidence index can be used directly to display FRET maps with color-coded FRET values and brigthness-coded W . To go one step further, the confidence index can be inserted in a spatial filter. Indeed, FRET maps often need to be spatially averaged, the actual resolution being limited by the diffusion of the FRET species and larger than the pixel size. A weighted gaussian filter was therefore designed where the effect of a gaussian kernel (typically 7x7 pixels2) was locally weigthed with W (Fig.3G) as follows: (W ◦ E)∗ G W ∗ G , E f ilt = (17) where ∗ denotes a convolution and ◦ the Hadamard product, E and W are dealt as matrices corresponding to the raw FRET image and the weights as defined in equation (16), E f ilt being the filtered FRET map. As the gaussian distribution never reaches 8/13 β X QuanTI-FRET 1.167± 0.008 1.13± 0.01 Lee et al.25 1.135± 0.005 Chen at al.29 γM 2.10± 0.02 2.37± 0.05 2.19± 0.02 C5V 50.3± 0.4 47.5± 0.4 49.4± 0.4 C17V 41.7± 0.8 38.9± 0.8 40.8± 0.8 C32V 35.1± 0.8 32.4± 0.8 34.2± 0.8 Table 1. Systematic comparison of QuanTI-FRET method with previous work from Lee et al.25 and Chen et al.29 Dataset [A] was analyzed with the three methods, the resulting correction factors and FRET probabilities for C5V , C17V and C32V are given in this table, with the uncertainty on β X and γM resulting from a different bootstrap analysis. zero, an additional threshold was applied based on the local weight of the considered pixel. An example is shown on Figure3E, the application of the weighted gaussian filter (σS = 0.1,σGauss = 1.5 and threshold on W Wth = 0.5) totally eliminates the background around the cells and also very dim areas inside cells as well as the bright vesicle with anomalous stoichiometry (Fig.3H). Discussion The definitions of FRET probability and stoichiometry used in QuanTI-FRET are mathematically equivalent to what was introduced previously by Chen et al.29 (γM ≡ G and β X ≡ 1/(G· k)) and Lee et al.25 (γM ≡ γ and β X ≡ β ). Therefore, we compared the performances of QuanTI-FRET to these two particular other methods. In the work by Chen et al.29, the physical origin of the parameters was not described in detail as γM was already introduced by Zal and Gascoigne21 and the second parameter, k, was rationally defined from the γM-corrected intensities to account for the stoichiometry. The proposed calibration was achieved in two separated steps. First, two constructs with defined and well-separated FRET efficiencies were needed to determine γM (a.k.a G). Second, a FRET standard with known stoichiometry was measured to calculate the other parameter, k, using G determined in step 1. In Chen's work, the calibration was achieved by imaging the FRET standards C5V and CTV, where the linker T is the 229 amino-acid TRAF domain of the TRAF2 protein24. However, the observation of the 3D representation of all the standards, including CTV, imaged in the present work, shows that CTV does not lie on the same plane as C5V, C17V and C32V (Supplementary Fig.S5). This is also visible in the E-S 2D histogram where the CTV cloud is tilted (Supplementary Fig.S5). These observations are in agreement with the later work of Koushik and Vogel32 and demonstrate the utility of the 3D representation of the fluorescence intensities as well as the E-S 2D histogram to proofread the quality of the experimental data. The analysis of the experimental dataset [A] with Chen's method gave results close to the QuanTI-FRET method: G = 2.19± 0.02 to compare with γM = 2.10± 0.02 and 1/(G· k) = 1.135± 0.005 to compare with β X = 1.167± 0.008 (see Table 1). However, the analysis of the second dataset [B] gave different results between the two methods: G = 3.63± 1 to compare with γM = 1.35± 0.07 and 1/(G· k) = 1.02± 7 to compare with β X = 2.03± 0.07 yielding less reliable FRET probabilities (respectively 16%, 24% and 30% for C32V, C17V and C5V). This discrepancy results from the dataset being less homogeneous and the limited number of cell-containing pixels where the two-step calibration of G and k is less robust than the single-step fit of the QuanTi-FRET method. In the work of Lee et al.25, the calibration consists of first calculating Eraw and Sraw with only spectral crosstalk corrections and then fitting the linear relation between 1/Sraw and Eraw, hereby assuming a 1:1 stoichiometry (see Supplementary Information). This method yielded very similar results to QuanTI-FRET: γ = 2.37± 0.05 to compare with γM = 2.10± 0.02 and β = 1.13± 0.01 to compare with β X = 1.167± 0.008, resulting in FRET values slightly lower for the FRET standards (∆E = 3%). The second dataset ([B]) was also used to test Lee's method leading to a decrease in the FRET values of ∆E = 8% with a relative difference of 11% for β X and 28% for γM. The correction factors and resulting FRET for the three FRET standards are summarized in Table 1. The average FRET probabilities are in very good agreement between QuanTI-FRET and Chen's methods, a systematic difference of about 3% is observed with Lee's method. The three methods can all be considered as quantitative. To further test the relative robustness of the three methods, a systematic bootstrap testing on experimental data ([A] with C5V, C17V and C32V) was performed. The whole experimental dataset was randomly divided to produce artificially smaller datasets and give access to statistical errors on the correction factors determination (as given so far). The standard deviation of γM was around 0.12 (QuanTI-FRET and Chen's) and 0.23 (Lee's) for the minimum tested sample sizes between 1000 and 1300 points. The standard deviation of β X was found to be around 0.04 (QuanTI-FRET and Chen's) and 0.07 (Lee's) for the same range of sample sizes. Over the whole range of sample sizes (from 103 to 105), the standard deviation of both correction factors obtained by Lee's method remained larger than the ones obtained by Chen's and QuanTI-FRET (see Supplementary Fig.S6). This analysis demonstrates that Lee's method is less robust to dataset length, probably due to the fitting of 1/S which diverges for small S values. A different test was performed by reducing the FRET range of the calibration dataset by taking alternatively only two standards (C5V-C17V, C17V-C32V and C5V-C32V) into account. In this case, Chen's method was not valid anymore for 9/13 C5V-C17V and C17V-C32V couples resulting in relative variations of 76% for G and 38% for β X (see Supplementary Fig.S6). Indeed, Chen's method relies purely on the comparison between the average intensities of two populations, the uncertainty grows as the FRET distance decreases. QuanTI-FRET and Lee's methods, by fitting the total distribution, perform well in this bench test (relative variations of 14% and 22% for γM respectively with QuanTI-FRET and Chen's, and 7% and 12% respectively for β X, see Supplementary Fig.S5). All in all, even if the three methods are quantitative in the best case scenario, QuanTI-FRET was demonstrated to be more robust to dataset dispersity, length and FRET range. The single-step calibration in a 3D IDD,Icorr DA ,IAA representation, on a continuous distribution of FRET efficiencies allows for the calibration on-the-fly of the sample of interest itself, provided a defined stoichiometry and a distribution of FRET efficiencies in the range of the bench test (at least 5 %). Taking inspiration from single-molecule literature, we can further exploit stoichiometry to provide a quality check of the experimental data and thereby filter the resulting FRET images. Conclusion Building upon the previous contributions from live-cell and single-molecule FRET experiments, we present a new framework allowing for quantitative FRET imaging in living cells with a simple multi-channel epifluorescence microscope. Here, we demonstrated the consistency of the method on two different microscopy systems in different laboratories. The QuanTI-FRET method does not require specific instrument for determining spectra or lifetime nor specific hardware development. Image- splitting devices and LED excitation are now commercially available and allow for the same image acquisition protocols as the experimental system used in this work. The QuanTI-FRET calibration does not require acceptor photobleaching, purified proteins or known FRET samples. The only requirement is a known stoichiometry sample (as other quantitative methods) with a broad FRET distribution, which can be obtained directly from the FRET construct of interest (intramolecular-FRET-based biosensors for instance). Nevertheless, an independent calibration using FRET standards is recommended as it allows one to evaluate FRET efficiency and stoichiometry independently. The QuanTI-FRET method was demonstrated to be quantitative and robust, with the additional benefit of having an inherent data quality check. Methods Cells and plasmids All plasmids were gifts from Steven Vogel: C5V (Addgene plasmid # 26394), C17V (Addgene plasmid # 26395), C32V (Addgene plasmid # 26396), mVenus N1 (Addgene plasmid # 27793), mCerulean C1 (Addgene plasmid # 27796), VCV (Addgene plasmid # 27788), CVC (Addgene plasmid # 27809) and CTV (Addgene plasmid # 27803). Plasmids were amplified in E.Coli (DH5α) and purified using the NucleoBond R(cid:13)Xtra kit from Macherey-Nagel GmbH (http://www.mn-net.com). Hela cells were cultured in Dulbecco's Modified Eagle Medium high glucose supplemented with Foetal Bovine Serum (10%), GlutaMAXTM (GibcoTM) and Penicillin/ Streptomycin (1%). Cells were transfected with Lipofectamine R(cid:13)2000 (InvitrogenTM) and Opti-MEM (GibcoTM), then incubated in Fluorobrite DMEM medium (GibcoTM) overnight and finally imaged in Leibovitz's L-15 medium (GibcoTM) without phenol red. Microscopic image acquisition, Grenoble, setup [A] Imaging was done with a system based on an Olympus IX83 body equipped with a home-made image splitting coupled to a sCMOS camera (ORCA Flash V2, Hamamatsu) as sketched in Fig.1. Excitation was done by a supercontinuum white laser (Fianium) coupled to a high power AOTF (Fianium), which was controlled through an FPGA-RT unit (National Instruments) coded with Labview. This unit synchronized the alternated laser excitation with the camera acquisition. Images were acquired at 37◦C with Micromanager and a 40x objective. The donor fluorophore was excited at 442nm, the acceptor at 515nm. The fluorescence emission was first separated from the excitation via a triple line beamsplitter (Brightline R442/514/561 Semrock) in the microscope body. The fluorescence emission was further splitted with a beamsplitter at 510nm (Chroma) and filtered with a 475/50 filter (BrightLine HC, Semrock) for the donor channel and a 519/LP longpass filter (BrightLine HC, Semrock) for the acceptor channel. Hence, in two camera snapshots, four images were obtained with all combinations of donor/acceptor excitation and donor/acceptor emission. Microscopic image acquisition, Munich, setup [B] Images were acquired on a Nikon Eclipse Ti microscope with home-built excitation and detection pathways. A 100x oil immersion objective (Apo-TIRF 100x Oil/NA 1.49, Nikon) was used for all measurements. Samples were excited with 445nm (MLD, Cobolt) and 514nm (Fandango, Cobolt) diode lasers coupled to an AOTF (PCAOM LFVIS5, Gooch & Housego) controlled by a FPGA unit (cRIO-9074, National Instruments). The fluorescence emission was separated from excitation pathway with a triple line 445/514/594 beamsplitter. The donor and acceptor emission were separated using an additional 514LP beamsplitter and were then spectrally filtered using 480/40 and 555/55 bandpass filters respectively before being detected 10/13 on separate EMCCD cameras (DU-897, Andor). Each cell was excited for 300 ms at 445 nm followed by 300 ms at 514 nm. The camera exposure was synchronized to laser excitation through the FPGA unit and a self-written Labview program. This produced four images over two exposure periods capturing donor and acceptor emissions at each excitation wavelength. Image analysis All the image analysis calculations were coded in Python, figures and plots were done in Python except for the boxplots obtained with PlotofPlots33. Raw fluorescence images were pre-treated by substracting the dark count of the camera and flattened by dividing with a fluorescence image obtained from a uniform fluorescent sample (Chroma slide). An essential step is then the registration between the two channels obtained on each half of the camera or between cameras. Brightfield images of beads randomly and densely spread on a coverslip were used for calibration. By calculating the image cross-correlations in local regions of the image between the two channels, a displacement map was obtained and hence a transformation matrix was calculated (accounting for translation, rotation, shear and magnification). This transformation matrix was systematically applied to IDD to match IDA and IAA before any calculation. Calibration of the system with QuanTI-FRET was done as explained in the main text. Visualization of the 3D fit was done in Paraview to explore all view angles. All calculations were done pixelwise. Parameters for the weighted gaussian filter are chosen as for gaussian filtering depending on the pixel intensity. Here, the spatial filtering is principally used to filter out pixels with an aberrant stoichiometry, i.e. S larger than 0.6 or smaller than 0.4 as estimated from the S-E histograms. The spatial gaussian enveloppe is designed to avoid adding noise in this operation, as S is subjected to stochastic pixel-to-pixel noise as E. The data that support the findings of this study are available from the corresponding author upon reasonable request. References 1. Ha, T. et al. Single-molecule fluorescence spectroscopy of enzyme conformational dynamics and cleavage mechanism. Proc. Natl. Acad. Sci. USA 6 (1999). 2. Weiss, S. Fluorescence Spectroscopy of Single Biomolecules. Science 283, 1676 -- 1683, DOI: 10.1126/science.283.5408. 1676 (1999). 3. Erickson, M. G., Alseikhan, B. a., Peterson, B. Z. & Yue, D. T. Preassociation of calmodulin with voltage-gated Ca 2+ channels revealed by FRET in single living cells. Neuron 31, 973 -- 985, DOI: 10.1016/S0896-6273(01)00438-X (2001). 4. Zhang, J., Ma, Y., Taylor, S. S. & Tsien, R. Y. Genetically encoded reporters of protein kinase A activity reveal impact of substrate tethering. Proc. Natl. Acad. Sci. 98, 14997 -- 15002, DOI: 10.1073/pnas.211566798 (2001). 5. Ting, A. Y., Kain, K. H., Klemke, R. L. & Tsien, R. Y. Genetically encoded fluorescent reporters of protein tyrosine kinase activities in living cells. Proc. Natl. Acad. Sci. 98, 15003 -- 15008, DOI: 10.1073/pnas.211564598 (2001). 6. Pertz, O., Hodgson, L., Klemke, R. L. & Hahn, K. M. Spatiotemporal dynamics of RhoA activity in migrating cells. Nature 440, 1069 -- 72, DOI: 10.1038/nature04665 (2006). 7. Miyawaki, A. et al. Fluorescent indicators for Ca 2 + based on green fluorescent proteins and calmodulin. Nature 388, 882 -- 887 (1997). 8. Grashoff, C. et al. Measuring mechanical tension across vinculin reveals regulation of focal adhesion dynamics. Nature 466, 263 -- 6, DOI: 10.1038/nature09198 (2010). 9. Meng, F., Suchyna, T. M. & Sachs, F. A fluorescence energy transfer-based mechanical stress sensor for specific proteins in situ: Mechanical stress sensor. FEBS J. 275, 3072 -- 3087, DOI: 10.1111/j.1742-4658.2008.06461.x (2008). 10. Ringer, P. et al. Multiplexing molecular tension sensors reveals piconewton force gradient across talin-1. Nat. Methods 14, 1090 -- 1096, DOI: 10.1038/nmeth.4431 (2017). 11. Padilla-Parra, S. & Tramier, M. FRET microscopy in the living cell: different approaches, strengths and weaknesses. Bioessays 34, 369 -- 76, DOI: 10.1002/bies.201100086 (2012). 12. Chen, Y., Mauldin, J. P., Day, R. N. & Periasamy, A. Characterization of spectral FRET imaging microscopy. J. Microsc. 228, 139 -- 152 (2007). 13. Wlodarczyk, J. et al. Analysis of FRET Signals in the Presence of Free Donors and Acceptors. Biophys. J. 94, 986 -- 1000, DOI: 10.1529/biophysj.107.111773 (2008). 14. Arsenovic, P. T., Mayer, C. R. & Conway, D. E. SensorFRET: A Standardless Approach to Measuring Pixel-based Spectral Bleed-through and FRET Efficiency using Spectral Imaging. Sci. Reports 7, 15609, DOI: 10.1038/s41598-017-15411-8 (2017). 11/13 15. Berney, C. & Danuser, G. FRET or no FRET: a quantitative comparison. Biophys. J. 84, 3992 -- 4010, DOI: 10.1016/ S0006-3495(03)75126-1 (2003). 16. Zeug, A., Woehler, A., Neher, E. & Ponimaskin, E. G. Quantitative intensity-based FRET approaches - A comparative snapshot. Biophys. J. 103, 1821 -- 1827, DOI: 10.1016/j.bpj.2012.09.031 (2012). 17. Dahan, M. et al. Ratiometric measurement and identification of single diffusing molecules. Chem. Phys. 247, 85 -- 106, DOI: 10.1016/S0301-0104(99)00132-9 (1999). 18. Kapanidis, A. N. et al. Fluorescence-aided molecule sorting: analysis of structure and interactions by alternating-laser excitation of single molecules. Proc. Natl. Acad. Sci. U. S. A. 101, 8936 -- 8941, DOI: 10.1073/pnas.0401690101 (2004). 19. Youvan, D. C. et al. Fluorescence Imaging Micro-Spectrophotometer (FIMS). Biotechnol. et alia 1, 1 -- 16 (1997). 20. Gordon, G. W., Berry, G., Liang, X. H., Levine, B. & Herman, B. Quantitative fluorescence resonance energy transfer measurements using fluorescence microscopy. Biophys. J. 74, 2702 -- 2713, DOI: 10.1016/S0006-3495(98)77976-7 (1998). 21. Zal, T. & Gascoigne, N. R. J. Photobleaching-corrected FRET efficiency imaging of live cells. Biophys. J. 86, 3923 -- 39, DOI: 10.1529/biophysj.103.022087 (2004). 22. Hochreiter, B., Kunze, M., Moser, B. & Schmid, J. A. Advanced FRET normalization allows quantitative analysis of protein interactions including stoichiometries and relative affinities in living cells. Sci. Reports 9, 8233, DOI: 10.1038/ s41598-019-44650-0 (2019). 23. Hoppe, A., Christensen, K. & Swanson, J. a. Fluorescence resonance energy transfer-based stoichiometry in living cells. Biophys. J. 83, 3652 -- 64, DOI: 10.1016/S0006-3495(02)75365-4 (2002). 24. Thaler, C., Koushik, S. V., Blank, P. S. & Vogel, S. S. Quantitative Multiphoton Spectral Imaging and Its Use for Measuring Resonance Energy Transfer. Biophys. J. 89, 2736 -- 2749, DOI: 10.1529/biophysj.105.061853 (2005). 25. Lee, N. K. et al. Accurate FRET measurements within single diffusing biomolecules using alternating-laser excitation. Biophys. J. 88, 2939 -- 53, DOI: 10.1529/biophysj.104.054114 (2005). 26. Koushik, S. V., Chen, H., Thaler, C., Iii, H. L. P. & Vogel, S. S. Cerulean , Venus , and VenusY67c FRET Reference Standards. Biophys. J. 91, L99 -- L101, DOI: 10.1529/biophysj.106.096206 (2006). 27. Hellenkamp, B. et al. Precision and accuracy of single-molecule FRET measurements -- a multi-laboratory benchmark study. Nat. Methods 15, 669 -- 676, DOI: 10.1038/s41592-018-0085-0 (2018). 28. Ha, T. et al. Probing the interaction between two single molecules: fluorescence resonance energy transfer between a single donor and a single acceptor. Proc. Natl. Acad. Sci. 93, 6264 -- 6268, DOI: 10.1073/pnas.93.13.6264 (1996). 29. Chen, H., Puhl, H. L., Koushik, S. V., Vogel, S. S. & Ikeda, S. R. Measurement of FRET Efficiency and Ratio of Donor to Acceptor Concentration in Living Cells. Biophys. J. 91, L39 -- L41, DOI: 10.1529/biophysj.106.088773 (2006). 30. Frommer, W. B., Davidson, M. W. & Campbell, R. E. Genetically encoded biosensors based on engineered fluorescent proteins. Chem. Soc. Rev. 38, 2833, DOI: 10.1039/b907749a (2009). 31. Sizaire, F. & Tramier, M. FRET-based biosensors: genetically encoded tools to track kinase activity in living cells. In Protein phosphorylation, 179 (IntechOpen, 2017). 32. Koushik, S. V. & Vogel, S. S. Energy migration alters the fluorescence lifetime of Cerulean: implications for fluorescence lifetime imaging Forster resonance energy transfer measurements. J. Biomed. Opt. 13, 031204, DOI: 10.1117/1.2940367 (2008). 33. Postma, M. & Goedhart, J. PlotsOfData -- A web app for visualizing data together with their summaries. PLOS Biol. 17, e3000202, DOI: 10.1371/journal.pbio.3000202 (2019). Acknowledgements The authors are grateful to S. Pinel and N. Scaramozzino (M2Bio platform) for advices and help in plasmid amplification and purification. We thank B. Arnal and I. Wang for fruitful discussions and numerical advices. P.Moreau is acknowledged for his technical participation in building and maintaining the setup. This work was funded by the Agence Nationale de la Recherche (ANR, grant n◦ ANR-13-PDOC-0022-01), and supported by the Université Grenoble Alpes (UGA, AGIR-POLE program 2015, project ACTSUB). C.M.B., A.C. and A.D. are part of the GDR 3070 CellTiss. D.C.L. gratefully acknowledges the financial support of the Deutsche Forschungsgemeinschaft (DFG) via the collaborative research center (SFB1035, Project A11) and the Ludwig-Maximilians-Universität through the Center for NanoScience (CeNS) and the BioImaging Network (BIN). CAR and CO are supported by FRM and ANR (CE17 CE13 0022 01). 12/13 Author contributions statement A.D. conceived the experiments and the theory. D.C.L contributed to the concept. C.O and C.A-R contributed to the design of the biological protocols. A.C., C.M.B, C.Q. conducted the experiments. A.D and A.C analyzed the results. F.W. and C.M.B built and interfaced the hardware. A.D drafted the manuscript. All authors reviewed the manuscript and approved the final version. Additional information The author(s) declare no competing interests. 13/13
1902.05780	1	1902	2019-02-15T11:59:46	Thermodynamics of force-induced B-DNA melting: single-strand discreteness matters	[ "physics.bio-ph", "cond-mat.soft", "q-bio.BM" ]	Overstretching of B-DNA is currently understood as force-induced melting. Depending on the geometry of the stretching experiment, the force threshold for the overstretching transition is around 65 or 110 pN. Although the mechanisms behind force-induced melting have been correctly described by Rouzina and Bloomfield \cite{RouzinaBloomfield2001a}, neither force threshold has been exactly calculated by theory. In this work, a detailed analysis of the force-extension curve is presented, based on a description of single-stranded DNA in terms of the discrete Kratky-Porod model, consistent with (i) the contour length expected from the crystallographically determined monomer distance, and (ii) a high value of the elastic stretch modulus arising from covalent bonding. The value estimated for the ss-DNA persistence length, $\lambda = 1.0 $ nm, is at the low end of currently known estimates and reflects the intrinsic stiffness of the partially, or fully stretched state, where electrostatic repulsion effects are expected to be minimal. A detailed analysis of single- and double-stranded DNA free energies provides estimates of the overstretching force thresholds. In the unconstrained geometry, the predicted threshold is 64 pN. In the constrained geometry, after allowing for the entropic penalty of the plectonemic topology of the molten state, the predicted threshold is 111 pN.	physics.bio-ph	physics	Thermodynamics of force-induced B-DNA melting: single-strand discreteness matters Nikos Theodorakopoulos1,2 1Theoretical and Physical Chemistry Institute, National Hellenic Research Foundation, Vasileos Constantinou 48, 116 35 Athens, Greece 2Fachbereich Physik, Universitat Konstanz, 78457 Konstanz, Germany (Dated: February 18, 2019) Overstretching of B-DNA is currently understood as force-induced melting. Depending on the geometry of the stretching experiment, the force threshold for the overstretching transition is around 65 or 110 pN. Although the mechanisms behind force-induced melting have been correctly described by Rouzina and Bloomfield [1], neither force threshold has been exactly calculated by theory. In this work, a detailed analysis of the force-extension curve is presented, based on a description of single-stranded DNA in terms of the discrete Kratky-Porod model, consistent with (i) the contour length expected from the crystallographically determined monomer distance, and (ii) a high value of the elastic stretch modulus arising from covalent bonding. The value estimated for the ss-DNA per- sistence length, λ = 1.0 nm, is at the low end of currently known estimates and reflects the intrinsic stiffness of the partially, or fully stretched state, where electrostatic repulsion effects are expected to be minimal. A detailed analysis of single- and double-stranded DNA free energies provides estimates of the overstretching force thresholds. In the unconstrained geometry, the predicted threshold is 64 pN. In the constrained geometry, after allowing for the entropic penalty of the plectonemic topology of the molten state, the predicted threshold is 111 pN. PACS numbers: 36.20.-r, 87.14.Gg, 87.15.Aa I. INTRODUCTION Understanding the elastic properties of single-stranded (ss) DNA is an important problem in a variety of bio- physical contexts, e.g. molecular recognition, folding into hairpin-like structures, unzipping and/or overstretching of the double helix. The latter process, where an ap- plied force of about 65 pN (or 110 pN, depending on the boundary conditions) destroys the B-DNA structure [2] is - following some speculation over a possible distinct double-stranded (ds) intermediate state - currently un- derstood as force-induced melting [1, 3]. Current estimates of ss-DNA flexibility, as expressed by the persistence length, vary, mainly according to the salinity of the solution, between 0.75 and 2 nm [4 -- 7]. The lowest value corresponds to a theoretical limit of high salt concentration which acts to shield out any elec- trostatic repulsion and therefore reflects in some sense the intrinsic stiffness of the bonds which form the poly- mer chain. On the other hand, the distance between successive monomers has been determined by a careful averaging of available crystallographic data to be 0.63 nm [8], i.e. comparable to the persistence length. This renders the use of a continuum approximation, and hence the use of the wormlike chain (WLC) model in analyzing ss-DNA flexibility data rather questionable - although in fact many of the present estimates have been based upon this model. In this paper, I analyze the force-extension data on DNA overstretching [3] in terms of the discrete Kratky- Porod (KP) model [9]. The analysis allows for elastic stretching of ss-DNA beyond its contour length, using an estimate of the stretch modulus based on ab initio cal- culations [10], which is in line with AFM measurements [10, 11]. The data can be well described in terms of a monomer distance equal to 0.63 nm, the average crystallographic value, and a persistence length equal to 1.0 nm. This value of the persistence length lies at the low end of cur- rent estimates and is consistent with the physics of the stretched state, which is expected to be less responsive to the electrostatic repulsion, and therefore reflects the intrinsic entropic stiffness of the chain. I present a simple thermodynamic analysis of the over- stretching transition, along the lines developed in [1]. In the case of unconstrained geometry, using the elastic energies of ss- and ds-DNA, along with standard DNA melting parameters as the only input, I calculate the threshold of the overstretching force, which turns out to be in excellent agreement with the experimentally deter- mined one. The case of constrained geometry is slightly more complicated. The ends of both strands of the DNA molecule are held fixed. This means that the molten state does not release the full amount of entropy corre- sponding to two free chains. Assuming that, in the over- stretched state, the two ss-DNA strands wrap around each other with a linking number "inherited" from the original, double-helical structure [3], one can calculate the entropic penalty arising from the constraint. Taking this into account predicts a force threshold very close to the observed one and supports the view of the plectone- mic structure of the overstretched state. The paper is divided in four parts. Section II provides some necessary background material on force-extension relationship, based on both the KP model and its con- tinuum WLC limit, and a comparative analysis of the two approaches in the case where the continuum approx- imation breaks down. Section III compares the results of theoretical calculations with experiment. A brief discus- sion of the results is presented in Section IV. II. THE FORCE-EXTENSION RELATIONSHIP A. Kratky-Porod, model essentials The KP model describes chain polymers with nonzero rigidity. A conformation of the chain with segments j = 1, · · · N , each of which has a fixed length a, and a direction vector ~tj has a total elastic energy H = − κ a N −1 (~tj · ~tj+1 − 1) − a ~f · Xj=1 N Xj=1 ~tj , (1) 2 can be computed by using in (3) the standard expansion of the isotropic interaction in terms of spherical harmon- ics eb~tj ·~tj+1 = ∞ l Xl=0 Xm=−l il(b) Ylm(Ωj)Y ∗ lm(Ωj+1) , (6) where il is the modified spherical Bessel function of lth order. The result of performing the angular integrations in (3) can be expressed as the leading diagonal element of the matrix product ZN (f ) = (UN )00 with the elements of the real, symmetric matrix U given by [12, 13] 1 2 Ull′ ≡ [il(b)il′ (b)(2l + 1)(2l′ + 1)]1/2Fll′ , l, l′ = 0, 1, · · · , (7) where il(b) = il(b)/i0(b) are ratios of the modified spher- ical Bessel functions, the Pl's are Legendre polynomials with the normalization Pl(0) = 1, where κ is a measure of the chain's stiffness and ~f repre- sents an externally applied force. The stretched chain has a fixed contour length L0 = N a. The WLC elastic energy can be obtained from (1) in the limit a = L0/N, N → ∞, as Fll′ ( f ) = Z 1 −1 dxPl(x)Pl′ (x)e f x (8) l+l′ = Xk=l−l′,k+l+l′=2r (2k + 1)(−1)k 1 r + 1/2 2 − ~f ·Z L0 0 ds ~t(s) , (2) · Ψ(r − k)Ψ(r − l)Ψ(r − l′) Ψ(r) ik( f ) , H = κ 2 Z L0 0 ds(cid:12)(cid:12)(cid:12)(cid:12) ∂~t ∂s(cid:12)(cid:12)(cid:12)(cid:12) in terms of the continuous direction field ~t(s) and the local curvature. The canonical partition function in the presence of the external force is ZN (f ) = Z dΩ1 · · · dΩN N −1 Yj=1 eb(~tj ·~tj+1−1) N Yj=1 eβa ~f ·~tj , (3) where β = 1/(kBT ), kB is the Boltzmann constant, T the temperature, b = βκ/a and Ωj ≡ (θj, φj) the solid angle specifying the orientation of ~tj. The persistence length, defined as the characteristic scale at which orientational correlations decay, is given by λ = − a ln(cid:0)cothb − 1 b(cid:1) . (4) The quantity of interest is the average extension L =< ( ~RN − ~R0) · f >= 1 β ∂ ∂f ln ZN (f ) , where Ψ(n) = Γ(n + 1 2 ) Γ(n + 1)Γ( 1 2 ) = n Yj=1 (cid:18)1 − 1 2j(cid:19) , (9) and f = βaf . The above equations provide a basis for a full numer- ical calculation of the force-extension curve in the case of discrete KP chains. If the number of monomers N is large, (5) will be dominated by the largest eigenvalue, µ0, of the matrix U. The force-dependent part of free energy will be given by G = −N kBT ln µ0 , (10) and the average extension, as a fraction of the contour length, will be L L0 = ∂ ln µ0 ∂ f . (11) Numerical diagonalization of the infinite matrix U ne- cessitates its truncation to finite dimension lmax. In the cases considered in the present paper an lmax of the order of 10 is quite sufficient, leading to very rapid computa- tion. As a practical matter, it is convenient to divide all elements Fll′ by an overall factor i0( f ) and add an extra term ZN (f ) = ZN (f ) ZN (0) , (5) to the right hand side of (11). ℓ0( f ) = coth f − 1/ f 3 B. The WLC limit C. The freely-jointed chain (FJC) limit In order to obtain the force-extension curve in the con- tinuum (WLC) limit, we keep terms of order a = L0/N and let N → ∞. There are two contributions of this order. The first comes from the asymptotic form of the modified spherical Bessel functions [14] for large argu- ments b = βκ/a, il(b) ∼ 1 2b eb(cid:26)1 − l(l + 1) 2b + O(1/b2)(cid:27) resulting in il(b) ∼ 1 − l(l + 1) 2b and generates a contribution of order a from the prefactor in (9). The second comes from expanding the exponential eβaf x ≈ 1 + aβf x. Combining the two results in U = I − 1 N L0 λ J , where I is the unit matrix, Jll′ = l(l + 1) 2 δll′ − βλf l′δl′,l+1 + lδl′,l−1 [(2l + 1)(2l′ + 1)]1/2 , (12) and, λ = βκ, the WLC persistence length. possible to take the limit It is now lim N→∞ UN = lim N→∞(cid:18)1 − 1 N L0 λ J(cid:19)N = e−L0/λJ . Using the eigenvector expansion of the J matrix results in ZW LC(f ) = lim N→∞ ZN (f ) = Xν 2 e−L0/λΛν (cid:12)(cid:12)A0 ν(cid:12)(cid:12) , (13) where Λν is the νth eigenvalue and A0 ν =< ν0 > is the l = 0 component of the νth eigenvector. The extension can, in general, be obtained by differentiating with re- spect to f . If the chain is flexible, i.e. the contour length L0 is much larger than the persistence length λ, (13) will be dominated by the smallest eigenvalue Λ0. In this case, the force-dependent part of the free energy will be G ≈ kBT L0 λ Λ0 if L0 ≫ λ , (14) and the relative extension L L0 ≈ − ∂Λ0 ∂ ¯f if L0 ≫ λ , (15) where ¯f = λβf . Note that the above condition is always satisfied in the case of long, genomic samples, e.g. such as used in the overstretching experiments considered here. On the other hand, it is clearly violated by DNA chains at the 100 nm scale, where contributions from the whole eigenvalue spectrum must be taken into account. For the sake of completeness, I also consider the limit of a freely-jointed chain with a segment length (Kuhn length) equal to 2λ. The relative extension is then given by L L0 = ℓ0(2βλf ). (16) D. Enthalpic corrections As the applied force increases beyond a few pN, stan- dard mechanical restoring forces come into play. In the case in double-stranded DNA, π-electron stacking forces provide a common physical origin of the stretching and bending modulus; enthalpic and entropic elasticity are both important near the fully stretched state. The esti- mated [15, 16] stretch modulus is about F0 = 1000 pN. For covalently bonded polymers this threshold may not set in until very high forces are applied. Early AFM measurements performed in the case of ss-DNA, which is bound by covalent bonds, show that it can sustain forces at least up to 800 pN, with an elastic stretch of the order of 15% beyond the crystallographic monomer distance [11]. This suggests an elastic stretch modulus of the order of 5 nN, implying a correction of the order of 1% in the context of a 100 pN experiment. Further work combining AFM measurements with ab initio cal- culations [10] estimates a value of the elastic modulus F0 = 8.4 nN for ss-DNA. I will use this estimate in the present analysis. Elastic stretching can be explicitly taken into account in the form L0(cid:19)total (cid:18) L = L L0 ·(cid:18)1 + f F0(cid:19) , (17) where the first factor may originate from any of the en- tropic elasticity alternatives (KP, WLC, FJC). E. Why discreteness matters I will now compare the force-extension curves pro- vided by three different models describing a flexible system with a large number of monomers (N=1000), and monomer distance a comparable to the persistence length. This corresponds closely to the situation of stretching experiments with ss-DNA. The first model is the Kratky-Porod model with N = 1000 segments, a monomer distance a = 0.63 nm and a stiffness constant κ/a = 8.0 pN nm, which from (4) implies λ = 1.0 nm at a temperature 20 C. The second model is a WLC with the same persistence length λ = 1.0 nm and a contour length L0 = 1000 × 0.63 nm = 630 nm. The third model is a freely-jointed chain (FJC) with a 4 L0 = N a computed with a = 0.34 nm and N = 8400 [3] and a persistence length λ = 53.2 nm. Only the latter parameter has been adjusted in order to obtain a visual fit to the data; its value is in excellent agreement with the one obtained from earlier force-extension measurements [17]. At forces beyond the 65 pN plateau, the data can be fitted to the KP model force-extension relationship (11) for ss-DNA, corrected for elastic stretching according to (17). I have used the monomer distance a′ = 0.63 nm, as determined by averaging over crystallographic data [8], the number of bases N = 8400 as given in [3], and a stretch modulus equal to 8.4 nN [10]. Again, the only adjustable parameter used to obtain a visual fit is the local stiffness κ/a′ = 8.0 pN nm, which corresponds to a persistence length λ = 1.0 nm. expt data WLC, =53.2 nm KP, =1.0 nm L/L_0 = a'/a fc = 65 pN 120 100 80 60 40 20 0 ) N p ( e c r o f 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 L/L0 FIG. 2: The DNA force-extension curve at forces up to 120 pN. Extension is measured relative to the ds-DNA contour length. Continuous line, experimental data redrawn from [3]. Dashed-dotted curve, theoretical WLC calculation with λ = 53.2 nm, contour length 8400 × 0.34 nm and elastic (stretch) modulus F0 = 1000 pN. Dashed curve, discrete Kratky-Porod calculation with N = 8400, monomer distance a′ = 0.63 nm, stiffness constant κ/a′ = 8 pN nm, corresponding to λ = 1.0 nm, and elastic stretch modulus 8.4 nN [10]. A dotted horizontal line at 65 pN denotes the overstretch threshold. A vertical, short-dashed line shows the limit of ss-DNA contour length. 0 L / L 1.0 0.8 0.6 0.4 0.2 0.0 N=1000 a = 0.63 nm =1.00 nm L0= 630 nm L m n ( ) KP WLC FJC 600 400 200 0 0 KP WLC WLC L0=670 50 100 f (pN) 0 20 40 60 80 100 120 force(pN) FIG. 1: The force-extension curve calculated for three dif- ferent models in the case where the persistence length λ = 1.0nm, is comparable i to the monomer distance a = 0.63 nm. The solid line is the KP calculation, according to (11), the dashed line a WLC calculation according to (15) and the dotted line the result of the FJC. The inset displays results for the same KP and WLC calculations on an absolute extension scale, where the contour length is 630 nm for both; the third curve, a dotted line, is the result of a WLC calculation with the same persistence length and an increased contour length, 670 nm, corresponding to a monomer distance a = 0.67 nm, which successfully mimicks the force-extension curve of the shorter KP chain. Kuhn length equal to 2λ = 2.0 nm and a number of segments L0/(2λ). Fig. 1 displays the force-extension relationship in all three cases. The difference between the KP and WLC cases is fairly significant, of the order of 7% at 70 pN. More importantly however, from a data analysis point of view, use of the WLC enforces a bias towards higher contour length, i.e. monomer distances systematically larger than those mandated by crystallographic data (cf. inset). III. OVERSTRETCHING OF DNA A. The force-extension curve Fig. 2 displays the DNA force-extension curve as mea- sured in an unconstrained geometry, i.e. where the chain is held fixed at the 3'-3' opposite ends [3]. The extension is defined relative to the contour length of ds-DNA. At low force levels, the data can be well described by the WLC model (15), corrected for elastic stretch- ing according to (17) with an elastic (stretch) modulus F0 = 1000. The WLC calculation uses a contour length B. Thermodynamics (unconstrained geometry) The plateau in the force-extension curve characterizes the coexistence of the two phases, double-stranded and single-stranded. In this geometry only one of the two single strands of the molten phase is under an applied force. Coexistence occurs when the difference in elastic free energies between single- and double-stranded phases fully compensate the stability free energy of the force-free double helix. 1. Elastic free energies The force-dependent parts of the elastic free energies per unit are and gss(f ) = −kBT ln µ0 + 1 2 a′ F0 f 2 gds(f ) = kBT a λ Λ0 + 1 2 a F0 f 2 (18) (19) for the single- and double-stranded cases, respectively, where the second term expresses the elastic stretch en- ergy. 2. Thermal stability of the double helix The double helix is generally stable at room tempera- ture. This is usually expressed in terms of the free energy difference per base pair between duplex and molten state ∆G = −(∆H − T ∆S) = (T − Tm)∆S , (20) where ∆H > 0 and ∆S > 0 denote, respectively, average values for the energy and the entropy of dissociation per base pair and Tm = ∆H/∆S is the melting temperature. A detailed analysis of DNA melting thermodynamics [18] suggests that the constant value ∆S ≈ 12.5 kB consis- tently describes, in a mean-field fashion, the cooperative melting behavior of a large number of samples of vary- ing sequence and composition.The above value has been used in [1] and will also be used here. The melting temperature Tm can be calculated from the empirical Marmur-Schildkraut-Doty [19 -- 21] equation Tm(C) = 193.67 − (3.09 − x) · (34.47 − 6.52 log10[c]) , (21) where x is the GC fraction and c the salt concentration. At x = 0.5 and c = 150 mM, this leads to Tm = 90.4 C. It should be noted that (20) accounts for all effects involved in the relative stability of ds- and ss-DNA, in- cluding entropic polymer contributions, except those ex- plicitly originating in the applied force and described by (18) and (19). 3. Overstretching threshold Fig. 3 displays the free energy difference gss(f )−gds(f ) at T = 20 C. After an initial positive phase at very low forces (< 10 pN), the difference becomes increasingly negative, until, at 64 pN it compensates fully for the room temperature duplex stability (20). This marks the onset of the overstretching transition. 5 gss-gds S(T-Tm) ) T B k ( y g r e n e e e r f 1 0 -1 -2 -3 -4 -5 -6 -7 0 20 40 60 80 100 120 force(pN) FIG. 3: Free energy balance resulting in DNA overstretching (unconstrained case).The solid line denotes the difference in elastic free energy between single- and double-stranded state. The dashed line denotes the duplex stability free energy at 20 C. Free energies are in units of kBT . The intersection defines the overstretch threshold at 64 pN. C. Thermodynamics (constrained geometry) In the constrained geometry version of the overstretch- ing experiment all four ends of the 3'5'-5'3' chain are attached; each single strand is subjected to half the ex- ternal force. The elastic free energy difference is now 2gss(f /2) − gds(f ). According to the discussion of the previous subsection, overstretching should occur at the force level where the elastic free energy difference fully compensates duplex stability. The geometry of the constraint however prohibits a full entropy release in the overstretched state. The two strands remain stretched, attached at both ends. More- over, available conformational space is further restricted by the entanglement brought about by the previous dou- ble helical order. An "inherited" structure seems to persist in the overstretched state for which "the most straightforward explanation ... is that the new struc- ture, generated during overstretching at 110 pN, consists of two single DNA strands lacking hydrogen bonds be- tween the bases, wrapped around each other with a link- ing number close to that of relaxed dsDNA"[3]. The quantity of interest in this context is the en- tropy difference between the molten state with the ge- ometrical constraint and the one without it, or, in other words, the entropic penalty ∆S∗ < 0 required to maintain the molten DNA with the inherited plectonemic struc- ture, compared to the reference state of the two free strands. Geometrically constrained force-induced melt- ing does not release the full amount of DNA melting en- tropy ∆S = 12.5 kB (cf. previous subsection), but a reduced amount ∆S + ∆S∗. This modifies the relative thermodynamic stability of the double helix (20), which now reads ∆G = (T − Tm)∆S + T ∆S∗ (22) per base pair, and results in an enhanced stability of the double helix. Overstretching will now occur when the elastic free energy difference compensates (22). In what follows I will calculate the entropic penalty associated with the plectonemic structure in two different ways, based on the FJC and the KP model, respectively. 1. FJC estimate of the constraint entropic penalty The overstretched DNA with the plectonemic structure can be visualized as consisting of the two strands joined at successive points with a spacing 10a, equal to the pitch of the original double helix. In the FJC model this cor- responds to ν = 10a′/d segments in each strand, where a′ = 0.63 nm and d = 2λ = 2.0 nm, are, respectively, the monomer distance and the Kuhn length of ss-DNA. The total number of such plectonemic joints will be N/10. Neglecting any effects arising from excluded volume, the probability of two distinct, partially stretched strand pieces, each with ν elements, starting from a common origin and meeting at a certain point ~r in space within an axial distance b and a radial distance ρ, is p2 ν, where pν = πbρ2 Pν(~r) and Pν (~r) = (cid:18) 3 2πνd2(cid:19)3/2 e− 3r2 2νd2 (cid:18)1 − 3 4ν + · · ·(cid:19) (23) (24) is the vector end-to-end distance probability distribution of the FJC chain. The conformation with all N/10 plectonemic joints has ν )N/10, i.e. an entropic a probability of occurrence (p2 penalty ∆S∗ = kB ln pν/5 per original base pair. In- serting some plausible values for the contact in terms of the inherited double helix, e.g. b = ρ = a/2, results in an estimate of ∆S∗ = −2.16 kB. 2. KP estimate of the constraint entropic penalty The calculation is quite analogous with that of the FJC, the only difference now being that ν = 10 and pν = πbρ2 a′3 Pν (cid:16)10 a a′(cid:17) (25) where Pν (x) is now the vector end-to-end distance distri- bution function, measured in units of the segment length, computed for the KP model with the stiffness parameter of ss-DNA (cf. section III A). 6 Computation of the KP end-to-end distribution func- tion can be performed, for relatively small systems, by direct matrix multiplication and Fourier transform in- version [12, 13]. Noting that the Fourier transform of the end-to-end distribution function, PN (q), can be for- mally obtained from ZN (f ) (cf. (3) and (5) above) by the substitution βf → iq, it is possible to write PN (q) = ( UN )00 , where the matrix U is defined as in (7), with F substi- tuted by Fll′ ( f ) = Z 1 −1 dxPl(x)Pl′ (x)eiqax (26) l+l′ = Xk=l−l′,k+l+l′=2r (2k + 1)(−i)k 1 r + 1/2 · Ψ(r − k)Ψ(r − l)Ψ(r − l′) Ψ(r) jk(qa) , where jk are now spherical Bessel functions, e.g. j0 = sin x/x. The result of the numerical computation, for r/a′ = 10a/a′ = 5.4, is P10 = 5.50 × 10−4, leading, for the same values of the parameters b and ρ as in the FJC case, to an entropic penalty of ∆S∗ = −2.06 kB per original base pair. 3. Estimation of the overstretching threshold Fig. 4 displays the difference 2gss(f /2) − gds(f ) in the elastic free energies as a function of the applied force. Also shown is the stability free energy of the double he- lix, with and without the entropic penalty of the con- straint, as given by (22) and (20), respectively. Both effects lead to an enhanced stability of double-stranded DNA, in comparison with the unconstrained geometry. The overstretching force threshold, as estimated from the KP model calculation, with lies at 111 pN, in excellent agreement with experiment [3]. IV. DISCUSSION The present analysis of the overstretching transition involves a number of material parameters for both ds- and ss-DNA. It is important to keep in mind that only two of these parameters originate in fitting the data of the experimental force-extension curve, namely the per- sistence lengths. All others are either standard accepted values (e.g. the crystallographically obtained values for the monomer distances a and a′, melting temperature Tm and melting entropy ∆S per base pair) or, at the very least (e.g. elastic stretch moduli for ss-DNA and ds-DNA), physically plausible published values. Not sur- prisingly, the limited space of the remaining free parame- ters improves the quality of the estimates obtained. Both ) T B k ( y g r e n e e e r f 2gss(f/2)-gds(f) S(T-Tm) S(T-Tm)+S*T 1 0 -1 -2 -3 -4 -5 -6 0 20 40 60 80 100 120 force(pN) FIG. 4: Free energy balance resulting in DNA overstretch- ing (constrained case).The solid line denotes the difference in elastic free energy between single- and double-stranded state. The dotted line denotes the duplex stability free energy at 20 C. The dashed line is the sum of duplex stability and entropic penalty ∆S = −2.06 kB due to the constraint (KP calcula- tion). Free energies are in units of kBT . The intersection of dashed snd full line defines the overstretch threshold at 111 pN. persistence lengths are consistent with the values ob- tained by other researchers. In particular the value λ = 1 nm obtained here for the persistence length of ss-DNA deserves a brief com- ment. The value lies at the low end of current estimates of 0.75 - 2 nm. Specifically, it compares well with the zero loop origami-based estimates of [7], in accordance with the expectation of loops unwinding under the influence of the external force. On the other hand, it is signifi- cantly lower than the SAXS result of [5], 1.7 nm at 150 mM NaCl concentration. Although a small part of the discrepancy may be accounted for by the use of the WLC continuum model in [5], it is instructive to consider the physics underlying the different experiments. The par- tially - or fully - stretched state of [3], is probably far less responsive to the electrostatic repulsion forces act- 7 ing between segments. As a result, observed persistence lengths may be lower than those observed in force-free solutions and actually approach values corresponding to high salt concentrations, where screening of electrostatic repulsion occurs. Independent sets of measurements [4, 5] suggest that there is an intrinsic lower limit of ss-DNA persistence length, of approximately 1 nm and 0.75 nm, respectively, which is approached at high salt concentra- tions. It is quite possible that experiments performed near the stretched state effectively detect this limit of extreme chain flexibility. The results of the thermodynamic analysis confirm the conjecture made [3] about the plectonemic nature of the overstretched state in the constrained geometry. It should be noted that this conclusion does not depend on the detailed choice of geometrical parameters specifying what exactly is considered as a contact. For example, increasing the linear dimension of a contact from a/2 to a in both the radial and axial directions would result in a decreased entropy penalty by (1/5) ln 8 = 0.41kB, changing the predicted threshold to 105 pN. Are there any plausible alternatives to the plectonemic structure of the geometrically constrained molten state? A radically weaker variant would be to consider solely the entropic effect of holding the ends of both strands stretched, without imposing any further structural mo- tifs. Such an alternative calculation results in a much smaller entropic penalty of the order −0.18kB per base pair, roughly ten times less than that of the plectone- mic structure, and cannot account for the observed force threshold. In other words, entropic considerations ef- fectively demand some sort of internal structure in the molten state. The "inherited" structure with knots fol- lowing the pitch of the double helix, as conjectured in [3] is probably the simplest model satisfying this demand. I would like to thank Michel Peyrard for reading an earlier version of the manuscript. I acknowledge support by the project "Advanced Ma- terials and Devices (MIS 5002409)", implemented under the Action for the Strategic Development on the Re- search and Technological Sector, funded by the Opera- tional Program "Competitiveness, Entrepreneurship and Innovation (NSRF 2014-2020)" and cofinanced by Greece and the European Union (European Regional Develop- ment Fund). [1] I. Rouzina and V. A. Bloomfield, Force-induced melt- ing of the DNA double helix 1. Thermodynamic analysis, Biophysical Journal 80, 882 (2001). [2] S. B. Smith, Y. Cui, and C. Bustamante, Overstretch- ing B-DNA: the elastic response of individual double- stranded and single-stranded DNA molecules, Science 271, 795 (1996). [3] J. van Mameren, P. Gross, G. Farge, P. Hooijman, M. Modesti, M. Falkenberg, G. J. L. Wuite, and E. J. G. Peterman, Unraveling the structure of DNA during over- stretching by using multicolor, single-molecule fluores- cence imaging, Proceedings of the National Academy of Sciences 106, 18231 (2009). [4] H. Chen, S. P. Meisburger, S. A. Pabit, J. L. Sutton, W. W. Webb, and L. Pollack, Ionic strength-dependent persistence lengths of single-stranded RNA and DNA, Proceedings of the National Academy of Sciences 109, 799 (2012). [5] A. Y. L. Sim, J. Lipfert, D. Herschlag, and S. Doniach, Salt dependence of the radius of gyration and flexibility of single-stranded DNA in solution probed by small-angle x-ray scattering, Physical Review E 86, 021901 (2012). [6] A. Bosco, J. Camunas-Soler, and F. Ritort, Elastic properties and secondary structure formation of single- stranded DNA at monovalent and divalent salt condi- tions, Nucleic Acids Research 42, 2064 (2014). [7] E. Roth, A. Glick Azaria, O. Girshevitz, A. Bitler, and Y. Garini, Measuring the conformation and persistence length of single-stranded DNA using a DNA origami structure, Nano Letters 18, 6703 (2018). [8] M. C. Murphy, I. Rasnik, W. Cheng, T. M. Lohman, and T. Ha, Probing single-stranded DNA conformational flex- ibility using fluorescence spectroscopy, Biophysical Jour- nal 86, 2530 (2004). [9] O. Kratky and G. Porod, Rontgenuntersuchung geloster Fadenmolekule, Recueil des Travaux Chimiques des Pays-Bas 68, 1106 (1949). [10] T. Hugel, M. Rief, M. Seitz, H. E. Gaub, and R. R. Netz, Highly stretched single polymers: Atomic-force- microscope experiments versus ab-initio theory, Phys. Rev. Lett. 94, 048301 (2005). [11] H. Clausen-Schaumann, M. Rief, C. Tolksdorf, and H. E. Gaub, Mechanical stability of single DNA molecules, Bio- physical Journal 78, 1997 (2000). [12] J. Errami, M. Peyrard, and N. Theodorakopoulos, Mod- 8 eling DNA beacons at the mesoscopic scale, The Euro- pean Physical Journal E 23, 397 (2007). [13] J. Yan, R. Kawamura, and J. F. Marko, Statistics of loop formation along double helix DNAs, Phys. Rev. E 71, 061905 (2005). [14] M. Abramowitz and I. Stegun, Handbook of Mathematical Functions (Dover 1964). [15] C. Bustamante, S. B. Smith, J. Liphardt, and D. Smith, Single-molecule studies of DNA mechanics, Current Opinion in Structural Biology 10, 279285 (2000). [16] J. F. Marko and S. Cocco, The micromechanics of DNA, Physics World 16, 37 (2003). [17] C. Bustamante, J. F. Marko, E. D. Siggia, and S. B. Smith, Entropic elasticity of lambda phage DNA, Science 265, 1599 (1994). [18] S. G. Delcourt and R. D. Blake, Stacking energies in DNA, Journal of Biological Chemistry 266, 15160 (1991). [19] J. Marmur and P. Doty, Determination of the base com- position of deoxyribonucleic acid from its thermal denat- uration temperature, Journal of Molecular Biology 5, 109 (1960). [20] C. Schildkraut, J. Marmur, and P. Doty, Determination of the base composition of deoxyribonucleic acid from its buoyant density in CsCl, Journal of Molecular Biology 4, 430 (1962). [21] R. D. Blake and S. G. Delcourt, Thermal stability of DNA, Nucleic Acids Research 26, 3323 (1998).
1002.1420	1	1002	2010-02-06T23:45:25	Stokesian jellyfish: Viscous locomotion of bilayer vesicles	[ "physics.bio-ph", "physics.flu-dyn" ]	Motivated by recent advances in vesicle engineering, we consider theoretically the locomotion of shape-changing bilayer vesicles at low Reynolds number. By modulating their volume and membrane composition, the vesicles can be made to change shape quasi-statically in thermal equilibrium. When the control parameters are tuned appropriately to yield periodic shape changes which are not time-reversible, the result is a net swimming motion over one cycle of shape deformation. For two classical vesicle models (spontaneous curvature and bilayer coupling), we determine numerically the sequence of vesicle shapes through an enthalpy minimization, as well as the fluid-body interactions by solving a boundary integral formulation of the Stokes equations. For both models, net locomotion can be obtained either by continuously modulating fore-aft asymmetric vesicle shapes, or by crossing a continuous shape-transition region and alternating between fore-aft asymmetric and fore-aft symmetric shapes. The obtained hydrodynamic efficiencies are similar to that of other low Reynolds number biological swimmers, and suggest that shape-changing vesicles might provide an alternative to flagella-based synthetic microswimmers.	physics.bio-ph	physics	Stokesian jellyfish: Viscous locomotion of bilayer vesicles Arthur A. Evans1,∗ Saverio E. Spagnolie2, and Eric Lauga2† 1 Department of Physics, University of California San Diego, 9500 Gilman Drive, La Jolla CA 92093-0354 and 2 Department of Mechanical and Aerospace Engineering, University of California San Diego, 9500 Gilman Drive, La Jolla CA 92093-0411. (Dated: October 24, 2018) Abstract Motivated by recent advances in vesicle engineering, we consider theoretically the locomotion of shape- changing bilayer vesicles at low Reynolds number. By modulating their volume and membrane composition, the vesicles can be made to change shape quasi-statically in thermal equilibrium. When the control param- eters are tuned appropriately to yield periodic shape changes which are not time-reversible, the result is a net swimming motion over one cycle of shape deformation. For two classical vesicle models (spontaneous curvature and bilayer coupling), we determine numerically the sequence of vesicle shapes through an en- thalpy minimization, as well as the fluid-body interactions by solving a boundary integral formulation of the Stokes equations. For both models, net locomotion can be obtained either by continuously modulating fore- aft asymmetric vesicle shapes, or by crossing a continuous shape-transition region and alternating between fore-aft asymmetric and fore-aft symmetric shapes. The obtained hydrodynamic efficiencies are similar to that of other low Reynolds number biological swimmers, and suggest that shape-changing vesicles might provide an alternative to flagella-based synthetic microswimmers. PACS numbers: 47.63.-b, 47.15.G-, 87.16.D-, 87.19.ru 0 1 0 2 b e F 6 ] h p - o i b . s c i s y h p [ 1 v 0 2 4 1 . 2 0 0 1 : v i X r a ∗Electronic address: [email protected] †Electronic address: [email protected] 1 I. INTRODUCTION The preeminence of viscous dissipation over inertial effects at low Reynolds numbers leads to many interesting consequences for life and engineering efforts at the micron-scale. In particular, swimming at zero Reynolds number is impossible using time-reversible motions, a result known as the Scallop theorem [1]. As a result, at least two actuation degrees of freedom are necessary to generate locomotion. The breaking of this time-reversal symmetry has been studied both from a mathematical point of view, and in the context of modeling real organisms [2 -- 6]. Unlike in high Reynolds number flows, such as those relevant in describing the swimming of fish and flying of birds, fluid motion at low Reynolds numbers is set almost instantaneously by the time-dependent geometries of the immersed bodies. Thus it is natural to inquire about the shapes of immersed (and possibly fluctuating) cell membranes, and their relationships to locomotion. Membranes composed of lipid bilayers are ubiquitous in nature, and the study of bilayer vesicles as a model system for biological cells has yielded significant insight into their behavior [7, 8]. In addition to the biological relevance of lipid bilayer vesicles, or liposomes, advances in self- assembly have paved the way for other types of vesicles to be developed experimentally [9, 10]. Vesicles assembled from block copolymers [11], liquid crystal amphiphiles [12], and membranes with embedded proteins or anchored polymers [13 -- 17] all have tunable material properties which can be manipulated with unprecedented control [18, 19]. It is also well known that many biological cells actively modify or maintain the shapes of their membranes [20, 21], either for developmental [22] or locomotive processes [23, 24]. Recently, synthetic microswimmers inspired by the locomotion of eukaryotic cells have been successfully designed in experiments [25], exploiting the planar beating of a flagellum-like organelle. Beyond biomimetic engineering, other small-scale synthetic swimmers or swimming strategies have also been proposed, both theoretically and experimentally [1, 6, 26 -- 32]. One recently-studied example is a self-propelled colloidal particle which exploits asymmetrically-distributed chemical reactions to swim in a viscous fluid [33, 34]. In the same spirit, we consider theoretically in this paper a novel swimming mechanism based on prescribed shape transformations of a bilayer vesicle. By modulating only its volume and membrane composition, the vesicle can be made to change shape quasi-statically in thermal equilibrium. For two different theoretical vesicle models, we determine numerically the vesicle shapes through an enthalpy minimization, and the fluid-body interactions by solving a boundary integral formulation of the Stokes equations. When the control parameters are tuned appropriately to yield periodic but 2 not time-reversible shape changes, we show that net locomotion can be obtained. Swimming arises either by continuously modulating fore-aft asymmetric vesicle shapes, or by crossing a continuous shape-transition region and alternating between fore-aft asymmetric and fore-aft symmetric shapes. In addition, the calculated hydrodynamic efficiencies are shown to be similar to that of other common low Reynolds number propulsive mechanisms. Our paper is organized as follows. We begin with a general discussion of the practical realiza- tion of controlled shape-changing vesicles, in particular the relevant time scales, and the possible actuation mechanisms. Two classical curvature-mediated vesicle models (spontaneous curvature and bilayer coupling) are presented, and the formulations used for the shape calculation and the numerical fluid-interaction model are introduced. We then discuss examples of vesicle shape cycles that yield a swimming motion, examine the fluid flow that develops around the vesicles during their deformation cycles, and compute the corresponding swimming speeds and hydrodynamic efficiencies. II. A ROADMAP TO VESICLE LOCOMOTION A vesicle immersed in a viscous fluid experiences a highly coupled array of forces, such as those generated by membrane tension, internal pressure, membrane (bending) elasticity, and the surrounding viscous fluid dynamics. In the microscopic, viscous environments relevant to our consideration, the Reynolds number, Re, is very small: Re = ρ UcLc/µ ≪ 1, where ρ is the fluid density, µ is the fluid shear viscosity, and Uc and Lc are characteristic velocity and length scales of the vesicle. The fluid behavior at low Reynolds number is highly dependent upon the immersed boundary geometry, and the resultant forces include not only local, but also non-local responses to its motion. A general study of vesicle dynamics should take non-equilibrium shapes into account, as even simple liposomes that can be created in situ can interact relatively quickly with the environment. It is possible to design experiments where carefully constructed initial conditions and lipid species lead to equilibrated vesicle shapes that are non-trivial, but in order to apply morphological changes and induce locomotion, a reversible parameter-changing mechanism is desirable. For our first approach to vesicle swimming, we consider in this paper a "stiff membrane" regime. The characteristic time of membrane relaxation in a viscous fluid is given by trel = µr3 0/κ, where κ is the elastic bending modulus of the membrane, and r0 is a characteristic radius of curvature. If we choose the maximum radius of the vesicle for the characteristic length scale Lc, then r0 . Lc. 3 For parameter variation significantly slower than the membrane relaxation rate, i.e. for a cycle time scale tcycle ≫ trel, then we operate safely within the decoupled regime. In this case, we may thus assume that there are no hydrodynamically induced shape changes, and that the shapes are determined quasi-statically in equilibrium. Using this time scale trel, we can also set a maximum swimming velocity scale, Uc = κ/µr2 0. Similar scaling arguments have been made in Refs. [35, 36]. For biologically relevant systems in water, κ ≈ 100 kBT, µ ≈ 10−3Pa s, ρ = 1 g/cm3, and r0 . 1 − 10 µm, leading to Lc ≈ 1 − 10 µm, Uc ≈ 1 − 10 µm/s, trel ≈ 0.01 − 1s and Re ≈ 10−4. For a vesicle with length scale Lc = 10 µm, diffusive time scales are approximately 104s, and thus negligible for the time being. In addition, we neglect thermal fluctuations in the determination of the vesicle shape, as they come in as a perturbation about the mean equilibrium shape of order (kBT /κ)2, which is very small under most conditions [7]. There are a number of different physical means by which a vesicle shape can be changed in a controlled fashion, and the methods could be different depending on the type of vesicle considered. We will consider two such means, internal volume changes and local membrane compositional changes. One experimentally feasible example of a possible volume-changing mechanism is a light-induced osmotic change. In an ordinary biological membrane the bilayer is embedded with numerous proteins, many of which are sensitive to mechanical forces, chemical gradients, or light. The protein bacteriorhodopsin, for example, is sensitive to green light, and in response to a signal the protein opens and closes like a valve [37]. The presence of such ion channels or active proteins on the surface of a membrane can cause osmotic changes of the fluid volume contained within the vesicle [19]. Recently, vesicle volume control was demonstrated via pH modulation of block copolymer networks along the surface of membrane [38]. The vesicles in this study were well separated from regimes associated with morphological transition, and thus changes in osmotic pressure induced only a volume change, leading to a "breathing" vesicle. Adjusting the membrane composition requires a more indirect experimental approach. Some bilayers are composed of different species of constituent parts, leading to an inherent mismatch between the intrinsic curvatures. In other words, there is an intrinsic curvature that would develop across the bilayer in the absence of other considerations. Because of the inherent difficulty in measuring these quantities it is likely to be more difficult to specify an exact change from one value of intrinsic curvature to another. However, the actual process of changing the intrinsic curvature can be achieved through inducing chemical changes of the lipid constituents of the membrane [38], or by conformational changes of polymers grafted to the surface of the vesicle [39]. 4 FIG. 1: Schematic illustration of a possible control mechanism for vesicle shape-change and swimming: Axisymmetric bilayer vesicle with embedded reactive polymers and polymers grafted to its surface. (a) → (b): Short frequency light impinges on the vesicle, catalyzing a de-polymerization reaction amidst the particle chains, and increasing the fluid volume available to the vesicle. (b) → (c): A second frequency of light induces the grafted polymers to coil up, inducing an entropic repulsion from the membrane and changing the macroscopic morphology. (c) → (d): The dispersed particles begin to polymerize back to their initial configuration, deflating the vesicle. (d) → (a): A third frequency of light is used to uncoil the polymers, relaxing the entropically induced curvature and returning the vesicle to its initial state. By combining two shape-changing mechanisms, it would in theory be possible to achieve a periodic shape cycle which is not time-reversible, yielding a net locomotion. One of many possible configurations that could produce a cycle in shape space is displayed schematically in Fig. 1, where we consider a bilayer vesicle with embedded reactive polymers and with polymers grafted to its surface. In the first step (Fig. 1a→b), a photo-chemical polymerization reaction is catalyzed by green (short wavelength) light, and the polymer chains in the interior of the vesicle disperse into a solution of particles, thus increasing the available volume within the vesicle. At a later time, another frequency of light (red, or longer wavelength) impinges on the vesicle, and the grafted polymers change from a distended to a coiled conformation, inducing an entropic repulsion and changing the curvature of the membrane (Fig. 1b→c). Over time the dispersed particles will polymerize and return the vesicle to its original volume (Fig. 1c→d), and finally a third frequency of light (blue or very short wavelength) can be used to change the conformation of the polymers to distended once more, returning the vesicle to its original state (Fig. 1d→a). While osmotic volume change or chemical-induced composition alteration are two possible ex- 5 perimental methods, not only these examples in no way constitute the full set of possibilities, but also they might be difficult to implement experimentally. Other experimental techniques already exist (see Refs. [19, 38]), or may be developed in the near future that could be more suited for controlled two-parameter change. Rather than suggest specific experimental methodology whose specifics would depend not only on the particular material of the bilayer vesicles, but also on the parameter alternation methods, we adopt in this paper a simplified modeling approach that highlights the qualitative pieces that are required in order to transform a motionless vesicle into a locomotive cargo-carrier. In parallel to the the various practical mechanisms that could be used to implement such shape changes experimentally, it is of fundamental interest to ask theoretically the question of prediction and performance. Would shape change indeed lead to locomotion of the vesicle? How efficient would it be? Can we quantitatively predict the resulting swimming speed and the work done against the fluid to achieve it? This is the approach taken in this paper. Considering two simplified vesicle models, and for slow modulations of the vesicle shapes, we introduce below a computational framework able to quantitatively predict swimming kinematics and performance. III. DYNAMICS OF COUPLED FLUID-BODY SYSTEM A. Vesicle physics While real biological membranes have multiple constituents, all interacting in non-trivial ways, minimal models can still help to illuminate the fundamental physics of such systems. For length scales on which a membrane is approximately flat a Monge parameterization can be employed [40 -- 42], but for a closed bilayer vesicle the curvatures can become very large and the small geometric gradient assumption may break down. In order to characterize the shapes of such objects, an enthalpy must be extremized and the full nonlinear shape equations so generated must be solved. There are many models that could be used to describe the physics of curvature-mediated vesicle morphology. In this paper we will consider two classical models as case studies. These formulations, known respectively as the spontaneous curvature and bilayer coupling models, have both been used in classical work [43] and correspond to different interaction dynamics between the membrane monolayers. Both of these models also include exactly two free parameters, which enable us to explore the breaking of the Scallop theorem, and the generation of locomotion via a change in morphology. 6 The enthalpy functional, F , in the spontaneous curvature model takes the following form [43] F = κ 2 ZS(t) (C1 + C2 − C0)2 dS + Σ A + P V, (1) where C1 and C2 are the principal membrane curvatures, and Σ and P are Lagrange multipliers which constrain the surface area A and volume V (physically they correspond to the membrane tension and pressure difference across the interface). In Eq. (1), S(t) denotes the time-dependent surface boundary, and C0 is the spontaneous curvature, which introduces an inherent mismatch in equilibrium preference of the membrane curvature. This quantity along with a fixed volume and surface area completely specifies the ensemble. Thus the spontaneous curvature model has area, volume, and integrated spontaneous curvature constrained, and we select as the control parameters the volume V and the spontaneous curvature C0 (the fixed surface area merely selects the overall size of the vesicle). In contrast, in the bilayer coupling model, the enthalpy functional G assumes the area difference ∆A between the membrane monolayers to be constant. One possible representation of this area difference is in terms of the integrated mean curvature, M =ZS(t) (C1 + C2) dS. (2) Then the area difference is ∆A = 2hM + O(h2/A), where h is the distance between monolayers [43]. The enthalpy then takes the form G = κ 2 ZS(t) (C1 + C2)2 dS + Σ′ A + P V + Q M, (3) where Σ′, P , and Q are Lagrange multipliers associated with A (area), V (volume) and M (in- tegrated mean curvature) respectively. We select as control parameters the volume V and the integrated mean curvature M . It is important to note that the functionals F and G are related via a Legendre transform, (Σ′, Q) →(cid:0)Σ + κ C 2 ically, the spontaneous curvature model corresponds to a bilayer in which the monolayer admits 0 /2, −2κ C0(cid:1), and thus describe the same system in a different ensemble. Phys- stretching or compression during bending, and thus finds an equilibrium distribution that has a preferred curvature. If the bilayer is composed of more than one species of lipid, each of which has a different preferred curvature (i.e. radius of gyration), it is likely that the membrane will actually prefer to be in a non-flat state. Conversely, the bilayer coupling model corresponds to a system that enforces that both monolayers are incompressible. The area difference between monolayers stays approximately constant on the timescales relevant to our consideration, and as long as the 7 z n t ψ r ψ s (r(s), z(s)) FIG. 2: Parameterization of an axisymmetric bilayer vesicle. We assume axisymmetry about the z-axis. The surface is described by x = (r(s, t), z(s, t)) in cylindrical coordinates, with s an arc-length parameter, t the unit tangent vector, n the outward pointing normal vector, and ψ the angle between the x-axis and t. distance between layers remains very small this implies that the integrated mean curvature also remains constant. B. Determination of the vesicle shape Assuming an axisymmetric vesicle shape, the body surface S(t) is parameterized at each time t as illustrated in Fig. 2. The arc-length measured along the surface in the x-z plane is denoted by s ∈ [0, L], with t the unit tangent vector, n the outward pointing normal vector, and ψ the angle between the x-axis and t. The body surface is represented in cylindrical polar coordinates, x(s, φ, t) = x(s, φ, t) + z0(t)z = (r(s, t) cos(φ), r(s, t) sin(φ), z(s, t) + z0(t)), (4) where φ ∈ [0, 2π) is the azimuthal angle, the surface x is taken to have its center of volume at the origin, and z0(t) is a translation of that center of volume which depends upon the fluid in- teraction. Under this parameterization, the principal membrane curvatures are C1 = ∂ψ/∂s and C2 = sin ψ/r. Upon insertion into either of the enthalpy functionals F or G, and performing a vari- ational extremization, we obtain the following system of first-order ordinary differential equations to describe the energetically stationary vesicle shapes at time t [43] ψs = K, sin ψ + cos ψ sin ψ r2 + 1 2 P r cos ψ, Ks = − K r γs = cos ψ + γ r (K − C0)2 − sin2 ψ 2r2 + P r sin ψ + Σ, 2 rs = cos ψ. (5) (6) (7) (8) Here K is an auxiliary function used to make the system of equations first-order (physically it corresponds to the curvature), γ is the Lagrange multiplier that enforces the interdependence of ψ 8 and r, and the subscript s denotes a derivative with respect to the arclength. The vesicle shape at time t is set by Eqs. (5-8), subject to the four boundary conditions r(0, t) = r(L, t) = ψ(0, t) = 0 and ψ(L, t) = π. Once the angle ψ is determined from the above, z(s, t) is set by an integration of zs = sin(ψ), where the constant of integration is chosen such that the center of volume of the surface x is at the origin. The vertical position z0(t) has no bearing on the vesicle shape determination, and we hold off further discussion on its dynamics until the following section. For the spontaneous curvature model, constraints on the unknown integration length L, the surface area A, the volume V , and the two constant Lagrange multipliers P and Σ are imposed as As = 2πr, Vs = πr2 sin ψ, Ps = 0, Σs = 0, Ls = 0. (9) (10) Defining R0 as the radius of the sphere with surface area A, the boundary conditions for the five constraint equations above are A(0) = V (0) = 0, A(L) = 4πR2 0, V (L) = 4πR3 0v/3, where v is a dimensionless "reduced volume." Due to the Lagrange function being independent of the arc- length s, the "Hamiltonian" is a conserved quantity and we have γ(0) = 0 (see Refs. [43, 44]). Also defining a reduced spontaneous curvature c0 = C0R0, we finally obtain the vesicle morphology as set by the two parameters (v, c0). In the bilayer coupling model, Eqs. (5-10) are solved with two additional constraints. First, the integrated mean curvature M is controlled, Ms = π(rK + sin ψ), and second, a new Lagrangian constraint enters, Qs = 0. The system is now closed with boundary conditions on the integrated mean curvature: M (0) = 0 and M (L) = 4πR0∆a, where ∆a is the reduced surface area difference between monolayers, ∆a = ∆A/8πR0h. In this case the vesicle morphology is set by the two parameters (v, ∆a), and the reduced spontaneous curvature c0 has been removed from the shape equations via the Legendre transform given above. Equations (5-10) are solved numerically. Due to coordinate singularities in the derivatives of r and z at the poles, the shape is determined on the contracted interval s ∈ [L δ, L (1 − δ)] for (L δ) ≪ 1, and Taylor-expanded versions of the boundary conditions are applied. For example, r(L δ, t) = r(0, t) + (L δ) rs(0, t) + O(cid:0)(Lδ)2(cid:1) = (L δ) rs(L δ, t) + O(cid:0)(Lδ)2(cid:1) = (L δ) cos(ψ(L δ)) + O(cid:0)(Lδ)2(cid:1) ≈ L δ. (11) To compute the shapes using either model, the arc-length is discretized using m uniformly spaced grid points, si, with s1 = Lδ and sm = L(1 − δ). A collocation method is then applied in a formulation and implementation similar to that recently used by Jiang et al. [44]. We employ a 9 standard continuation scheme in order to interpolate solutions from one point in the parameter space (v, c0) or (v, ∆a) to neighboring points. By extremizing the enthalpies F or G, the shape equations give only stationary solutions, not necessarily the lowest energy solutions. A numerically determined shape may correspond to an energy saddle point, maximum, or minimum. Although it is possible that the lowest energy state may not be achievable for a non-equilibrium shape change, for our purposes we will examine the minimum energy shapes, and thus a "phase diagram" for the possible shapes is of great use. Just as in a more conventional phase transition, shape transformations correspond to transitions between different symmetry states. Since we consider only axisymmetric shapes here, spherical solutions have the highest symmetry state. For small perturbations around spherical shapes, the solution can be represented as r(s, t) = R0 1 + ∞ Xℓ=0 Bℓ0Y 0 ℓ (θ(s), φ = 0, t)! , (12) where the functions Y 0 ℓ are the spherical harmonics, and the constants Bℓ0 can generate symmetry breaking. Because we consider only axisymmetric vesicles, only the m = 0 spherical harmonics (of the Y m ℓ ) contribute to the sum, and the angle θ is given by tan θ = r/z. While it is not possible to produce an analytical solution using this formulation, it is useful for understanding the morphological transitions in terms of symmetry breaking. For example, breaking ℓ = 2 symmetry (B20 6= 0) leads to a prolate or oblate shape, while breaking ℓ > 2 symmetry can give more complicated shapes, such as the so-called "pear" or "stomatocyte" shapes [7]. In our numerical investigation, symmetry is frequently exploited in order to efficiently compute the equilibrium shape. In regions of multiple stability the solution branches that correspond to lowest energy shapes must be chosen, and by inserting numerically an initial symmetry breaking the algorithm used can more readily converge upon the appropriate solution. C. Fluid-body interaction Modulation of the dimensionless parameter set (v, c0) or (v, ∆a) generates quasi-static defor- mations which in turn lead to motion in the surrounding fluid medium. Given that the Reynolds number is small, the dynamics of the fluid surrounding the vesicle is effectively governed by viscous dissipation and is well modeled by the incompressible Stokes equations, ∇ · σ = 0, ∇ · u = 0, (13) 10 where σ = −pI + 2µE is the Newtonian stress tensor with p the pressure, u the fluid velocity, and E the symmetric rate-of-strain tensor, E = 1 2 (∇u + (∇u)T ). The fluid equations are made dimensionless by scaling velocities upon Uc, lengths upon Lc, and time upon trel = Lc/Uc. Since the surface area A = 4πR2 0 is constant, we define the characteristic length scale by this radius, i.e. Lc = R0. Henceforth, the swimming velocity is understood to be dimensionless, and each shape cycle occurs over a unit in dimensionless time. A no-slip condition is applied on the body surface. For a given path through the parameter space (v, c0) or (v, ∆a), the resulting sequence of instantaneously determined shapes set uniquely the "surface deformation velocity" ud(x, t); namely, ud(x(s, φ, t), t) = ∂x ∂t (s, φ, t). (14) In addition, the surface moves as a rigid body along the z direction due to axisymmetry, with velocity U = Uz = z′ 0(t)z. The no-slip condition is thus written as u(x, t) = Uz + ud(x, t). To close the system of equations describing the fluid-body interaction, we assume that no external forces are acting upon the vesicle, and thus force and torque balance give ZS(t) σ(x) · n(x) dS = 0, ZS(t) x × [σ(x) · n(x)] dS = 0. (15) The computation of the swimming velocity is performed using a standard double-layer boundary integral formulation of the Stokes equations. The details of this formulation and numerical method are presented in the appendix. In addition to computing the swimming velocity, we consider a possibly more important quan- tity, the hydrodynamic efficiency. This swimming efficiency is defined as (see Ref. [45]) , (16) ηH = DZS(t) DU · FE (U + ud) · f dSE = DU · FE ud · f dSE DZS(t) where f = −σ · n is the force density acting on the fluid at the body surface, h·i denotes a time- average over a full shape cycle, and F = 6πµ a Uz is the force required to move a sphere of radius a at a speed U . At each time we use the maximum vesicle radius, a(t) = kr(s, t)k∞. The first term in the denominator of Eq. (16) integrates to zero due to the zero-net force condition (Eq. (15)). The computation of the fluid stress σ is significantly more involved than the computation of the swimming velocity. We employ a numerical method for computing σ based on the evaluation of a hypersingular integral which may be derived from the double-layer formulation of the fluid velocity. The framework and numerical approach are described in the appendix, and a more detailed description of the method and examples of its use will be featured in a subsequent paper. 11 Physically, ηH measures the proportion of work done by the vesicle against the surrounding fluid which is used for swimming purposes, and is typically on the order of 1% for biological cells. Note that the swimming efficiency only measures the hydrodynamic efficiency, not a total efficiency. For example, the bending energy of the vesicle is not captured in this measure. The inclusion of bending costs into swimming efficiency measures has recently been proposed to study optimal locomotion strategies in flagellated cells, but presents an avenue of inquiry beyond the scope of this paper [46]. IV. VESICLE LOCOMOTION BY SHAPE-CHANGE As stated in the introduction, due to the linearity and time-reversibility of Eqs. (13), any time-reversible geometrical surface deformations cannot result in a net locomotion. This result is known as the Scallop theorem, in reference to the sole, time-reversible motions available to a small scallop (opening and closing) [1]. As a consequence of this constraint, a single degree of freedom is insufficient for swimming. Two degrees of freedom are however sufficient to generate a swimming motion, as first described in Ref. [1], and as we shall show presently for the systems of interest. A. Spontaneous curvature model We begin by presenting a characteristic shape cycle that can be generated by adjusting the reduced volume and spontaneous curvature, (v, c0), in a periodic fashion. By selecting a specific elliptical path in the (v, c0) parameter space, namely v(t) = 0.425 + 0.125 cos(2πt), c0(t) = −0.1 + 0.3 sin(2πt), the resulting shape cycle is not time-reversible; hence, the constraints of the Scallop theorem are bypassed, and locomotion may be achieved. For these parameters the vesicle shapes are always stomatocytes, and the neck separating the internal sphere of fluid from the external fluid is very small. Figure 3 shows the corresponding minimal energy vesicle shapes at four times, along with the vorticity generated in the surrounding fluid by the body deformation, ω = ∇ × u. Positive vorticity, corresponding to counter-clockwise rotation, is shown in red, and negative vorticity, corresponding to clockwise rotation, is shown in blue. Hollow arrows indicate the instantaneous swimming velocity of the vesicle, while the plain arrows indicate the direction of time. At zero Reynolds number the swimming velocity, external flow, and swimming efficiency are determined uniquely by the time-dependent surface geometry and surface deformation velocity, so we need not consider the internal flow dynamics (which may in general depend upon the means of modulating 12 t = 0 t = 1/4 t = 3/4 t = 1/2 z x FIG. 3: (color online) Stomatocyte shapes and vorticity profiles produced using the spontaneous curvature model, with v(t) = 0.425 + 0.125 cos(2πt), c0(t) = −0.1 + 0.3 sin(2πt). Positive vorticity, corresponding to counter-clockwise rotation, is shown in red, and negative vorticity, corresponding to clockwise rotation, is shown in blue. Hollow arrows indicate the instantaneous swimming velocity. During one cycle, the vesicle experience net locomotion in the −z direction. the parameters (v, c0)). From t = 0 to t = 1/4 the vesicle volume is decreasing while the spontaneous curvature is increasing. The decrease in volume draws fluid into the stomatocyte cavity, while the surface material near the opening to the cavity moves inward nearly tangentially to the surface itself. While the deformation velocity is normal to the surface near the north and south poles (s = 0 and s = π), the deformations are elsewhere primarily tangential, and vorticity is created as the fluid is sheared accordingly. At t = 1/2 the vesicle volume is minimal, and the fluid volume inside the stomatocyte cavity is beginning to decrease. From t = 1/2 to t = 3/4, the vesicle volume increases while the spontaneous curvature continues to decrease to its minimum value. This can best be understood by observing that when c0 < 0 the membrane prefers a total negative curvature, and as can be seen at t = 3/4, the internal cavity of the vesicle takes its smallest value, maximizing 13 (a) c0 2 1 0 −1 −2 Self-intersecting stomatocyte Stomatocyte-Oblate L Limit shape stomatocyte 0.2 0.3 0.4 0.5 0.6 v 0.7 (b) 0.4 0.3 0.2 0.1 U 0 !0.1 !0.2 !0.3 !0.4 0 !UL" = −0.008 !U " = −0.005 !U " = −0.002 0.25 0.5 t 0.75 1 FIG. 4: (color online) (a) Phase diagram for the spontaneous curvature model in the (v, c0) parameter space. Solid lines are our numerically-calculated lines that denote morphological transitions, while the dashed lines are qualitative, and adapted from Ref. [43]. (b) Three velocity profiles, corresponding to the elliptic paths through parameter space indicated in (a), with the largest velocities achieved along the elliptic path enclosing the greatest area. negative curvature. The increasing volume expels fluid from the cavity, and leads to a reversing of the sign of the vorticity. The overall sequence of asymmetric shapes is not time-reversible, leading to a net swimming velocity taking place in the −z direction. A phase diagram for the minimal energy shapes using the spontaneous curvature model is presented in Fig. 4a. The limit lines correspond to discontinuous morphological transitions, and therefore cannot be crossed in our quasi-static shape-change approach. One critical line corre- sponds to vesicles whose north and south poles self-intersect, and a second line corresponds to stomatocyte shapes that have a vanishing opening between the external fluid and the cavity within (i.e. the shapes are two spheres, one contained entirely within the other). A third line marks the discontinuous phase transition between stomatocyte and oblate shapes. More details may be found in Ref. [43]. Beyond the symmetry constraints imposed by the Scallop theorem, other symmetry breaking is necessary in order for a body to achieve a net motion from a periodic shape cycle. Namely, the body surface must express fore-aft asymmetry in order to swim preferentially in any direction. Hence, parameter paths in the regions of phase space corresponding to prolate or oblate vesicle shapes cannot yield a net motion. However, paths which correspond to stomatocyte or pear shapes are fore/aft asymmetric and can swim. Since the area in phase space that contains pear shapes is very small, we will only examine the swimming stomatocytes. The largest elliptic path shown 14 in Fig. 4a corresponds to the shape cycle shown in Fig 3. The associated time-evolution of the vesicle center of mass velocity is shown in Fig. 4b, along with two other velocities corresponding to elliptic paths enclosing smaller areas in Fig. 4a. We see in Fig. 4 that the larger the area of the cycle in parameter space, the faster the vesicle swims. In fact, the mean velocity roughly scales as the square-root of the area enclosed by the elliptic path of phase space. Drawing on an analogy with thermodynamics, cycles with larger area in the appropriate ensemble space do more work, and thus we might expect that the transduction of shape deformation into mechanical work would exhibit similar behavior. Although our equivalent to an equation of state is too complicated to show a simple relationship between swimming velocity and the area enclosed in this phase space, the basic idea appears to remain valid. We finally note that the net translation during each shape cycle in each case is small compared to the amplitude of the motion, and even smaller when compared to the maximum vesicle radius. The swimming velocities and hydrodynamic efficiencies of shape cycles in the spontaneous curvature model are also small. The maximum velocity achieved for the cycles shown is hU i = −0.008, while we calculate an efficiency of ηH = 0.4%. B. Bilayer coupling model We now consider the bilayer coupling model, for which a schematic phase diagram is shown in Fig. 5a. Although in the spontaneous curvature model there are no continuous transitions between oblate and stomatocyte shapes, the interesting feature of the bilayer coupling model is the presence of a continuous stomatocyte-oblate transition. The upper (solid) line in Fig. 5a denotes a limit line between oblate and prolate shapes, while the lower (dashed) line represents a continuous transition between stomatocyte and oblate shapes. In order to examine how breaking or restoring oblate (ℓ = 2) symmetry relates to swimming, we now consider two shape cycles with equal enclosed area in phase space, as shown in Fig. 5a. The upper cycle crosses the continuous transition line, while the lower cycle remains in the stomatocyte region. The vesicle shapes in the lower cycle of Fig. 5a are displayed in Fig. 6. They correspond to a modulation of the volume and surface area difference between monolayers for the vesicle as v(t) = 0.775 + 0.075 sin(2πt), ∆a(t) = −0.14 cos(2πt) + 0.86. From t = 0 to t = 1/4 the vesicle volume is increasing, expelling fluid from the cavity and pushing fluid away from the surface of the membrane. Due to the larger amount of surface area facing the aft end of the vesicle, the net 15 Oblate-Prolate (a) 1.05 Stomatocyte- Oblate (b) 1 0.5 ∆a U 0 0.7 0.7 t = 0 0.8 v !U " = −0.055 !0.5 !U " = −0.048 !1 0 0.9 0.2 0.4 0.6 0.8 1 t FIG. 5: (color online) (a) Phase diagram for the bilayer coupling model in the (v, ∆a) parameter space, adapted from Ref. [43]. The dashed line indicates a continuous transition, while the solid line indicates a limit shape. Both lines are shown schematically in order to exaggerate the difference between the shape cycles. The two elliptical cycles considered enclose the same area in phase space, but one crosses the transition line. (b) Swimming velocity of the vesicle as a function of time, for the two shape cycles shown in (a). The squares denote the continuously varying velocity of the lower cycle in (a), which is similar to what we observed for the spontaneous curvature model. The circles correspond to the upper cycle in (a) and involves a shape transition, and there is a portion of the cycle during which the vesicle has zero swimming velocity due to fore-aft symmetry. motion during this quarter-cycle is forward. From t = 1/4 to t = 1/2, the "lobes" of the vesicle move downwards, propelling the vesicle upwards, albeit at a decreasing rate. This portion of the motion resembles the characteristic undulatory shape of a jellyfish, albeit one at zero Reynolds number. Between t = 1/2 and t = 3/4, the vesicle deflates and the lobes begin to move upwards again, with the material points of the lobes moving almost completely tangentially to the surface. This creates a vortex dipole at the lobes, leading to the stagnation point that can be seen in the figure. Finally, in the last quarter cycle, the vesicle encloses itself and returns to the starting position. We calculate a mean swimming velocity of hU i = −0.048, and a hydrodynamic efficiency of ηH = 0.6%. The upper elliptical cycle of Fig. 5a, with shapes illustrated in Fig. 7, follows the parameter path v(t) = 0.775 + 0.075 sin(2πt), ∆a(t) = −0.14 cos(2πt) + 0.89, which lies above the continuous stomatocyte-oblate phase transition line from t ≈ 0.45 to t ≈ 0.55. During this portion of the cycle the vesicle has exactly zero swimming velocity due to the fore/aft symmetry of oblate shapes. 16 t = 0 = 0 = 0 = 0 = 0 = 1 = 1 = 1 = 1 t = 1/4 t = 3/4 = 3 = 3 = 3 = 3 t = 1/2 = 1 = 1 = 1 = 1 048 048 048 048 048 z x FIG. 6: (color online) Vesicle shape cycle using the bilayer coupling model, with v(t) = 0.775+0.075 sin(2πt), ∆a(t) = −0.14 cos(2πt) + 0.89, corresponding to the lower cycle of Fig. 5a. This vesicle does not change morphological symmetry states during the swimming cycle and remains within the stomatocyte domain. Hollow arrows denote the instantaneous swimming velocity. Between t = 0 and t = 1/4, the volume and area difference are decreasing, leading the nearly oblate shape into a clearly stomatocyte configuration. For our purposes, we will not address the spontaneous symmetry breaking that is associated with crossing a transition line, but simply assume that once broken, the cycle will break the symmetry in the same way during each cycle. In the example shown, the stomatocyte inflates as it assumes a more oblate shape, expelling fluid from the cavity and producing vorticity along the lobes. As the vesicle continues to deflate from t = 1/4 to t = 1/2, the lobes sweep downwards, moving the stomatocyte upwards as it assumes a perfectly oblate shape. At t ≈ 0.42 the shape transitions into an oblate shape, precluding any net swimming by symmetry. The swimming velocity as a function of time is shown in Fig. 5b. As 17 t = 0 = 0 = 0 = 0 = 0 = 1 = 1 = 1 = 1 t = 1/4 t = 3/4 = 3 = 3 = 3 = 3 t = 1/2 = 1 = 1 = 1 = 1 055 055 055 055 055 z x FIG. 7: (color online) Vesicle shape cycle using the bilayer coupling model across a continuous phase transition, with v(t) = 0.775 − 0.075 sin(2πt), ∆a(t) = 0.14 cos(2πt) + 0.89, and corresponding to the upper cycle of Fig. 5a. This vesicle is oblate for a small part of the cycle, precluding swimming by symmetry, but a net locomotion occurs over the entire cycle. Hollow arrows denote instantaneous swimming velocity. the oblate vesicle deflates, at t ≈ 0.58 the stomatocyte symmetry state is entered once more, the lobes sweep upwards, and the vesicle moves downwards. Despite the presence of a becalmed period during the vesicle does not move, the cycle that involves the shape transition yields a larger mean velocity than the lower cycle, hU i = −0.055, and an increased hydrodynamic efficiency, ηH = 0.7%. As previously noted, crossing the shape transition line between stomatocyte and oblate shapes indicated in Fig. 5a yields a continuous shape change. However, if we exploit the analogy with phase transitions, we note that some quantities must be discontinuous across the transition. Without exploring the details of a dynamic phase transition in the context of vesicle locomotion, although the order parameter is continuous, derivatives of the order parameter need not be so. In other 18 words, the material at a given point s along the boundary experiences a continuous positional change, but a discontinuous velocity relative to the center of mass of the body as the parameters are varied continuously through the transition line. The discontinuous relative material velocity then generates the discontinuous swimming velocity seen in Fig. 5b for the body which exhibits the oblate shapes for part of its periodic cycle. Interestingly, even though the area enclosed in phase space by the two cycles illustrated in Fig. 5a is the same, the relationship between parameter space, efficiency, and swimming velocity is not evident. The upper cycle shown in Fig. 7 has a larger mean swimming speed and is more efficient than the cycle shown Fig. 6, suggesting that the vesicle can increase its efficiency by passing through a phase transition. V. DISCUSSION In this paper, we have shown computationally that it is possible for a bilayer vesicle to swim under a prescribed shape change using two different vesicle models. By modulating the vesicle volume and either its preferred curvature (spontaneous curvature model) or the surface area differ- ence between membrane monolayers (bilayer coupling model), the vesicle can be made to undergo deformations which are not time-reversible, yielding therefore a net swimming motion. Net lo- comotion can be obtained either by continuously modulating fore-aft asymmetric vesicle shapes (stomatocytes), or by crossing a continuous shape-transition region with fore-aft symmetric shapes, and alternating therefore between fore-aft asymmetric and fore-aft symmetric shapes. At first sight, the swimming efficiencies obtained in this paper appear to be low. For the swim- ming stomatocyte shown in Fig. 4, the efficiency is on the order of 0.4%, while for the bilayer coupling model we calculate an efficiency of 0.6% for a non-transitioning vesicle, and 0.7% for a vesicle that undergoes a transition from stomatocyte to oblate. However, it is known from many theoretical studies that the hydrodynamic efficiency of swimming microorganisms, such as flagel- lated bacteria or spermatozoa, is on the order of 1 to 2% (see Ref. [6] and references therein). Our results indicated therefore that the equilibrium morphologies of bilayer vesicles, together with their appropriate modulations as is done in this paper, lead to locomotion means which are almost as efficient as those displayed by biological cells, and might therefore provide an interesting alternative to flagella-based synthetic micro-swimmers. Further optimization of the size and shape of cycle in parameter space will likely lead to swimming vesicle outperforming the efficiency of flagellated cells. In addition, a swimming vesicle has the advantage that the swimmer and the cargo can be 19 one and the same. Let us now discuss the typical time and velocity scales obtained in our simulations. A typical vesicle size is approximately 10 µm, and for liposomes κ ≈ 10−19 Nm. Except for very curved vesicles, the typical radius of curvature r0 is approximately 10 µm as well, leading to a velocity scale of 10 µm/s. This gives calculated mean velocities on the order of 0.1 µm/s for the sponta- neous curvature model, and 0.5 µm/s for the bilayer coupling model. Translational and rotational diffusion constants for vesicles this size at room temperature are D ≈ 10−14 m2/s and Dr ≈ 10−3 s−1, respectively. This implies a time scale for translational diffusion of approximately 104 s, and a time scale for diffusive reorientation of approximately 103 s. Since the actuation proposed in this paper can be implemented faster than both of these time scales, significant diffusion will take place only after many actuation cycles. For time scales much larger than D−1 r , the effective vesicle diffu- sion will then be given by Def f ≈ U 2/Dr [47], which accounts for both swimming and orientation loss. The ratio Def f /D ≈ 103 is large, which implies that locomotion will lead to a substantially enhanced diffusion of the vesicles over long time scales. We have considered only two minimal models for vesicle shape change, and many possible av- enues exist to expand upon this basic model, including a study non-axisymmetric vesicles, more advanced curvature models, and arc length-dependent spontaneous curvature. Since we have as- sumed a quasi-static deformation, non-equilibrium effects would also have to be taken into account for fast deformations, and the shape should be fully determined as a balance between elastic and fluid forces. In addition, swimming is just one example of behavior that could be exhibited by a membrane that is actively deformed. It is perhaps the simplest transduction of geometrical defor- mation into mechanical work, and one that we hope provides further inspiration for the combined study of membrane physics and low Reynolds number fluid mechanics. Acknowledgements This research was funded in part by the NSF (grant CBET-0746285). Appendix: Velocity and Stress Computation The swimming velocity is computed at each time by solving a standard boundary integral formulation of the Stokes equations. As an application of the Lorentz reciprocal identity, the 20 solution to Eqs. (13) may be written as integrations upon the surface velocity and the fluid stress, u(x) = 1 8πµZS(t) G(x, y) · (σ(y) · n(y)) dSy + 1 8π ZS(t) u(y) · T(x, y) · n(y) dSy (17) where Gij(x, y) = δij x − y + (xi − yi)(xj − yj) x − y3 , Tijk(x, y) = −6 (xi − yi)(xj − yj)(xk − yk) x − y5 , (18) (19) are the singular Stokeslet and Stresslet tensors, respectively (see Ref. [48]). By introducing a complementary flow u′ which has the same values of the surface force σ · n as the flow u on the surface S(t), Eq. (17) may be written solely in terms of the second, double-layer integral, u(x) =ZS(t) q(y) · T(x, y) · n(y) dSy, (20) where q(x) is an unknown density of the singular Stresslet tensor. In the limit as the x approaches the body surface S(t), inserting the no-slip condition for the surface velocity there, we find the expression U + ud(x) =ZS(t) (q(y) − q(x)) · T(x, y) · n(y) dSy. The vertical swimming velocity U = U · z is related to the Stresslet density as U = − 4π A ZS(t) z · q(x) dS (21) (22) (recall that A is the vesicle surface area). Equation (21) is a well-posed Fredholm integral equation of the second kind for the unknown density q(x), and has a unique solution. This approach is numerically better conditioned than those based on first-kind equations. The Stresslet integral operator in Eq. (21) has a six-dimensional nullspace corresponding to rigid body motion, and in the presence of external body forces or torques this representation must be closed by a range completion technique (see Ref. [49]). However, in the swimming problem where the deformation velocity ud(x) is specified and there are no body forces or torques, Eqs. (21-22) are closed and prescribe uniquely the swimming velocity U . The integrand in Eq. (21) is discontinuous at the singularity but finite, so that the integrals are computed to second-order in the surface mesh element size using a standard trapezoidal quadrature (setting the quadrature weight to zero at the singularity). The axisymmetry of the problem is inserted into the definition of the body surface as well as the density q(x). The number of gridpoints 21 is chosen to be sufficiently large such that further resolution does not significantly alter the density q(x) or the swimming velocity U . At each time, the curve (r(s, t), z(s, t)) is discretized uniformly in s. Application of a Nystrom collocation method produces a linear system of equations for the density q(x) at the gridpoints, which is then solved iteratively using the method GMRES [50], with an inversion error tolerance such that the only errors are due to discritization. Finally, the body position z0(t) is updated at each time using a second-order Runge-Kutta method. Both convergence tests and comparison with known exact solutions were used to validate the code [51 -- 54]. Computing the hydrodynamic or swimming efficiency (which requires pointwise information about the stress σ) is more difficult. Here we compute σ(x) using the approach outlined below, though a more detailed description of the method and examples of its use will be featured in a subsequent paper. Many common methods for computing the stress are developed using a first-kind boundary integral formulation of the Stokes equations, and hence can suffer from the ill-posedness of the underlying equations [48]. Instead, we solve for the surface stress by evaluating a hypersingular integral which may be derived from the second-kind integral equation for the velocity (see Ref. [48]), 1 µ σim(x) =ZS qj(y)Lijkm(x, y)nk(y) dSy, (23) where Lijkm(x, y) = −4 δimδjk x − y3 − 6 (xk − yk)[δjm(xi − yi) + δij(xm − ym)] x − y5 − 6 (xj − yj)[δmk(xi − yi) + δik(xm − ym)] x − y5 + 60 (xi − yi)(xj − yj)(xk − yk)(xm − ym) x − y7 , (24) and we have set S(t) = S for clarity. The expression L(x, y) is achieved by differentiating the double-layer integral for the fluid velocity and including the pressure term which may also be written as an integration against q(x), with σ = −pI + µ(cid:0)∇u + ∇uT(cid:1) (see Ref. [48]). The stress is determined on the same spatial grid as used to determine the swimming velocity, a uniform discritization in s of the curve (r(s, t), z(s, t)) (with polar angle φ = 0). The integration of Eqn (23) is performed in local polar coordinates, and the singular contributions are handled analytically as follows. The procedure follows the work of Guiggiani et al. [55]. The integration of Eq. (23) is performed on a modified surface S = sǫ + (S − eǫ) and is taken in two parts: the portion of a sphere of radius ǫ centered at the singular point x which is internal to the body surface (sǫ) and intersects the surface S at its boundary, and the body surface punctured 22 by the sphere (S − eǫ). The modified surface limits to the body surface S as ǫ → 0. For a point x ∈ S, Eq. (23) is written as a small ǫ limit, 1 µ σim(x) = lim ǫ→0nZS−eǫ qj(y)Lijkm(x, y)nk(y) dSy +Zsǫ qj(y)Lijkm(x, y)nk(y) dSyo· (25) Under the assumption that q(x) is differentiable, with a derivative which is Holder continuous, we subtract and add the density q(x) and its gradient at the singular point in the second integral of Eq. (25), 1 µ σim(x) = lim qj(y)Lijkm(x, y)nk(y) dSy ǫ→0nZS−eǫ +Zsǫ(cid:16)qj(y) − qj(x) − (xh − yh)qj,h(x)(cid:17)Lijkm(x, y)nk(y) dSy + qj,h(x)Zsǫ + qj(x)Zsǫ Lijkm(x, y)nk(y) dSy,o (xh − yh)Lijkm(x, y)nk(y) dSy (26) (27) (28) (29) where qj,h = ∂qj/∂xh. As shown in Ref. [55], the above integration may be reduced to a final formula upon the introduction of a local polar coordinate system (ρ, η) about the target point x(s, φ), with φ′ = φ + ρ cos(η), s′ = s + ρ sin(η), where η ∈ [0, 2π), ρ ∈ [0, ¯ρ(η)], and dSy = J(s′)ds′ dφ′ = J(s′(ρ, η))ρ dρ dη, (30) (31) with J(s′) = x′ s × xφ the surface Jacobian. ρ = ¯ρ(η) is the equation in the local polar coordinate system of the edge of the semi-periodic domain, (s, φ) ∈ ([0, L] × [0, 2π]). The integration is assisted by the extra factor of ρ in the surface area element, and the final expression for the fluid stress may be reduced to 1 µ σim(x) =Z 2π Z ¯ρ(η) +Z 2π 0 nF (−1) 0 0 nFijk(ρ, η) −h F (−2) ρ2 ijk (η) F (−1) ijk (η) ρ + iodρ dη ijk (η) ln ¯ρ(η) − F (−2) ijk (η)h 1 ¯ρ(η)io dη, ijk (η) and F (−2) (32) (33) where Fijk(x, y) = qi(x)Lijk(x, y)nk(y) [55]. The functions F (−1) ijk (η) are the singular parts of an expansion of Fijk(ρ, η) about ρ = 0. The integrals above all have finite integrands, and are treated using adaptive quadrature methods. 23 Convergence tests and comparisons with known exact solutions were used to validate the code. In particular, we have checked to ensure that the surface deformation relation of Samuel & Stone (1996) is satisfied [52, 54]. With the stress σ in hand, the efficiency ηH (Eq. (16)) is determined to second-order in the grid-spacing by a simple trapezoidal quadrature. The stress need only be computed for φ = 0 due to axisymmetry. As a final note, at zero Reynolds number the swimming velocity and efficiency are entirely determined by the surface deformation velocity. Other more general measures of energetic expen- diture and total efficiency have been considered for other swimming systems (see Ref. [46]), but in this case the total efficiency will depend significantly upon the means used to produce the vesicle shape-change. In addition, should there be a fluid internal to the vesicle, for example, internal dissipation costs would be relevant in a more general measure of energetic expenditure. [1] E. Purcell, Am. J. Phys 45, 11 (1977). [2] C. Brennen and H. Winet, Ann. Rev. Fluid Mech. 9, 339 (1977). [3] S. Childress and R. Dudley, J. Fluid Mech. 498, 257 (2004). [4] E. Lauga, Phys. Fluids 19, 061703 (2007). [5] H. C. Fu, C. W. Wolgemuth, and T. R. Powers, Phys. Rev. E 78, 041913 (2008). [6] E. Lauga and T. Powers, Rep. Prog. Phys. 72, 096601 (2009). [7] U. Seifert, Advances in Physics 46, 13 (1997). [8] J. Kas and E. Sackmann, Biophys J 60, 825 (1991). [9] M. Antonietti and S. Forster, Adv. Mater. 15, 1323 (2003). [10] M. Li and P. Keller, Soft Matter 5, 927 (2009). [11] H. Kukula, H. Schlaad, M. Antonietti, and S. Forster, J. Am. Chem. Soc 124, 1658 (2002). [12] J. Yang, R. Pinol, F. Gubellini, and D. Levy, Langmuir 22, 7907 (2006). [13] R. Lipowsky, Europhys. Lett. 30, 197 (1995). [14] J. Wang, K. Guo, F. Qiu, H. Zhang, and Y. Yang, Phys. Rev. E 71, 041908 (2005). [15] K. Guo, J. Wang, F. Qiu, H. Zhang, and Y. Yang, Soft Matter 5, 1646 (2009). [16] M. Breidenich, R. Netz, and R. Lipowsky, Molecular Physics 103, 3169 (2005). [17] F. E. Antunes, E. F. Marques, M. G. Miguel, and B. Lindman, Advances in Colloid and Interface Science 147-148, 18 (2009). [18] H. Dobereiner, Ph.D. thesis, Simon Fraser University (1995). [19] P. Petrov, J. Lee, and H. Dobereiner, Europhys. Lett. 48, 435 (1999). [20] H. McMahon and J. Gallop, Nature 438, 590 (2005). [21] A. Veksler and N. Gov, Biophys. J 93, 3798 (2007). 24 [22] K. Huang, R. Mukhopadhyay, and N. Wingreen, PLoS Comput Biol 2, e151 (2006). [23] H. Grimm, A. Verkhovsky, and A. Mogilner, Eur Biophys J 32, 563 (2003). [24] D. Bottino, A. Mogilner, T. Roberts, and M. Stewart, J Cell Sci 115, 367 (2002). [25] R. Dreyfus, J. Baudry, M. L. Roper, M. Fermigier, H. A. Stone, and J. Bibette, Nature 437, 862 (2005). [26] L. E. Becker, S. A. Koehler, and H. A. Stone, J. Fluid Mech. 490, 15 (2003). [27] A. Najafi and R. Golestanian, Phys. Rev. E 69, 062901 (2004). [28] I. M. Kulic, R. Thaokar, and H. Schiessel, Europhys. Lett. 72, 527 (2005). [29] A. Leshansky, O. Kenneth, O. Gat, and J. Avron, New J. Phys. 9, 145 (2007). [30] M. Leoni, J. Kotar, B. Bassetti, P. Cicuta, and M. C. Lagomarsino, Soft Matter 5, 472 (2009). [31] A. M. Leshansky and O. Kenneth, Phys. Fluids 20, 063104 (2008). [32] S. E. Spagnolie, Phys. Rev. E 80, 046323 (2009). [33] J. R. Howse, R. A. L. Jones, A. J. Ryan, T. Gough, R. Vafabakhsh, and R. Golestanian, Phys. Rev. Lett. 99, 048102 (2007). [34] R. Golestanian, T. B. Liverpool, and A. Ajdari, New J. Phys. 9 (2007). [35] R. Lipowsky, Statistical Mechanics of Biocomplexity (Springer Berlin/Heidelberg, 1999). [36] F. Brochard and J. Lennon, J. Phys. France 36, 1035 (1975). [37] G. Karp, Cell and Molecular Biology (John Wiley and Sons, Inc., 1999). [38] S. Yu, T. Azzam, I. Rouiller, and A. Eisenberg, J. Am. Chem. Soc. 131, 10557 (2009). [39] V. Nikolov, R. Lipowsky, and R. Dimova, Biophys. J 92, 4356 (2007). [40] S. Ramaswamy, J. Toner, and J. Prost, Phys. Rev. Lett. 84, 3494 (2000). [41] R. Reigada, J. Buceta, and K. Lindenberg, Phys. Rev. E 72, 051921 (2005). [42] F. Campelo and A. Hernandez-Machado, Eur Phys J 143, 101 (2007). [43] U. Seifert, K. Berndl, and R. Lipowsky, Phys. Rev. A 44, 1182 (1991). [44] H. Jiang, G. Huber, R. Pelcovits, and T. Powers, Phys. Rev. E 76 (2007). [45] S. Childress, Mechanics of Swimming and Flying (Cambridge University Press, 1981). [46] S. E. Spagnolie and E. Lauga, Phys. Fluids (to appear) (2010). [47] H. Berg, Random walks in biology (Princeton University Press, 1993). [48] C. Pozrikidis, Boundary Integral and Singularity Methods (Cambridge University Press, 1992). [49] H. P. G. Miranda, Singular Integrals in B.E. Methods, SIAM 47, 689 (1987). [50] Y. Saad and M. H. Schultz, SIAM J. Sci. Stat. Comp. 7, 859 (1986). [51] G. B. Jeffery, Proc. Roy. Soc. Lond. A 102, 161 (1922). [52] J. Happel and H. Brenner, Low Reynolds Number Hydrodynamics (Prentice-Hall, 1965). [53] A. J. Goldman, R. G. Cox, and H. Brenner, Chem. Eng. Sci. 21, 1151 (1966). [54] H. Stone and A. Samuel, Phys. Rev. Lett. 77, 4102 (1996). [55] M. Guiggiani, in: Singular Integrals in B. E. Methods, V. Sladek and J. Sladek (eds.) (1998). 25
1503.02929	1	1503	2015-03-10T14:35:26	Motor-free actin bundle contractility driven by molecular crowding	[ "physics.bio-ph", "cond-mat.soft", "q-bio.SC" ]	Modeling approaches of suspended, rod-like particles and recent experimental data have shown that depletion forces display different signatures depending on the orientation of these particles. It has been shown that axial attraction of two rods yields contractile forces of 0.1pN that are independent of the relative axial shift of the two rods. Here, we measured depletion-caused interactions of actin bundles extending the phase space of single pairs of rods to a multi-particle system. In contrast to a filament pair, we found forces up to 3pN . Upon bundle relaxation forces decayed exponentially with a mean decay time of 3.4s . These different dynamics are explained within the frame of a mathematical model by taking pairwise interactions to a multi-filament scale. The macromolecular content employed for our experiments is well below the crowding of cells. Thus, we propose that arising forces can contribute to biological force generation without the need to convert chemical energy into mechanical work.	physics.bio-ph	physics	Motor-free actin bundle contractility driven by molecular crowding Jörg Schnauss1, Tom Golde1, Carsten Schuldt1, B. U. Sebastian Schmidt1, Martin Glaser1, Dan Strehle1, Claus Heussinger2, and Josef A. Käs1 1 Institute for Experimental Physics I, University of Leipzig, Linnéstrasse 5, 04103 Leipzig, Germany and 2 Institute for Theoretical Physics, Georg-August University of Göttingen, Friedrich-Hund Platz 1, 37077 Göttingen, Germany (Dated: October 9, 2018) Modeling approaches of suspended, rod-like particles and recent experimental data have shown that depletion forces display different signatures depending on the orientation of these particles. It has been shown that axial attraction of two rods yields contractile forces of 0.1 pN that are independent of the relative axial shift of the two rods. Here, we measured depletion-caused interactions of actin bundles extending the phase space of single pairs of rods to a multi-particle system. In contrast to a filament pair, we found forces up to 3 pN. Upon bundle relaxation forces decayed exponentially with a mean decay time of 3.4 s. These different dynamics are explained within the frame of a mathematical model by taking pairwise interactions to a multi-filament scale. The macromolecular content employed for our experiments is well below the crowding of cells. Thus, we propose that arising forces can contribute to biological force generation without the need to convert chemical energy into mechanical work. PACS numbers: 87.15.H-, 87.16.Ln, 87.15.kr, 87.16.Ka Interactions of actin and its molecular motor myosin are known as the fundamental process for biological force generation. These interactions convert chemical energy into mechanical work by ATP hydrolysis [1]. However, we show an alternative mechanism of force generation in the absence of any molecular motors or actin accessory proteins. The system is not driven by ATP hydrolysis and relies solely on minimization of the free energy based on filament - filament interactions. Interactions are induced by a crowded environment in a regime well below the macromolecular content of biological cells [2]. These so-called depletion forces were originally de- scribed by spherical colloidal particles suspended in a polymeric solution [3, 4]. However, this effect inherently appears in crowded solutions independent of the geometry of colloidal particles. Besides lateral particle attraction, the influence of depletion forces on axially shifted rod-like colloids has already been described by theoretical ap- proaches [5 -- 7]. All these approaches describe the relative shift of two rod-like particles due to the induced interac- tion. Arising forces are found to be linear in the axial shift since the energy gain per unit length is constant. Recently, Hilitski et al. experimentally verified these approaches by investigating the overlap of single micro- tubuli filaments [8]. They found a constant force driving these two rods towards a maximized overlap of their ex- cluded volumes. Additionally, force components sum up in a pairwise manner when introducing a third rod to the system [8]. We describe a different, emerging behavior of rod-like colloids in a multi-filament system, in our case actin bundles formed by the depletant methyl cellulose [9 -- 12]. These bundles are formed without influences of additional accessory proteins. We used a mesoscopic approach allowing to deflect bundles from their energetic minimum by pulling forces exerted by optical tweezers. We investigated kinetics and restoring forces arising from the relative, axial sliding of single rod-like filaments within the bundle. Observed responses did not yield a constant force - in contrast to a two filament system - but an exponential force decay. These unexpected, complex dynamics can be explained by a mathematical model when increasing pairwise, linear interactions to a multi-filament scale. Additionally, the model is verified by simulations. These emergent dynam- ics can exert forces corresponding to a regime of weak active behavior of single myosin motors [13, 14]. To probe these contractions, we used two different experimental approaches. A dual-trap configuration was used to maneuver one bead while the other bead was held at a constant position. This pulling process resulted in a stretched bundle exceeding its former contour length. After releasing the deflected bead from the trap, the bundle started to contract (Fig. 1 (a)). This process was recorded as an image series of the fluorescent signals (Fig. 1 (b), Movie S 1). A bead tracking algorithm - computed with adapted MATLAB routines provided by Pelletier et al. [15] - was used to transform these image series to trajectories. Following data evaluations were conducted with self-written MATLAB scripts as described previously [16]. Alternatively, an arrangement of only one bead attached to a bundle was employed. By displacing the bead, the bundle was dragged through the solution. The longitudinal viscous drag elongated the bundle (Fig. 1 (c)). Immediately after the movement was stopped bundles started to contract and fluorescent signals were recorded (Movie S 2). To evaluate these experiments, a kymograph (picture series joined in one image) was used to visualize the bundle length over time (Fig. 1 (d)). Standard edge detections routines of MATLAB were employed to extract the bundle length at given times. arXiv:1503.02929v1 [physics.bio-ph] 10 Mar 2015 2 ical model below. Resulting exponential decay func- tions (bundle length(t) = a · exp(−t/τ) + c) yield a dis- tribution of decay times τ showing the consistency of the effect with a median of 3.4 s (Fig. 2 (a)) inset). De- termined exponential decays were used to calculate the velocity of contractions showing maximal speeds in the range from 0.10 to 0.65 µm s−1 (Fig. S2). Resulting max- imal forces were evaluated by Stokes' law and typically range from 0.5 to 3.0 pN (Fig. S2). In three cases we observed contractions involving ad- ditional accelerating events. These rendered one single exponential decay inappropriate (Fig. 2 B) to describe the whole contraction process. Those contractions, however, can be well described by a series of exponential decay functions. Interestingly, the decay times of these individ- ual exponential decay functions are consistent. In general, we attribute these accelerating events to split bundle structures. A part of the bundle with originally overlap- ping filaments was fully detached during the stretching process. Filaments in the "main bundle" still shared ex- cluded volumes to cause contractions. When releasing the external stress these bundles started to contract lacking the contribution of the non-overlapping filaments. Their formerly attractive potential was not involved in the cu- mulative energy balance of the starting contraction. At a certain point non-overlapping filaments came close enough to share excluded volumes with the already contracting bundle. New overlaps changed the attractive potential and accordingly the energy balance. Thus, a second or third additional internal contraction process set in and the overall contraction was accelerated again (Fig. 2 (b)). To model the results of our experiments we extend the depletion-induced interaction between filaments from in- dividual filament pairs to a multi-filament scale. Within the model a bundle is represented as a two-dimensional arrangement of N rigid rods of length L. The only degree of freedom of the rods is their relative axial shift xi (Fig. 3). The Hamiltonian is given by H = −u (L − xi) − f xi, (1) N−1Xi=1 N−1Xi=1 where the first term represents the depletion-induced attraction between filaments; assumed pair-wise additive and of strength u. The second term is the work done by the external pulling force f. The free energy F and the force-extension relation hR − Li = −∂F/∂f can easily be calculated numerically. In the large-N limit one obtains in linear response hR − Li = f N(cid:10)x2(cid:11) kBT , (2) which is a consequence of the law of large numbers (similar as, for example, in the Gaussian chain model). The value (cid:10)x2(cid:11) represents fluctuations of a single filament pair in the absence of force (f=0). Neglecting end-effects Figure 1. (Color online) (a) Optical tweezers were used to stretch bundles exceeding normal elastic deformations. After releasing one bead from the trap the bundle started to contract (Movie S 1). In a stretched bundle, the overlap of excluded volumes was not maximized anymore. When the pulling force was switched off, filaments tended to maximize this overlap again and contractions appeared (magnification). A bead tracking algorithm was used to transform recorded image series to bead trajectories giving the bundle length between these two beads over time. (b) After the pulling process the right bead was released and the bundle relaxed to a position maximizing the overlap of the excluded volumes again. (c) The bead can be trapped and moved through the viscous solution stretching the bundle due to friction. When the movement is stopped, the bundle started to contract. (d) Displayed are the first frame of the picture series and the kymograph of a contractile actin bundle attached to one bead with detected bundle length over time. These methods are suitable to investigate dynamics of the system. However, multi-filament systems involve a variety of parameters (such as molecular content, bundle thickness, filament length distributions and more) yielding diverse starting conditions for every experiment. In our experiments we tested responses of filament bun- dles under stress and recorded strains exceeding normal elastic deformations (up to 175 % of the initial contour length). Due to actin's rigidity these elongations can neither be attributed to thermal fluctuations of single filaments nor stretching of the filament backbone. Thus, within the pulling process filaments were pulled apart and overlapping excluded volumes of filaments were not maximized anymore. After stress release bundles started to contract. This behavior can be attributed to filaments restoring a maximal overlap of their excluded volumes. Although previous studies revealed a constant force for pairwise overlapping filaments, the decreasing bun- dle length over time in our experiments is well de- scribed by an exponential decay (Fig. 2 (a)). Thus, bundle dynamics correspond to an overdamped relax- ation in a harmonic free energy landscape. These dynamics arise due to the multi-filament nature of probed actin bundles as described in the mathemat- (a) (c) (b) t = 0 s t = 100 s 5 µm (d) t = 0 s t = 0 s 5 µm t = 10.5 s 3 Figure 2. (Color online) (a) The recorded bundle length over time is well described by an exponential decay allowing an evaluation of the contraction velocity (red graph). Decay times (inset) are consistent with a median of 3.4 s. (b) A contracting bundle can exhibit multiple contraction events (due to a split bundle structure) described by a series of exponential decay functions with consistent decay times. The black curve displays the bundle contraction overlaid with according single exponential decay functions. The red graph is the according velocity for the contractions and accelerating events. it is given by (cid:10)x2(cid:11) = 2/(βu)2. As a result we find the force to be proportional to the extension with a spring constant k = u2/NkBT. In Fig. 3 (a) we numerically calculate the free energy of this model as a function of the bundle extension R − L and the number of filaments N that are arranged laterally. With two filaments in the arrangement (one pair), the free energy is a linear function of bundle extension, as expected from the definition of the model. However, this linear relation does not persist for multi-filament arrangements. Already bundles with four filaments display approximately a harmonic free energy that very closely resembles the asymptotic form (N → ∞). Thus, within the analytical model a combination of several linear force pairs in an additive manner yields a relation describing a harmonic potential. The origin of this transition is the addition of more and more internal degrees of freedom (in our case the relative sliding of each individual pair) contributing an entropic term to the free energy. Extended bundles have a much smaller entropy, because the accessible configuration space for Figure 3. (Color online) (a) Schematic of idealized 2d-scenario, where forces are applied at the first (i = 1) and at the last fila- ment (i = N). (b) Free energy FN(R) vs. end-to-end distance R. A two-filament bundle (N = 2) has a linear energy land- scape, but with only a few filaments (N = 4) the asymptotic harmonic form (dashed) is nearly reached. bundle conformations is highly reduced. While for the formulation of the theory several simplify- ing assumptions have been made, we expect this scenario to be generic and also useful to understand the complex experimental bundle contraction. The internal degrees of freedom in this case may also include filament bending fluctuations which are suppressed by extension [17], thus decreasing the entropy. The depletion force, in general, cannot be written as a sum over two-body contributions [18]. We test this assumption of the model by running molecular dynamics simulations where the depletant is modeled explicitly via soft spheres (see supplement). Sim- ulations and theoretical model are in excellent agreement indicating that many-body effects for the depletion inter- action in our case are indeed negligible. Although other studies revealed a constant force for pairwise overlapping filaments [5 -- 7], we are able to show that pairwise linear, additive forces create a harmonic potential. This poten- tial for attractive filament-filament interactions in our system supports the approach to describe actin bundle contractions by an overdamped harmonic motion. Within the frame of this model decreasing bundle lengths over time can be well described by an exponential decay. Our approach explains the spring-like elastic behavior of the bundle under elongation predicting a spring constant k = c · u2/(kBT). The prefactor c ∝ N3d/N can be esti- mated to depend on the number of filament pairs N in the two-dimensional bundle element ( Fig. 3 (a)) and velocity (µm/s) 1 0.8 0.6 0.4 .2 00 velocity (µm/s) velocity (µm/s) 0.6 0.4 .2 00 Median 5 10 Decay time constant (s) 15 20 012345 0 Counts 2 4 6 8 10 Time (s) 20 40 Time (s) 60 (a) 15 14 13 Bundle length (µm) 12 0 23 21 19 0 Bundle length (µm) (b) 25 x4 x3 (>0) x2 x1 R = L + ∑ixi N = 2 N = 3 N = 4 Harmonic potential (N ∞) -0.4 -0.2 0.0 0.2 0.4 Extension (R-L)/L (a) (b) Free energy NFN(R) 0.4 0.3 0.2 0.1 0.0 the number N3d of these elements that are coupled in parallel. The precise value of c depends on the internal bundle structure and may vary with the experimental situation. With c = O(1) we are able to estimate a spring constant by using filament-filament interaction energies of 30 kB T /µm as measured previously [12]. The resulting k=3.6 pN µm−1 is in good agreement with the magnitudes of our data. We were not able to quantitatively compare this theoretical approach to our data any further due to experimental uncertainties. Unfortunately, there is no technique known to us to determine the exact amount of filaments within the bundle in situ and values can be only estimated roughly [10]. Furthermore, packing effects within the bundle cannot be resolved, which would be es- sential to extend our model from a simple two-dimensional arrangement to fully three dimensional structures and to determine N3d. Computer simulations are underway to elucidate the specific role of packing effects. Evaluations of the dynamic behavior of contractions enable us to test influences of the contraction process on the bundle itself. We monitored the bundle thick- ness over time (Fig. 4 (a)) and found that a bundle becomes thicker during the contraction since overlapping filaments are driving the system to a shorter structure. These measurements aimed at comparing different bundle thicknesses to contractile dynamics. Bundle widths were evaluated via a Radon transform along the bundle back- bone [19]. Some of these measurements were discarded since these evaluations showed a dependency on input parameters. Presented data, however, were consistent and evaluations were done with the same parameters. The full width at half maximum of the intensity profiles was chosen to compare thicknesses of different bundles. Due to the direct correlation to interacting filament pairs we expect an influence of the overall bundle thickness on exerted forces and kinetics during contractions of differ- ent bundles. However, we found no apparent correlation within the limits of our measurements (Fig. 4 (a)). This data is in good agreement with consistent decay times for accelerating events, where a bundle thickening has seemingly a minor influence. Possibly, effects due to thicker bundles are superimposed by unavoidable viscos- ity variations for differing experiments. These variations correspond to different macromolecular contents of the depletion agent influencing the contraction process. We observed that a higher macromolecular content inherently slows down contractions due to higher friction forces and shows approximately a linear dependency with respect to the decay times of contraction processes (Fig. 4 (b)). For this investigation the viscosity was measured by optical tweezers as described in [20] with a bead in vicinity to the bundle. Viscosity values were directly translated into macromolecular contents with data sheets provided by Sigma-Aldrich. As a further test of influences of the contraction process to the bundle, we investigated the persistence of a bun- 4 Figure 4. (Color online) (a) The bundle thickness has no detectable scaling behavior within the limits of our measure- ments. (Inset) In the course of a contraction process (red dots) a thickening of the contracting bundle (blue dots) can be observed . (b) With an increasing molecular content of the depletion agent an increase of decay times was observed. Thus, a more viscous medium yields a slower bundle contraction due to higher friction. dle's contraction behavior. In that course we deformed a single bundle multiple times and recorded its contraction behavior. For a better experimental realization bundles attached solely to one bead were probed. Our experi- ments revealed a degenerating effect after consecutive expansions and contractions. As displayed in Fig. 5, later contractions display lower maximal velocities and reach a higher baseline representing an increased relaxed bun- dle length. We attribute this fact to potential filament annealing (two filaments concatenate) yielding a change in the energy balance [21]. For further contractions these merged filaments would have to buckle and thus hinder the overall contraction process. In conclusion, we developed an optical tweezers based technique to investigate the contractile behavior induced by depletion forces [9] of a multi-filament actin bundle. In comparison to previous theoretical as well as experimental studies [5 -- 8] we found a fundamentally different, dynamic behavior. These earlier studies described that a relative, axial sliding of single rod-like filaments induced by deple- 0.45 0.55 0.65 0.75 Bundle thickness (µm) Bundlelength (µm) 14 10 50 150 250 Time (s) 350 0.6 0.5 Bundle thickness (µm) 12345678 (a) Decay time (s) (b) 10 6 2 Decay time (s) 80 100 120 140 160 180 200 220 Macromolecular content (µM) 5 acknowledges the support of the German Science Founda- tion (DFG) via the Emmy Noether fellowship He 6322/1-1 as well as via the collaborative research center SFB 937, project A16. [1] H. F. Lodish, Molecular cell biology, 4th ed. (W.H. Free- man, New York, op. 2000). [2] R. Ellis, Trends in Biochemical Sciences 26, 597 (2001). [3] S. Asakura and F. Oosawa, Journal of Polymer Science 33, 183 (1958). [4] S. Asakura and F. Oosawa, The Journal of Chemical Physics 22, 1255 (1954). [5] M. Kinoshita, Chemical Physics Letters 387, 47 (2004). [6] W. Li and H. R. Ma, The European physical journal. E, Soft matter 16, 225 (2005). [7] J. Galanis, R. Nossal, and D. Harries, Soft matter 6, 1026 (2010). [8] F. Hilitski, A. R. Ward, L. Cajamarca, M. F. Hagan, G. M. Grason, and Z. Dogic, arXiv:1408.5068 (2014). [9] M. Hosek and J. Tang, Physical Review E 69 (2004), 10.1103/PhysRevE.69.051907. [10] D. Strehle, J. Schnauss, C. Heussinger, J. Alvarado, M. Bathe, J. Käs, and B. Gentry, European Biophysics Journal 40, 93 (2011). [11] Lau, A. W. C., A. Prasad, and Z. Dogic, EPL (Euro- physics Letters) 87, 48006 (2009). [12] M. Streichfuss, F. Erbs, K. Uhrig, R. Kurre, A. E.-M. Clemen, Böhm, Christian H J, T. Haraszti, and J. P. Spatz, Nano letters 11, 3676 (2011). [13] K. Carvalho, F.-C. Tsai, E. Lees, R. Voituriez, G. H. Koenderink, and C. Sykes, Proceedings of the National Academy of Sciences 110, 16456 (2013). [14] A. E.-M. Clemen, M. Vilfan, J. Jaud, J. Zhang, M. Bär- mann, and M. Rief, Biophysical Journal 88, 4402 (2005). and M. Kilfoil, Physical Review Letters 102 (2009), 10.1103/Phys- RevLett.102.188303. [15] V. Pelletier, N. Gal, P. Fournier, [16] T. Golde, C. Schuldt, J. Schnauss, D. Strehle, M. Glaser, and J. Käs, Physical Review E 88 (2013), 10.1103/Phys- RevE.88.044601. [17] F. MacKintosh, J. Käs, and P. Janmey, Physical Review Letters 75, 4425 (1995). [18] M. Dijkstra and R. van Roij, Physical Review Letters 89 (2002), 10.1103/PhysRevLett.89.208303. [19] Q. Zhang and I. Couloigner, IEEE Transactions on Image Processing 16, 310 (2007). [20] S. F. Tolić-Nørrelykke, E. Schäffer, J. Howard, F. S. Pavone, F. Jülicher, and H. Flyvbjerg, Review of Scien- tific Instruments 77, 103101 (2006). [21] E. Andrianantoandro, L. Blanchoin, D. Sept, J. McCam- mon, and T. D. Pollard, Journal of Molecular Biology 312, 721 (2001). [22] J. T. Finer, R. M. Simmons, and J. A. Spudich, Nature 368, 113 (1994). [23] S. J. Kron and J. A. Spudich, Proceedings of the National Academy of Sciences of the United States of America 83, 6272 (1986). [24] J. M. Scholey, Motility assays for motor proteins, Methods in cell biology, Vol. v. 39 (Academic Press, San Diego, 1993). Figure 5. (Color online) Contractions show a decaying behavior when one bundle is deformed multiple times consecutively (Movie S 2). Dynamic behavior becomes slower with every expansion and contraction event and the bundle relaxes to other baselines. Numbers refer to the specific contraction process. Image series were evaluated in form of a kymograph. The relative bundle length describes the actual bundle length normalized by the maximal outstretched configuration in the according experiment. tion forces leads to a linear force exertion. Dynamics of contractions would then proceed with a constant velocity, at odds with our findings of an exponentially decreasing velocity. We are able to describe this behavior as an emergent phenomenon of rod-like colloids in an actin bun- dle when taking pairwise interactions to a multi-filament scale. To further understand the results of our experi- ments we model the bundle as a simple two-dimensional arrangement of N-1 laterally stacked pairs of rigid fila- ments. The arising harmonic potential and accordingly the exponential force decay were verified by simulations. To measure absolute force values, different techniques have to be applied like in [8], but these methods hardly allow evaluations of dynamics. Molecular crowding effects represent a fundamental physical interaction, which cannot be switched off even in active systems such as cells. The cytoplasm itself is a dense environment filled with macromolecules [2]. In the experiments presented here, however, the amount of macromolecules is well below the macromolecular content of a cell emphasizing the biological relevance (see supple- ment). Additionally, kinetics and force generation are in a regime of active processes but without the need to convert chemical energy into mechanical work [1, 14, 22 -- 24]. We like to thank Jessica Lorenz helping to set up the SDS-PAGE and David M. Smith as well as Tina Händler for proofreading this manuscript and helpful discussions. This work was supported by the graduate school "Building with Molecules and Nano-Objects" (BuildMoNa -- GSC 185/1) and the DFG Forschergruppe (FOR 877). CH 13 11 2 1 12 7 16 1 2 7 1 1 13 0 4 8 Time (s) 1 0.9 0.8 0.7 0.6 Relative bundle length Motor-free actin bundle contractility driven by molecular crowding Supplementary Material Jörg Schnauss1, Tom Golde1, Carsten Schuldt1, B. U. Sebastian Schmidt1, Martin Glaser1, Dan Strehle1, Josef A. Käs1, and Claus Heussinger2 1 Institute for Experimental Physics I, University of Leipzig , Linnéstrasse 5, 04103 Leipzig , Germany and 2 Institute for Theoretical Physics, Georg-August University of Göttingen, Friedrich-Hund Platz 1, 37077 Göttingen, Germany glucose/glucose oxidase as antiphotobleaching agent, and 1.6 % methyl cellulose in F-buffer conditions. This solution was deposited into a sample chamber as described previously [2]. Fluorescence was induced by a mercury vapor lamp and a N2.1 filter cube transmitting the appropriate green light (Leica 11513882, excitation filter from 515 to 560 nm) on the sample. In this spectrum actin bundles as well as beads were visualized. To manipulate beads and actin structures an epi- fluorescence Leica DM IRB microscope equipped with a 100×oil-immersion objective (Leica 11506168) and a custom-built optical tweezers setup was used (similar to the setup described by Koch et al. [3] equipped with a Manlight ML3-CW-P-OEM-OTS laser). Ob- servations and image acquisitions were realized by a Hamamatsu Orca ER digital CCD camera (Hamamatsu Photonics). All components of the setup were controlled and integrated by a self-written LabVIEW (National Instruments) program. Experiments were observed via image sequences with a typical frame rate of 20 Hz. Figure S1. SDS-PAGE was used to show the absence of any myosin motors or other accessory proteins in solution. Only the actin signal at 42 kDa was observed verifying the purity of the solutuion. I BIOCHEMICAL PREPARATION AND EXPERIMENTAL METHODS G-actin was prepared from rabbit muscle as described previously [1]. This procedure involved size-exclusion chromatography to ensure the purity of the actin and the absence of other proteins. This purity was verified by SDS-PAGE (Fig. S1). We observed the actin signal at 42 kDa but no other signals, which would indicate additional components. To ensure a complete independence of myosins, we verified the persistence of the effect under ADP conditions rendering active processes impossible. The presence of any crosslinking proteins would have inhibited the initial experimental step since sliding motions would have been suppressed. Monomeric actin was polymerized and labelled by adding 1/10 volume fraction of 10 times concen- trated F-buffer (1.0 M KCl, 10 mM MgCl2, 50 mM HEPES, 2.0 mM ATP or ADP, 10 mM DTT) and Phalloidin -- Tetramethylrhodamine B isothiocyanate (Phalloidin-TRITC - Sigma-Aldrich Co.) at a concentra- tion of 5 µM. Biotinylated actin (5 µM - Cytoskeleton Inc.) was added to the solution to decorate filament ends after polymerization (ratio of 3 to 50). Further actin preparation was done as described above and polymerization was induced by ADP-F-Buffer conditions. For experiments under ADP conditions, all buffers were prepared with ADP instead of ATP. To exchange the former G-Buffer (5 mM Tris-HCl, 0.1 mM CaCl2, 0.2 mM ATP, 1.0 mM DTT, 0.01 % NaN3), G-actin was polymerized under ADP-F-Buffer conditions and centrifuged at 100,000 g for 3.5 h at 4°C. The pellet was resuspended in ADP-G-Buffer (5 mM Tris-HCl, 0.1 mM CaCl2, 0.2 mM ADP, 1.0 mM DTT, 0.01 % NaN3) and dialyzed against ADP-G-Buffer overnight. 2 µm fluorescent (Streptavidin streptavidin beads Fluoresbrite® YG Microspheres - Polysciences Inc.) were used to allow binding of the prepared, labelled actin via biotin-streptavidin bonds. Beads were captured by optical tweezers enabling contact-free manipulations and measurements of dynamics within the system. Depletion forces were induced by adding methyl cellulose (400 cP (2 % in aqueous solution) - Sigma-Aldrich Co.) to arrange actin filaments into bundles without the need of accessory proteins [2]. The final solution contained 0.2 µM TRITC-labelled, biotinylated actin, arXiv:1503.02929v1 [physics.bio-ph] 10 Mar 2015 260 kDa 70 kDa 40 kDa 260 kDa 70 kDa 40 kDa Sample solution Reference Reference II FORCES AND VELOCITIES III SIMULATION OF THE CROSS-OVER TO A MULTI-FILAMENT SYSTEM 2 Exponential decay functions describing the bundle length over time during the contraction process can be used to further evaluate these processes. The fitting func- tion can be analytically differentiated yielding the velocity of the contraction. Resulting maximal velocities of the contractions typically ranged from 0.10 to 0.65 µm/s (Fig. S2). If two beads were attached to a bundle, the bead's geometry was used to determine contractile forces. The released bead was dragged through the solution by the con- traction. Forces, necessary to displace a bead in a viscous medium, can be approximated by Stokes' law (F = 6πηrv, with η being the viscosity, r the radius of the bead, and v the velocity). This approach, however, systematically underestimates forces since longitudinal friction at the bundle surface is neglected. For this evaluation the vis- cosity of the surrounding medium was measured with microrheological methods. Measurements yielded a vis- cosity of η = 0.24 ± 0.03 Pa s. Our data have shown maximal forces typically range from 0.5 to 3.0 pN (Fig. S2). A spread of maximal forces is evident since bundles are very heterogeneous structures and uniform starting conditions are not feasible. Bundle thicknesses, for in- stance, naturally vary and the number of filaments within a bundled structure can only be approximated [2]. The amount of free filament ends can vary drastically from bundle to bundle. Additionally, maximal velocities as well as forces depend on the extension of the bundle during the stretching process, which is limited by the strength of the optical tweezers setup. Due to the exponential length decay, generated forces converge to zero when bundles are fully contracted ap- proaching the thermal noise regime. Forces due to thermal noise, however, are not directed. Contractile forces in our experiments are directed and thus bundles still relax even in this low force regime until reaching their energetic minimum. The analytical model describes the free energy of the system as a function of the end-to-end distance R quickly reaching an asymptotic form already for a few filaments. We are able to confirm this behavior by molecular dy- namics simulations, where the depletant is modeled ex- plicitly via soft spheres. Filaments are modeled as rigid rods and excluded volume effects via shifted and trun- cated Lennard-Jones interactions for filament-filament and filament-depletant pairs. The ratio between filament and depletant diameter is chosen to be five. The filament length is 50 in units of the depletant diameter, and the volume fraction of the depletant is taken to be 0.5. In Fig. S3 the probability distribution for the offset be- tween first and last filament in the bundle (no force, f = 0) is shown. For an N = 2 bundle an exponential distribution arises as given by the Boltzmann weight (exp(-βu x)). Upon addition of more filaments to the bundle a cross- over to a Gaussian distribution can be observed. This distribution corresponds to a harmonic energy profile. Distributions of more filaments (e.g. N = 4, 6) can be well described by Gaussian fits, in particular for small offsets. Furthermore, the fit for N = 6 resembles also the behavior for larger offsets as expected from the theory. The relative width σ of the Gaussian fits for the two cases behavior is as expected from the N-dependence of the of N = 4,6 filament bundles yield σ6/σ4 ≈p6/4. This spring constant k ∼ 1/N and σ ∼p1/k. Figure S3. Probability distributions P(R) for the extensions (end-to-end distance) of bundles of N = 2, 4, 6 filaments are displayed. Points are direct results of the simulation and lines are corresponding fits. For a bundle of two filaments the probability distribution follows an exponential distribution as given by the Boltzmann weight. Probability distributions of a bundle with more filaments follow a Gaussian distribution. This transition is illustrated by simulating N = 2, 4, 6 filament bundles. Already bundles formed by six filaments yield a Gaussian probability distribution corresponding to a harmonic potential. Figure S2. Maximal velocities (typically ranging from 0.10 to 0.65 µm/s) and forces (typically ranging from 0.5 to 3.0 pN) of contractile bundles attached to two beads. vmax (µm/s) 0 0.22 0.44 0.66 0.88 0 1 2 3 4 Forcemax (pN) 02468 Counts N = 2 N = 4 N = 6 0.1 0.2 0.3 0.4 Extension (R-L)/L 1 0.1 0.01 PN(R) 0.001 0 IV BIOLOGICAL RELEVANCE Especially for forces in the piconewton range, the passive process we found cannot be ignored. These contributions minimize the free energy of the bundle if filaments interact in an attractive fashion. Our experiments only rely on filament-filament interactions induced by the environment and no other proteins besides actin are involved. In addition, contractions do not require a specific orientation of actin filaments concerning their plus and minus ends. Dynamics of the energetically driven processes, however, are slow in comparison to velocity ranges of some myosins. Myosin II motors, for instance, have been reported to move up to 4.5 µm/s [4 -- 7]. Myosin types moving actively in a lower velocity range have also been described [8]. Besides dynamic differences, myosin activity is switchable while filament interactions cannot be influenced easily [9]. Although dynamics of the energy minimizing processes are slow, they can be employed to contribute to contrac- tions. The passive contractility process we found may act independently of molecular motors but may also act in concert with motor activity. Cellular actin systems, for instance, are prestressed by myosin minifilaments [10]. These motors maintain tensile strains of the bundle and the overall cellular strain can be preserved. However, myosin motors bind transiently and minifilaments can be very small containing only three to four myosins. Thus, there are states when all molecular motors are detached from the bundle. Due to its prestress a bundle would extend its length decreasing its tension. In extreme cases bundles would even fully dissolve. Nevertheless prestressed bundles remain intact since filament-filament interactions antagonize bundle disintegration. Tensile strains of bundles and consequently of whole cells can be sustained. On a cellular scale this mechanism may play a role in the retraction phase of filopodia, which are among the most molecularly crowded structures in cells. Actin filament bundles stretching from the tip to the base of filopodia are highly contractile although they are not directly contracted by the myosin motors located at the basal interface to the interior actin cortex [11]. This is further supported by the fact that bundled actin filaments within filopodia are highly parallel in polarity due to their polymerization from the tip, thereby rendering active myosin contractions implausible. In contrast, the fundamental mechanism underlying contraction presented here is entirely independent of the orientation of interacting filament pairs. Stall forces for retracting filopodia are on the same order of magnitude as the crowding-induced contractile forces measured in our experiments [11]. Their kinetics, however, seem to be slower [12], which might be attributed to crosslinkers such as fascin or fimbrin regulating the depletion force 3 effect. Fascin itself might inhibit these depletion driven contractions due to its properties as a relatively stable crosslinker on the time scale of retraction events [13]. However, fascin's binding affinity can be influenced by phosphorylation, which drastically reduces its actin binding affinity [14, 15]. This mechanism, which itself is regulated by interactions between the surface of the cell and its surrounding environment, might act as a trigger to initiate bundle, and thereby filopodia contraction via crowding-induced forces. The in vivo situation in cellular systems, however, is far more complex with many interacting components and interwoven functions. Examinations of a minimal system of actin filaments in a crowded environment initially allowed us to investigate distinct biophysical properties, which are hardly accessible in in vivo systems. However, extensive in vivo studies are necessary to determine processes and magnitudes of entropic forces induced by a crowded environment in active systems such as cells. In conclusion, these contractions are based on funda- mental physical laws and are very robust. Entropic crowding effects can generate forces in the piconewton range which are relevant in biological systems since they are comparable to myosin activity. This statement is emphasized by the macromolecular content employed for our experiments, which is well below the macromolecular content of cells. Furthermore, crowding effects cannot be switched off and bundle contractions always appear if filaments can slide against each other. Thus, depletion forces and accordingly these contractile actin structures should appear in cellular systems as well. They are independent of any active structural arrangements and parameters such as filament orientation. [1] B. Gentry, D. Smith, and J. Käs, Physical Review E 79 (2009), 10.1103/PhysRevE.79.031916. [2] D. Strehle, J. Schnauss, C. Heussinger, J. Alvarado, M. Bathe, J. Käs, and B. Gentry, European Biophysics Journal 40, 93 (2011). [3] D. Koch, T. Betz, A. Ehrlicher, M. Gogler, B. Stuhrmann, J. Kas, K. Dholakia, and G. C. Spalding, in Optical Science and Technology, the SPIE 49th Annual Meeting, SPIE Proceedings (SPIE, 2004) pp. 428 -- 436. [4] H. F. Lodish, Molecular cell biology, 4th ed. (W.H. Free- man, New York, op. 2000). [5] K. Carvalho, F.-C. Tsai, E. Lees, R. Voituriez, G. H. Koenderink, and C. Sykes, Proceedings of the National Academy of Sciences 110, 16456 (2013). [6] J. T. Finer, R. M. Simmons, and J. A. Spudich, Nature 368, 113 (1994). [7] A. E.-M. Clemen, M. Vilfan, J. Jaud, J. Zhang, M. Bär- mann, and M. Rief, Biophysical Journal 88, 4402 (2005). [8] J. M. Scholey, Motility assays for motor proteins, Methods in cell biology, Vol. v. 39 (Academic Press, San Diego, 1993). [9] M. R. Stachowiak, P. M. McCall, T. Thoresen, H. E. Bal- cioglu, L. Kasiewicz, M. L. Gardel, and B. O'Shaughnessy, Biophysical Journal 103, 1265 (2012). [10] M. L. Gardel, F. Nakamura, J. H. Hartwig, J. C. Crocker, T. P. Stossel, and D. A. Weitz, Proceedings of the Na- tional Academy of Sciences 103, 1762 (2006). [11] S. Romero, A. Quatela, T. Bornschlogl, S. Guadagnini, P. Bassereau, and Tran Van Nhieu, G., Journal of Cell Science 125, 4999 (2012). [12] T. Bornschlögl, S. Romero, C. L. Vestergaard, J.-F. Joanny, Van Nhieu, Guy Tran, and P. Bassereau, Proceed- ings of the National Academy of Sciences of the United States of America 110, 18928 (2013). 4 [13] D. Vignjevic, S.-i. Kojima, Y. Aratyn, O. Danciu, T. Svitk- ina, and G. G. Borisy, The Journal of Cell Biology 174, 863 (2006). [14] Y. Yamakita, S. Ono, F. Matsumura, and S. Yamashiro, The Journal of biological chemistry 271, 12632 (1996). [15] S. Ono, Y. Yamakita, S. Yamashiro, P. T. Matsudaira, J. R. Gnarra, T. Obinata, and F. Matsumura, The Jour- nal of biological chemistry 272, 2527 (1997).
1506.05686	1	1506	2015-06-18T14:14:08	The free energy cost of accurate biochemical oscillations	[ "physics.bio-ph", "q-bio.MN" ]	Oscillation is an important cellular process that regulates timing of different vital life cycles. However, in the noisy cellular environment, oscillations can be highly inaccurate due to phase fluctuations. It remains poorly understood how biochemical circuits suppress phase fluctuations and what is the incurred thermodynamic cost. Here, we study four different types of biochemical oscillations representing three basic oscillation motifs shared by all known oscillatory systems. We find that the phase diffusion constant follows the same inverse dependence on the free energy dissipation per period for all systems studied. This relationship between the phase diffusion and energy dissipation is shown analytically in a model of noisy oscillation. Microscopically, we find that the oscillation is driven by multiple irreversible cycles that hydrolyze the fuel molecules such as ATP; the number of phase coherent periods is proportional to the free energy consumed per period. Experimental evidence in support of this universal relationship and testable predictions are also presented.	physics.bio-ph	physics	The free energy cost of accurate biochemical oscillations Yuansheng Cao1, Hongli Wang1, Qi Ouyang1,2,∗ and Yuhai Tu3,2† 1The State Key Laboratory for Artificial Microstructures and Mesoscopic Physics, School of Physics, Peking University, Beijing, 100871, China 2Center for Quantitative Biology and Peking-Tsinghua Center for Life Sciences, AAIC, Peking University, Beijing, 100871, China and 3IBM T. J. Watson Research Center, Yorktown Heights, New York 10598, USA Oscillation is an important cellular process that regulates timing of different vital life cycles. However, in the noisy cellular environment, oscillations can be highly inaccurate due to phase fluctuations. It remains poorly understood how biochemical circuits suppress phase fluctuations and what is the incurred thermodynamic cost. Here, we study four different types of biochemical oscillations representing three basic oscillation motifs shared by all known oscillatory systems. We find that the phase diffusion constant follows the same inverse dependence on the free energy dissipation per period for all systems studied. This relationship between the phase diffusion and energy dissipation is shown analytically in a model of noisy oscillation. Microscopically, we find that the oscillation is driven by multiple irreversible cycles that hydrolyze the fuel molecules such as ATP; the number of phase coherent periods is proportional to the free energy consumed per period. Experimental evidence in support of this universal relationship and testable predictions are also presented. 5 1 0 2 n u J 8 1 ] h p - o i b . s c i s y h p [ 1 v 6 8 6 5 0 . 6 0 5 1 : v i X r a ∗ [email protected] † [email protected] I. INTRODUCTION 2 Living systems are dissipative, consuming energy to perform key functions for their sur- vival and growth. While it is clear that free energy [1 -- 3] is needed for physical functions, such as cell motility [4] and macromolecule synthesis [5], it remains poorly understood whether and how regulatory functions are enhanced by free energy consumption. The relationship between biological regulatory functions and nonequilibrium thermodynamics has been an active area in biophysics [6 -- 11]. For example, recent studies in different cellular adaptation processes demonstrated that the cost-performance trade-off follows a universal relationship, independent of the detailed biochemical circuits [8, 9]. Oscillatory behaviors exist in many biological systems, e.g., glycolysis [12], cyclic AMP signaling [13], cell cycle [14 -- 16], circadian rhythms [12, 17], and synthetic oscillators [18, 19]. These biochemical oscillations are crucial in controlling the timing of life processes. Much is known now about the structure of biochemical circuits responsible for these oscillatory behaviors. There are a few basic network motifs, illustrated in Figure 1a, which are respon- sible for all known biochemical and genetic oscillations [12, 13, 16 -- 18]. These network motifs share a few essential features, such as nonlinearity, negative feedback, and a time delay, as summarized by Novak and Tyson in [20]. However, in small systems such as a single cell, the dynamics of oscillations are subject to large fluctuations from the environment, due to their small sizes. Thus, one may ask how biological systems maintain coherence of oscillations amidst these fluctuations [21]. Here, we study the thermodynamic cost of controlling oscil- lation coherence in different representative oscillatory systems and investigate whether there is a general (universal) relation between the accuracy of the oscillation and its minimum free energy cost that may apply to all biochemical oscillations. We study four specific models, the activator-inhibitor (AI) model, the repressilator model, the brusselator model, and the allosteric glycolysis model, chosen to exemplify the three different basic oscillation motifs, as shown in Fig. 1. For all the systems studied, a finite (critical) amount of free energy is needed to drive them to oscillate. Beyond the onset of oscillation, extra free energy dissipation is used to reduce the phase diffusion constant and thus enhance the coherence time and phase accuracy of the oscillations. A general inverse relationship between the phase diffusion constant and the free energy dissipation is found in all the four models studied, suggesting that the relation may hold true for all biochemical oscillations. The energy-accuracy relation for noisy oscillations is also verified analytically in the noisy complex Stuart-Landau equation. In the following, we report these results followed by a in-depth discussion of a plausible general microscopic mechanism/strategy for energy-assisted noise suppression. 3 II. MODELS AND RESULTS A. Four biochemical oscillators representing the basic network motifs All known biochemical and genetic oscillators contain at least one of the basic motifs (or their variance) in network topology [20? ]. To search for general principles in these noisy oscillatory systems, we study four biochemical systems (Fig. 1), each representing one of the three basic network motifs responsible for oscillatory behaviors. The first one is the activator-inhibitor (AI) system, where a negative feedback is interlinked with a positive feedback (Left panel, Fig. 1a). This regulatory motif is common in biological oscillators, like the circadian clock in cyanobacteria [22, 23], cell cycle in frog egg [24, 25], cAMP signal- ing in Dictyostelium, and genetic oscillators in synthetic biology [19, 26, 27]. We implement this motif in a simplified biological network with a phosphorylation-dephosphorylation cycle (Fig. 1b). The second model is a repressilator, which consists of three components connected in a negative feedback loop, such that each component represses the next one in the loop, and is itself repressed by the previous one (Middle panel, Fig. 1a). The first synthetic ge- netic oscillator was built with this motif [18]. Many important transcriptional-translational oscillators also use this motif as their backbone, such as circadian clock in mammalian cells [17], NF-κB signaling [28], and the p53-mdm2 oscillations in cancer cells [29]. Here, we take the repressilator composed of CDK1, Plk1, and APC in a cell cycle as our case study (Fig. 1c). The third model we chose is the brusselator, which is one of the simplest two-component systems that can generate sustained oscillations (Right panel, Fig. 1a). The concentration fluctuations due to small molecule numbers were analyzed in [30], here, we aim to study the effect of noise on the phase of the oscillation. The brusselator (Fig. 1d) is a special kind of substrate-depletion model [31], where substrate S is converted by a process that is amplified autocatalytically by the product P . Examples of substrate-depletion motif are oscillations in glycolysis [12, 32] and Calcium signaling [33]. Here, we examine the noise effect in glycolysis 4 oscillations, where the allosteric enzyme PFK catalyzes substrate to product in a network shown in Fig. 1e. In our study, we introduced a parameter γ to characterize the reversibility of the biochem- ical networks. In a reaction loop, γ corresponds to the ratio of the product of the reaction rates in one direction (e.g., counter-clock-wise) and that in the other direction (e.g., clock- wise). When γ = 1, the system is in equilibrium without any free energy dissipation. For γ 6= 1, free energy is dissipated. Here, we study the relationship between the dynamics and the energetics of the biochemical networks by varying γ. The mathematical details of the four models are described in the Supplemental Information (SI), all parameters (e.g., reaction rates, concentrations, time, and volumes) are shown here as dimensionless numbers with their units explained in SI. B. Phase diffusion reduces the coherence time In all four models that we studied, there is an onset of oscillation as γ decreases below a critical value γc (< 1). This means that a finite critical free energy dissipation (Wc > 0) is needed to generate an oscillatory behavior (see Fig. S1 in SI). In Fig. 2a, two trajectories of the concentration of the inhibitor X are shown for γ < γc in the activator-inhibitor model, where γc = 2 × 10−3. As evident in Fig. 2a, biochemical oscillations are noisy. To characterize the coherence of the oscillation in time, we computed the auto-correlation function C(t) for a given concentration variable x in the network. As shown in Fig. 2b, C(t) follows a damped oscillation: C(t) ≡ h(x(t + s) − hxi)(x(s) − hxi)is hx2i − hxi2 = exp(−t/τc) × cos(2πt/T ), (1) where T is the period and τc defines a coherence time for the oscillation. The oscillatory state breaks time translation invariance (symmetry) of the underlying biochemical system. As a result, the phase of the oscillation is a soft mode and follows diffusive dynamics in the presence of noise. To quantify the phase diffusion, we simulated many trajectories in the model(s) with the same parameters and the same initial conditions. In Fig. 2c, the peak times for 500 trajectories in the AI model are shown in a raster plot together with the peak time distributions (red lines). The variance (σ2) of the distribution versus the average peak time is shown in Fig. 2d. It is clear that the variance goes lin- 5 early with time, confirming the diffusive nature of the phase, and the linear slope defines a peak time diffusion constant D. It is easy to show that the coherence time τc is inversely proportional to D: τc = αT 2/D, (2) where α is a constant dependent on the waveform (α = (2π2)−1 for a sine wave). C. Free energy dissipation suppresses phase diffusion As γ decreases below γc, more free energy is dissipated. What is the effect of the additional free energy dissipation beyond the onset of oscillation? From the chemical reaction rates, we can compute the free energy dissipation rate [34]: W = Xi (J + i − J − i ) ln J + i J − i (3) where J + i and J − i are the forward and backward fluxes of the ith reaction, and free energy is in units of kBT , set to unity here. For the activator-inhibitor and glycolysis models, we calculated the energy dissipation rate using Eq. 3. For systems with continuum stochas- tic dynamics described by Langevin equations (e.g., the brusselator and the repressilator models), we can obtain the steady-state distribution P (~x) by solving the corresponding Fokker-Planck equation or by direct stochastic simulations (see Fig. S2 in SI for an ex- ample). From P (~x), we computed the phase space fluxes and the free energy dissipation rate following [9] (see the Methods section and SI for details). For oscillatory systems, the W dt to characterize the free dissipation rate W varies in a period T . We define ∆W ≡ R T 0 energy dissipation per period per volume. For each of the four models, ∆W and the dimensionless peak time diffusion constant D/T were computed for different parameter values (reaction rates, protein concentrations) in the oscillatory regime γ < γc and for different volume V . As shown in Fig. 3, for all the four models considered, D/T decreases as the energy dissipation ∆W increases and eventually saturates to a fixed value when ∆W → ∞ (i.e., γ = 0). The phase diffusion constants scale inversely with the volume V . As shown in the insets of Fig. 3, the scaled D/T (by the volume V ) collapsed onto a simple curve, which can be approximated by the same simple form in all the four models studied: V × D T ≈ C + W0 ∆W − Wc , (4) where Wc is the critical free energy, and W0 and C are intensive constants (independent of volume), whose values in different systems (models) are given in the legend of Fig. 3. 6 D. The free energy sources and experimental evidence What is the free energy source driving the biochemical oscillations? For the activator- inhibitor model, the free energy is provided by ATP hydrolysis in the phosphorylation- dephosphorylation (PdP) cycle (see Fig. 1a). Besides the standard free energy ∆G0 of ATP hydrolysis, the total free energy dissipation per period ∆W also depends on (and thus can be controlled by) the concentrations of ATP, ADP and the inorganic phosphate Pi. These concentrations ([ATP], [ADP], and [Pi]) directly affect the biochemical reaction rates in our model and consequently the phase diffusion of the oscillation. In Fig. 4a, we show the phase diffusion constant (D/T ) versus the dissipation per period (∆W ) for 300 randomly chosen points in the oscillatory regime of the ([AT P ], [ADP ], [Pi]) space (see Fig. 4b). Remarkably, all the points lie above an envelope curve (the dotted line), which follows Eq.3. This envelope curve defines the best performance of the biochemical network, i.e., the minimum free energy ∆Wm needed to achieve a given level of phase coherence. For each choice of the concentrations ([AT P ], [ADP ], [Pi]), a functional efficiency E can be defined as the ratio of ∆Wm and the actual cost ∆W for the same performance (D/T ). The efficiency is represented by color in Fig. 4a&b. We investigated how efficiency depends on the three concentrations. As shown in Fig. 4c, the efficiency E does not simply increase with the ATP concentration; instead it peaks near a particular level of [ATP], at which the phosphorylation and dephosphorylation fluxes are matched. Similarly, E does not have any clear dependence on [ADP ] or [Pi] level, it is high near a fixed ratio of [ADP ]/[Pi], when the kinetic rates of the phosphorylation and dephosphorylation parts of the PdP cycle are matched. These predicted dependence of oscillatory behaviors on [ATP], [ADP], and [Pi] concen- trations, as shown in Fig. 4, may be tested experimentally by measuring peak-to-peak time variations or equivalently the correlation time for different nucleotide concentrations. As re- ported in two recent studies [35, 36], the oscillatory dynamics of the phosophorylated KaiC protein in a reconstituted circadian clock from cyanobacteria (the Kai system) have been measured in media with different ATP/ADP ratios. We analysed the data according to Eq. 7 1 and obtained the correlation time (τc) and the period (T ) for different ATP/ADT ratios (see SI and Fig. S3 in SI for details). In Fig. 5, we plotted the period and the phase diffusion versus ln([AT P ]/[ADP ]), which is the entropic contribution to the free energy dissipation. As the ATP/ADP ratio increases, the period changes little. In contrast, the phase diffusion T /τc ≡ α−1D/T decreases significantly and eventually saturates at high ATP/ADP ratios, consistent with the relationship between energy dissipation and phase diffusion discovered here. E. Analytical results from the noisy Stuart-Landau equation To understand the relationship between phase accuracy and energy dissipation, we con- sider the noisy Stuart-Landau equation for a complex order parameter Z: dZ dt = (a + ib)Z − (c + id)Z2Z + ηZ, (5) where a, b, c(> 0), d are real variables, i = √−1, and ηZ is a complex noise term. For a > 0, the system starts to oscillate with a mean amplitude rs = p a c . Eq. (5) can be decomposed into two Langevin equations for the amplitude r and the phase θ of Z = reiθ: dr dt = ar − cr3 + ηr(t) , dθ dt = b − dr2 + ηθ(t), (6) where ηr and ηθ are the white noises of the amplitude and the phase. For simplicity, we consider the case where ηr and ηθ are uncorrelated with constant strength ∆r and ∆θ re- spectively. The average phase velocity is ω(r) ≡ hdθ/dti = b − dr2. It is clear from Eq. 6 that detailed balance is broken and the system is dissipative. To compute the free energy dissipation, we first determine the phase-space probability distri- bution function P (r, θ, t), which follows the Fokker-Planck equation: ∂P ∂t 1 r = − ∂ ∂r(cid:2)(ar2 − cr4)P − ∆rr 2 ∂P ∂r (cid:3)− ∂ ∂θ(cid:2)(b − dr2)P − ∆θ 2 ∂P ∂θ (cid:3) ≡ − 1 r ∂(rJr) ∂r − ∂Jθ ∂θ , (7) where Jr and Jθ are the probability density fluxes in phase space. Since ω(r) does not depend on θ, the steady state probability distribution Ps(r, θ) only depends on r: Ps(r, θ) = P (r) = A exp [− where A = (cid:2)2πR exp [−2(cr4/4 − ar2/2)/∆r]rdr(cid:3)−1 Eq. 8, the flux vanishes in the r-direction Jr = 0. However, there is a finite flux in the 2(cr4/4 − ar2/2) ∆r ] (8) is the normalization constant. From 8 θ-direction Jθ(r) = ω(r)P (r). We compute the system's entropy production rate S [37, 38], from which we obtain the minimum free energy dissipation (see SI for details): W = kBTeZ Z [ J 2 r ∆rP + J 2 θ ∆θP ]rdrdθ = kBTehω2i ∆θ , (9) where Te is an (effective) temperature of the environment, we set kBTe = 1 here. The phase diffusion constant is determined by expanding the phase velocity ω(r) around r = rs, the most probable amplitude from P (r). This leads to dθ/dt = ω(rs) + βδr(t) + ηθ(t), with β ≡ ∂ω(rs)/∂r = −2dpa/c. The period of the oscillation is T = 2π/ω(rs), and the phase fluctuation δθ ≡ θ − ω(rs)t follows diffusion with the diffusion constant given by: Dθ = β2∆r 4a2 + ∆θ. From Eq. 9&10, the relation between phase diffusion and energy dissipation emerges: Dθ = D0 + hω2iT ∆W ≡ C + W0 ∆W − Wc , (10) (11) where Wc = 0 because Jr = 0, something that is not generally valid (see SI and Fig. S4 in SI for a more general case of the noisy Stuart-Landau equation). The two constants, C = D0 = β2∆r 4a2 and W0 = hω2iT , depend on the details of the system. Eq.11 has the same form as Eq. 4 obtained empirically from studying different biochem- ical networks. Analysis of the noisy Stuart-Landau equation clearly shows that free energy dissipation is used to suppress phase diffusion to increase the coherence of the oscillation. Though parameters in this relation may depend on the details of the system, the inverse dependence of phase diffusion on energy dissipation appears to be universal. III. DISCUSSION Oscillations are critical for many biological functions that require accurate time control, such as circadian clock, cell cycle, and development. However, biological systems are inher- ently noisy. The phase of a noisy oscillator fluctuates (diffuses) without bound and eventually destroys the coherence (accuracy) of the oscillation. Specifically, the number of periods Nc in which the oscillation maintains its phase coherence is given by Nc = τc/T = αT /D, which decreases with the phase diffusion constant. Here, our study shows that free energy dissipation can be used to reduce phase diffusion and thus prolong the coherence of the 9 oscillation. A general relationship between the phase diffusion constant and the minimum free energy cost, as given in Eq. 4, holds true for all the oscillatory systems we studied here. The amplitude fluctuations also decrease with free energy dissipation (see Fig. S5 in SI for details), as fluctuations in phase and amplitude are coupled in realistic systems. Our study thus establishes a cost-performance tradeoff for noisy biochemical oscillations. How do biological systems use their free energy sources (e.g., ATP) to enhance the accu- racy of the biochemical oscillations? As illustrated in Fig. 5a, a biochemical oscillation can be considered as a clock, which goes through a series of time-ordered chemical states (green dots) during each period. These chemical states are characterized by the conformational and chemical modification (e.g., phosphorylation) states of the key proteins or protein complexes in the system. The forward transition from one state to the next is coupled to a PdP cycle (blue arrowed circle) driven by hydrolysis of one ATP molecule. For each forward step, the reverse transition introduces a large error in the clock. The system suppresses these backward transitions by utilizing the ATP hydrolysis free energy. However, this is just one half of the story. Even in the absence of the reverse transition, the time duration between two consecutive states is highly variable due to the stochastic nature (Poisson process) of the chemical transitions. A general strategy of increasing accuracy is averaging [? ]. In the case of biochemical oscillations, each period may consist of multiple steps, each powered by at least one ATP molecule. As a result of averaging, the error in the period should go down as the number of steps increases. Specifically, we expect that the variance of the period σ2 T (= D/T ) should be inversely proportional to the total number of ATP hydrolyzed NAT P ∝ T /τcyc in each period T , where τcyc is the average PdP cycle time, which is essen- tially the ATP turnover time. Consequently, the number of coherent period Nc = αT /D should be proportional to the number of ATP hydrolyzed in each period. We checked this prediction by varying the kinetic rates in the PdP cycle to change τcyc (see Methods section for details). In Fig. 5b, it is shown that the accuracy of the oscillation (clock), as measured by Nc, is enhanced by the number of ATP molecules hydrolyzed in each period. This re- sult reveals a general strategy for oscillatory biochemical networks to enhance their phase coherence by coupling to multiple energy consuming cycles in each period. Interestingly, ap- proximately 15 ATP molecules are consumed per KaiC molecule per period in the circadian clock of cyanobacteria [39]. Biological systems need to function robustly against variations in its underlying biochem- 10 ical parameters (rates, concentrations) [40, 41]. For oscillatory networks, the free energy dissipation needs to reach a critical value (Wc) to drive the system to oscillate. We showed here that additional free energy cost in excess of Wc is needed to make the oscillation more accurate, as demonstrated explicitly in Eq. 4. In addition to this accuracy-energy tradeoff, we found that larger energy dissipation can also enhance the system's robustness against its parameter variations. Take the activator-inhibitor model, for example: the concentrations of enzyme E (ET ) and phosphatase K (EK) may vary from cell to cell. We search for the existence of oscillation in the (ET , KT ) space for different values of γ. Robustness is defined as the area of the parameter space where oscillation exists. As shown in Fig. S6 in the SI, the robustness increases as the system becomes more irreversible, i.e., when more free energy dissipation is dissipated. This suggests a possible general tradeoff between the functional robustness and energy dissipation in biological networks. IV. ACKNOWLEDGEMENT We thank Dr. Michael Rust for sharing the experimental data in Ref. [35]&[36] with us. This work is partly supported by a NIH grant (R01GM081747 to YT). V. METHODS Simulation Methods. The Gillespie algorithm [42] is used for the stochastic simulations of the reaction kinetics. For given kinetic rates and the volume V , we simulated 1000 trajectories starting with the same initial condition. For the jth trajectory, we obtained its ith peak time tij from the trajectory xj(t) after smoothing (smooth function in MATLAB was used). The peak positions for two trajectories are shown in Fig. 2a. For all the σ2 trajectories, we computed the mean of their ith peak time mi = Pj tij/N, and variance i = Pj(tij −mi)2/(N −1), where N is the total number of trajectories. The average period T is given by T = mi/i. Asymptotically, σ2 i depends linearly on mi (Fig. 2d), and the slope of this linear dependence is the peak time diffusion constant D, which has the dimension of time. The phase diffusion constant Dφ is linearly proportional to D: Dφ = (2π)2D/T . For the repressilator and the brusselator models, we simulated the stochastic kinetic equations to a sufficiently long time (10000 periods) to obtain the time-averaged distribution P (~x), where ~x represents the phase space. We used P (~x) to compute free energy dissipation. 11 Random Sampling in the ([AT P ], [ADP ], [Pi]) space is performed (in log scale) in the [Pi]0 ∈ [−3, 1] by using Latin hyper- cube sampling (the lhsdesign function in MATLAB). The reference concentrations [AT P ]0, [ADP ]0 ∈ [−3, 1], log10 [AT P ]0 ∈ [2, 5], log10 region log10 [ADP ] [AT P ] [Pi] [ADP ]0, and [Pi]0 are set to unity and their actual values are absorbed into the baseline reaction rates a1,0, f−1,0 and f−2,0, which are given in the legend of Fig. 4. RAT P = V (J + ATP consumption. In the activator-inhibitor model, the ATP consumption rate is p are the fluxes for the E → Ep and Ep → E reactions, respectively. We varied the overall reaction kinetics, e.g., τcyc and the ATP consumption rate, p ), where J + p −J − p and J − by introducing a timescale factor B for all four rates d1 = d2 = Bd, f1 = f2 = Bf , where d = 15, f = 15 are the original values used in this paper (see SI). By changing the rates this way, the free energy release of ATP hydrolysis ∆G = − ln γ = ln(a1f1a2f2/(d1f−1d2f−2)) is unchanged. We varied B ∈ [0.2, 2], and computed the total number of ATP consumed per period NAT P ≡ R T 0 RAT P dt and Nc for Fig. 5b. VI. ACKNOWLEDGEMENT We thank Dr. Michael Rust for sharing the experimental data in Ref. [35]&[36] with us. This work is partly supported by a NIH grant (R01GM081747 to YT). [1] Eisenberg, E. & Hill, T. L. Muscle contraction and free energy transduction in biological systems. Science 227, 999 -- 1006 (1985). [2] Hill, T. L. Free energy transduction and biochemical cycle kinetics (Academic Press, New York, 1977). [3] Qian, H. & Beard, D. A. Chemical biophysics :quantitative analysis of cellular systems. Cam- bridge texts in biomedical engineering (Cambridge University Press, Cambridge, 2008). [4] Julicher, F., Ajdari, A. & Prost, J. Modeling molecular motors. Rev. Mod. Phys. 69, 1269 (1997). [5] Nelson, D. L., Lehninger, A. L. & Cox, M. M. Lehninger principles of biochemistry (Macmillan Publisher, New York, 2008). 12 [6] Bialek, W. & Setayeshgar, S. Physical limits to biochemical signaling. Proc Natl Acad Sci U S A 102, 10040 -- 5 (2005). [7] Hu, B., Chen, W., Rappel, W. J. & Levine, H. Physical limits on cellular sensing of spatial gradients. Phys Rev Lett 105, 048104 (2010). [8] Lan, G., Sartori, P., Neumann, S., Sourjik, V. & Tu, Y. The energy-speed-accuracy tradeoff in sensory adaptation. Nat Phys 8, 422 -- 428 (2012). [9] Lan, G. & Tu, Y. The cost of sensitive response and accurate adaptation in networks with an incoherent type-1 feed-forward loop. J R Soc Interface 10, 20130489 (2013). [10] Skoge, M., Naqvi, S., Meir, Y. & Wingreen, N. S. Chemical sensing by nonequilibrium cooperative receptors. Phys Rev Lett 110, 248102 (2013). [11] Lang, A. H., Fisher, C. K., Mora, T. & Mehta, P. Thermodynamics of statistical inference by cells. Phys Rev Lett 113, 148103 (2014). [12] Goldbeter, A. Biochemical oscillations and cellular rhythms: the molecular bases of periodic and chaotic behaviour (Cambridge University Press, Cambridge, 1996). [13] Martiel, J. L. & Goldbeter, A. A model based on receptor desensitization for cyclic amp signaling in dictyostelium cells. Biophys J 52, 807 -- 28 (1987). [14] Pomerening, J. R., Sontag, E. D. & Ferrell, J. E. Building a cell cycle oscillator: hysteresis and bistability in the activation of cdc2. Nature Cell Biology 5, 346 -- 351 (2003). [15] Tsai, T. Y.-C. et al. Robust, tunable biological oscillations from interlinked positive and negative feedback loops. Science 321, 126 -- 129 (2008). [16] Ferrell, J. J., Tsai, T. Y. & Yang, Q. Modeling the cell cycle: why do certain circuits oscillate? Cell 144, 874 -- 85 (2011). [17] Hogenesch, J. B. & Ueda, H. R. Understanding systems-level properties: timely stories from the study of clocks. Nat Rev Genet 12, 407 -- 16 (2011). [18] Elowitz, M. B. & Leibler, S. A synthetic oscillatory network of transcriptional regulators. Nature 403, 335 -- 8 (2000). [19] Stricker, J. et al. A fast, robust and tunable synthetic gene oscillator. Nature 456, 516 -- 519 (2008). [20] Novak, B. & Tyson, J. J. Design principles of biochemical oscillators. Nat Rev Mol Cell Biol 9, 981 -- 91 (2008). [21] Barkai, N. & Leibler, S. Circadian clocks limited by noise. Nature 403, 267 -- 8 (2000). 13 [22] Nakajima, M. et al. Reconstitution of circadian oscillation of cyanobacterial kaic phosphory- lation in vitro. Science 308, 414 -- 415 (2005). [23] Rust, M. J., Markson, J. S., Lane, W. S., Fisher, D. S. & O'Shea, E. K. Ordered phosphory- lation governs oscillation of a three-protein circadian clock. Science 318, 809 -- 12 (2007). [24] Goldbeter, A. A minimal cascade model for the mitotic oscillator involving cyclin and cdc2 kinase. Proc Natl Acad Sci U S A 88, 9107 -- 11 (1991). [25] Pomerening, J. R., Kim, S. Y. & Ferrell Jr, J. E. Systems-level dissection of the cell-cycle os- cillator: bypassing positive feedback produces damped oscillations. Cell 122, 565 -- 578 (2005). [26] Danino, T., Mondragon-Palomino, O., Tsimring, L. & Hasty, J. A synchronized quorum of genetic clocks. Nature 463, 326 -- 30 (2010). [27] Prindle, A. et al. A sensing array of radically coupled genetic biopixels. Nature 481, 39 -- 44 (2012). [28] Krishna, S., Jensen, M. H. & Sneppen, K. Minimal model of spiky oscillations in nf-kappab signaling. Proc Natl Acad Sci U S A 103, 10840 -- 5 (2006). [29] Geva-Zatorsky, N. et al. Oscillations and variability in the p53 system. Mol Syst Biol 2, 2006.0033 (2006). [30] Qian, H., Saffarian, S. & Elson, E. L. Concentration fluctuations in a mesoscopic oscillating chemical reaction system. Proc Natl Acad Sci U S A 99, 10376 -- 81 (2002). [31] Szallasi, Z., Stelling, J. & Periwal, V. System modeling in cell biology: from concepts to nuts and bolts (MIT Press, Cambridge, Mass., 2006). [32] Goldbeter, A. & Lefever, R. Dissipative structures for an allosteric model. application to glycolytic oscillations. Biophys J 12, 1302 -- 15 (1972). [33] Dupont, G., Berridge, M. & Goldbeter, A. Signal-induced Ca2+ oscillations: Properties of a model based on Ca2+-induced Ca2+ release. Cell calcium 12, 73 -- 85 (1991). [34] Qian, H. Phosphorylation energy hypothesis: open chemical systems and their biological functions. Annu Rev Phys Chem 58, 113 -- 42 (2007). [35] Rust, M. J., Golden, S. S. & O'Shea, E. K. Light-driven changes in energy metabolism directly entrain the cyanobacterial circadian oscillator. Science 331, 220 -- 3 (2011). [36] Phong, C., Markson, J. S., Wilhoite, C. M. & Rust, M. J. Robust and tunable circadian rhythms from differentially sensitive catalytic domains. Proc Natl Acad Sci U S A 110, 1124 -- 1129 (2013). 14 [37] Tome, T. & de Oliveira, M. J. Entropy production in irreversible systems described by a fokker-planck equation. Phys Rev E 82, 021120 (2010). [38] Seifert, U. Entropy production along a stochastic trajectory and an integral fluctuation the- orem. Phys Rev Lett 95, 040602 (2005). [39] Terauchi, K. et al. Atpase activity of kaic determines the basic timing for circadian clock of cyanobacteria. Proc Natl Acad Sci U S A 104, 16377 -- 16381 (2007). [40] Barkai, N. & Leibler, S. Robustness in simple biochemical networks. Nature 387, 913 -- 7 (1997). [41] Ma, W., Trusina, A., El-Samad, H., Lim, W. A. & Tang, C. Defining network topologies that can achieve biochemical adaptation. Cell 138, 760 -- 73 (2009). [42] Gillespie, D. T. Exact stochastic simulation of coupled chemical reactions. J Chem Phys 81, 2340 -- 2361 (1977). 15 f2 EpK a2 c APC d f-2 d1 a1 1/γ ER R Ep f1 d2 f-1 X CDK1 XErp1 A B X Y 2 X + Y 3 X Plk1 X b E a C R X S P A B e S vi R0j R02 a2P d2 R01 a2P d2 R00 T0 T1 T2 a1S d1 R1j R12 R11 R10 a1S d1 R2j R22 R21 R20 k k'P k k'P R0j+P ksP P R1j+P 16 FIG. 1 (preceding page). Different network motifs and the corresponding biochemical oscilla- tory systems. (a) Illustrations of three network motifs for oscillation: activator-inhibitor, re- pressilator, and substrate-depletion. (b) The activator-inhibitor model with a phosphorylation- dephosphorylation (PdP) cycle. R and K catalyse two opposing reactions E ↔ Ep (phosphoryla- tion and dephosphorylation) through different intermediate complexes ER and EpK. Ep activates both R (activator) and X (inhibitor). X inhibits R by enhancing its degradation. Parameter γ = d1f−1d2f−2/(a1f1a2f2) is introduced to characterize the reversibility of the system. (c) The "repressilator" model of cell cycle in eukaryotic cells. In the simplified network, CDK1 activates Plk1, Plk1 activates APC, and APC degrades CDK1 (dashed line), forming the mutually ac- tiving/inhibiting loop. Other intermediates are ignored here. (d) The brusselator model with detailed reactions. A and B are constant sources. (e) The glycolysis network. The allosteric en- zyme's protomer has two states, R (binding with P) and T (unbinding with P), and only R has the catalysis activity. Each Ri,j, with i = 1, 2,··· , ni and j = 1, 2,··· , nj represent the num- ber of S and P bound to R, here we used ni = nj = 2. Each Ri,j can undergo reactions of Ri,j + S ↔ Ri,j+1 ↔ Ri,j + P . Detailed descriptions and rate values are given in SI. a n o i t a r t n e c n o C 10 8 6 4 2 0 0 c s e i r o j t c e a r T 500 400 300 200 100 0 0 17 1 0.5 0 −0.5 −1 0 b n o i t l a e r r o c o t u A d 2 σ e c n a i r a v k a e P 25 20 15 10 5 0 0 C(t)=exp(−t/τ c) 50 Time 100 σ2=Dt 50 Time 100 50 Time 100 50 Time 100 σ(cid:19) FIG. 2. Correlation and phase diffusion in the activator-inhibitor model with V = 50, γ = 10−5. (a) Two noisy oscillation trajectories, with the peaks labeled by circles and squares. (b) Auto- correlation function (defined in Eq. 1) of the inhibitor X. C(t) decays exponentially with cor- relation time τc = 37.7. (c) Raster plot of the peak times for 500 different trajectories starting with the same initial condition. The distributions of the peak times for each consecutive peaks are shown by red lines. The peak time variance σ2 is shown. (d) Peak time variance σ2 goes linearly with the average peak time, with the linear coefficient defined as the peak time diffusion constant. Here, the diffusion constant D = 0.2 and α ≡ τcD/T 2 ≈ 0.07. a T / D 0.1 0.08 0.06 0.04 0.02 0 c 0.04 T / D * V 4 2 0 500 1000 ∆W V=50 V=100 V=200 400 ∆W (Dissipation per period) 600 800 1000 20 10 T / D * V 0 0 1000 500 ∆W 0.03 V=400 V=600 T / D 0.02 V=800 0.01 0 0 500 ∆W 1000 18 b 0.08 0.07 0.06 V=400 V=500 V=600 T / D 0.05 0.04 0.03 0.02 T / D * V 30 20 10 3 3.5 ∆W 4 3 3.5 4 ∆W d T / D 0.05 0.04 0.03 0.02 0.01 0 T / D * V 3 2 1 0 0 400 200 ∆W V=50 V=100 V=200 100 200 300 400 ∆W FIG. 3. Relation between the dimensionless diffusion constant (D/T ) and free energy dissipation per period per volume (∆W , in units of kBT ) for the four oscillatory systems. Detailed descriptions of the models and parameters can be found in SI. The relationships for different volumes collapse onto the same curve when the peak time diffusion constant is scaled by V , as shown in the insets. The black dashed line in the activator-inhibitor model indicates the value of ∆W if we assume that hydrolysis of one ATP molecule provides ≈ 12kBT energy, which corresponds to γ ≈ 10−5.2. All the data can be well fitted with Eq. 4: V × D/T = C + W0/(W − Wc) (lines in insets), where Wc is determined from the critical value γc, W0 and C are from fitting. The parameters are: (a) activator-inhibitor, Wc = 360.4, W0 = 447.3 ± 55.8, C = 0.28 ± 0.16; (b) repressilator, Wc = 1.9, W0 = 17.7 ± 3.9, C = 5.5 ± 2.5; (c) brusselator, Wc = 93.1, W0 = 1135 ± 142, C = 0.27 ± 0.11; (d) glycolysis, Wc = 67.4, W0 = 135.8 ± 3.4, C = 0.025 ± 0.019. a 0.06 0.05 0.04 / T D 0.03 0.02 0.01 1 b i 0.5 P 0 1 g o l 1 0 −1 −2 −3 1 0 500 1000 ∆W 1500 2000 0 −1 log10ADP 19 c 15 10 5 0 2 20 10 0 t n u o C t n u o C 1 0.8 0.6 0.4 3 4 log10ATP 5 2 0 log10(ADP/Pi) 4 −2 −3 2 3 5 4 log10ATP FIG. 4. The dependence of phase diffusion on the ATP, ADP, and Pi concentrations. We studied the activator-inhibitor model with 300 randomly chosen parameters of dimensionless [ATP],[ADP] and [Pi] (see Methods). The affected kinetic rates are a1 = a1,0[AT P ], f−1 = f−1,0[ADP ], f−2 = f−2,0[Pi] with a1,0 = 0.1, f−1,0 = f−2,0 = 1. We chose V = 100. (a) D/T versus ∆W for the 300 different parameter choices. All the points lie above an envelope curve, which follows Eq. 4 with D/T = 1.94/(∆W − 400) + 0.0036. Points in square indicate the points shown in Fig. 3. For a given value of D/T , the corresponding minimal energy dissipation ∆Wmin is computed according to the fitted envelope curve. The efficiency is defined as E ≡ ∆Wmin/∆W . Colors (b) Distribution of the 300 randomly sampled points in of the points indicate the efficiency. the parameter space ([AT P ], [ADP ], [Pi]). Colors of the points indicate the efficiency as in (a). Points with high efficiency are clustered. (c) Distribution of [ATP], and [ADP]/[Pi] for parameter choices with high efficiency E ≥ 0.75. The most probable parameter values for high efficiency are [AT P ] ≈ 103, [ADP ] ≈ [Pi], which corresponds to a1 = a2, f−1 = f−2 in the kinetic equations. This result indicates that high efficiency is achieved when the kinetic rates in the two halves of the PdP cycle (phosphorylation and dephosphorylation) are matched. 20 a T 28 26 24 22 −2 −1 1 2 0 ln(ATP/ADP) 3 ∞ b 1 0.8 0.6 τ / T 0.4 0.2 0 −2 −1 1 2 0 ln(ATP/ADP) 3 ∞ FIG. 5. Experimental evidence from Ref.[36] (blue curve) and Ref.[35] (Red curve). The two experiments measured the oscillation of KaiC phosphorylation in vitro in media with different ATP/ADP ratios. The autocorrelation functions were calculated from the original data and fitted by a exponential decay cosine function A cos (2πt/T )e−t/τc , where T is the period, and τc is the correlation time. ln(AT P/ADP ) represents the entropic contribution to the free energy. (a) The period T is robust against changes in the ATP/ADP ratio. (b) T /τc ≡ α−1D/T decreases with ln(AT P/ADP ) and eventually saturates at large ATP/ADP ratio, consistent with our theoretical prediction. 21 a + i ∆+ P G ATP → ADP cycτ T b ) D / T = ( c N 90 80 70 60 50 40 30 2000 4000 6000 8000 10000 12000 N ATP FIG. 6. Oscillation coherence increases with the number of ATP hydrolyzed per period. (a) Illustration of a biochemical oscillation as a clock in phase space. The intermediate states (green dots) are represented as the "hour ticks" of the clock. The transition from one tick to the next is coupled with a ATP hydrolysis cycle. The free energy release ∆G from the hydrolysis cycle powers the forward transition (thick solid arrow) and/or suppresses the backward transition (thin dotted arrow). The number of ATP consumed per enzyme molecule in each period T is given by T /τcyc, where τcyc is the average cycle time. (b) The accuracy of the oscillation, characterized by the number of correlated (coherent) periods Nc, increases linearly with the total number of ATP consumed per period NAT P before saturating at very high NAT P . We varied NAT P by changing the cycle time (see Methods for details), we used V = 100 here. 5 1 0 2 n u J 8 1 ] h p - o i b . s c i s y h p [ 1 v 6 8 6 5 0 . 6 0 5 1 : v i X r a Supplementary Information for "The free energy cost of accurate biochemical oscillations" by Cao et al. 1 I. DESCRIPTIONS OF THE FOUR MODELS Here, we describe the mathematical details of the four models of biochemical oscillations studied in this paper. For the activator-inhibitor and the glycolysis model, we only give the ordinary differential equations (ODE's) to describe the deterministic part of the chemical reactions. The actual simulations of the stochastic reactions were done using Gillespie algorithm (see Methods in the main text). For the repressilator and brusselator models, we use the chemical master equation (or Langevin equation with Poisson noise) and solve the corresponding Fokker-Planck equation: ∂P (~x, t) ∂t = −∇(FP − D∇P ) = −∇J, (S1) where F is the force vector, and D is the noise matrix. J is the flux vector for each direction. In all our models, the units of the parameters are composed of concentration (c, arbitrary), time (t, arbitrary) and volume (V , arbitrary). The molecule number unit c × V represents the real counts of a given molecule in the system. For example, if the concentration of enzyme E is ET = 10(c), and the volume is V = 100(V ), then the total number of enzyme E in our system is ET V = 1000. A. Activator-inhibitor model The main components of the model are the activator R and its inhibitor X. R and X are linked in a feedback loop through a phosphorylation-dephosphorylation (PdP) cycle (main text Fig. 1b). R activates the synthesis of both R and X through phosphorylated enzyme E, thus forms a positive feedback; at the same time, X degrades R, thus forms a negative feedback. The parameter γ = d1d2f−1f−2/(a1a2f1f2) is introduced to distinguish wether the system is in equilibrium (γ = 1) or non-equilibrium (0 < γ < 1). The kinetics is described by: d[R] dt d[X] = k0[Ep] + k1S − k2[X][R] = k3[Ep] − k4[X] = f2[EpK] + d1[ER] − a1[E]([R] − [ER]) − f−2[E][K] = a1[E]([R] − [ER]) + f−1[Ep]([R] − [ER]) − (f1 + d1)[ER] = f1[ER] + d2[EpK] − a2[Ep][K] − f−1[Ep]([R] − [ER]) dt d[E] dt d[ER] dt d[Ep] dt 2 (S2) d[EpK] dt = a2[Ep][K] + f−2[E][K] − f2[EpK] − d2[EpK] with two mass conservation constraints: [E] + [Ep] + [ER] + [EpK] = ET , [EpK] + [K] = KT , where ET and KT are the total concentrations of enzyme E and phosphatase K. Each term in the equations represents one of the reactions in the main text (see Fig. 1b). We take symmetric parameters: k0 = k1 = k3 = 1(t−1), k2 = 1(c−1t−1), k4 = 0.5(t−1), S = 0.4(c), KT = 1(c), ET = 10(c), a1 = a2 = 100(c−1t−1), f1 = f2 = d1 = d2 = 15(t−1), f−1 = f−2 = √γa1f1/d1(c−1t−1). The oscillation onset point is at γc = 2 × 10−3. Notice that the PdP cycle's reaction rates are much larger than the reactions of synthesis and degradation of R and X, so the total R is almost unchanged in the time scale of the PdP cycles. B. Repressilator We use the simplified cell cycle model in[1], where CDK activates Plk1, and Plk1 activates APC, which inhibits CDK in return. The (deterministic) negative feedback loop kinetics are governed by the following ODE's, with CDK, Plk1, APC concentrations represented by x, y, z, respectively: dx dt dy dt dz dt zn1 = α1 − β1 K n1 = fx − dx 1 + zn1 xn2 2 + xn2 − β2y = fy − dy K n2 yn3 3 + yn3 − β3z = fz − dz K n3 = α2(1 − y) = α3(1 − z) (S3) where fi, di are the synthesis and decay rates of each component. We chose α1 = 0.1(ct−1), α2 = 3(t−1), β1 = 3(ct−1), β2 = 1(t−1), β3 = 1(t−1), K1 = 0.5(c), K2 = 0.5(c), K3 = 0.5(c), n1 = 3 8, n2 = 8, n3 = 8, and α3(t−1) is taken as the control parameter ranges from 1.0 to 3.0. The oscillation onset point is α3 = 0.8. The full stochastic dynamics is described by the Fokker-Planck equation (Eq. S1) with the force vector and noise matrix given by: F = [fx − dx, fy − dy, fz − dz], D = 1 2V diag[fx + dx, fy + dy, fz + dz] . C. Brusselator The deterministic equation of brusselator is dx dt dy dt = a − x + x2y, = b − x2y. (S4) The Fokker-Planck equation was derived from chemical master equations (CME) in [2], which gave: F =   a − x + x2y b − x2y D = 1 2V 1 2V   −1/2 − 2xy + x2/2   +  a + x + x2y −x2y  b + x2y 2xy − x2/2   −x2y   (S5a) (S5b) where V is the volume of the system. We used b = 0.4, and varied a ∈ [0.12, 0.18] in our study. Fig. S1 gives two examples of probability distribution P (~x, t) and fluxes J in the brusselator model. D. Glycolysis oscillation The glycolysis model in our study is taken from [3], but we have introduced finite reverse reaction rates for the catalysis processes to study the free energy dissipation in glycolysis. The enzyme PFK are composed of n protomers, and undergoes allosteric regulation by its product P. Each protomer exists two states, R, which has catalytic activity for converting substrate S to P , and T, which is inactive. Assuming a quasi-equilibrium of the allosteric states of PFK [3], the dynamics of S and P can be written as: dS dt dP dt = vi − = nD(1 + α)n−1(1 + θ)n(kα − k′P ) L + (1 + α)n(1 + θ)n nD(1 + α)n−1(1 + θ)n(kα − k′P ) L + (1 + α)n(1 + θ)n − ksP 4 (S6) where α = (a1S + k′P )/(k + d1), θ = a2P/d2. Parameters were chosen as D = 500(c), n = 2, a1 = a2 = 10(c−1t−1), d1 = d2 = 10(t−1), k = 1(t−1), vi = 0.2(ct−1), ks = 0.1(t−1), L = 7.5 × 106. k′(c−1t−1) is the reverse reaction rate of P to S. The oscillation onset point is k′ = 4 × 10−1. Stochastic simulations were performed for 4 reactions: synthesis of S, degradation of P, catalysis of S to P, reverse reaction of P to S. Here, we combined all the enzymatic reactions into two reactions since the transitions between different allosteric states of the enzyme are much faster than the slow reactions of substrate injection (vi) and product removal (ks). The dissipation of the system can be directly calculated by summing up the dissipation of all the enzymatic reactions: ∆W (t) = nD(1 + α)n−1(1 + θ)n(kα − k′P ) L + (1 + α)n(1 + θ)n × log ka1S k′d1P = NS × ∆G, (S7) where NS quantifies how many molecules of S are catalysed to P. II. ENERGY DISSIPATION DETERMINED FROM SOLVING THE FOKKER-PLANCK EQUATION Consider a general Fokker-Planck equation ∂P (~x, t) ∂t = −∇(FP − D∇P ) = −∇J (S8) where F is the force vector, and D is the noise matrix. J is the flux vector for each direction. The system's entropy is S(t) = −Z P (~x, t) ln P (~x, t)d~x The entropy production rate is[4]: dS(t) dt = −Z [ln P (~x, t) + 1] ∂P (~x, t) ∂t d~x = Z [ln P (~x, t) + 1]∇Jd~x Integrated by parts: dS(t) dt = −Z J T∇ ln P (~x, t)d~x (S9) (S10) (S11) where J T is the transposition of J. By definition J = FP − D∇P , we have Finally J T∇ ln P = J T D−1F − J T D−1J P dS(t) dt = −Z J T D−1Fd~x +Z J T D−1J P d~x 5 (S12) (S13) The second term is the free energy dissipation rate (also called entropy production rate[5]) in unit of kBT . III. ONSET OF OSCILLATION Systems at the onset of oscillation is dissipative. In Fig. S2, we show that at the onset of oscillation, i.e., when amplitude is zero, the free energy dissipation is finite positive. The free energy dissipation at onset defines Wc that we used in fitting main text Fig3. IV. EXPERIMENTAL DATA ANALYSIS The experimental data were obtained from Ref[6] and Ref[7]. The data were processed to calculate the autocorrelation function and fitted the autocorrelation function with expo- nentially decay function A cos(2πt/T )e−t/τ , from which the period T and correlation time τ could be obtained. See FigS3. V. AMPLITUDE FLUCTUATION AND PHASE DIFFUSION IN THE STUART-LANDAU EQUATION We first derive the amplitude and phase variance in the simplified Langevin equations from the main text (Eq. 6). The deviation in r can be defined as δr = r − rs, (S14) where rs = pa/c. Perturbation δr near rs follows the equation (in first order approxima- tion): d(δr) dt = −2aδr + ηr(t). (S15) Following [8], as t → ∞, we obtain the amplitude fluctuation: hδr2(∞)i = ∆r 4a , which only depends on ∆r. 6 (S16) 0 ω(r(τ ))dτ . With the expanding expression ω(r(τ )) = ω(rs) + βδr(τ ) + ηθ(τ ), we have the phase variance: The phase is θ(t) = R t hθ2i − hθi2 = Z t 0 Z t (hω(τ1)ω(τ2)i − hω(τ1)ihω(τ2)i)dτ1dτ2 = β2Z t 0 Z t Following [8], as t → ∞, we have: 0 [hδr(τ1)δr(τ2)i − hδr(τ1)ihδr(τ2)i]dτ1dτ2 + ∆θt. 0 (S17) (S18) hθ2i − hθi2 = (β2 ∆r 4a2 + ∆θ)t ≡ Dθt. Clearly, the phase diffusion depends on both ∆θ and ∆r. Next, we show a general case of the Stuart-Landau equation described by the Langevin equations for the two real variables x and y: Z = x + iy. dx dt dy dt = (ax − by) − (cx − dy)(x2 + y2) +p∆1η1(t) = (bx + ay) − (cy + dx)(x2 + y2) +p∆2η2(t) (S19) We studied these two equations numerically. We found that if ∆1 6= ∆2, then Wc 6= 0. In Fig. S4, we show a case with fixed ∆1 = 0.1 while varying ∆2 ∈ [0.05, 0.2]. We found that the onset, which corresponds to D → ∞, occurs at a finite energy dissipation Wc 6= 0 (Fig. S4a). However, the general relationship of D/T = W0/(∆W − Wc) + C, with Wc 6= 0 from Fig. S4a, holds true (Fig. S4b). VI. AMPLITUDE FLUCTUATION The amplitude fluctuation can be defined as the dispersion of the stochastic trajectories departing from the deterministic trajectory in phase-space: d2 = Z min (~x − ~xd)2P (~x)d~x (S20) where ~xd are the points on the deterministic trajectory. In the four models we studied, the amplitude fluctuations decrease with energy dissipation and scale with 1/√V , as shown in Fig. S5. 7 VII. ROBUSTNESS AND ENERGY DISSIPATION For the activator-inhibitor model, the total number of enzyme E and phosphatase K may vary in real systems (e.g., from cell to cell). Here we search the parameter space of (ET , KT ) in the region ET ∈ [0, 1000] and KT ∈ [0, 2], and check wether the system oscillates (with amplitude larger than 0.1) for different values of γ. Robustness is defined as the area in the parameter space where oscillation exists. We found robustness increases as the system becomes more irreversible or equivalently dissipates more free energy, as shown in Fig. S6. [1] Ferrell, J. J., Tsai, T. Y. & Yang, Q. Modeling the cell cycle: why do certain circuits oscillate? Cell 144, 874 -- 85 (2011). [2] Qian, H., Saffarian, S. & Elson, E. L. Concentration fluctuations in a mesoscopic oscillating chemical reaction system. Proc Natl Acad Sci U S A 99, 10376 -- 81 (2002). [3] Goldbeter, A. & Lefever, R. Dissipative structures for an allosteric model. application to glycolytic oscillations. Biophys J 12, 1302 -- 15 (1972). [4] Tome, T. & de Oliveira, M. J. Entropy production in irreversible systems described by a fokker-planck equation. Phys Rev E 82, 021120 (2010). [5] Ge, H. & Qian, H. Physical origins of entropy production, free energy dissipation, and their mathematical representations. Phys Rev E 81, 051133 (2010). [6] Phong, C., Markson, J. S., Wilhoite, C. M. & Rust, M. J. Robust and tunable circadian rhythms from differentially sensitive catalytic domains. Proc Natl Acad Sci U S A 110, 1124 -- 1129 (2013). [7] Rust, M. J., Golden, S. S. & O'Shea, E. K. Light-driven changes in energy metabolism directly entrain the cyanobacterial circadian oscillator. Science 331, 220 -- 3 (2011). [8] Van Kampen, N. G. Stochastic processes in physics and chemistry (Elsevier, Amsterdam, 1992). 0.6 0.4 0.2 b e d u t i l p m A d d o i r e P 10 9 8 7 6 14 400 400 500 500 600 600 ∆W 700 700 800 800 2 2.5 3.0 3.5 ∆W 0 4 2 12 d o i r e P e d u t i l p m A 8 4.5 4 3.5 3 d o i r e P 125 120 d o i r e P 200 200 400 400 600 600 ∆W 800 800 10 1000 1000 0 0 0 100 100 115 300 300 200 200 ∆W a e d u t i l p m A 2 1.5 1 0.5 0 c e d u t i l p m A 4 2 0 FIG. S1. The dependence of amplitude (A) and period (T ) on energy dissipation for the four models: (a) activator-inhibitor, (b)repressilator, (c)brusselator, (d)glycolysis. For all the four cases, the critical free energy dissipation per period Wc is finite at the onset of the oscillation, i.e., A = 0. a 4 3 Y 2 1 0 0 1 2 X 3 4 9 b 4 3 Y 2 1 0 0 1 2 X 3 4 FIG. S2. Probability distributions and fluxes (red arrows) in the brusselator model. (a)a = 0.18, near the onset (bifurcation point), (b) a = 0.12, far from the bifurcation point. a b 10 FIG. S3. Autocorrelation of experimental data. The autocorrelation function were calculated from the original data and fitted with A cos(2πt/T )e−t/τ , where T is the period, and τ is the correlation time. (a). Data from Ref[7] with different ATP ratio. (b).Data from Ref[6] with different ADP ratio. a W ∆ 12000 10000 8000 6000 4000 Wc 2000 11 b 0.015 0.01 T / D 0.005 D/T=W 0 /(∆W−W )+C c 0 5 10 1/∆ 2 15 20 0 4000 6000 8000 10000 12000 ∆W FIG. S4. Numerical simulation results of of the general Stuart-Landau equation (Eq. S24) with a = 1, c = 1, b = 2, d = 1, ∆1 = 0.1. We varied ∆2 ∈ [0.05, 0.2]. (a) Relationship between energy dissipation ∆W and noise strength ∆2. When ∆2 is large, ∆W decreases linearly with 1/∆2. The dashed line shows when ∆2 → ∞, ∆W → Wc ≈ 2967 (black circle). (b) The peak time diffusion constant D/T versus ∆W . The red dashed curve is the fitting inverse proportional relation, with parameters W0 = 10.2, Wc = 2967, C = 0.0011. a 0.15 V=50 0.1 A / A δ V=100 V=200 0.05 0 2 / 1 V * A / A δ 1 0.5 0 500 1000 ∆W 400 600 800 1000 ∆W c 0.25 0.2 V=400 V=600 0.15 A / A δ V=800 0.1 0.05 0 2 / 1 V * A / A δ 6 4 2 0 500 ∆W 1000 500 ∆W 12 b A / A δ d A / A δ 0.2 0.18 0.16 0.14 0.12 0.1 0.08 0.3 0.25 0.2 0.15 0.1 2 / 1 V * A / A δ 4 3 2 3 3.5 4 ∆W V=400 V=500 V=600 3 3.5 4 4.5 ∆W 2 / 1 V * A / A δ 2 1.5 1 0.5 100 200 300 400 ∆W V=50 V=100 V=200 1000 0.05 100 200 300 400 ∆W FIG. S5. Relation between relative amplitude fluctuation δA/A and free energy dissipation ∆W for the four models. (a) activator-inhibitor; (b)repressilator; (c)brusselator; (d)glycolysis. Data for different volumes collapses (insets) when we scaled amplitude fluctuation with 1/√V . a 3 10 T M 2 10 13 b 3 10 i i e z S n o g e R y r o a t 2 10 0 0.5 1.5 2 1 K T l l i c s O 1 10 3 4 5 6 −log 10 γ FIG. S6. The relationship between functional robustness and free energy dissipation. (a) The green, blue and red curves in the (ET , KT ) space correspond to the boundaries inside which oscillations exist for γ = 10−3, γ = 10−4 and γ = 10−5, respectively. The star indicates the parameters in main text Fig1 a. (b) Robustness, defined as the area of oscillation in the parameter space, increases as γ decreases. This means that higher free energy consumptions (on average) is needed for higher robustness against parameter variations in achieving oscillatory behaviors.
1508.03509	1	1508	2015-08-14T14:13:28	Programmable DNA-mediated multitasking processor	[ "physics.bio-ph" ]	Because of DNA appealing features as perfect material, including minuscule size, defined structural repeat and rigidity, programmable DNA-mediated processing is a promising computing paradigm, which employs DNAs as information storing and processing substrates to tackle the computational problems. The massive parallelism of DNA hybridization exhibits transcendent potential to improve multitasking capabilities and yield a tremendous speed-up over the conventional electronic processors with stepwise signal cascade. As an example of multitasking capability, we present an in vitro programmable DNA-mediated optimal route planning processor as a functional unit embedded in contemporary navigation systems. The novel programmable DNA-mediated processor has several advantages over the existing silicon-mediated methods, such as conducting massive data storage and simultaneous processing via much fewer materials than conventional silicon devices.	physics.bio-ph	physics	Source: DOI: Journal of Physical Chemistry B, Vol. 119, No. 17, pp. 5639‐5644, 2015; 10.1021/acs.jpcb.5b02165 Programmable DNA-Mediated Multitasking Processor Jian-Jun SHU,†, Qi-Wen WANG†, Kian-Yan YONG†, Fangwei SHAO‡, and Kee Jin LEE† †School of Mechanical & Aerospace Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798 ‡School of Physical & Mathematical Sciences, Nanyang Technological University, 21 Nanyang Link, Singapore 637371 KEYWORDS: DNA; processor; material; programmable biochemical operator. ABSTRACT: Because of DNA appealing features as perfect material, including minuscule size, defined structural repeat and rigidity, programmable DNA-mediated processing is a promising computing paradigm, which employs DNAs as information storing and processing substrates to tackle the computational problems. The massive parallelism of DNA hybridization exhibits transcendent potential to improve multitasking capabilities and yield a tremendous speed-up over the conventional electronic processors with stepwise signal cascade. As an example of multitasking capability, we present an in vitro programmable DNA-mediated optimal route planning processor as a functional unit embedded in contemporary navigation systems. The novel programmable DNA-mediated processor has several advantages over the existing silicon- mediated methods, such as conducting massive data storage and simultaneous processing via much fewer materials than conventional silicon devices. ■ INTRODUCTION For a conventional silicon-mediated digital computer, the growing demands for computational ability, processing speed and parallelism, require the size of individual transistor elements to be significantly reduced and hence to allow additional elements to be packed onto the same chip. The increasing packing density has led to many essential problems, including power consumption and heat dissipation. More alarmingly, the entire semiconductor industry is quickly approaching the physical constraints as predicted by Moore’s law.1 In principle, any device endowed with three fundamental functions – processing, storing and displaying information – can be regarded as a computer. Therefore, researchers from various disciplines are engaged in exploring alternatives to silicon-mediated digital computer.2-4 Among various intriguing approaches, DNA-mediated computing seems to be the feasible strategy to compete with silicon- mediated counterpart, and ultimately brings the entire field into a new era. DNA has appealing features,5 including minuscule size, defined structural repeat and rigidity. Programmable DNA- mediated processing is a promising computing paradigm, which employs DNAs as information storing and processing substrates to tackle the computational problems. Since the demands of monolithic parallel computing ability have grown rapidly due to specific computational algorithm,6 the massive parallelism of DNA hybridization exhibits transcendent potential to improve multitasking capabilities and yield a tremendous speed-up over the traditional electronic 2 processors with stepwise signal cascade. The novel programmable DNA-mediated processor has several advantages over the existing silicon-mediated methods, such as conducting massive data storage7 and simultaneous processing8 via much fewer materials than conventional silicon devices. As compared with the history of evolving silicon-mediated computer, the development of DNA-mediated computer remains at relatively early stage.9 DNA molecules have been successfully utilized to demonstrate the solutions of various problems, such as Hamiltonian path problem,10 maximal clique problem11 and strategic assignment problem.12 As an example of multitasking capability, an in vitro programmable DNA-mediated optimal route planning processor is designed and experimentally demonstrated. ■ METHODS Optimal route planning processor behaves as a functional unit embedded in contemporary GPS (Global Positioning System) navigator. By assigning any two physical locations, present location and final destination, to the processor, it automatically routes an optimal path, or a shortest path, between the selected places, based on the information pre-stored within its database. The programmable DNA-mediated optimal route planning processor is employed to perform the equivalent function by means of using DNA molecules as information storing and processing instrument. For demonstration purpose, the selected case study is an arbitrary map in GPS navigation system, which contains exactly six physical locations connected by the bi-paths with opposite directions, as shown in Figure 1. For the sake of simplicity, the key information on the map can be represented by using an abstract graph S, as shown in the right hand side of Figure 1. Each vertex i and weighted edge (i, j) represent an associated physical location and the connecting path from location i to j, respectively, in the original map, where i, j  {1, 2, 3, 4, 5, 3 6}. The numbers placed on the side of the edges are the lengths (or weights) of the corresponding connecting path. The entire case study can be converted to a mathematical problem: Given a graph S, made up of vertices and weighted edges, what is the shortest path between any two user-specified vertices? Figure 1. Scheme of an optimal route planning problem: The original map with six locations (left) can be converted to an abstract graph (right) with each vertices indicated as colored circles and the length of each bi-path annotated by red digits, respectively. The programmable DNA-mediated optimal route planning apparatus is composed of six stages as shown in Figure 2. 1). The constituent elements, vertices and weighted edges in the graph S, are converted to different DNA sequences by problem encoder. 2). Based on the information obtained from problem encoder, DNA solution required for subsequent processes is prepared in 4 DNA solution bay. 3). In mixing controller, the appropriate DNA solution of vertices and weighted edges is mixed and ligased to generate a template pool of DNA duplexes that represent the possible routes from any given start to destination. In the perspective of mathematics, all non-restricted random walks are initiated within the graph S upon the completion of hybridization. 4). The obtained DNA template containing the optimal path is isolated from the oligonucleotides of incomplete hybridization, enzymes and other possible impurities. 5). The optimal path is subsequently amplified by using PCR (Polymerase Chain Reaction). The input commands to the processor are translated into the forward DNA primer representing the initial location and the reverse DNA primer representing the destination. As the efficiency of PCR is proportional to the length and copy number of the DNA templates, the DNA duplex that represents the optimal path (usually the shortest path) becomes the dominate products from PCR amplifier. 6). The PCR result is subjected to native polyacrylamide gel electrophoresis to gauge the lengths of the dominate products as the optimal (shortest) path. In this stage, the length of DNA duplex is deciphered based upon the path lengths between each pair of locations, to determine the optimal route from the given start location and destination. 5 Figure 2. Process flow chart of programmable DNA-mediated optimal route planning processor Problem encoder – For silicon-mediated computer, the very beginning stage is to digitize information in terms of binary expressions. Analogous to digital computing, the first step for DNA-mediated computing is to encode information in terms of a combination of DNA sequences – A, T, G and C. Each vertex in the graph S is encoded into a 20-mer ssDNA (single-stranded DNA). To avoid undesired hybridization during mixing stage, the generated DNA strands have 50% CG content and melting temperature, mT = 70°C. Based on the specified rule and the sequence designing methodology,13 the encoded DNA sequences with respect to their corresponding vertices are listed in Table 1. 6 Table 1. Vertices and the corresponding ssDNA sequences vertex ssDNA sequence (5' to 3') 1 2 3 4 5 6 CCGTGTCTACAACAGAAGGA CCCTCATTTGACGAGGAATG TGGTTAGACGCAGAGAGTTC AGGCACACGATTATGGACAG CGAATTTAGCACCGCATGTG GGGGTCTTCAACTATTGTCC As described in graph theory, the edge within the graph S is used to connect two adjacent vertices. The entire graph S is a bi-graph which means there is no specific restriction on the direction of edges. Therefore, it requires sixteen assorted DNA templates to symbolize eight corresponding edges. The template used to encode the edges is specifically designed as the dsDNA (double-stranded DNA) duplex with two overhangs (i.e. exposed ssDNA). As the example shown in Figure 3, each edge DNA contains a duplex portion (Segment 2) with overhangs extended from 5’- and 3’-overhangs (Segment 1 and Segment 3, respectively). Segments 1 and 3 are employed to connect from vertex i to vertex j, whereas Segment 1 is complementary to the rear 10-mer ssDNA of vertex i and Segment 3 is complementary to the former 10-mer ssDNA of vertex j, where i, j  {1, 2, 3, 4, 5, 6}. Segment 2 is acting as the biasing parameter to count the length of the edges. The length of Segment 2 is determined by subtracting 20bp (base pair) from the length of the corresponding edges in graph S. For instance, edge (1, 5) has 32 units of length as depicted in S. Therefore, the corresponding length of 7 Segment 2 of DNA template is 12bp. In short, the length of edge is proportional to that of the assigned DNA template. Figure 3. Schematic representation of edge encoding scheme. The backbones of the vertex DNA sequences are presented in red or navy, which those of the edge DNA sequences are presented in blue. DNA solution bay – In order to prevent cross-contamination, DNA solution required for the subsequent processes is prepared within an isolated and decontaminated area, called DNA solution bay. The DNA strands representing vertex 1~6, sixteen edge paths and six pairs of primers for the amplification of optimal solution are listed in Tables 1, 2, and 3, respectively. The vertex and primer DNA strands are prepared as single strands and stored as 100µM aqueous stock solution, while the edge DNA strands are annealed to duplex and prepared into 50µM stock solution. Table 2. Edges and the corresponding dsDNA sequences edge dsDNA sequence# 8 (1,2) (2,1) (1,5) (5,1) (2,3) (3,2) (2,4) (4,2) (2,6) (6,2) (3,4) (4,3) 5’-CCCAGTTAATGCAACCTTGTAGCGGT–3’ 3’-TTGTCTTCCTGGGTCAATTACGTTGGAACATCGCCAGGGAGTAAAC-5’ 5’-CCCAGTTAATGCAACCTTGTAGCGGT-3’ 3’-TGCTCCTTACGGGTCAATTACGTTGGAACATCGCCAGGCACAGATG-5’ 5’-TATAAGTGCCCG-3’ 3’-TTGTCTTCCTATATTCACGGGCGCTTAAATCG-5’ 5’-TATAAGTGCCCG-3’ 3’- TGGCGTACACATATTCACGGGCGGCACAGATG-5’ 5’-TATGCTGACCATCCTACCTTCC-3’ 3’-TGCTCCTTACATACGACTGGTAGGATGGAAGGACCAATCTGC-5’ 5’-TATGCTGACCATCCTACCTTCC-3’ 3’-GTCTCTCAAGATACGACTGGTAGGATGGAAGGGGGAGTAAAC-5’ 5’-AAAGCTCTAGGG-3’ 3’-TGCTCCTTACTTTCGAGATCCCTCCGTGTGCT-5’ 5’-AAAGCTCTAGGG-3’ 3’-AATACCTGTCTTTCGAGATCCCGGGAGTAAAC-5’ 5’-ACATGACATCTCCGCTGA-3’ 3’-TGCTCCTTACTGTACTGTAGAGGCGACTCCCCAGAAGT-5’ 5’-ACATGACATCTCCGCTGA-3’ 3’-TGATAACAGGTGTACTGTAGAGGCGACTGGGAGTAAAC-5’ 5’-ATCTGGTATTCGTCCC-3’ 3’-GTCTCTCAAGTAGACCATAAGCAGGGTCCGTGTGCT-5’ 5’-ATCTGGTATTCGTCCC-3’ 9 3’-AATACCTGTCTAGACCATAAGCAGGGACCAATCTGC-5’ 5’-CCCCTAGTCATCGTTACT-3’ 3’-GTCTCTCAAGGGGGATCAGTAGCAATGAGCTTAAATCG-5’ 5’-CCCCTAGTCATCGTTACT-3’ 3’-TGGCGTACACGGGGATCAGTAGCAATGAACCAATCTGC-5’ 5’-GCAAGTTTGG-3’ 3’-AATACCTGTCCGTTCAAACCGCTTAAATCG-5’ 5’-GCAAGTTTGG-3’ 3’-TGGCGTACACCGTTCAAACCTCCGTGTGCT-5’ (3,5) (5,3) (4,5) (5,4) #The duplex segments proportional to the length of the edges are highlighted in red. Table 3. DNA primer sequence design vertex forward primer sequence (5’ to 3’) reverse primer sequence (5’ to 3’) 1 2 3 4 5 6 CCGTGTCTACAACAGAAGGA TCCTTCTGTTGTAGACACGG CCCTCATTTGACGAGGAATG CATTCCTCGTCAAATGAGGG TGGTTAGACGCAGAGAGTTC GAACTCTCTGCGTCTAACCA AGGCACACGATTATGGACAG CTGTCCATAATCGTGTGCCT CGAATTTAGCACCGCATGTG CACATGCGGTGCTAAATTCG GGGGTCTTCAACTATTGTCC GGACAATAGTTGAAGACCCC Mixing controller – After all DNA strands, ssDNA and dsDNA, as stated above, are prepared in individual eppendorf on DNA solution bay, a certain amount of solution, representing vertices and edges in the graph S, is aliquoted out and mixed in a new eppendorf to conduct the 10 computing. For six vertex ssDNA sequences, which represent the vertices in graph S, 1µL of solution is taken from each stock solution. For the sixteen dsDNA sequences, which represent the edges, the amount of solution to be added is inversely proportional to their lengths as indicated in graph S. Such a strategy promotes the formation of the optimal path. The individual volume of dsDNA solution is detailed in Table 4. The resultant 22µL mixture is then subjected to a process known as hybridization, whereas the complementary ssDNA and DNA duplexes with sticky ends are readily annealed to each other through hydrogen bonding. To optimize the result of hybridization, 2.4µL of 10× ligation buffer is added into the mixture. The hybridization occurs instantly once all the DNA sequences are placed in mixing controller. Hence the incubation of mixing controller at room temperature takes only 15 minutes to ensure the completion of hybridization process. Table 4. DNA solution pool (during hybridization) Test Tube Solution Volume (µL) 1 to 6 7 and 8 9 and 10 11 and 12 13 and 14 15 and 16 17 and 18 19 and 20 21 and 22 (ssDNA, 20mer) (dsDNA, 46bp) (dsDNA, 32bp) (dsDNA, 42bp) (dsDNA, 32bp) (dsDNA, 38bp) (dsDNA, 36bp) (dsDNA, 38bp) (dsDNA, 30bp) 1.00 0.78 1.13 0.86 1.13 0.95 1.00 0.95 1.20 11 After the completion of hybridization process, DNA strands are held together through relative weak hydrogen bonds. In order to connect the optimal pathway into a one-piece DNA fragment as the template required by the next step, PCR amplifier, the adjacent DNA strands arranged by overhang annealing in the previous step are ligased by sequentially adding 2µL NEB kinase and 2µL T4 DNA liganse. In each step, hybridized DNA fragments are incubated with the enzyme for 20 minutes at 37°C. In a DNA pool with as low as 50 (cid:1868)(cid:1865)(cid:1867)(cid:1864) solution, there are approximately 1.5(cid:3400)10(cid:2877) copies of associated DNA strands encoding each vertex and weighted edges, which is plenty enough to generate non-restricted random walks during hybridization among vertex and edge DNAs. Therefore, it is believed that all possible paths can be generated at the end of ligation process. Solution purifier – In order to obtain the optimal outcome from the subsequent PCR amplifier, DNA solution from the previous stage is immediately subjected to solution purifier. In DNA- mediated computing, solution purifier is acting as a “noise signal remover” to isolate all the possible paths as dsDNA templates and ultimately to remove the impurities including incompletely hybridized oligonucleotides, various enzymes, and salt from desired solution. Contemporarily, there are several optimized purification kits available in the market, which are manufactured to satisfy the various requirements in accord with the range of DNA fragment binding-size. In this experiment, QIAquick® PCR purification kit is used for purification process. 12 After 2µL of final solution is taken for concentration analysis, the remaining solution of 26.4µL is pipetted into an empty test tube. A total volume of 132µL of PB buffer from QIAquick® PCR purification kit, or 5 times of the amount of the remaining solution, is added into the test tube. The well-mixed solution is poured into a QIAquick column and attached to a 2mL collection tube. Subsequently, the sample is centrifuged at 13,000 rpm (rounds per minute) for 60 seconds. The QIAquick column is placed back into a new test tube. The PE buffer of 0.75mL is added into the QIAquick column and centrifuged for another 60 seconds. Again, the flow through is discarded and the QIAquick column is placed back in the same tube. The QIAquick column is centrifuged in a 2mL collection tube for another 1 minute to remove the residual buffer. The QIAquick column is then placed in a new 1.52mL micro-centrifuge tube. Finally, 50µL EB buffer is added into the membrane at the center of the column, retained at room temperature for another 1 minute, and followed by one additional centrifugal process to elute the DNA solution. The final elute contains roughly 40 to 50µL of purified dsDNA duplex. PCR amplifier – PCR is a frequently employed laboratory technique in DNA-mediated computing to amplify specific “signal” – DNA template, by varying the “input signal” – various synthesized DNA primers, which is specified by the user of optimal route planning processor. To demonstrate the DNA-mediated processor for optimal path question in GPS devices, two cases are studied here. It is desired to determine the optimal path traveling from two distinct locations, home (vertex 1) and company (vertex 6), to the same destination – hospital (vertex 4). In the first case, the optimal path starts from home (vertex 1) and terminates at hospital (vertex 4). The selected primers for PCR amplifier are the forward primer of vertex 1 and the reverse 13 primer of vertex 4, as specified in Table 3. Similarly, in the second case, the optimal path starts from company (vertex 6) and terminates at hospital (vertex 4). The forward primer of vertex 6 and the reverse primer of vertex 4 are used. The selected primer and purified DNA solution from the previous stage are mixed together with the reagents of QIAGEN fast cycling PCR kit in eppendorf. Each PCR sample contains 14µL PCR reagents (10µL of PCR master mix and 4µL of Q solution), 2µL of the corresponding forward primer, 2µL of the corresponding reverse primer, and 2µL of purified DNA solution from the previous stage, resulting in a mixture of 20µL prior to PCR. The detailed PCR protocol is described as follows: Step 1: Increase solution temperature to 95°C and maintain for 5 minutes. Step 2: Increase solution temperature to 96°C and maintain for 5 seconds. Step 3: Gradually reduce solution temperature to 48°C and maintain for 5 seconds. Step 4: Increase solution temperature to 68°C and maintain for 5 seconds. Step 5: Repeat Step 2 to Step 4 for 29 cycles. Step 6: Increase solution temperature to 72°C and maintain for 1 minute. Step 7: Gradually reduce solution temperature to room temperature 25°C. Based upon the different primers selected for these two cases above, the PCR samples are labelled with “1→4” and “6→4”, respectively. As the result of PCR, for “ 1→4”, only the paths which starts with vertex 1 and terminates at vertex 4 are amplified. The same condition 14 applies to “6→4”. After PCR amplifier, it is recommended to repeat the purification process to remove any possible impurities. However, the purification process is not compulsory. Gel electrophoresis – Gel electrophoresis employs native polyacrylamide gel as sieving material to distribute the linear DNA molecules with respect to their lengths. In this experiment, 12% gel is prepared according to the recipe as specified in Table 5. After the completion of electrophoresis, the gel is stained in the TBE buffer containing 3× GelRed for 30 minutes. Subsequently, the results are observed in a 2-dimentional multi-fluorescence scanner Typhoon 9410. Under the green-excited mode, the DNA bands in gel slab have a strong emission subjected to excitation at 532nm wavelength UV (ultraviolent) light. The 2D image of 12% native acrylamide gel of four lanes is displayed in Figure 4(a). Table 5. 12% native polyacrylamide gel recipe Ingredient Volume (µL) H2O 30% acrylamide 5(cid:3400) TBE buffer 10% ammonium persulfate TEMED 15.7 16 8 0.28 0.026 15 Figure 4. The optimal paths from home or company to hospital on 12% PAGE (a: lane 3 and lane 4) and illustrative map (b and c). Lane 1: DNA ladder; Lane 2: 82bp DNA. ■ RESULTS Figure 4 shows the PCR outcome of the optimal path from either home (vertex 1) or company (vertex 6) to hospital (vertex 4) selected by the DNA-mediated processor. Four wells are built in gel slab to contain different solution – well 1: DNA ladders in 10bp interval; well 2: dsDNA template of 82bp; well 3: PCR product of the sample “1→4”; well 4: PCR product of the sample “6→4”. Well 1 is used to display DNA ladders, ranging from 50bp to 150bp, in 10bp interval. According to its working principle, the migration distance of DNA solution toward anode electrode is inversely proportional to the length of linear dsDNA strands. DNA ladders in 16 well 1 can be used to gauge the length of PCR outcome from the two selections in lane 3 and lane 4. For the same purpose, well 2 contains the dsDNA duplex of a specified length of 82bp. The gel slab immersed in TBE buffer is subjected to gel electrophoresis process by setting the power of 18W with variable voltage and current at 560V and 32mA, respectively. The entire electrophoresis process takes roughly one and a half hours. After the gel electrophoresis process, it is possible to determine the total length of optimal path by comparing the migration lengths of the most prominent band in lane 3 and lane 4 with respect to the bands in reference lanes, which are lane 1 and lane 2. The maximum migration length toward the anode electrode of the most prominent band along lane 3 or lane 4 is the dsDNA duplex of the shortest length. Therefore, the DNA-mediated processor provides the solution to the optimal path problem for both user inputs discussed above. ■ In both cases of “1→4” and “6→4” (lane 3 and lane 4, respectively), only one prominent DISCUSSION band is observed to indicate the most optimal path selected by the GPS. The length of optimal path obtained from the DNA processor is gauged by the DNA ladders in lane 1 ranging from 50bp to 150bp in the interval of 10bp, and/or a dsDNA duplex in lane 2 of a specified length 82bp, as the expected shortest distance from home (vertex 1) to hospital (vertex 4), that is, the path “1→5→4”. Such length and sequence of the reference DNA are determined by using the encoding scheme together with Figure 1. The reference dsDNA duplex in lane 2 is used to determine the migration length of the most prominent band in lane 3. In simple words, it is going to testify the following hypothesis: The resultant DNA from optimal route planning 17 processor, which represents the shortest path from home to hospital as shown in lane 3, has the same migration distance as predicted in lane 2. Lane 3 is used to migrate the PCR product, which is selected by the primers representing the starting and terminating points as vertex 1 and vertex 4, respectively. As observed in Figure 4(a), a uniquely visible band in lane 3 has the same migration distance as that in lane 2. From the entire template pool of all possible routes between any two desired locations, the pair of input primers is able to amplify specifically one template of 82bp as the only output signal from the DNA processor. Therefore, it is believed the programmable DNA-mediated optimal route planning processor has “selected” the expected path, which is the path “1→5→4”, as shown in Figure 4(b). Analogously, a unique band in lane 4 can be observed by using the forward primer of vertex 6 and the reverse primer of vertex 4. As compared with the DNA ladder in lane 1, the migration distance of band in lane 4 is equivalent to that of dsDNA duplex of 90bp in lane 1. As depicted in Figure 1, it is possible to determine that the optimal path from company (vertex 6) to hospital (vertex 4) is likely to be the path “6 →2→4” as shown in Figure 4(c), which is 90bp based upon the DNA encoder listed in Tables 1 and 2. In the two cases studied here, all of the possible paths between any of two locations are generated in the DNA-encoded processor simultaneously. Once that all the vertex and edge DNA strands are hybridized in the step of mixing controller, the readout of the outcome, the 18 optimal path, is unambiguous if the specific pair of PCR primers is used as an input command from the user. ■ CONCLUSIONS The novel programmable DNA-mediated processor is developed to encode a complete bi- directional road map including six locations. The optimal route between any of two locations can be simultaneously selected by the programmable DNA-mediated processor in parallel and selectively revealed by the conventional biochemistry assays, PCR amplifier and gel electrophoresis. This programmable DNA-mediated processor has several advantages over the existing silicon-mediated methods in solving similar problems, such as conducting massive data storage and simultaneous processing for multiple path selections. In theory, it requires much fewer materials and much lower space than conventional silicon devices. The utilizing length of DNA strands, as the biasing parameter, is believed to be superior to the other alternatives, such as varying the pH value or concentration of DNA solution. Furthermore, with the proper encoding of the map into DNA edge duplexes, the optimal path can be directly deduced by the length of PCR outcome. Hence the current processor avoids the tremendous processing cost and time in DNA sequencing, which is required by many precedent DNA-based systems to address mathematical problems. It is worth to mention to this end that the DNA GPS system proposed in this paper may provide a new way to understand the working mechanism of a brain’s GPS system due to the Nobel Prize winning discovery of the place14 and grid15 cells. AUTHOR INFORMATION Corresponding Author 19 (J.-J.S.) E-mail: [email protected]. Notes The authors declare no competing financial interest. REFERENCES (1) Moore, G. E. Cramming More Components onto Integrated Circuits. Electronics 1965, 38, 114–117. (2) Kuhnert, L.; Agladze, K. I.; Krinsky, V. I. Image-Processing Using Light-Sensitive Chemical Waves. Nature 1989, 337, 244–247. (3) Divincenzo, D. P. Quantum Computation. Science 1995, 270, 255–261. (4) Brooks, R. Artificial Life - From Robot Dreams to Reality. Nature 2000, 406, 945–947. (5) Seeman, N. C. DNA in a Material World. Nature 2003, 421, 427–431. (6) Yurke, B.; Turberfield, A. J.; Mills, A. P.; Simmel, F. C.; Neumann, J. L. A DNA-Fuelled Molecular Machine Made of DNA. Nature 2000, 406, 605–608. (7) Church, G. M.; Gao, Y.; Kosuri, S. Next-Generation Digital Information Storage in DNA. Science 2012, 337, 1628–1628. (8) Benenson, Y.; Gil, B.; Ben-Dor, U.; Adar, R.; Shapiro, E. An Autonomous Molecular Computer for Logical Control of Gene Expression. Nature 2004, 429, 423–429. (9) Reif, J. H. Scaling up DNA Computation. Science 2011, 332, 1156–1157. 20 (10) Adleman, L. M. Molecular Computation of Solutions to Combinatorial Problems. Science 1994, 266, 1021–1024. (11) Ouyang, Q.; Kaplan, P. D.; Liu, S. M.; Libchaber, A. DNA Solution of the Maximal Clique Problem. Science 1997, 278, 446–449. (12) Shu, J.-J.; Wang, Q.-W.; Yong, K.-Y. DNA-Based Computing of Strategic Assignment Problems. Phys. Rev. Lett. 2011, 106, 188702. (13) Tanaka, F.; Kameda, A.; Yamamoto, M.; Ohuchi, A. Design of Nucleic Acid Sequences for DNA Computing Based on a Thermodynamic Approach. Nucleic Acids Res. 2005, 33, 903–911. (14) O’Keefe, J.; Dostrovs, J. The Hippocampus as a Spatial Map. Preliminary Evidence from Unit Activity in the Freely-Moving Rat. Brain Res. 1971, 34, 171–175. (15) Hafting, T.; Fyhn, M.; Molden, S.; Moser, M.-B.; Moser, E. I. Microstructure of a Spatial Map in the Entorhinal Cortex. Nature 2005, 436, 801–806. 21 Table of Contents (TOC) Graphic 22
1806.00261	1	1806	2018-06-01T09:59:48	Fluid-structure interaction study of spider's hair flow-sensing system	[ "physics.bio-ph", "cond-mat.soft", "physics.app-ph" ]	In the present work we study the spider's hair flow-sensing system by using fluid-structure interaction (FSI) numerical simulations. We observe experimentally the morphology of Theraphosa stirmi's hairs and characterize their mechanical properties through nanotensile tests. We then use the obtained information as input for the computational model. We study the effect of a varying air velocity and a varying hair spacing on the mechanical stresses and displacements. Our results can be of interest for the design of novel bio-inspired systems and structures for smart sensors and robotics.	physics.bio-ph	physics	Fluid-structure interaction study of spider's hair flow-sensing system Roberto Guarinoa, Gabriele Grecoa,b, Barbara Mazzolaib, Nicola M. Pugnoa,c,d, F* aLaboratory of Bio-Inspired & Graphene Nanomechanics, Department of Civil, Environmental and Mechanical Engineering, University of bCenter for Micro-BioRobotics@SSSA, Istituto Italiano di Tecnologia, Viale Rinaldo Piaggio 34, 56025 Pontedera, Italy cKet Lab, Edoardo Amaldi Foundation, Italian Space Agency, Via del Politecnico snc, 00133 Rome, Italy bSchool of Engineering and Materials Science, Queen Mary University of London, Mile End Road, London E1 4NS, United Kingdom Trento, Via Mesiano 77, 38123 Trento, Italy Abstract In the present work we study the spider's hair flow-sensing system by using fluid-structure interaction (FSI) numerical simulations. We observe experimentally the morphology of Theraphosa stirmi's hairs and characterize their mechanical properties through nanotensile tests. We then use the obtained information as input for the computational model. We study the effect of a varying air velocity and a varying hair spacing on the mechanical stresses and displacements. Our results can be of interest for the design of novel bio-inspired systems and structures for smart sensors and robotics. © 2018 Elsevier Ltd. All rights reserved. Selection and Peer-review under responsibility of 1st International Conference on Materials, Mimicking, Manufacturing from and for Bio Application (BioM&M). Keywords: sensing system; smart sensor; spider hair; numerical simulation; fluid-structure interaction 1. Introduction The ability of spiders in colonizing many habitats is also due to the fact that they interact efficiently with other animals for catching preys, escaping from predators and detecting potential partners also during the night. Spiders present a very complex and advanced sensing system based on tactile (or mechanical) [1-4], chemical [5,6] and air flow receptors [7-9] such as the hairs. These latter structures interact with the air velocity field and provide the spider with information regarding the surrounding environment [10,11], representing one of the most sensitive biosensors in Nature [12]. Since their extraordinary sensibility and efficiency, the interest in designing bio-inspired sensing systems and structures for soft robotics and high-tech applications has increased in the last decade [13-15]. Several numerical simulation works are available in the literature, but they are mainly related, for instance, to the engineering design and analysis of bio-inspired micro electro-mechanical systems (MEMS) (see, e.g., Ref. [16]). Numerical studies on the biological structures, in fact, are still poorly addressed and further investigations are needed in order to characterize the performance of spiders' hairs. * Corresponding author. Tel.: +39-0461-282525; fax: +39-0461-282599. E-mail address: [email protected] 2214-7853 © 2018 Elsevier Ltd. All rights reserved. Selection and Peer-review under responsibility of 1st International Conference on Materials, Mimicking, Manufacturing from and for Bio Application (BioM&M). 2 In this work, we employ fluid-structure interaction (FSI) simulations to study the behavior of spider's hairs in air. We investigate the effect of a varying flow velocity on the sensing capabilities, quantified in terms of displacements and von Mises stresses; and of a varying spacing of the hairs. The morphology of the hairs and their elastic properties are obtained experimentally, through Scanning Electron Microscopy (SEM), optical microscopy and nanotensile tests, and are used as inputs in the numerical simulations. 2. Materials and methods 2.1. Sample preparation The hairs are obtained from an exoskeleton of Theraphosa stirmi kept under controlled feeding and environmental conditions. The tested samples are prepared following the same procedure reported by Blackledge et al. [17]. We stick the hair samples on a paper frame provided with a square window of 5 mm side. The hair sample is fixed on the paper frame with a double-sided tape coupled with a glue. 2.2. Morphology and mechanical properties For the morphology characterization by SEM, we use a Zeiss Supra 40 (Carl Zeiss AG, Germany) at 2.30 kV. The metallization is made by using a sputtering machine Q150T (Quorum Technologies Ltd, UK) and the sputtering mode is Pt/Pd 80:20 for 5 minutes. In addition, we employ optical microscopy (Carl Zeiss AG, Germany) at 10x magnification for the extraction of an average value of the hair diameter. For the mechanical characterization, we use a T150 UTM nanotensile machine (Agilent Technologies Inc., USA) with a 500 mN load cell. The displacement speed is 10 µm s-1 with the frequency load at 20 Hz. The declared sensitivity of the machine is 10 nN for the load and 1 Å for the displacement in the dynamic configuration. 2.3. Numerical simulations The FSI simulations are carried out in COMSOL Multiphysics® [18]. We employ two-way coupling in order to have a comprehensive understanding of the interaction between fluid flow and mechanical displacements [19]. A single spider's hair is considered and it is approximated with a truncated cone of height L = 1000 µm, base diameter Db = 30 µm and tip diameter Dt = 15 µm. The computational domain for the fluid flow is composed of a 3D box of width 100 µm and height 1200 µm, with inlet and outlet positioned 1000 µm before and after the hair, respectively. Moreover, a chitin of thickness h = 100 µm is considered below the hair. We assign linear elastic material properties, using a Young's modulus E = 600 MPa (i.e. in the order of the measured value), a Poisson's ratio ν = 0.25 and a density ρh = 1425 kg m-3 (taken close to the respective values for chitin [20]). For the air, we consider its properties at 300 K, i.e. density ρa = 1.177 kg m-3 and dynamic viscosity µa = 1.85·10-5 kg m-1 s-1 [21]. A constant fluid velocity is imposed at the inlet and the domain is delimited by walls with full-slip boundary conditions. The whole domain is discretized with about 4.5·105 tetrahedral elements. The inlet velocity is chosen from 0.1 to 1 m s-1, which is in the range of the velocity field measured around flying insects [10]. The corresponding maximum von Mises stress σmax and maximum displacement max is measured. An additional analysis is carried out on the hair spacing, considering two hairs on the same transversal coordinate, but positioned from 50 to 1000 µm from each other. In this case, the differences in σmax and max are monitored, for a constant mean inlet velocity uavg = 1 m s-1. 3. Results 3.1. Morphology and mechanical properties Fig. 1 shows the SEM images of different hair typologies found on the spiders' exoskeleton. We can observe a typical length from a few hundreds of µm to 1-2 mm and a variable diameter, with a tip size usually in the range 1-20 3 µm. The typical hair spacing is a few mm, i.e. in the order of the hair length. Usually, shorter hairs not involved in flow-sensing are also present on the spider's cuticle and are not shown here, for brevity. The stress-strain curves extracted from the nanotensile test on different hairs are shown in Fig. A1. The mechanical properties of the hairs are intrinsically variable, because of their biological nature, and we have measured an initial Young's modulus in the range 500-800 MPa using linear regression, with a fracture strain in the order of 0.5 mm mm- 1. In Fig. A2, we show a Theraphosa stirmi of the collection and an optical microscopy investigation. The average diameter of the hairs is around 30 µm. Fig. 1. SEM morphology of different hairs found on the spider's exoskeleton. 3.2. Effect of a varying air flow Fig. 2 shows the values of σmax and max as function of the mean inlet velocity, as well as the von Mises stress distribution together with the qualitative flow streamlines. We can observe that both the stress and the displacement scale almost linearly with uavg. 4 Fig. 2. Maximum von Mises stress and maximum hair displacement as function of the mean inlet velocity. Inset: qualitative visualization of the displacement field of the structure and of the flow streamlines. A rough estimation of the mechanical load can be obtained by considering a beam, here approximated as a cylinder . We can assume that the fluid flow exerts a clamped at one end, with average diameter constant force per unit length, which, denoting by CD the dimensionless drag coefficient, is given by: (1) Neglecting the shear deformation, the displacement at the tip of the hair is given by: (2) where we have inserted the expression of the maximum bending moment, equal to , and the moment of inertia of the section, i.e. . For a cylinder immersed in a laminar flow, for low Reynolds numbers Re and considering constant fluid properties, the drag coefficient depends on the flow velocity roughly as [22]: (3) Therefore, inserting Eq. (3) into Eq. (2), we get the observed linear relationship between maximum hair displacement and mean inlet velocity, i.e. . An analogous reasoning can be made on the von Mises stress, except for 2avgbtDDD212DaDavgavgfCDu442max348aDDavgavgCLfLuEIED22DfL464avgID11ReDDavgavgCCuumaxavgu 5 the fact that its expression is more complicated because it must include, together with the bending stress, also a shear term and the stresses deriving from the substrate. For simplicity, considering only the maximum bending stress σb,max: (4) and again, from Eq. (3), we get a law of the type as shown in Fig. 2. Since Eq. (2) and Eq. (4) are obtained considering a rigid substrate, we can observe that they underestimate the maximum displacement at the hair tip and overestimate the maximum stress at the hair base, respectively. We can fit the values of the maximum displacement with a law of the type , obtaining K ≈ 1.943·10- 6 s (R2-value 0.9944). Consequently, the numerically-derived expression of the drag coefficient is: (5a) Considering the expression of the maximum bending stress in Eq. (4), instead, we can employ a law of the type for fitting, obtaining Kσ ≈ 67840 kg m-2 s-1 (R2-value 0.9917). In this case, the drag coefficient is: (5b) Both expression of CD are in the same order of magnitude of the data available in the literature, e.g. for perfect spheres or cylinders [22]. Furthermore, we can observe that Eq. (5a) can be considered as a lower bound, while Eq. (5b) as an upper bound, because of the rigid substrate assumption described above. This result represents a good estimation of the aerodynamic performance of the spider's hair at low Re, if considering the introduced approximation (i.e. cylindrical instead of conical geometry and rigid constraint) and the presence of the hair tip (i.e. finite-length body). Note that a more general power law can be employed for fitting, e.g. for the displacement, and we get K2 ≈ 1.125 (R2-value 0.9999) thus the linear law assumed above can be considered sufficiently precise. Usually sensors are based on linear responses, since only one parameter is necessary to correlate the quantity to measure to an electric signal, and we observe that the spider's sensing system present a similar behavior. Thus, this effect can be particularly interesting for the design of new sensors and MEMS for fluid flow measurements. 3.3. Effect of a varying hair spacing The effect of the hair spacing, here denoted with Δs, is quantified by monitoring the response of the second hair with respect to the first one. Specifically, we measure the variation of the maximum displacement at the hair tip and of the maximum von Mises stress at the hair base. As shown in Fig. 3, for small values of Δs the effect of the aerodynamic wake behind the first hair is to reduce the stresses and displacements to which the second hair is subjected. Therefore, for small values of the spacing there is a significant variation of σmax and max, while it becomes negligible (or, at least, acceptable for the sensing system, being below 10%) when Δs is in the order of half-length of the hair (i.e. around 500 µm for the two 1000 µm hairs considered here). Note that this result has been obtained considering the maximum flow velocity, i.e. uavg = 1 m s-1. For lower values of uavg, the variations Δσmax and Δmax are expected to be lower due to a shorter aerodynamic wake, thus the sensing system is effective also for lower values of the hair spacing. This result suggests that the relative position of flow-sensing hairs is probably dictated by the mutual interference. Therefore, we have observed that there is an optimal configuration to maximize the spider's sensing capabilities: this happens when there is the maximum density of hairs (i.e. number of hairs per unit length of the substrate) capable of sensing a certain level of stress. This optimal density of hairs is compatible to what is observed experimentally, as shown in Fig. A3. 222,max2822avgaDDbavgavgDCLfLuID,maxbavgumaxavgKu112.7ReDC,maxbavgKu116.4ReDC2max1KavgKu6 Fig. 3. Difference on the maximum von Mises stress and the maximum displacement between the first and the second hair as function of the hair spacing. Inset: schematic view of the considered geometry. 4. Conclusions We have used FSI numerical simulations to investigate the behavior of Theraphosa stirmi's hairs under a fluid flow. We have extracted the morphology and the elastic properties of the hairs from experimental tests. The simulation study has allowed to compute the mechanical response of the hair under a varying flow velocity, suggesting a linear relationship between maximum displacement and mean inlet velocity, as well as between maximum von Mises stress and uavg. In addition, we have considered the spacing between two hairs, showing that when the spacing is low the wake of the first hair affects the sensing capability of the second one. Therefore, this suggests that the optimal hair spacing in spiders might derive from fluid dynamic reasons and from the need of maximizing the sensing capabilities. These results can be of interest for the optimal design of novel sensors and MEMS for fluid flow measurements. Acknowledgements RG is supported by Bonfiglioli Riduttori SpA. NMP is supported by the European Commission under the Graphene Flagship Core 2 No. 785219 (WP14 "Polymer Composites") and FET Proactive "Neurofibres" grant No. 732344. References [1] R.F. Foelix, I.-W. Chu-Wang, Tissue Cell 5 (1973) 451-460. [2] J.T. Albert, O.C. Friedrich, H.-E. Dechant, F.G. Barth, J. Comp. Physiol. A 187 (2001) 303-312. [3] P. Fratzl, F.G. Barth, Nature 462 (2009) 442-448. [4] F.G. Barth, Curr. Opin. Neurobiol. 14 (2004) 415-422. [5] R.F. Foelix, J. Morphol. 132 (1970) 313-333. [6] R.F. Foelix, I.-W. Chu-Wang, Tissue Cell 5 (1973) 461-468. [7] B. Bathellier, F.G. Barth, J.T. Albert, J.A.C. Humphrey, J. Comp. Physiol. A 191 (2005) 733-746. 7 [8] J.A.C. Humphrey, F.G. Barth, Adv. Insect Physiol. 34 (2007) 1-80. [9] A.-S. Ganske, G. Uhl, Arthtropod Struct. Dev. 47 (2018) 144-161. [10] C. Klopsch, H.C. Kuhlmann, F.G. Barth, J. R. Soc. Interface 9 (2012) 2591-2602. [11] C. Klopsch, H.C. Kuhlmann, F.G. Barth, J. R. Soc. Interface 10 (2013) 20120820. [12] F.G. Barth, Naturwissenschaften 87 (2000) 51-58. [13] M. Dijkstra, J.J. van Baar, R.J. Wiegerink, T.S.J. Lammerink, J.H. de Boer, G.J.M. Krijnen, J. Micromech. Microeng. 15 (2005) S132-S138. [14] G.J.M. Krijnen, M. Dijkstra, J.J. van Baar, S.S. Shankar, W.J. Kuipers, R.J.H. de Boer, D. Altpeter, T.S.J. Lammerink, R. Wiegerink, Nanotechnology 17 (2006) S84-S89. [15] J. Casas, T. Steinmann, G. Krijnen, J. R. Soc. Interface 7 (2010) 1487-1495. [16] B. Yang, D. Hu, L. Wu, Sensors 16 (2016) 1056. [17] T.A. Blackledge, J.E. Swindeman, C.Y. Hayashi, J. Exp. Biol. 208 (2005) 1937-1949. [18] COMSOL Multiphysics® v. 5.2a. www.comsol.com. COMSOL AB, Stockholm, Sweden. [19] F.-K. Benra, H.J. Dohmen, J. Pei, S. Schuster, B. Wan, J. Appl. Math. 2011 (2011) 853560. [20] H. Fabritius, C. Sachs, D. Raabe, S. Nikolov, M. Friák, J. Neugebauer, in: N.S. Gupta (Ed.), Chitin: Formation and Diagenesis, Springer Science+Business Media B.V., Dordrecht, 2011, pp. 35-60. [21] Dry Air Properties, Engineering ToolBox (2005) [online], https://www.engineeringtoolbox.com/dry-air-properties-d_973.html. [22] J.D. Anderson, Fundamentals of Aerodynamics, second ed., McGraw-Hill, New York, 1991. Appendix A. Mechanical and optical characterization Fig. A1. Example stress-strain curves of different Theraphosa stirmi's hairs obtained through nanotensile tests. Inset: example linear fit on the first part of a curve, with extracted Young's modulus E ≈ 705.8 MPa (R2-value = 0.921). 8 Fig. A2. Left: a picture of a Theraphosa stirmi of the collection. Right: optical microscope image of a hair. Fig. A3. Picture of a Theraphosa stirmi's leg with flow-sensing hairs and typical spacing in the order of the half-length of a hair (white arrows).
1801.01376	1	1801	2018-01-04T14:49:16	A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing	[ "physics.bio-ph", "physics.med-ph" ]	Classical homogenization theory based on the Hashin-Shtrikman coated ellipsoids is used to model the changes in the complex valued conductivity (or admittivity) of a lung during tidal breathing. Here, the lung is modeled as a two-phase composite material where the alveolar air-filling corresponds to the inclusion phase. The theory predicts a linear relationship between the real and the imaginary parts of the change in the complex valued conductivity of a lung during tidal breathing, and where the loss cotangent of the change is approximately the same as of the effective background conductivity and hence easy to estimate. The theory is illustrated with numerical examples, as well as by using reconstructed Electrical Impedance Tomography (EIT) images based on clinical data from an ongoing study within the EU-funded CRADL project. The theory may be potentially useful for improving the imaging algorithms and clinical evaluations in connection with lung EIT for respiratory management and monitoring in neonatal intensive care units.	physics.bio-ph	physics	A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing Sven Nordebo1, Mariana Dalarsson1, Davood Khodadad1, Beat Muller2, Andreas Waldman2, Tobias Becher3, Inez Frerichs3, Louiza Sophocleous4, Daniel Sjoberg5, Nima Seifnaraghi6, Richard Bayford6 1 Department of Physics and Electrical Engineering, Linnaeus University, 351 95 Vaxjo, Sweden. E-mail: {sven.nordebo,mariana.dalarsson,davood.khodadad}@lnu.se. 2 Swisstom AG, Schulstrasse 1, CH-7302 Landquart, Switzerland. E-mail: {bmu,awa}@swisstom.com 3 Department of Anaesthesiology and Intensive Care Medicine, University Medical Centre Schleswig-Holstein, Campus Kiel, 24105 Kiel, Germany. E-mail: {Tobias.Becher,Inez.Frerichs}@uksh.de. 4 The KIOS Research Center, Department of Electrical and Computer Engineering, University of Cyprus, Nicosia, Cyprus. E-mail: [email protected]. 5 Department of Electrical and Information Technology, Lund University, Box 118, 221 00 Lund, Sweden. E-mail: [email protected]. 6 Department of Natural Sciences, Middlesex University, Hendon campus, The Burroughs, London, NW4 4BT, United Kingdom. E-mail: [email protected], [email protected]. 8 1 0 2 n a J 4 ] h p - o i b . s c i s y h p [ 1 v 6 7 3 1 0 . 1 0 8 1 : v i X r a A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 2 Abstract. Classical homogenization theory based on the Hashin-Shtrikman coated ellipsoids is used to model the changes in the complex valued conductivity (or admittivity) of a lung during tidal breathing. Here, the lung is modeled as a two-phase composite material where the alveolar air-filling corresponds to the inclusion phase. The theory predicts a linear relationship between the real and the imaginary parts of the change in the complex valued conductivity of a lung during tidal breathing, and where the loss cotangent of the change is approximately the same as of the effective background conductivity and hence easy to estimate. The theory is illustrated with numerical examples, as well as by using reconstructed Electrical Impedance Tomography (EIT) images based on clinical data from an ongoing study within the EU-funded CRADL project. The theory may be potentially useful for improving the imaging algorithms and clinical evaluations in connection with lung EIT for respiratory management and monitoring in neonatal intensive care units. 1. Introduction spatial it benefits Electrical Impedance Tomography (EIT) is a non- radiative, inexpensive technique that can facilitate real time dynamic monitoring of regional lung aeration and ventilation for clinical use [1]. The approach lacks largely from its high temporal resolution and is therefore currently emerging as a technique that can potentially reduce complications and disability in preterm babies by continuous bedside monitoring and respiratory management [2, 3]. resolution, but From a mathematical/physical point of view, EIT constitutes an ill-posed inverse problem [4–7], and the use of EIT as a successful imaging modality relies therefore on the effectiveness of creating difference conductivity images rather than to generate absolute reconstructions, see e.g., [8, 9]. The difference imaging approach benefits from the linearization of the forward problem, and it alleviates much of the sensitivity to sensor imperfections as well as the unknown background parameter values. In lung EIT, tidal images can be created by using a breath detector [10] providing difference voltage data that is perfectly synchronized with the time of end-expiration and end- inspiration, defining the breathing cycle. For these tidal images, EIT related lung function parameters such as Center of Ventilation, Silent Spaces and ventilation distribution, etc., can then be calculated [3]. Early EIT systems were usually designed for operation at very low frequencies, typically in the kilohertz range [11, 12], where the conductivity of biological tissue usually is considered to be purely resistive (even though this is not entirely true [13]). However, the need of improved image quality in both the spatial and the temporal domains is nowadays driving the development of the commercial EIT systems and their data acquisition hardware to higher speeds (typically in the range of several hundreds of kilohertz) and hence their performance requirements imply that the imaginary part of the complex valued conductivity (or admittivity) of human tissue no longer can be disregarded. Notably, this constitutes a technical challenge, but it can also be a great asset due to the additional clinical information carried by the permittivity of the tissue. To this end, there is no fundamental limitation associated with the reconstruction of the complex valued conductivity using standard optimization algorithms such as in [4,5,7,9,11], nor with the new developments such as the D-bar methods [14, 15]. In this paper, we investigate a mathematical/physical mechanism that can model the changes in the complex valued conductivity of a lung during tidal breathing. This is particularly interesting as it is already empirically known that the EIT pixel sum of (real valued) conductivity changes within a particular region of interest is almost linearly related to the changes in lung volume, see e.g., [12]. tissue, A comprehensive overview on the dielectric spectral properties of biological including modeling, measurements and literature is given in [13, 16, 17]. In particular, as a realistic estimate of the background conductivity of an inflated lung we have used here the real valued conductivity and permittivity data from [13, Fig. 2e on p. 2257] which is of bovine origin. Measurement data from human tissue, and in particular from neonatal patients is obviously very difficult to obtain. Hence, in order to estimate the typical volume fraction of air during tidal breathing we have used data regarding the lung volume of an adult male and the corresponding condensed matter weight from [18] and [19], respectively. A basic model predicting the dielectric properties of lung tissue as a function of air content was given in [20], and which has been followed by several other works, see e.g., [21] and [22] with references. The basic work in [20] was followed by a comprehensive study based on EIT spectroscopy measurements and an improved model taking both the air content and tissue dispersion into account [21]. In [22] is given an experimental study of dielectric properties of human A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 3 lung tissue in vitro and its dependency on the air filling factor. Human lung tissue specimens from more than 100 patients were investigated with respect to the differences in the impedance spectrum for cancerous and normal tissue, and where the cancerous tissue typically show much larger conductivity values due to the deterioration of the alveolar structure. The same can also be said about interstitial pneumonia [20], where the increased conductivity is explained by the increased thickness of the alveolar walls. Notably, the investigations on the dielectric properties of lung tissue as a function of air content in [20] and [22] are in vitro, with volume fractions of air up to about 58-60 %, whereas our estimates of the air content with reference to [18] and [19] are rather in the range 75-78 % for tidal breathing in vivo. The dielectric measurements in [20] were made on the excised lungs of slaughtered calves in the frequency range of 5 kHz to 100 kHz, and a simple theory was developed to model the complex valued conductivity as a function of air filling. The theoretical model in [20] is based on the deformation of a cube-shaped alveolus where the conductivity is given approximately as the conductivity of the enclosing walls of a square cylinder where the volume of the (periodic) wall system is kept fixed and the cube size and the wall thickness are variable to account for the deformation of the epithelial cells and blood vessels through the expansion of the alveoli. The parameters of the model are adjusted by histological investigations. In this way, the thinning of the alveolar walls can explain the decrease of effective conductivity and permittivity of the lung as a function of an increased air filling. In this paper, we derive an alternative physical model based on classical homogenization theory [23, 24] to predict the changes in the complex valued conductivity of a lung during tidal breathing. The lung is modeled here as a two-phase composite material where the alveolar air-filling corresponds to the inclusion phase and the exterior phase is due to the blood and tissue. The parametric model is based on classical Hashin-Shtrikman/Maxwell-Garnett theory [23, 24] and is simple and well-suited for an analytical study on the changes in the effective conductivity of the lung due to the corresponding small changes in the volume fraction of air during tidal breathing. This model is in many ways similar to that of (an effective conductivity model with thinning of the alveolar walls, etc.), but it has a rigorous foundation in homogenization theory. In particular, the parametric model based on the Hashin-Shtrikman (HS) assemblage [24] comprises a fully three-dimensional homogenization of the lung tissue with ellipsoidal shaped alveoli constituting the inclusion phase. The HS assemblage furthermore [20] for the random nature of the alveolar accounts [20, Figs. 14 and 15 on structure, as seen in e.g., p. 711], by the inherent variation of the HS scaling and positioning of the prototype ellipsoid. Hence, it may be argued from the figures in [20] that the shape of the alveoli is typically more round than square and that the close packing of alveoli at high air content can be achieved due to a large variation in their sizes, similar to the HS assemblage. The presented parametric model predicts an almost linear relationship between the real and the imaginary parts of the changes in the complex valued conductivity of a lung during tidal breathing, and where the loss cotangent of the change is approximately the same as of the effective background conductivity and hence easy to estimate. It is expected that this a priori knowledge can be useful in the development of new improved image reconstruction algorithms exploiting complex valued measurement data, and/or to define new clinically useful outcome parameters in lung EIT. The theoretical study is based on realistic parameter choices for the conductivity of an inflated lung [13], and hence the corresponding loss cotangent is estimated to be in the order of about cot δ = 0.2 at 200 kHz. It should be noted that, even at low frequencies, the loss cotangent of the inflated lung may not be negligible (due to the very high relative permittivity of tissue at low frequencies), and is expected to be in the order of about cot δ = 0.1 at 1 kHz. At higher frequencies, it will increase to about cot δ = 0.3 at 1 MHz [13]. The theoretical study is illustrated with numerical examples, as well as with clinical data from the ongoing EU Horizon 2020 CRADL (Continuous Regional Analysis Device for neonate Lung) project registered at ClinicalTrials.gov (NCT02962505). 2. Homogenization theory based on Hashin-Shtrikman coated ellipsoids A brief review on the classical homogenization theory based on Hashin-Shtrikman coated ellipsoids is given in this section, see [24] for an in depth derivation of the corresponding results. 2.1. Notation and conventions The following notation and conventions will be used below. Classical electromagnetic theory is considered based on SI-units [25], and with time convention ejωt for time harmonic fields where ω is the angular frequency. Let µ0, 0 and c0 denote the permeability, the permittivity and the speed of light in vacuum, respectively, and where c0 = 1/√µ00. A passive, homogeneous and isotropic dielectric material with complex valued relative permittivity = (cid:48) − j(cid:48)(cid:48) A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 4 and real valued conductivity σr ≥ 0 has complex valued conductivity (or admittivity) σ given by σ = σR + jσI where σI = ω0(cid:48) represents the lossless capacitive part and σR = σr + ω0(cid:48)(cid:48) ≥ 0 includes both the conduction and the dielectric losses. The complex valued conductivity is conveniently written as σ = σR (1 + jη) where η = cot δ is the loss cotangent (corresponding to the more commonly used loss tangent tan δ = σR/σI = 1/η). Both parameters σR and σI may depend on frequency, in which case they must also satisfy the associated Kramers- Kronig relations, see e.g., [26, 27]. The cartesian unit vectors are denoted ( x1, x2, x3). Finally, the real and imaginary part and the complex conjugate of a complex number ζ are denoted (cid:60){ζ}, (cid:61){ζ} and ζ∗, respectively. 2.2. The Hashin-Shtrikman coated ellipsoid assemblage Figure 1. The Hashin-Shtrikman coated ellipsoid assemblage. The figure illustrates a partial assemblage in a process that is completed when all space has been filled with ellipsoids. Here, Ej denotes the applied electric field, σeff the homogenized j (effective) conductivity parameter and σ1 and σ2 the core and the exterior conductivity of the prototype ellipsoid, respectively. Consider a general anisotropic two-phase compos- ite material consisting of an inclusion phase with con- ductivity σ1 and an exterior phase with conductivity σ2 and where the volume fraction of the inclusion phase is given by the parameter f . A classical homogeniza- tion approach to model such a material is given by the Hashin-Shtrikman coated ellipsoid assemblage [24], as illustrated in figure 1. All the coated ellipsoids are scaled and translated versions of a single prototype coated ellipsoid consisting of a core (inclusion) phase with conductivity σ1 and an exterior (coating) phase with conductivity σ2. The semi-axis lengths of the pro- totype core and exterior ellipsoids are denoted lcj and lej , respectively, and where j = 1, 2, 3 refers to the cartesian coordinates x1, x2 and x3, respectively. The prototype core and exterior ellipsoids are confocal in the sense that they can both be represented in the same system of elliptical coordinates as + + = 1, x2 3 c2 3 + ρ x2 2 c2 2 + ρ x2 1 c2 1 + ρ where cj are constants and ρ the elliptical coordinate playing the role of the "radius". Hence, with ρc and ρe denoting the "radius" of the two confocal ellipsoids, we have (1) = (2) = 1, cj + α, l2 l2 ej cj c2 c2 j + ρe j + ρc which implies that l2 ej = l2 (3) where α = ρe − ρc > 0 and j = 1, 2, 3. The volume of the two prototype ellipsoids are Vc = 4πlc1lc2lc3 /3 and Ve = 4πle1le2le3 /3, and hence the volume fraction between the core phase and the total volume of the coated ellipsoid is given by f = Vc Ve = lc1 lc2lc3 le1 le2le3 . (4) The Hashin-Shtrikman coated ellipsoid assemblage is obtained when the whole space is filled with scaled ellipsoids as indicated in figure 1. In this way, the resulting assemblage models the general anisotropic two-phase composite material where the volume fraction f of the inclusion phase is preserved and parameterized according to (4). 2.3. The effective conductivity Consider a single prototype coated ellipsoid with core and exterior conductivities σ1 and σ2, respectively, as described in section 2.2 above. Suppose further that the prototype ellipsoid is embedded in an arbitrary homogeneous auxiliary medium with conductivity σeff , and excited with an external static electric field Ej = E0 xj aligned with the jth axis of the ellipsoid, as illustrated in figure 1. The fundamental equations to be solved are given by ∇ × E(r) = 0, ∇ · J (r) = 0, J (r) = σ(r)E(r), (5)  where E(r) and J (r) are the electric field intensity and the electric current density, respectively, and where σ(r) is the complex valued conductivity which is assigned the appropriate constant values (σ1, σ2, σeff ) inside and outside the prototype ellipsoid, respectively. The electric field outside the prototype ellipsoid may furthermore be denoted E(r) = Ej(r) + Es(r) where Aparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing2Abstract.ClassicalhomogenizationtheorybasedonHashin-Shtrikmancoatedellipsoidsisusedtomodelthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing.Atwo-phasecompositematerialmodelisconsideredwherethealveolarair-fillingcorrespondstotheinclusionphase.Thetheorypredictsalinearrelationshipbetweentherealandtheimaginarypartsofthechangesinconductivityofthelungduringtidalbreathing,andwhichisillustratedbynumericalexamples.ExperimentaldatafromtheCRADLprojectisalsousedinacomparisonwiththetheoreticalprediction.1.IntroductionTobewritten.2.HomogenizationtheorybasedonHashin-ShtrikmancoatedellipsoidsInthissectionisgivenabriefreviewontheclassicalhomogenizationtheorybasedonHashin-Shtrikmancoatedellipsoids,see[1]foranindepthderivationofthecorrespondingresults.2.1.NotationandconventionsThefollowingnotationandconventionswillbeusedbelow.ClassicalelectromagnetictheoryisconsideredbasedonSI-units[2],andwithtimeconventionej!tfortimeharmonicfieldswhere!istheangularfrequency.Letµ0,✏0andc0denotethepermeability,thepermittivityandthespeedoflightinvacuum,respectively,andwherec0=1/pµ0✏0.Apassive,homogeneousandisotropicdielectricmaterialwithcomplexvaluedrelativepermittivity✏=✏0j✏00andrealvaluedconductivityr0hascomplexvaluedconductivitygivenby=R+jIwhereI=!✏0✏0representsthelosslesscapacitivepartandR=r+!✏0✏000includesboththeconductionandthedielectriclosses.Thecomplexvaluedconductivityisconvenientlywrittenas=R(1+j⌘)where⌘isthelosscotangent(correspondingtothecommonlyuseddielectriclosstangenttan=R/I=1/⌘).BothparametersRandImaydependonfrequency,inwhichcasetheymustalsosatisfytheassociatedKramers-Kronigrelations,seee.g.,[3,4].Thecartesianunitvectorsaredenoted(x1,x2,x3).Finally,therealandimaginarypartandthecomplexconjugateofacomplexnumber⇣aredenoted<{⇣},={⇣}and⇣⇤,respectively.2.2.TheHashin-ShtrikmancoatedellipsoidassemblageEje↵j21Figure1.TheHashin-Shtrikmancoatedellipsoidassemblage.Thefigureillustratesapartialassemblageinaprocessthatiscompletedwhenallspacehasbeenfilledwithellipsoids.Here,Ejdenotestheappliedelectricfield,e↵jthehomogenized(e↵ective)conductivityparameterand1and2thecoreandtheexteriorconductivityoftheprototypeellipsoid,respectively.Considerageneralanisotropictwo-phasecompos-itematerialconsistingofaninclusionphasewithcon-ductivity1andanexterior(matrix)phasewithcon-ductivity2andwherethevolumefractionofthein-clusionphaseisgivenbytheparameterf.AclassicalhomogenizationapproachtomodelsuchamaterialisgivenbytheHashin-Shtrikmancoatedellipsoidassem-blage[1],asillustratedinfigure1.Allthecoatedellip-soidsarescaledandtranslatedversionsofasinglepro-totypecoatedellipsoidconsistingofacore(inclusion)phasewithconductivity1andanexterior(coating)phasewithconductivity2.Thesemi-axislengthsoftheprototypecoreandexteriorellipsoidsaredenotedlcjandlej,respectively,andwherej=1,2,3referstothecartesiancoordinatesx1,x2andx3,respectively.Theprototypecoreandexteriorellipsoidsareconfocalinthesensethattheycanbothberepresentedinthesamesystemofellipticalcoordinatesasx21c21+⇢+x22c22+⇢+x23c23+⇢=1,(1)wherecjareconstantsand⇢theellipticalcoordinateplayingtheroleofthe'radius'.Hence,with⇢cand⇢eA parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 5 Es(r) may be thought of as the scattered field. The boundary conditions supplementing the equations in (5) are obtained from the continuity of the normal component of the current density J (r) at the media interfaces. The equations in (5) can be solved by introducing the scalar potential Φ(r) where E(r) = −∇Φ(r), and where Φ(r) satisfies the Laplace equation ∇2Φ(r) = 0, together with the continuity of Φ(r) as well as the continuity of the normal current σ(r) ∂ ∂n Φ(r) at the media interfaces. Since the elliptical coordinates constitute one of the very special coordinate systems for which the Laplace operator can be separated [28], the electrostatic problem above can be solved analytically involving ordinary functions and elliptic integrals, see e.g., [24, 28, 29]. In particular, by investigating these analytic solutions, it is found that one can choose the auxiliary medium parameter σeff in such a way as to render the scattered field Es(r) = 0 (and at the same time there is a uniform non-zero field inside the ellipsoid core with the same polarization direction as the applied field). Hence, in this situation the prototype ellipsoid is in some sense cloaked as it does not interfere with the surrounding uniform current field. The resulting auxiliary, or effective, medium parameter σeff j depend in general on the polarization direction xj, and is given by σeff j = σ2 + f σ2 (σ1 − σ2) σ2 +(cid:0)dcj − f dej(cid:1) (σ1 − σ2) where the depolarizing factors (cf., the geometrical factors in [29], and the demagnetizing factors in [30]) dcj and dej are given by dcj = dj(lc1, lc2, lc3 ), dej = dj(le1, le2, le3 ) where , (6) dj(l1, l2, l3) l1l2l3 = 2 (cid:90) ∞ 0 dy 1 + y) (l2 2 + y) (l2 3 + y) , (7) j + y(cid:1)(cid:112)(l2 (cid:0)l2 and where l1, l2 and l3 are the semi-axis lengths of the corresponding ellipsoids, and j = 1, 2, 3, see [24, p. 129]. The depolarizing factors are normalized in the sense that d1 + d2 + d3 = 1. Since further scaled and translated coated ellip- soids can be inserted into the effective medium with- out disturbing the surrounding uniform current field, the resulting Hashin-Shtrikman coated ellipsoid assem- blage can finally be viewed from a macroscopic scale to have the homogeneous and anisotropic constitutive re- lation J = σeff · E, (8) where the effective conductivity dyadic σeff is given by 3(cid:88)j=1 σeff j xj xj. σeff = (9) The coated sphere is a special case of the coated ellipsoid with dcj = dej = 1/3 for j = 1, 2, 3, and hence σeff sph = σ2 + 3f σ2 (σ1 − σ2) 3σ2 + (1 − f ) (σ1 − σ2) , (10) which is identical with the classical Maxwell-Garnett mixing formula, see e.g., [23, 24]. 2.4. Spheroidal inclusions Spheroids are particularly simple ellipsoidal shapes having rotational symmetry, and for which the corresponding depolarizing factors as well as their surface areas can be expressed by explicit formulas involving simple functions. Without loss of generality, it will be assumed here that the axis of rotation is defined by the x1-axis. A prolate spheroid is characterized by its semi- axis properties l1 > l2 = l3 = l, and hence with depolarizing factors d1 < d2 = d3 = d where d1 < 1/3 and d = (1 − d1)/2. The eccentricity of the prolate spheroid is defined by , (11) ε =(cid:115)1 −(cid:18) l l1(cid:19)2 ε2 (cid:18) 1 1 − ε2 d1 = 2ε and the corresponding depolarizing factor d1 is given by ln(cid:18) 1 + ε 1 − ε(cid:19) − 1(cid:19) , (12) cf., e.g., [24, p. 132] and [30, pp. 352–354]. The surface area of the prolate spheroid is furthermore given by S = 2πl2 + 2π ll1 ε see [31, p. 364]. arcsin ε, (13) An oblate spheroid is characterized by its semi- axis properties l1 < l2 = l3 = l, and hence with depolarizing factors d1 > d2 = d3 = d where d1 > 1/3 and d = (1 − d1)/2. The eccentricity of the oblate spheroid is defined by , (14) and the corresponding depolarizing factor d1 is given by ε =(cid:115)1 −(cid:18) l1 l(cid:19)2 ε2(cid:32)1 − d1 = 1 √1 − ε2 ε arcsin ε(cid:33) , cf., e.g., [24, p. 132] and [30, pp. 352–354]. The surface area of the oblate spheroid is furthermore given by S = 2πl2 + π l2 1 ε see [31, p. 364]. ln(cid:18) 1 + ε 1 − ε(cid:19) , (15) (16) A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 6 3. A parametric model for the changes in the conductivity of a lung during tidal breathing the changes is considered to model A Hashin-Shtrikman homogenization approach based on (6) in the complex valued conductivity of a lung during tidal breathing. The lung is modelled as a two- phase composite material where the air-filled alveoli constitute the inclusion phase with volume fraction f and conductivity σ1 = jω0 corresponding to the electric displacement current in vacuum (similar as in air). The conductivity σ2 of the exterior phase can be identified from some a priori information regarding the effective conductivity σEI of an inflated lung. Hence, it is assumed that the conductivity of the inflated lung obtained from measurements such as in [13] can be used as an effective conductivity of the lung corresponding to a certain maximum volume fraction fEI at end- inspiration, and where the alveoli have a maximally extended spherical shape. The conductivity of the exterior phase can then be obtained by solving the Maxwell-Garnett equation (10) with respect to σ2 when the effective parameter σeff sph = σEI is given. This is equivalent to finding the roots of the following second order polynomial equation σ2 22 (1 − fEI) + σ2 (σ1 (1 + 2fEI) − σEI (2 + fEI)) −σEI (1 − fEI) σ1 = 0, (17) and where the root of physical interest has (cid:60){σ2} > 0 and (cid:61){σ2} > 0 (unless the exterior phase is a Drude material with (cid:61){σ2} < 0, etc.). Tidal breathing is then considered with small changes in alveolar air-filling where the corresponding volume fraction f changes from its maximum value fEI at end-inspiration to its minimum value fEE at end-expiration. Two fundamentally different physical modes of alveolar air-fillings are considered for the tidal breathing. • Spherical shaped alveoli with fixed shape and varying surface area: The alveoli are assumed to have a fixed spherical shape and the volume change is obtained by a change of its radius. This implies that the surface area of the alveoli as well as the whole structure of the lung is stretched during the tidal breathing. • Spheroidal shaped alveoli with varying shape and fixed surface area: The alveoli are assumed to have a varying spheroidal shape and the volume change is obtained by a change of its spheroidal eccentricity while keeping its surface area fixed. In this mode, the volume change of the alveoli is due solely to a change in the alveolar shape (prolongation or flattening of the spheroid) with a minor stretch in the lung structure. With a tidal breathing based on spherical shaped alveoli (fixed shape and varying surface area) the change in conductivity is obtained from (10) as ∆σeff sph(fEI) where f denotes the volume fraction of air-filled alveoli during tidal breathing and fEI the corresponding value at end- inspiration. Hence sph(f ) − σeff sph = σeff ∆σeff sph = 3f σ2 (σ1 − σ2) 3σ2 + (1 − f ) (σ1 − σ2) 3fEIσ2 (σ1 − σ2) , 3σ2 + (1 − fEI) (σ1 − σ2) − (18) where f = fEI − ∆f and ∆f > 0. For a comparison with the spheroidal case below, the Hashin-Shtrikman prototype core sphere is defined to have unit radius at maximal volume fraction fEI corresponding to the surface area S0 = 4π. j = σeff With a tidal breathing based on spheroidal shaped alveoli (varying shape and fixed surface area) the change in conductivity is obtained from (6) and defined by ∆σeff j (fEI). The prototype core spheroids are furthermore assumed to be unit spheres at the maximal volume fraction fEI at end-inspiration, and hence from (6) and (10) f σ2 (σ1 − σ2) j (f ) − σeff ∆σeff j = σ2 +(cid:0)dcj − f dej(cid:1) (σ1 − σ2) 3fEIσ2 (σ1 − σ2) , (19) 3σ2 + (1 − fEI) (σ1 − σ2) − where f = fEI − ∆f and ∆f > 0. Here, dcj and dej are the depolarizing factors of the Hashin- Shtrikman prototype core and exterior spheroids at volume fraction f , respectively. Since the Hashin- Shtrikman prototype core spheroid coincides with the unit sphere at maximal volume fraction fEI, the following relation is obtained from (4) fEI = 1 le1le2 le3 , (20) is proportional to the where the product le1 le2le3 volume of the prototype exterior spheroid. Finally, the eccentricity ε of the prototype core spheroid as defined in (11) or (14) is used as a parameter to control the shape of the spheroid, as well as its volume fraction f < fEI. 3.1. The prototype core spheroids Consider a prolate core spheroid with lc1 being the length of the semi-axis of rotation and lc = lc2 = lc3 the length of the orthogonal axes. Let the spheroidal eccentricity = √1 − t2 be fixed, where t = lc/lc1, 0 < t < 1 and 0 < ε < 1. The surface area of the prototype core spheroid is fixed at S = 4π, and hence (13) yields the equation 2πl2 c + 2π lclc1 ε arcsin ε = 4π, (21) A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 7 and which can be solved for lc to yield 3.3. Sensitivity analysis for high contrast inclusions 2εt εt + arcsin ε , (22) A first order Taylor series approximation of the conductivity changes in (18) and (19) is given by lc =(cid:114) and lc1 = lc/t. The corresponding depolarizing factor dc1 is given by (12) and dc = (1 − dc1 )/2. Consider similarly an oblate core spheroid with lc1 being the length of the semi-axis of rotation and lc = lc2 = lc3 the length of the orthogonal axes. Let the spheroidal eccentricity = √1 − t2 be fixed, where t = lc1 /lc, 0 < t < 1 and 0 < ε < 1. The surface area of the prototype core spheroid is fixed at S = 4π, and hence (16) yields the equation (23) (24) 2πl2 c + π l2 c1 ε ln(cid:18) 1 + ε 1 − ε(cid:19) = 4π, and which can be solved for lc to yield lc =(cid:118)(cid:117)(cid:117)(cid:116) 4ε 2ε + t2 ln(cid:16) 1+ε 1−ε(cid:17) , and lc1 = tlc. The corresponding depolarizing factor dc1 is given by (15) and dc = (1 − dc1 )/2. 3.2. The prototype external spheroids By increasing the eccentricity ε > 0 of the prototype core spheroid while keeping its surface area fixed, its volume will decrease. Hence, by defining the volume of the prototype exterior spheroid Ve = 4πle1le2le3 /3 to be fixed, the corresponding volume fraction f of the inclusion phase will decrease. The volume fraction f for a given eccentricity ε is hence obtained from (4) and (20) as f = fEIlc1 l2 c . (25) To find the semi-axes of the prototype exterior spheroid at volume fraction f , the relations (3) of the confocal ellipsoids are now inserted into the following equation based on the definition (4) f 2l2 (26) e2l2 e3 = l2 c1l2 c2 l2 c3, e1l2 and which is equivalent to finding the real and positive root of the algebraic equation c1l2 c1 + 2l2 c + l4 α3 + α2(cid:0)l2 c(cid:18)1 − c1l4 +l2 c(cid:1) + α(cid:0)2l2 f 2(cid:19) = 0. 1 c(cid:1) (27) Once the correct real valued and positive root α has been identified, the semi-axes lengths lej are given by (3). The eccentricity and the depolarizing factors dej of the prototype exterior spheroid are now given by (11) and (12) for the prolate spheroid, or by (14) and (15) for the oblate. ∆σeff j ≈ dσeff j df (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)f =fEI df, (28) where σeff is given by (6), and where the spherical case j (18) is a special case of (19) for which dcj = dej = 1/3 and (10) is used. Note that in the spheroidal case as defined above, the depolarizing factors dcj and dej depend on f via ε. σeff Assume that the conductivity σ1 of the inclusion phase is very small and negligible in comparison to the conductivity σ2 of the exterior phase. The exact expression (6) can then be approximated by j ≈ σ2(cid:32) 1 − dcj + f(cid:0)dej − 1(cid:1) 1 + f dej − dcj j and its derivative (cid:33) . df (cid:40) 1 − dcj + f(cid:0)dej − 1(cid:1) Hence, both σeff 1 + f dej − dcj dσeff j df ≈ σ2 are proportional to the complex valued conductivity σ2 of the exterior medium, and where the constant of proportionality is real valued. in the spherical case the following simple expressions are obtained In particular, (cid:41) , (30) (29) d σeff 2 + f (cid:19) , sph ≈ σ2(cid:18) 2 − 2f 6 and dσeff sph df ≈ −σ2 Note that for the a priori effective conductivity σEI of the inflated lung, (31) yields (2 + f )2 . (32) σEI ≈ σ2(cid:18) 2 − 2fEI 2 + fEI (cid:19) , which approximates the solution to (17). By writing σ2 = (cid:60){σ2} (1 + jη), and by employing (28) and (30), it is concluded that ∆σeff j ≈ (cid:60){∆σeff j } (1 + jη) , where η is the loss cotangent associated with the exterior phase having complex valued conductivity σ2. Note finally that (33) implies that σEI ≈ (cid:60){σEI} (1 + jη) , expressing that the loss cotangent of σEI is approxi- mately the same as of σ2. (35) (31) (33) (34) A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 8 4. Numerical examples and clinical data 4.1. Theoretical study of tidal breathing A theoretical study on the changes in the complex valued conductivity of a lung during tidal breathing is described below. As a prerequisite for this modeling an a priori estimate of the complex valued conductivity σEI of an inflated lung is needed. Here, the data is taken from [13, Fig. 2e on p. 2257] with σEI = σa R + jω0a r = 2000 at 200 kHz yielding σEI = 0.1 (1 + j0.22). The resulting conductivity of the exterior phase is obtained from (17) and is given by σ2 = 0.6318 (1 + j0.22) S/m. The complex valued conductivity of the air-filled alveoli is given by σ1 = j1.1127 · 10−5 S/m, and which hence can be regarded negligible in comparison to the conductivity σ2 of the exterior phase. R = 0.1 S/m and a r where σa Another a priori input needed for this modeling is the range of volume fractions [fEE, fEI] during tidal breathing. Consider e.g., the lung volume of an adult male with a functional residual capacity of 2.3 l and tidal volume 0.5 l, cf., Suppose further that the weight of the two lungs is about 0.8 kg [19], corresponding approximately to 0.8 l of blood and tissue. The following volume fractions are then obtained [18]. fEE = fEI = 2.3 2.3 + 0.8 2.8 2.8 + 0.8 = 0.75, = 0.78, (36)  where the results have been rounded to two digits. The theoretical study based on (18) and (19) (spheres and spheroids) is illustrated in figures 2 through 4 where the changes in conductivity are plotted in the complex plane and parameterized by the volume fraction f = fEI − ∆f with ∆f ∈ [0, 0.03] and fEI = 0.78. In figures 2 and 3, it is noted that there is a fixed, almost linear relationship between the changes in the real part of the conductivity (cid:60){∆σeff} and the imaginary part (cid:61){∆σeff}. This observation is in full agreement with the theory predicted by the Taylor series approximation (28) and (34) assuming that the changes in volume fraction df = −∆f < 0 are small. Hence, the linear slopes seen in figures 2 and 3 are consistent with the theory predicted by (34) and where η = 0.22 is the loss cotangent associated with the exterior phase having complex valued conductivity σ2 = 0.6318 (1 + j0.22). Note that this loss cotangent is also approximately the same as the one associated with the a priori (effective) background conductivity σEI = 0.1 (1 + j0.22) according to the theory expressed in (35). As predicted by the linearization theory expressed in (30), the linear slopes seen in figures 2 and 3 are independent of the assumed alveoli model (spherical, Figure 2. Change in the effective lung conductivity parameter ∆σeff , plotted in the complex plane. The resulting change in complex conductivity ∆σeff = σeff (f ) − σeff (fEI) is plotted for volume fractions ranging from fEI = 0.78 to fEE = 0.75 where the last value at fEE is indicated with the corresponding plot symbol. The plot shows results with a prolate spheroidal alveoli (constant surface area and increasing eccentricity, hence decreasing volume) in comparison with a spherical alveoli (decreasing volume). Figure 3. Change in the effective lung conductivity parameter ∆σeff , plotted in the complex plane. The resulting change in complex conductivity ∆σeff = σeff (f ) − σeff (fEI) is plotted for volume fractions ranging from fEI = 0.78 to fEE = 0.75 where the last value at fEE is indicated with the corresponding plot symbol. The plot shows results with an oblate spheroidal alveoli (constant surface area and increasing eccentricity, hence decreasing volume) in comparison with a spherical alveoli (decreasing volume). prolate spheroidal or oblate spheroidal), as well as of the assumed excitation polarization (excitation aligned along spheroid or orthogonal to it). It is only the magnitude of the conductivity changes that differ in between these different modes of alveolar air-filling and their anisotropy. In figure 4 is illustrated the different sensitivity slopes that are obtained with a prolate spheroidal alveoli model excitet along its symmetry axis, and with different a priori assumed background conductivities σEI = σa r . As before, the resulting change in complex valued conductivity ∆σeff = σeff (f )−σeff (fEI) is plotted for volume fractions ranging from fEI = 0.78 R + jω0a Aparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing84.Numericalexamplesandclinicaldata4.1.TheoreticalstudyoftidalbreathingAtheoreticalstudyonthechangesinthecomplexvaluedconductivityofalungduringtidalbreathingisdescribedbelow.AsaprerequisiteforthismodelinganaprioriestimateofthecomplexvaluedconductivityEIofaninflatedlungisneeded.Here,thedataistakenfrom[13,Fig.2eonp.2257]withEI=aR+j!✏0✏arwhereaR=0.1S/mand✏ar=2000at200kHzyieldingEI=0.1(1+j0.22).Theresultingconductivityoftheexteriorphaseisobtainedfrom(17)andisgivenby2=0.6318(1+j0.22)S/m.Thecomplexvaluedconductivityoftheair-filledalveoliisgivenby1=j1.1127·105S/m,andwhichhencecanberegardednegligibleincomparisontotheconductivity2oftheexteriorphase.Anotheraprioriinputneededforthismodelingistherangeofvolumefractions[fEE,fEI]duringtidalbreathing.Considere.g.,thelungvolumeofanadultmalewithafunctionalresidualcapacityof2.3landtidalvolume0.5l,cf.,[18].Supposefurtherthattheweightofthetwolungsisabout0.8kg[19],correspondingapproximatelyto0.8lofbloodandtissue.Thefollowingvolumefractionsarethenobtained8>><>>:fEE=2.32.3+0.8=0.75,fEI=2.82.8+0.8=0.78,(36)wheretheresultshavebeenroundedtotwodigits.Thetheoreticalstudybasedon(18)and(19)(spheresandspheroids)isillustratedinfigures2through4wherethechangesinconductivityareplottedinthecomplexplaneandparameterizedbythevolumefractionf=fEIfwithf2[0,0.03]andfEI=0.78.Infigures2and3,itisnotedthatthereisafixed,almostlinearrelationshipbetweenthechangesintherealpartoftheconductivity<{e↵}andtheimaginarypart={e↵}.ThisobservationisinfullagreementwiththetheorypredictedbytheTaylorseriesapproximation(28)and(34)assumingthatthechangesinvolumefractiondf=f<0aresmall.Hence,thelinearslopesseeninfigures2and3areconsistentwiththetheorypredictedby(34)andwhere⌘=0.22isthelosscotangentassociatedwiththeexteriorphasehavingcomplexvaluedconductivity2=0.6318(1+j0.22).Notethatthislosscotangentisalsoapproximatelythesameastheoneassociatedwiththeapriori(e↵ective)backgroundconductivityEI=0.1(1+j0.22)accordingtothetheoryexpressedin(35).Aspredictedbythelinearizationtheoryexpressedin(30),thelinearslopesseeninfigures2and3areindependentoftheassumedalveolimodel(spherical,10010203005<{e↵}(mS/m)={e↵}(mS/m)e↵forprolatespheroidalalveolimodelexcitationalignedalongspheroidsphericalmodelexcitationorthogonaltospheroidFigure2.Changeinthee↵ectivelungconductivityparametere↵,plottedinthecomplexplane.Theresultingchangeincomplexconductivitye↵=e↵(f)e↵(fEI)isplottedforvolumefractionsrangingfromfEI=0.78tofEE=0.75wherethelastvalueatfEEisindicatedwiththecorrespondingplotsymbol.Theplotshowsresultswithaprolatespheroidalalveoli(constantsurfaceareaandincreasingeccentricity,hencedecreasingvolume)incomparisonwithasphericalalveoli(decreasingvolume).10010203005<{e↵}(mS/m)={e↵}(mS/m)e↵foroblatespheroidalalveolimodelexcitationorthogonaltospheroidsphericalmodelexcitationalignedalongspheroidFigure3.Changeinthee↵ectivelungconductivityparametere↵,plottedinthecomplexplane.Theresultingchangeincomplexconductivitye↵=e↵(f)e↵(fEI)isplottedforvolumefractionsrangingfromfEI=0.78tofEE=0.75wherethelastvalueatfEEisindicatedwiththecorrespondingplotsymbol.Theplotshowsresultswithanoblatespheroidalalveoli(constantsurfaceareaandincreasingeccentricity,hencedecreasingvolume)incomparisonwithasphericalalveoli(decreasingvolume).prolatespheroidaloroblatespheroidal),aswellasoftheassumedexcitationpolarization(excitationalignedalongspheroidororthogonaltoit).Itisonlythemagnitudeoftheconductivitychangesthatdi↵erinbetweenthesedi↵erentmodesofalveolarair-fillingandtheiranisotropy.Infigure4isillustratedthedi↵erentsensitivityslopesthatareobtainedwithaprolatespheroidalalveolimodelexcitetalongitssymmetryaxis,andwithdi↵erentaprioriassumedbackgroundconductivitiesEI=aR+j!✏0✏ar.Asbefore,theresultingchangeinAparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing84.Numericalexamplesandclinicaldata4.1.TheoreticalstudyoftidalbreathingAtheoreticalstudyonthechangesinthecomplexvaluedconductivityofalungduringtidalbreathingisdescribedbelow.AsaprerequisiteforthismodelinganaprioriestimateofthecomplexvaluedconductivityEIofaninflatedlungisneeded.Here,thedataistakenfrom[13,Fig.2eonp.2257]withEI=aR+j!✏0✏arwhereaR=0.1S/mand✏ar=2000at200kHzyieldingEI=0.1(1+j0.22).Theresultingconductivityoftheexteriorphaseisobtainedfrom(17)andisgivenby2=0.6318(1+j0.22)S/m.Thecomplexvaluedconductivityoftheair-filledalveoliisgivenby1=j1.1127·105S/m,andwhichhencecanberegardednegligibleincomparisontotheconductivity2oftheexteriorphase.Anotheraprioriinputneededforthismodelingistherangeofvolumefractions[fEE,fEI]duringtidalbreathing.Considere.g.,thelungvolumeofanadultmalewithafunctionalresidualcapacityof2.3landtidalvolume0.5l,cf.,[18].Supposefurtherthattheweightofthetwolungsisabout0.8kg[19],correspondingapproximatelyto0.8lofbloodandtissue.Thefollowingvolumefractionsarethenobtained8>><>>:fEE=2.32.3+0.8=0.75,fEI=2.82.8+0.8=0.78,(36)wheretheresultshavebeenroundedtotwodigits.Thetheoreticalstudybasedon(18)and(19)(spheresandspheroids)isillustratedinfigures2through4wherethechangesinconductivityareplottedinthecomplexplaneandparameterizedbythevolumefractionf=fEIfwithf2[0,0.03]andfEI=0.78.Infigures2and3,itisnotedthatthereisafixed,almostlinearrelationshipbetweenthechangesintherealpartoftheconductivity<{e↵}andtheimaginarypart={e↵}.ThisobservationisinfullagreementwiththetheorypredictedbytheTaylorseriesapproximation(28)and(34)assumingthatthechangesinvolumefractiondf=f<0aresmall.Hence,thelinearslopesseeninfigures2and3areconsistentwiththetheorypredictedby(34)andwhere⌘=0.22isthelosscotangentassociatedwiththeexteriorphasehavingcomplexvaluedconductivity2=0.6318(1+j0.22).Notethatthislosscotangentisalsoapproximatelythesameastheoneassociatedwiththeapriori(e↵ective)backgroundconductivityEI=0.1(1+j0.22)accordingtothetheoryexpressedin(35).Aspredictedbythelinearizationtheoryexpressedin(30),thelinearslopesseeninfigures2and3areindependentoftheassumedalveolimodel(spherical,10010203005<{e↵}(mS/m)={e↵}(mS/m)e↵forprolatespheroidalalveolimodelexcitationalignedalongspheroidsphericalmodelexcitationorthogonaltospheroidFigure2.Changeinthee↵ectivelungconductivityparametere↵,plottedinthecomplexplane.Theresultingchangeincomplexconductivitye↵=e↵(f)e↵(fEI)isplottedforvolumefractionsrangingfromfEI=0.78tofEE=0.75wherethelastvalueatfEEisindicatedwiththecorrespondingplotsymbol.Theplotshowsresultswithaprolatespheroidalalveoli(constantsurfaceareaandincreasingeccentricity,hencedecreasingvolume)incomparisonwithasphericalalveoli(decreasingvolume).10010203005<{e↵}(mS/m)={e↵}(mS/m)e↵foroblatespheroidalalveolimodelexcitationorthogonaltospheroidsphericalmodelexcitationalignedalongspheroidFigure3.Changeinthee↵ectivelungconductivityparametere↵,plottedinthecomplexplane.Theresultingchangeincomplexconductivitye↵=e↵(f)e↵(fEI)isplottedforvolumefractionsrangingfromfEI=0.78tofEE=0.75wherethelastvalueatfEEisindicatedwiththecorrespondingplotsymbol.Theplotshowsresultswithanoblatespheroidalalveoli(constantsurfaceareaandincreasingeccentricity,hencedecreasingvolume)incomparisonwithasphericalalveoli(decreasingvolume).prolatespheroidaloroblatespheroidal),aswellasoftheassumedexcitationpolarization(excitationalignedalongspheroidororthogonaltoit).Itisonlythemagnitudeoftheconductivitychangesthatdi↵erinbetweenthesedi↵erentmodesofalveolarair-fillingandtheiranisotropy.Infigure4isillustratedthedi↵erentsensitivityslopesthatareobtainedwithaprolatespheroidalalveolimodelexcitetalongitssymmetryaxis,andwithdi↵erentaprioriassumedbackgroundconductivitiesEI=aR+j!✏0✏ar.Asbefore,theresultingchangeinA parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 9 To start with, we will give a brief motivation to why the reconstruction matrix optimized using the GREIT algorithm and real valued data and parameters, is also expected to be (suboptimally) useful for reconstruction of the imaginary part of the conductivity. Using the GREIT algorithm, a real valued reconstruction matrix R is chosen to optimize the criterion (cid:107)∆σ(k) R − R∆v(k)(cid:107)2 W (k) (37) minimize (cid:88)k where the index k refers to a sequence of training data, ∆σ(k) the chosen perturbations in the real R valued conductivity, ∆v(k) the corresponding changes in the real valued voltage data obtained by using a forward model based on Finite Element Modeling (FEM) and some particular stimulation patters, and W (k) a weighting matrix chosen according to some a priori weighting with respect to the training set, cf., [9]. Assuming that the changes in conductivity are small, a first order Taylor series approximation at the background conductivity profile σR can be used to express the changes as Figure 4. Change in the effective lung conductivity parameter ∆σeff for prolate spheroidal alveoli with excitation aligned along the spheroid, as in figure 2, plotted here for different values of a priori inflated lung parameter σEI = σa R + jω0a r . to fEE = 0.75 where the last value at fEE is indicated with the corresponding plot symbol. Note that the corresponding changes ∆σeff are directly proportional to σ2 by (30), or σEI by (33). 4.2. Suboptimal imaging of permittivity based on EIT trained for real valued data The clinical study that is referred to below has been performed within the EU-funded project CRADL, cf., ClinicalTrials.gov (NCT02962505). The EIT voltage data from each neonatal patient is acquired using the CRADL device (Swisstom Landquart, Switzerland) using a textile sensor belt around the chest comprising 32 electrodes, and measured at an operating frequency of about 200 kHz, down-converted and digitized at a sampling rate of 48 Hz. A four-point measurement strategy is employed using pairs of current stimulation and voltage measurement electrodes having an internal distance of 5 sensors. To avoid amplifier saturation the measurements are omitted when measurement electrodes are at a distance smaller than or equal to 2 sensors from the stimulation electrodes. The electronics of the measurement equipment allows for high precision in-phase and quadrature-phase (I/Q) measurements and hence complex valued data is stored. The aim of the present study is to demonstrate that there is a non-negligible imaginary part of the data that may have a potential for clinical use. In particular, the theoretical study above show that the change in the imaginary part of the conductivity is expected to be in the order of about 20 % of the corresponding change in the real part during tidal breathing (η = 0.22 in the theoretical study above). ∆σ(k) = ∆σ(k) R (cid:16)1 + jη(k)(cid:17) , R (cid:17) − v (σR) ∂v ∆v(k) = v(cid:16)σR + ∆σ(k) ∂σ(cid:12)(cid:12)(cid:12)(cid:12)σR ∆σ(k) R , ≈ where ∆v(k) and ∆σ(k) vectors, v(σR) is the forward model and ∂v corresponding Jacobian matrix. R can be interpreted as column the Suppose now that the background conductivity σR is changed by adding an imaginary part so that σ = σR + jσI. In a corresponding complex algorithm, a complex valued reconstruction matrix Rc would now be chosen to optimize the criterion ∂σ(cid:12)(cid:12)σR minimize (cid:88)k (cid:107)∆σ(k) − Rc∆v(k) c (cid:107)2 R + j∆σ(k) I W (k) , (39) where ∆σ(k) = ∆σ(k) denotes the chosen complex valued conductivity perturbations and ∆v(k) the corresponding changes in the voltage data obtained by using the same forward model v(σ). Again, by assuming that the changes in conductivity are small in relation to the real valued background conductivity σR, a first order Taylor series approximation at σR can be used to express the changes as c ∂v c = v(cid:16)σR + jσI + ∆σ(k)(cid:17) − v (σR + jσI) ∂σ(cid:12)(cid:12)(cid:12)(cid:12)σR ∆σ(k), ≈ where the Jacobian matrix is the same as in (38). Assuming that the same real valued perturbations ∆σ(k) R will be used as in (37), we can write ∆v(k) (38) (40) (41) Aparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing90204060051015<{e↵}(mS/m)={e↵}(mS/m)e↵forprolatespheroidalalveolimodelaR=0.1,✏ar=2000aR=0.1,✏ar=4000aR=0.05,✏ar=2000aR=0.2,✏ar=2000aR=0.1,✏ar=1000Figure4.Changeinthee↵ectivelungconductivityparametere↵forprolatespheroidalalveoliwithexcitationalignedalongthespheroid,asinfigure2,plottedherefordi↵erentvaluesofaprioriinflatedlungparameterEI=aR+j!✏0✏ar.complexvaluedconductivitye↵=e↵(f)e↵(fEI)isplottedforvolumefractionsrangingfromfEI=0.78tofEE=0.75wherethelastvalueatfEEisindicatedwiththecorrespondingplotsymbol.Notethatthecorrespondingchangese↵aredirectlyproportionalto2by(30),orEIby(33).4.2.SuboptimalimagingofpermittivitybasedonEITtrainedforrealvalueddataTheclinicalstudythatisreferredtobelowhasbeenperformedwithintheEU-fundedprojectCRADL,cf.,ClinicalTrials.gov(NCT02962505).TheEITvoltagedatafromeachneonatalpatientisacquiredusingtheCRADLdevice(SwisstomLandquart,Switzerland)usingatextilesensorbeltaroundthechestcomprising32electrodes,andmeasuredatanoperatingfrequencyofabout200kHz,down-convertedanddigitizedatasamplingrateof48Hz.Afour-pointmeasurementstrategyisemployedusingpairsofcurrentstimulationandvoltagemeasurementelectrodeshavinganinternaldistanceof5sensors.Toavoidamplifiersaturationthemeasurementsareomittedwhenmeasurementelectrodesareatadistancesmallerthanorequalto2sensorsfromthestimulationelectrodes.Theelectronicsofthemeasurementequipmentallowsforhighprecisionin-phaseandquadrature-phase(I/Q)measurementsandhencecomplexvalueddataisstored.Theaimofthepresentstudyistodemonstratethatthereisanon-negligibleimaginarypartofthedatathatmayhaveapotentialforclinicaluse.Inparticular,thetheoreticalstudyaboveshowthatthechangeintheimaginarypartoftheconductivityisexpectedtobeintheorderofabout20%ofthecorrespondingchangeintherealpartduringtidalbreathing(⌘=0.22inthetheoreticalstudyabove).Tostartwith,wewillgiveabriefmotivationtowhythereconstructionmatrixoptimizedusingtheGREITalgorithmandrealvalueddataandparameters,isalsoexpectedtobe(suboptimally)usefulforreconstructionoftheimaginarypartoftheconductivity.UsingtheGREITalgorithm,arealvaluedreconstructionmatrixRischosentooptimizethecriterionminimizeXkk(k)RRv(k)k2W(k)(37)wheretheindexkreferstoasequenceoftrainingdata,(k)Rthechosenperturbationsintherealvaluedconductivity,v(k)thecorrespondingchangesintherealvaluedvoltagedataobtainedbyusingaforwardmodelbasedonFiniteElementModeling(FEM)andsomeparticularstimulationpatters,andW(k)aweightingmatrixchosenaccordingtosomeaprioriweightingwithrespecttothetrainingset,cf.,[9].Assumingthatthechangesinconductivityaresmall,afirstorderTaylorseriesapproximationatthebackgroundconductivityprofileRcanbeusedtoexpressthechangesasv(k)=v⇣R+(k)R⌘v(R)⇡@v@R(k)R,(38)wherev(k)and(k)Rcanbeinterpretedascolumnvectors,v(R)istheforwardmodeland@v@RthecorrespondingJacobianmatrix.SupposenowthatthebackgroundconductivityRischangedbyaddinganimaginarypartsothat=R+jI.Inacorrespondingcomplexalgorithm,acomplexvaluedreconstructionmatrixRcwouldnowbechosentooptimizethecriterionminimizeXkk(k)Rcv(k)ck2W(k),(39)where(k)=(k)R+j(k)Idenotesthechosencomplexvaluedconductivityperturbationsandv(k)cthecorrespondingchangesinthevoltagedataobtainedbyusingthesameforwardmodelv().Again,byassumingthatthechangesinconductivityaresmallinrelationtotherealvaluedbackgroundconductivityR,afirstorderTaylorseriesapproximationatRcanbeusedtoexpressthechangesasv(k)c=v⇣R+jI+(k)⌘v(R+jI)⇡@v@R(k),(40)wheretheJacobianmatrixisthesameasin(38).AssumingthatthesamerealvaluedperturbationsA parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 10 where η(k) is a real valued parameter. Hence, by using (38), (40) and (41) it is seen that ∆v(k) ∂v ∆σ(k) c ≈ R (cid:16)1 + jη(k)(cid:17) ∂σ(cid:12)(cid:12)(cid:12)(cid:12)σR ≈ ∆v(k)(cid:16)1 + jη(k)(cid:17) , minimize (cid:88)k (cid:12)(cid:12)(cid:12)1 + jη(k)(cid:12)(cid:12)(cid:12) 2 which yields the following criterion based on (39) (42) (43) (cid:107)∆σ(k) R − Rc∆v(k)(cid:107)2 W (k) . Now, according to the theoretical results of section 3, and in particular (34) which are illustrated in section 4.1, it is expected that the real and imaginary parts of the changes in conductivity of a lung during tidal breathing will follow the linear relationship ∆σ = ∆σR (1 + jη) , (44) where η is a real valued constant. This means that that the factor 1 + jη(k) in (43) should be independent of k with η(k) = η, and hence that the optimal solution Rc in (43) coincides with the optimal solution R in (37). In conclusion, based on the assumptions of small changes with linear approximations as in (38) and (40), together with training sequences ∆σ(k) following the linear relationship (44), it is expected that the choice Rc = R will yield a suboptimal solution to (39) that can be useful also for the reconstruction of the imaginary part of the complex valued conductivity. 4.3. A comparison with clinical data As an illustration of its potential clinical use, the presented theory is employed here in a comparison with a small set of clinical data collected within the CRADL project, as mentioned above. The EIT data from one mechanically ventilated neonatal patient (weight 1400 g and gestational age 29 weeks) is included in the examples below, as shown in Figures 5 through 12. Two sequences of data have been included in these examples, one with a somewhat unstable breathing period shown in figures 5 through 8, and one with a more stable ventilation as shown in figures 9 through 12. The figures 5 and 9 illustrate the corresponding breathing signals comprising the sum of all reconstructed image pixels of real valued conductivity changes, and which is the input for the breath detection [10]. Based on the identified timing of end-expiration and end-inspiration, a tidal image at a particular breath is then obtained by applying a precalculated real valued reconstruction matrix to the corresponding voltage difference data (end-expiration data minus end-inspiration data), as described in section 4.2. Figures 6 through 8, and 10 through 12 show the image pixel values). Figure 5. Breath detection based on the breathing signal (sum of The lower red and upper green diamonds indicate the timing of end-expiration and end- inspiration, respectively. The black star and circle in the middle of the figure indicate breath no 88 corresponding to the tidal images shown in figure 6. Figure 6. Reconstructed pixel values of tidal image at breath no. 88. The upper plot shows a comparison with the presented homogenization theory (dashed line), and the lower plots show normalized tidal images of the real and imaginary conductivity changes, respectively, in arbitrary units. reconstructed pixel values and tidal images at breath no. 88-92 and 262-266, respectively, for both the real and the imaginary part of the complex valued conductivity changes. The upper figures show the complex valued pixels in a comparison with the presented homogenization theory of section 3, and in particular (34) where ∆σ = (cid:60){∆σ} (1 + jη) and where η = 0.22 (dashed line), as described in section 4.1. The red and blue circles indicate pixels from the left lung and the right lung, respectively. The lower plots show tidal images of the real and the imaginary parts of the reconstructed conductivity changes, respectively, with the corresponding contours of the left lung, the right lung and the heart indicated within the figure (the right lung contour appears on the left-hand side of the figures, etc.). Aparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing10(k)Rwillbeusedasin(37),wecanwrite(k)=(k)R⇣1+j⌘(k)⌘,(41)where⌘(k)isarealvaluedparameter.Hence,byusing(38),(40)and(41)itisseenthatv(k)c⇡@v@R(k)R⇣1+j⌘(k)⌘⇡v(k)⇣1+j⌘(k)⌘,(42)whichyieldsthefollowingcriterionbasedon(39)minimizeXk1+j⌘(k)2k(k)RRcv(k)k2W(k).(43)Now,accordingtothetheoreticalresultsofsection3,andinparticular(34)whichareillustratedinsection4.1,itisexpectedthattherealandimaginarypartsofthechangesinconductivityofalungduringtidalbreathingwillfollowthelinearrelationship=R(1+j⌘),(44)where⌘isarealvaluedconstant.Thismeansthatthatthefactor1+j⌘(k)in(43)shouldbeindependentofkwith⌘(k)=⌘,andhencethattheoptimalsolutionRcin(43)coincideswiththeoptimalsolutionRin(37).Inconclusion,basedontheassumptionsofsmallchangeswithlinearapproximationsasin(38)and(40),togetherwithtrainingsequences(k)followingthelinearrelationship(44),itisexpectedthatthechoiceRc=Rwillyieldasuboptimalsolutionto(39)thatcanbeusefulalsoforthereconstructionoftheimaginarypartofthecomplexvaluedconductivity.4.3.AcomparisonwithclinicaldataAsanillustrationofitspotentialclinicaluse,thepresentedtheoryisemployedhereinacomparisonwithasmallsetofclinicaldatacollectedwithintheCRADLproject,asmentionedabove.TheEITdatafromonemechanicallyventilatedneonatalpatient(weight1400gandgestationalage29weeks)isincludedintheexamplesbelow,asshowninFigures5through12.Twosequencesofdatahasbeenincludedintheseexamples,onewithasomewhatunstablebreathingperiodshowninfigures5through8,andonewithamorestableventilationasshowninfigures9through12.Thefigures5and9illustratethecorrespondingbreathingsignalscomprisingthesumofallreconstructedimagepixelsofrealvaluedconductivitychanges,andwhichistheinputforthebreathdetection[10].Basedontheidentifiedtimingofend-expirationandend-inspiration,atidalimageataparticularbreathisthenobtainedbyapplyingaprecalculatedrealvaluedreconstructionmatrixtothecorresponding1351401451500.500.5time(s)SumofpixelsBreath-detectionFigure5.Breathdetectionbasedonthebreathingsignal(sumofimagepixelvalues).Thelowerredanduppergreendiamondsindicatethetimingofend-expirationandend-inspiration,respectively.Theblackstarandcircleinthemiddleofthefigureindicatebreathno88correspondingtothetidalimagesshowninfigure6.4202400.51<{}={}at88thbreathLeftLungRightLungHom.Theory<{"<} at 88th breath1020305101520253000.20.40.60.81={"<} at 88th breath1020305101520253000.20.40.60.81Figure6.Reconstructedpixelvaluesoftidalimageatbreathno.88.Theupperplotshowsacomparisonwiththepresentedhomogenizationtheory(dashedline),andthelowerplotsshownormalizedtidalimagesoftherealandimaginaryconductivitychanges,respectively,inarbitraryunits.voltagedi↵erencedata(end-expirationdataminusend-inspirationdata),asdescribedinsection4.2.Figures6through8,and10through12showthereconstructedpixelvaluesandtidalimagesatbreathno.88-92and262-266,respectively,forboththerealandtheimaginarypartofthecomplexvaluedconductivitychanges.Theupperfiguresshowthecomplexvaluedpixelsinacomparisonwiththepresentedhomogenizationtheoryofsection3,andinparticular(34)where=<{}(1+j⌘)andwhere⌘=0.22(dashedline),asdescribedinsection4.1.Theredandbluecirclesindicatepixelsfromtheleftlungandtherightlung,respectively.ThelowerplotsshowtidalimagesoftherealandtheimaginarypartsAparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing10(k)Rwillbeusedasin(37),wecanwrite(k)=(k)R⇣1+j⌘(k)⌘,(41)where⌘(k)isarealvaluedparameter.Hence,byusing(38),(40)and(41)itisseenthatv(k)c⇡@v@R(k)R⇣1+j⌘(k)⌘⇡v(k)⇣1+j⌘(k)⌘,(42)whichyieldsthefollowingcriterionbasedon(39)minimizeXk1+j⌘(k)2k(k)RRcv(k)k2W(k).(43)Now,accordingtothetheoreticalresultsofsection3,andinparticular(34)whichareillustratedinsection4.1,itisexpectedthattherealandimaginarypartsofthechangesinconductivityofalungduringtidalbreathingwillfollowthelinearrelationship=R(1+j⌘),(44)where⌘isarealvaluedconstant.Thismeansthatthatthefactor1+j⌘(k)in(43)shouldbeindependentofkwith⌘(k)=⌘,andhencethattheoptimalsolutionRcin(43)coincideswiththeoptimalsolutionRin(37).Inconclusion,basedontheassumptionsofsmallchangeswithlinearapproximationsasin(38)and(40),togetherwithtrainingsequences(k)followingthelinearrelationship(44),itisexpectedthatthechoiceRc=Rwillyieldasuboptimalsolutionto(39)thatcanbeusefulalsoforthereconstructionoftheimaginarypartofthecomplexvaluedconductivity.4.3.AcomparisonwithclinicaldataAsanillustrationofitspotentialclinicaluse,thepresentedtheoryisemployedhereinacomparisonwithasmallsetofclinicaldatacollectedwithintheCRADLproject,asmentionedabove.TheEITdatafromonemechanicallyventilatedneonatalpatient(weight1400gandgestationalage29weeks)isincludedintheexamplesbelow,asshowninFigures5through12.Twosequencesofdatahasbeenincludedintheseexamples,onewithasomewhatunstablebreathingperiodshowninfigures5through8,andonewithamorestableventilationasshowninfigures9through12.Thefigures5and9illustratethecorrespondingbreathingsignalscomprisingthesumofallreconstructedimagepixelsofrealvaluedconductivitychanges,andwhichistheinputforthebreathdetection[10].Basedontheidentifiedtimingofend-expirationandend-inspiration,atidalimageataparticularbreathisthenobtainedbyapplyingaprecalculatedrealvaluedreconstructionmatrixtothecorresponding1351401451500.500.5time(s)SumofpixelsBreath-detectionFigure5.Breathdetectionbasedonthebreathingsignal(sumofimagepixelvalues).Thelowerredanduppergreendiamondsindicatethetimingofend-expirationandend-inspiration,respectively.Theblackstarandcircleinthemiddleofthefigureindicatebreathno88correspondingtothetidalimagesshowninfigure6.4202400.51<{}={}at88thbreathLeftLungRightLungHom.Theory<{"<} at 88th breath1020305101520253000.20.40.60.81={"<} at 88th breath1020305101520253000.20.40.60.81Figure6.Reconstructedpixelvaluesoftidalimageatbreathno.88.Theupperplotshowsacomparisonwiththepresentedhomogenizationtheory(dashedline),andthelowerplotsshownormalizedtidalimagesoftherealandimaginaryconductivitychanges,respectively,inarbitraryunits.voltagedi↵erencedata(end-expirationdataminusend-inspirationdata),asdescribedinsection4.2.Figures6through8,and10through12showthereconstructedpixelvaluesandtidalimagesatbreathno.88-92and262-266,respectively,forboththerealandtheimaginarypartofthecomplexvaluedconductivitychanges.Theupperfiguresshowthecomplexvaluedpixelsinacomparisonwiththepresentedhomogenizationtheoryofsection3,andinparticular(34)where=<{}(1+j⌘)andwhere⌘=0.22(dashedline),asdescribedinsection4.1.Theredandbluecirclesindicatepixelsfromtheleftlungandtherightlung,respectively.ThelowerplotsshowtidalimagesoftherealandtheimaginarypartsA parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 11 image pixel values). Figure 9. Breath detection based on the breathing signal (sum of The lower red and upper green diamonds indicate the timing of end-expiration and end- inspiration, respectively. The black star and circle in the middle of the figure indicate breath no 262 corresponding to the tidal images shown in figure 10. Figure 7. Same as in figure 6, at breath no. 90. Figure 8. Same as in figure 6, at breath no. 92. Figure 10. Reconstructed pixel values of tidal image at breath no. 262. The upper plot shows a comparison with the presented homogenization theory (dashed line), and the lower plots show normalized tidal images of the real and imaginary conductivity changes, respectively, in arbitrary units. As can be seen from these figures, the re- constructed pixel values agree quite well with the theory that predicts the linear relationship ∆σ = (cid:60){∆σ} (1 + jη) for a lung during tidal breathing. On the other hand, we will also stress here that these im- ages have been chosen deliberately with the purpose to illustrate the theory, and many other tidal images can not be interpreted in the same favorable way. However, one should also keep in mind that there can be exter- nal disturbances and/or other competing conductivity mechanisms that are present during the measurements. The figure 8 (at breath no. 92) is particularly interesting, as it is only the left lung pixels that show a clear linear "correlation" as predicted by theory, and it is also only the left lung that appears in full contrast in the imaging of both the real and the imaginary part of the conductivity change. A hypothetical interpretation of this image is that it is only the left lung that is properly ventilated during this particular breath, and what is seen in the real valued conductivity change in the right lung could instead be caused by a change in blood perfusion, gas concentration, muscle contractions, etc., without much alveoli ventilation. 4.4. Possible clinical applications The main conclusion from the theoretical analysis in this paper is that the loss cotangent of the change in the complex valued conductivity of a lung during tidal breathing is expected to be the same as of the effective background conductivity. Hence, the presence of a linear relationship between the corresponding real and imaginary components could serve as part Aparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing114202400.51<{}={}at90thbreathLeftLungRightLungHom.Theory<{"<} at 90th breath1020305101520253000.20.40.60.81={"<} at 90th breath1020305101520253000.20.40.60.81Figure7.Sameasinfigure6,atbreathno.90.4202400.51<{}={}at92thbreathLeftLungRightLungHom.Theory<{"<} at 92th breath1020305101520253000.20.40.60.81={"<} at 92th breath1020305101520253000.20.40.60.81Figure8.Sameasinfigure6,atbreathno.92.ofthereconstructedconductivitychanges,respectively,withthecorrespondingcontoursoftheleftlung,therightlungandtheheartindicatedwithinthefigure(therightlungcontourappearsontheleft-handsideofthefigures,etc.).Ascanbeseenfromthesefigures,there-constructedpixelvaluesagreequitewellwiththetheorythatpredictsthelinearrelationship=<{}(1+j⌘)foralungduringtidalbreathing.Ontheotherhand,wewillalsostressherethattheseim-ageshavebeenchosendeliberatelywiththepurposetoillustratethetheory,andmanyothertidalimagescannotbeinterpretedinthesamefavorableway.However,4104154204250.40.200.20.4time(s)SumofpixelsBreath-detectionFigure9.Breathdetectionbasedonthebreathingsignal(sumofimagepixelvalues).Thelowerredanduppergreendiamondsindicatethetimingofend-expirationandend-inspiration,respectively.Theblackstarandcircleinthemiddleofthefigureindicatebreathno262correspondingtothetidalimagesshowninfigure10.1.510.500.511.500.20.4<{}={}at262thbreathLeftLungRightLungHom.Theory<{"<} at 262th breath1020305101520253000.20.40.60.81={"<} at 262th breath1020305101520253000.20.40.60.81Figure10.Reconstructedpixelvaluesoftidalimageatbreathno.262.Theupperplotshowsacomparisonwiththepresentedhomogenizationtheory(dashedline),andthelowerplotsshownormalizedtidalimagesoftherealandimaginaryconductivitychanges,respectively,inarbitraryunits.oneshouldalsokeepinmindthattherecanbeexter-naldisturbancesand/orothercompetingconductivitymechanismsthatarepresentduringthemeasurements.Thefigure8(atbreathno.92)isparticularlyinteresting,asitisonlytheleftlungpixelsthatshowaclearlinear"correlation"aspredictedbytheory,anditisalsoonlytheleftlungthatappearsinfullcontrastintheimagingofboththerealandtheimaginarypartoftheconductivitychange.Ahypotheticalinterpretationofthisimageisthatitisonlytheleftlungthatisproperlyventilatedduringthisparticularbreath,andwhatisseenintherealvaluedconductivitychangeintherightlungcouldinsteadbecausedbyachangeinbloodperfusion,gasconcentration,muscleAparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing114202400.51<{}={}at90thbreathLeftLungRightLungHom.Theory<{"<} at 90th breath1020305101520253000.20.40.60.81={"<} at 90th breath1020305101520253000.20.40.60.81Figure7.Sameasinfigure6,atbreathno.90.4202400.51<{}={}at92thbreathLeftLungRightLungHom.Theory<{"<} at 92th breath1020305101520253000.20.40.60.81={"<} at 92th breath1020305101520253000.20.40.60.81Figure8.Sameasinfigure6,atbreathno.92.ofthereconstructedconductivitychanges,respectively,withthecorrespondingcontoursoftheleftlung,therightlungandtheheartindicatedwithinthefigure(therightlungcontourappearsontheleft-handsideofthefigures,etc.).Ascanbeseenfromthesefigures,there-constructedpixelvaluesagreequitewellwiththetheorythatpredictsthelinearrelationship=<{}(1+j⌘)foralungduringtidalbreathing.Ontheotherhand,wewillalsostressherethattheseim-ageshavebeenchosendeliberatelywiththepurposetoillustratethetheory,andmanyothertidalimagescannotbeinterpretedinthesamefavorableway.However,4104154204250.40.200.20.4time(s)SumofpixelsBreath-detectionFigure9.Breathdetectionbasedonthebreathingsignal(sumofimagepixelvalues).Thelowerredanduppergreendiamondsindicatethetimingofend-expirationandend-inspiration,respectively.Theblackstarandcircleinthemiddleofthefigureindicatebreathno262correspondingtothetidalimagesshowninfigure10.1.510.500.511.500.20.4<{}={}at262thbreathLeftLungRightLungHom.Theory<{"<} at 262th breath1020305101520253000.20.40.60.81={"<} at 262th breath1020305101520253000.20.40.60.81Figure10.Reconstructedpixelvaluesoftidalimageatbreathno.262.Theupperplotshowsacomparisonwiththepresentedhomogenizationtheory(dashedline),andthelowerplotsshownormalizedtidalimagesoftherealandimaginaryconductivitychanges,respectively,inarbitraryunits.oneshouldalsokeepinmindthattherecanbeexter-naldisturbancesand/orothercompetingconductivitymechanismsthatarepresentduringthemeasurements.Thefigure8(atbreathno.92)isparticularlyinteresting,asitisonlytheleftlungpixelsthatshowaclearlinear"correlation"aspredictedbytheory,anditisalsoonlytheleftlungthatappearsinfullcontrastintheimagingofboththerealandtheimaginarypartoftheconductivitychange.Ahypotheticalinterpretationofthisimageisthatitisonlytheleftlungthatisproperlyventilatedduringthisparticularbreath,andwhatisseenintherealvaluedconductivitychangeintherightlungcouldinsteadbecausedbyachangeinbloodperfusion,gasconcentration,muscleAparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing114202400.51<{}={}at90thbreathLeftLungRightLungHom.Theory<{"<} at 90th breath1020305101520253000.20.40.60.81={"<} at 90th breath1020305101520253000.20.40.60.81Figure7.Sameasinfigure6,atbreathno.90.4202400.51<{}={}at92thbreathLeftLungRightLungHom.Theory<{"<} at 92th breath1020305101520253000.20.40.60.81={"<} at 92th breath1020305101520253000.20.40.60.81Figure8.Sameasinfigure6,atbreathno.92.ofthereconstructedconductivitychanges,respectively,withthecorrespondingcontoursoftheleftlung,therightlungandtheheartindicatedwithinthefigure(therightlungcontourappearsontheleft-handsideofthefigures,etc.).Ascanbeseenfromthesefigures,there-constructedpixelvaluesagreequitewellwiththetheorythatpredictsthelinearrelationship=<{}(1+j⌘)foralungduringtidalbreathing.Ontheotherhand,wewillalsostressherethattheseim-ageshavebeenchosendeliberatelywiththepurposetoillustratethetheory,andmanyothertidalimagescannotbeinterpretedinthesamefavorableway.However,4104154204250.40.200.20.4time(s)SumofpixelsBreath-detectionFigure9.Breathdetectionbasedonthebreathingsignal(sumofimagepixelvalues).Thelowerredanduppergreendiamondsindicatethetimingofend-expirationandend-inspiration,respectively.Theblackstarandcircleinthemiddleofthefigureindicatebreathno262correspondingtothetidalimagesshowninfigure10.1.510.500.511.500.20.4<{}={}at262thbreathLeftLungRightLungHom.Theory<{"<} at 262th breath1020305101520253000.20.40.60.81={"<} at 262th breath1020305101520253000.20.40.60.81Figure10.Reconstructedpixelvaluesoftidalimageatbreathno.262.Theupperplotshowsacomparisonwiththepresentedhomogenizationtheory(dashedline),andthelowerplotsshownormalizedtidalimagesoftherealandimaginaryconductivitychanges,respectively,inarbitraryunits.oneshouldalsokeepinmindthattherecanbeexter-naldisturbancesand/orothercompetingconductivitymechanismsthatarepresentduringthemeasurements.Thefigure8(atbreathno.92)isparticularlyinteresting,asitisonlytheleftlungpixelsthatshowaclearlinear"correlation"aspredictedbytheory,anditisalsoonlytheleftlungthatappearsinfullcontrastintheimagingofboththerealandtheimaginarypartoftheconductivitychange.Ahypotheticalinterpretationofthisimageisthatitisonlytheleftlungthatisproperlyventilatedduringthisparticularbreath,andwhatisseenintherealvaluedconductivitychangeintherightlungcouldinsteadbecausedbyachangeinbloodperfusion,gasconcentration,muscleAparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing114202400.51<{}={}at90thbreathLeftLungRightLungHom.Theory<{"<} at 90th breath1020305101520253000.20.40.60.81={"<} at 90th breath1020305101520253000.20.40.60.81Figure7.Sameasinfigure6,atbreathno.90.4202400.51<{}={}at92thbreathLeftLungRightLungHom.Theory<{"<} at 92th breath1020305101520253000.20.40.60.81={"<} at 92th breath1020305101520253000.20.40.60.81Figure8.Sameasinfigure6,atbreathno.92.ofthereconstructedconductivitychanges,respectively,withthecorrespondingcontoursoftheleftlung,therightlungandtheheartindicatedwithinthefigure(therightlungcontourappearsontheleft-handsideofthefigures,etc.).Ascanbeseenfromthesefigures,there-constructedpixelvaluesagreequitewellwiththetheorythatpredictsthelinearrelationship=<{}(1+j⌘)foralungduringtidalbreathing.Ontheotherhand,wewillalsostressherethattheseim-ageshavebeenchosendeliberatelywiththepurposetoillustratethetheory,andmanyothertidalimagescannotbeinterpretedinthesamefavorableway.However,4104154204250.40.200.20.4time(s)SumofpixelsBreath-detectionFigure9.Breathdetectionbasedonthebreathingsignal(sumofimagepixelvalues).Thelowerredanduppergreendiamondsindicatethetimingofend-expirationandend-inspiration,respectively.Theblackstarandcircleinthemiddleofthefigureindicatebreathno262correspondingtothetidalimagesshowninfigure10.1.510.500.511.500.20.4<{}={}at262thbreathLeftLungRightLungHom.Theory<{"<} at 262th breath1020305101520253000.20.40.60.81={"<} at 262th breath1020305101520253000.20.40.60.81Figure10.Reconstructedpixelvaluesoftidalimageatbreathno.262.Theupperplotshowsacomparisonwiththepresentedhomogenizationtheory(dashedline),andthelowerplotsshownormalizedtidalimagesoftherealandimaginaryconductivitychanges,respectively,inarbitraryunits.oneshouldalsokeepinmindthattherecanbeexter-naldisturbancesand/orothercompetingconductivitymechanismsthatarepresentduringthemeasurements.Thefigure8(atbreathno.92)isparticularlyinteresting,asitisonlytheleftlungpixelsthatshowaclearlinear"correlation"aspredictedbytheory,anditisalsoonlytheleftlungthatappearsinfullcontrastintheimagingofboththerealandtheimaginarypartoftheconductivitychange.Ahypotheticalinterpretationofthisimageisthatitisonlytheleftlungthatisproperlyventilatedduringthisparticularbreath,andwhatisseenintherealvaluedconductivitychangeintherightlungcouldinsteadbecausedbyachangeinbloodperfusion,gasconcentration,muscleA parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 12 characterized by an increase in pulmonary fluid content [32]. Hence, a change in the loss cotangent of the linear relationship between the real and imaginary part might be clinically useful for the detection of pulmonary edema. 4.5. Uncertainties One parameter of major uncertainty in this study is the actual volume fraction f of air during tidal breathing. However, since there is a very high contrast between the complex valued conductivity of air and tissue, the sensitivity analysis of section 3.3 is valid showing that the result (34) is independent of f . Furthermore, the loss cotangent η of σ2 = (cid:60){σ2} (1 + jη) is essentially the same as the loss cotangent of the a priori parameter σEI, as expressed in (33) and (35). Hence, an uncertainty in the volume fraction fEI at end-inspiration (where σEI is given) will only affect the magnitude of the estimated parameter σ2, and it will not affect the estimated loss cotangent η nor the conclusion made in relation to (34). Another uncertainty related to the measured CRADL EIT data described in section 4.3 above is due to the fact that the measured I and Q data have been rotated already in the hardware in order to minimize the imaginary (Q) part of the data. The main purpose of this procedure is to minimize the effect of the phase shifts in the internal electronics as well as in the stray capacitances in leads and electrodes, etc. However, this means also that there is an unknown rotation angle φ present in the data relating to a "true" resistive background. Notably, since the reconstruction matrix in our case is real valued, the presence of the unknown rotation angle φ means simply that our actual estimates are also rotated as I sin φ, I cos φ, σI = σt where σt I are the desired "true" estimates that would have resulted if the calibration would have been carried out by using a "true" resistive phantom. Obviously, the two sets of estimates {σR, σI} and {σt I} are equivalent in the sense of the linear and bijective rotation operation. However, for an EIT instrumentation that is designed to image the real valued conductivity, it is desirable to have φ as small as possible. Hence, in a future investigation to exploit the complex valued data as described above, a thorough calibration procedure should be included to eliminate the phase uncertainty in the complex valued measurements. R, σt (cid:40) σR = σt R cos φ − σt R sin φ + σt R and σt (45) Figure 11. Same as in figure 10, at breath no. 264. Figure 12. Same as in figure 10, at breath no. 266. of a "quality indicator" for an EIT image, i.e., to determine if the image can be trusted as representing primarily tidal ventilation and not some sort of artifact. As was indicated above with reference to figure 8, this observation could also be used as an indicator to distinguish when the right and/or the left lung is properly ventilated in contrast to other competing physiological mechanisms of conductivity change (blood perfusion, muscle contractions, etc) and which are not expected to follow the same linear law as with the normal tidal breathing. Another potentially interesting application is based on the assumption that the actual background conductivity of the lung is correlated with pulmonary fluid content. Pulmonary edema, which is a clinically relevant pathology both in adults and neonates, is Aparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing121.510.500.511.500.20.4<{}={}at264thbreathLeftLungRightLungHom.Theory<{"<} at 264th breath1020305101520253000.20.40.60.81={"<} at 264th breath1020305101520253000.20.40.60.81Figure11.Sameasinfigure10,atbreathno.264.1.510.500.511.500.20.4<{}={}at266thbreathLeftLungRightLungHom.Theory<{"<} at 266th breath1020305101520253000.20.40.60.81={"<} at 266th breath1020305101520253000.20.40.60.81Figure12.Sameasinfigure10,atbreathno.266.contractions,etc.,withoutmuchalveoliventilation.4.4.PossibleclinicalapplicationsThemainconclusionfromthetheoreticalanalysisinthispaperisthatthelosscotangentofthechangeinthecomplexvaluedconductivityofalungduringtidalbreathingisexpectedtobethesameasofthee↵ectivebackgroundconductivity.Hence,thepresenceofalinearrelationshipbetweenthecorrespondingrealandimaginarycomponentscouldserveaspartofa"qualityindicator"foranEITimage,i.e.,todetermineiftheimagecanbetrustedasrepresentingprimarilytidalventilationandnotsomesortofartifact.Aswasindicatedabovewithreferencetofigure8,thisobservationcouldalsobeusedasanindicatortodistinguishwhentherightand/ortheleftlungisproperlyventilatedincontrasttoothercompetingphysiologicalmechanismsofconductivitychange(bloodperfusion,musclecontractions,etc)andwhicharenotexpectedtofollowthesamelinearlawaswiththenormaltidalbreathing.Anotherpotentiallyinterestingapplicationisbasedontheassumptionthattheactualbackgroundconductivityofthelungiscorrelatedwithpulmonaryfluidcontent.Pulmonaryedema,whichisaclinicallyrelevantpathologybothinadultsandneonates,ischaracterizedbyanincreaseinpulmonaryfluidcontent[32].Hence,achangeinthelosscotangentofthelinearrelationshipbetweentherealandimaginarypartmightbeclinicallyusefulforthedetectionofpulmonaryedema.4.5.UncertaintiesOneparameterofmajoruncertaintyinthisstudyistheactualvolumefractionfofairduringtidalbreathing.However,sincethereisaveryhighcontrastbetweenthecomplexvaluedconductivityofairandtissue,thesensitivityanalysisofsection3.3isvalidshowingthattheresult(34)isindependentoff.Furthermore,thelosscotangent⌘of2=<{2}(1+j⌘)isessentiallythesameasthelosscotangentoftheaprioriparameterEI,asexpressedin(33)and(35).Hence,anuncertaintyinthevolumefractionfEIatend-inspiration(whereEIisgiven)willonlya↵ectthemagnitudeoftheestimatedparameter2,anditwillnota↵ecttheestimatedlosscotangent⌘northeconclusionmadeinrelationto(34).AnotheruncertaintyrelatedtothemeasuredCRADLEITdatadescribedinsection4.3aboveisduetothefactthatthemeasuredIandQdatahavebeenrotatedalreadyinthehardwareinordertominimizetheimaginary(Q)partofthedata.Themainpurposeofthisprocedureistominimizethee↵ectofthephaseshiftsintheinternalelectronicsaswellasinthestraycapacitancesinleadsandelectrodes,etc.However,thismeansalsothatthereisanunknownrotationanglepresentinthedatarelatingtoa"true"resistivebackground.Notably,sincethereconstructionmatrixinourcaseisrealvalued,thepresenceoftheunknownrotationanglemeanssimplythatouractualestimatesarealsorotatedas(R=tRcostIsin,I=tRsin+tIcos,(45)wheretRandtIarethedesired"true"estimatesthatwouldhaveresultedifthecalibrationwouldhavebeencarriedoutbyusinga"true"resistivephantom.Aparametricmodelforthechangesinthecomplexvaluedconductivityofalungduringtidalbreathing121.510.500.511.500.20.4<{}={}at264thbreathLeftLungRightLungHom.Theory<{"<} at 264th breath1020305101520253000.20.40.60.81={"<} at 264th breath1020305101520253000.20.40.60.81Figure11.Sameasinfigure10,atbreathno.264.1.510.500.511.500.20.4<{}={}at266thbreathLeftLungRightLungHom.Theory<{"<} at 266th breath1020305101520253000.20.40.60.81={"<} at 266th breath1020305101520253000.20.40.60.81Figure12.Sameasinfigure10,atbreathno.266.contractions,etc.,withoutmuchalveoliventilation.4.4.PossibleclinicalapplicationsThemainconclusionfromthetheoreticalanalysisinthispaperisthatthelosscotangentofthechangeinthecomplexvaluedconductivityofalungduringtidalbreathingisexpectedtobethesameasofthee↵ectivebackgroundconductivity.Hence,thepresenceofalinearrelationshipbetweenthecorrespondingrealandimaginarycomponentscouldserveaspartofa"qualityindicator"foranEITimage,i.e.,todetermineiftheimagecanbetrustedasrepresentingprimarilytidalventilationandnotsomesortofartifact.Aswasindicatedabovewithreferencetofigure8,thisobservationcouldalsobeusedasanindicatortodistinguishwhentherightand/ortheleftlungisproperlyventilatedincontrasttoothercompetingphysiologicalmechanismsofconductivitychange(bloodperfusion,musclecontractions,etc)andwhicharenotexpectedtofollowthesamelinearlawaswiththenormaltidalbreathing.Anotherpotentiallyinterestingapplicationisbasedontheassumptionthattheactualbackgroundconductivityofthelungiscorrelatedwithpulmonaryfluidcontent.Pulmonaryedema,whichisaclinicallyrelevantpathologybothinadultsandneonates,ischaracterizedbyanincreaseinpulmonaryfluidcontent[32].Hence,achangeinthelosscotangentofthelinearrelationshipbetweentherealandimaginarypartmightbeclinicallyusefulforthedetectionofpulmonaryedema.4.5.UncertaintiesOneparameterofmajoruncertaintyinthisstudyistheactualvolumefractionfofairduringtidalbreathing.However,sincethereisaveryhighcontrastbetweenthecomplexvaluedconductivityofairandtissue,thesensitivityanalysisofsection3.3isvalidshowingthattheresult(34)isindependentoff.Furthermore,thelosscotangent⌘of2=<{2}(1+j⌘)isessentiallythesameasthelosscotangentoftheaprioriparameterEI,asexpressedin(33)and(35).Hence,anuncertaintyinthevolumefractionfEIatend-inspiration(whereEIisgiven)willonlya↵ectthemagnitudeoftheestimatedparameter2,anditwillnota↵ecttheestimatedlosscotangent⌘northeconclusionmadeinrelationto(34).AnotheruncertaintyrelatedtothemeasuredCRADLEITdatadescribedinsection4.3aboveisduetothefactthatthemeasuredIandQdatahavebeenrotatedalreadyinthehardwareinordertominimizetheimaginary(Q)partofthedata.Themainpurposeofthisprocedureistominimizethee↵ectofthephaseshiftsintheinternalelectronicsaswellasinthestraycapacitancesinleadsandelectrodes,etc.However,thismeansalsothatthereisanunknownrotationanglepresentinthedatarelatingtoa"true"resistivebackground.Notably,sincethereconstructionmatrixinourcaseisrealvalued,thepresenceoftheunknownrotationanglemeanssimplythatouractualestimatesarealsorotatedas(R=tRcostIsin,I=tRsin+tIcos,(45)wheretRandtIarethedesired"true"estimatesthatwouldhaveresultedifthecalibrationwouldhavebeencarriedoutbyusinga"true"resistivephantom.A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 13 5. Summary and conclusions the changes A theoretical model based on classical homogenization theory and the Hashin-Shtrikman coated ellipsoids has been derived to model in the complex valued conductivity of a lung during tidal breathing. Here, the alveolar air-filling corresponds to the inclusion phase of the two-phase composite material. The model predicts a linear relationship between the real and the imaginary parts of the changes in the complex valued conductivity of a lung during tidal breathing, and where the loss cotangent of the change is approximately the same as of the effective background conductivity. Hence, even though the magnitude of the change depend on the chosen ellipsoidal model, the loss cotangent of the change is virtually independent of the shape and orientation (anisotropy) of the alveoli. The theory has been illustrated with numerical examples, as well as by using reconstructed EIT images based on measurement data from the ongoing CRADL study. The new theory may have a potential for a development of new improved image reconstruction algorithms exploiting complex valued measurement data, and/or to define new clinically useful outcome parameters in lung EIT. Acknowledgments This project has received funding from the European Unions Horizon 2020 research and innovation pro- gramme under grant agreement No. 668259 (CRADL project). The work has also been supported by the Swedish Foundation for Strategic Research (SSF) un- der the programme Applied Mathematics and the project Complex Analysis and Convex Optimization for EM Design. References [1] I. Frerichs. Electrical impedance tomography (EIT) in applications related to lung and ventilation: a review of experimental and clinical activities. Physiological Measurement, 15(2):1, 2000. [2] H. R. Carlisle, R. K. Armstrong, P. G. Davis, A. Schibler, I. Frerichs, and D. G. Tingay. Regional distribution of blood volume within the preterm infant thorax during Intensive Care synchronised mechanical ventilation. Med, 36:2101–2108, 2010. [3] I. Frerichs, M. B. P. Amato, A. H. van Kaam, D. G. Tingay, Z. Zhao, B. Grychtol, M. Bodenstein, H. Gagnon, S. H. Bohm, E. Teschner, O. Stenqvist, T. Mauri, V. Torsani, L. Camporota, A. Schibler, G. K. Wolf, D. Gommers, S. Leonhardt, and A. Adler. Chest electrical impedance tomography examination, data analysis, terminology, clinical use and recommendations: consensus statement of the TRanslational EIT developmeNt stuDy group. Thorax, 72:83–93, 2017. doi:10.1136/thoraxjnl-2016- 208357. [4] Erkki Somersalo, Margaret Cheney, and David Isaacson. Existence and uniqueness for electrode models for electric current computed tomography. SIAM J. Appl. Math., 52(4):1023–1040, 1992. [5] M. Cheney, D. Isaacson, and J. C. Newell. Electrical impedance tomography. SIAM Review, 41(1):85–101, 1999. [6] R. Bayford. (electrical impedance tomography). Annu. Rev. Biomed. Eng., 8:63–91, 2006. Bioimpedance tomography [7] S. Nordebo, R. Bayford, B. Bengtsson, A. Fhager, M. Gustafsson, P. Hashemzadeh, B. Nilsson, T. Ry- lander, and T. Sjod´en. An adjoint field approach to Fisher information-based sensitivity analysis in electri- cal impedance tomography. Inverse Problems, 26, 2010. 125008. [8] A. Adler and R. Guardo. Electrical impedance tomography: IEEE regularized imaging and contrast detection. Transactions on Medical Imaging, 15(2):170–179, 1996. [9] A. Adler, J. H. Arnold abd R. Bayford, A. Borsic, B. Brown, P. Dixon, T. J. C. Faes, I. Frerichs, H. Gagnon, Y. Garber, B. Grychtol, G. Hahn, W. R. B. Lionheart, A. Malik, R. P. Patterson, J. Stocks, A. Tizzard, N. Weiler, and G. K. Wolf. GREIT: a unified approach to 2D linear EIT reconstruction of lung images. Physiological Measurement, 30(6):35–55, 2009. [10] D. Khodadad, S. Nordebo, N. Seifnaraghi, A. D. Waldmann, B. Muller, and R. Bayford. Breath detection using short-time Fourier transform analysis In 32nd URSI in Electrical Impedance Tomography. General Assembly & Scientific Symposium, pages 2185– 2187, Montreal, August 2017. [11] Margaret Cheney and David Isaacson. Issues in electrical IEEE Computational Science & impedance imaging. Engineering, pages 53–62, 1995. [12] A. Adler, R. Amyot, R. Guardo, J. H. T. Bates, and Y. Berthiaume. Monitoring changes in lung air and liquid volumes with electrical impedance tomography. Journal of Applied Physiology, 83(5):1762–1767, 1997. [13] S. Gabriel, R. W. Lau, and C. Gabriel. The dielectric properties of biological tissues: II. Measurements in the frequency range 10 Hz to 20 GHz. Phys. Med. Biol., 41:2251–2269, 1996. [14] D. Isaacson, J. L. Mueller, J. C. Newell, and S. Siltanen. Imaging cardiac activity by the D-bar method for electrical impedance tomography. Physiol. Meas., 27:43– 50, 2006. [15] S. J. Hamilton, J. L. Mueller, and M. Alsaker. Incorporat- ing a spatial prior into nonlinear D-bar EIT imaging for IEEE Transactions on Medical complex admittivities. Imaging, 36(2):457–466, 2017. [16] C. Gabriel, S. Gabriel, and E. Corthout. The dielectric I. Literature survey. properties of biological tissues: Phys. Med. Biol., 41:2231–2249, 1996. [17] S. Gabriel, R. W. Lau, and C. Gabriel. The dielectric properties of biological tissues: III. Parametric models for the dielectric spectrum of tissues. Phys. Med. Biol., 41:2271–2293, 1996. [18] William F. Ganong. Review of medical physiology. Lange Medical Books/McGraw-Hill, New York, 21st edition, 2003. [19] D. K. Molina and V. J. DiMaio. Normal organ weights in men: part II-the brain, lungs, liver, spleen, and kidneys. Am J Forensic Med Pathol., 33:368–372, 2012. [20] P. Nopp, E. Rapp, H. Pfutzner, H. Nakesch, and C. Ruhsam. Dielectric properties of lung tissue as a function of air content. Phys. Med. Biol., 38:699–716, 1993. [21] P. Nopp, N. D. Harris, T. X. Zhao, and B. H. Brown. Model for the dielectric properties of human lung tissue against frequency and air content. Medical & Biological Engineering & Computing, 35(6):695–702, 1997. A parametric model for the changes in the complex valued conductivity of a lung during tidal breathing 14 [22] J-R Wang, B-Y Sun, H-X Wang, S Pang, X Xu, and Q Sun. Experimental study of dielectric properties of human lung tissue in vitro. Journal of Medical and Biological Engineering, 34(6):598–604, 2014. [23] Ari Sihvola. Electromagnetic Mixing Formulae and IEE Electromagnetic Waves Series, 47. Applications. Institution of Electrical Engineers, 1999. [24] Graeme W. Milton. The Theory of Composites. Cambridge University Press, Cambridge, U.K., 2002. [25] J. D. Jackson. Classical Electrodynamics. John Wiley & Sons, New York, third edition, 1999. [26] H. M. Nussenzveig. Causality and dispersion relations. Academic Press, London, 1972. [27] F. W. King. Hilbert transforms vol. I–II. Cambridge University Press, 2009. [28] P. M. Morse and H. Feshbach. Methods of Theoretical Physics, volume 2. McGraw-Hill, New York, 1953. [29] C. F. Bohren and D. R. Huffman. Absorption and Scattering of Light by Small Particles. John Wiley & Sons, New York, 1983. [30] J. A. Osborn. Demagnetizing factors of the general ellipsoid. Phys. Rev., 67:351–357, 1945. [31] D. Zwillinger. CRC Standard Mathematical Tables and Formulae. Chapman & Hall/CRC Press LLC, 31st edition, 2003. [32] C. J. C. Trepte, C. R. Phillips, J. Sol`a, A. Adler, S. A. Haas, M. Rapin, S. H. Bohm, and D. A. Reuter. Electrical impedance tomography (EIT) for quantification of pulmonary edema in acute lung injury. Critical Care, 2(18):1–9, 2016.
1006.1069	1	1006	2010-06-05T21:23:34	Multimodal transition and stochastic antiresonance in squid giant axons	[ "physics.bio-ph", "q-bio.NC" ]	The experimental data of N. Takahashi, Y. Hanyu, T. Musha, R. Kubo, and G. Matsumoto, Physica D \textbf{43}, 318 (1990), on the response of squid giant axons stimulated by periodic sequence of short current pulses is interpreted within the Hodgkin-Huxley model. The minimum of the firing rate as a function of the stimulus amplitude $I_0$ in the high-frequency regime is due to the multimodal transition. Below this singular point only odd multiples of the driving period remain and the system is highly sensitive to noise. The coefficient of variation has a maximum and the firing rate has a minimum as a function of the noise intensity which is an indication of the stochastic coherence antiresonance. The model calculations reproduce the frequency of occurrence of the most common modes in the vicinity of the transition. A linear relation of output frequency vs. $I_0$ for above the transition is also confirmed.	physics.bio-ph	physics	Multimodal transition and stochastic antiresonance in squid giant axons Faculty of Physics, Adam Mickiewicz University, Umultowska 85, 61-614 Poznan, Poland L. S. Borkowski The experimental data of N. Takahashi, Y. Hanyu, T. Musha, R. Kubo, and G. Matsumoto, Physica D 43, 318 (1990), on the response of squid giant axons stimulated by periodic sequence of short current pulses is interpreted within the Hodgkin-Huxley model. The minimum of the firing rate as a function of the stimulus amplitude I0 in the high-frequency regime is due to the multimodal transition. Below this singular point only odd multiples of the driving period remain and the system is highly sensitive to noise. The coefficient of variation has a maximum and the firing rate has a minimum as a function of the noise intensity which is an indication of the stochastic coherence antiresonance. The model calculations reproduce the frequency of occurrence of the most common modes in the vicinity of the transition. A linear relation of output frequency vs. I0 for above the transition is also confirmed. PACS numbers: 87.19.ll,87.19.ln,87.19.lc The Hodgkin-Huxley (HH) model[1] is a prototypical resonant neuron with the main resonant frequency typi- cally of order 40 to 60 Hz. Its output interspike intervals (ISI) can be classified in terms of integer multiples of the driving period. The multimodality is revealed when the HH neuron is stimulated by noisy inputs, such as additive noise[2, 3], random synaptic inputs[2, 4, 5] or channel noise[6]. Such ISI histograms are encountered frequently in periodically forced sensory neurons. An explanation in terms of a two-state system with noise was put forward by Longtin et al.[7]. The multimodal character is manifest also in a deterministic HH model near excitation threshold[8, 9] and in regimes of irregular response between mode-locked states[9]. It was shown recently that also the parity of ISI plays a significant role[10]. Even (odd) modes dominate in the vicinity of even (odd) mode-locked states, respectively. The most significant manifestation of this effect is the multimodal odd-all transition between states 3:1 and 2:1[10], where the coefficient of variation (CV) has a maximum and the firing rate has a minimum. The notation p:q means p output spikes for every q input current pulses. Below this singularity only odd multiples of the input period exist and above it harmonics of both parities participate in the response. The transition may be crossed by vary- ing either the stimulus amplitude or the input period. The minimum of the firing rate occurs slightly above the transition. In earlier experiments in giant axons of squid stimu- lated periodically by a train of short rectangular current pulses the firing rate, defined as the ratio of the out- put and input frequency fo/fi, had a well pronounced minimum as a function of the interval between adjacent pulses[11] or the stimulus amplitude[12]. Even modes were absent below the minimum[12]. This effect occurred near the excitation threshold, between states 3:1 and 2:1. Another interesting result was the continuous relation be- tween the firing rate and the stimulus amplitude. This set of experimental and theoretical results deserves a more detailed comparison. The theory can be tested also by considering a pe- riodic drive in the presence of noise. Noisy biological systems[2, 5, 13 -- 16], including the HH neuron, are known to exhibit stochastic resonance (SR). This phenomenon is mainly, though not exclusively, characterized by a maxi- mum of the signal to noise ratio as a function of the noise intensity. Another effect associated with the presence of noise is the decrease of the firing threshold and the coher- ence resonance[17, 18], where the minimum variability of the output signal, expressed by CV in absence of a de- terministic drive, is achieved at some intermediate noise strength. Recently it was found experimentally[19, 20] theoretically[21, 22] that small amplitude noise may de- crease the firing rate or even turn it off. The nonlinear system in the vicinity of the multimodal transition is a natural candidate for finding interesting effects due to noise since the trajectories of different modes are very close in parameter space. In the following we compare experimental data to theoretical results for the determin- istic case and calculate the sytem's response to a periodic drive with additive Gaussian noise. In the experiment of Takahashi et al.[12] the squid axon was stimulated by periodic train of rectangular current steps of width 0.6ms. Fig. 1 shows the experimentally obtained firing rate as a function of stimulus amplitude scaled by the minimum current threshold It. On the left side of the minimum only odd modes were recorded. Even modes were present at the minimum point, with the 6 : 1 mode occurring more frequently than the 4 : 1 com- ponent, and 2 : 1 entirely absent. This is consistent with calculation results[10], where even modes disappear be- fore reaching the multimodal transition (which is slightly below the minimum of the firing rate), with the low order modes vanishing first, beginning with mode 2 : 1. We try to reproduce this type of dependence using the HH model with the classic parameter set and rate 2 TABLE I: Frequency of occurrence of the six lowest modes at the minimum of the firing rate. The upper row is based on Fig. 13e from the experimental data of Takahashi et al.[12]. The bottom row is the result of calculations, assuming Ti = 7ms and I0 = 18µA/cm2 (see Fig. 4). mode 2 0 0.002 3 0.66 0.74 4 0.04 0.07 5 0.12 0.11 6 0.09 0.04 7 0.04 0.02 the axon used by Hodgkin and Huxley was of poor qual- ity and in later studies significantly higher conductivities were obtained. Paydarfar et al.[19] in their recent study recorded firing periods in the range between 7 and 16 ms. The overall dynamics of Figs. 1 and 2 agrees very well, including the location and depth of the local min- ima. We verified that the form of Fig. 2 was unchanged for pulse widths between 0 and 1ms after dividing the current amplitude by R Ti 0 I(t)dt. Fig. 3 shows the response diagram in the high- frequency limit. The dotted line separates the monos- table firing solution from the silent state and bistable areas where the limit cycle coexists with a fixed point so- lution. Boundaries of bistability were determined using a continuation method starting from a region with a single solution. FIG. 1: The average firing rate, ¯To/Ti, as a function of the stimulus amplitude I from the work of Takahashi et al.[12]. It is the minimum current threshold obtained in the range Ti = 2.5ms to Ti = 6.5ms. The measurements were carried out at Ti = 3.8ms. constants[1], C dV dt = −IN a − IK − IL + Iapp, (1) where IN a, IK , IL, Iapp, are the sodium, potassium, leak, and external current, respectively. C = 1µF/cm2 is the membrane capacitance. The input current is a periodic set of rectangular steps of width 0.6ms and height I0. Equations are integrated within the fourth order Runge- Kutta scheme with a time step of 0.001ms. The data points are obtained from runs of 400 s, discarding the initial 4 s. The dependence of the firing rate on the stim- ulus amplitude is shown in Fig. 2, where Ti = 7ms. FIG. 2: The calculated average firing rate at Ti = 7ms with- out noise. The similarity to experiment is striking. Although the calculated minimum occurs at almost twice the experi- mental Ti, the other time scales such as the refractory period and the time span of the bifurcation diagram dif- fer by a similar factor. The entire dynamics of the axon from the study of Takahashi et al. is significantly faster than that of Hodgkin and Huxley. This difference of time scales is not unusual. Long time ago Best[23] noted that FIG. 3: The bifurcation diagram in the Ti-I0 plane, showing the main mode-locked states in the model without noise. The unmarked intrusion in the upper left corner is the 5:1 state. The bottom part of the figure is occupied by the silent state. In the firing part of the diagram there are two solutions below the dotted line. Here the limit cycle coexists with the fixed point. Full squares show the location of the minima of the firing rate. The borders of states below Ti = 4.5ms are shown in an approximate form. The detailed picture is less regular and somewhat more complex. The experimental local maximum on the plateau fo/fi = 0.4 is due to the state 10100, where modes 2:1 and 3:1 alternate. The other local maximum at fo/fi = 0.429 with tendency to lock into the (10)2100 was also reproduced. Fig. 4 shows the relative frequency of participation of the most common modes on a logarith- mic scale. Higher-order modes appear more frequently near the minimum of the firing rate. Experimental and calculated ISI histograms are compared in Table I. The overall agreement is quite remarkable. Also the calcu- lated evolution of individual modes as a function of I0 is close to measured values. In experiment the probability of appearance of mode 4:1 between I0/It = 1.2 and 1.3 remains in the range 0.06 to 0.08, which agrees well with Fig. 4 for I0 between 18µA/cm2 and 22µA/cm2. The published experimental runs[12] contain 80 to 100 output spikes for selected data points. On the basis of this data set we can conclude that the frequency of participation of the low order modes is approximately reproduced in simulations. Above the multimodal transition the exper- imental firing rate near the threshold rises linearly with the stimulus amplitude, see Fig. 5. The dependence of fo/fi vs. of I0 is well reproduced in Fig. 6, except in the vicinity of the 2:1 plateau, where an addition of a small amount of noise would improve the fit. FIG. 4: The relative frequency of occurrence of low-order even and odd modes for the parameter set of Fig. 2 The vertical line marks the position of the minimum of the firing rate. We now consider the model with a Gaussian white noise: C dV dt = −IN a − IK − IL + Iapp + Cξ(t), (2) where < ξi(t) >= 0, < ξ(t)ξ(t′) >= 2Dδ(t − t′), and D is expressed in mV2/ms. The HH equations are in- tegrated using the second-order stochastic Runge-Kutta algorithm[24]. The simulations are carried out with the time step of 0.01ms and are run for 400 s, discarding the initial 40 s. There is a tendency to assume that biological systems, including neurons, should always be treated as noisy sys- 3 FIG. 5: The linear relation of the firing rate vs. the stimulus amplitude above the multimodal transition point at Ti = 4ms. These are experimental results of Takahashi et al.[12]. FIG. 6: Calculated average firing rate vs. stimulus amplitude above the multimodal transition for three values of Ti. The current pulse width is 0.6ms. tems. While the neuron is sensitive to noise it is not ob- vious that single neuron dynamics should always include stochastic terms. Fig. 7 shows the quick disappearance of the fo/fi = 0.4 plateau in Fig. 2 with increasing noise. Comparing with the experimental data in Fig. 1 we con- clude that calculations reproduce experimental data only for D < 10−4. Certainly more experiments are needed to FIG. 7: Sensitivity of the fo/fi = 0.4 plateau from Fig. 2 to noise. understand the role of noise in neurons. Fig. 8 presents the firing rate as a function of D for three parameter sets from the 3 : 1 plateau of Fig. 2. For small noise fo/fi drops quickly below 1/3 over an en- tire plateau, with the biggest drop near the edges. This behavior should be contrasted with the resonant regime where the central part of each plateau maintains phase locking over much larger range of noise intensities and D ∼ 1 is needed to lower the firing rate of an entire plateau below the D = 0 value[9]. Another difference is the direction of frequency changes at the plateau edges. In the resonant state the frequency below (above) the plateau midpoint is lowered (increased), respectively[9]. In the antiresonant limit the entire plateau is unstable to even a small noise which slows down the system consid- erably. FIG. 8: The firing rate vs. the noise intensity. The middle curve was obtained for I0 = 15µA/cm2. Here Ti = 7ms. At D = 0 all three curves start in the 3:1 mode. CV as a function of D has a maximum for the same parameter set, see Fig. 9. The increased variability is associated with increased participation of higher-order modes and may be called a stochastic coherence antires- onance. A maximum of CV was found earlier in a leaky integrate-and-fire model with an absolute refractory pe- riod for suprathreshold base current[25]. A small local maximum of CV at intermediate noise level was also found by Luccioli et al.[5] in a HH model driven by a dc current in a bistable regime, where the neuron was stim- ulated by a large number of stochastic inhibitory and excitatory postsynaptic potentials. It was pointed out that the stochastic antiresonance may exist in regions of bistability[22], when the stable limit cycle coexists with other attractors. This typically occurs in the vicinity of a bifurcation when the value of the bifurcation parame- ter slightly exceeds the critical value. In the HH model near the multimodal transition there are many compet- ing limit cycles. Noise enhances trajectory switching and may even stop the firing entirely. A decrease of the fir- ing rate and an increase of incoherence may occur along much of the excitation threshold, where the deterministic system is bistable or responds irregularly[9]. 4 FIG. 9: The maximum of the coefficient of variation as a function of the noise intensity is a property of the stochastic coherence antiresonance. The maximum of CV and the min- imum of the firing rate occur at different noise levels. The irregularity of the I0 = 14µA/cm2 curve is a consequence of proximity to the excitation threshold. In conclusion, numerical solutions of the determinis- tic HH equations show that the minimum of the firing rate observed by Takahashi et al.[12] is due to the multi- modal transition[10]. The statistics of the experimental spike trains confirm that below the transition only odd modes remain. Even modes are present at the minimum of fo/fi, in agreement with theoretical calculations[10]. The calculated frequencies of occurrence of the most com- mon modes are close to experimental values. Also the location of the minimum of fo/fi in the vicinity of the 3:1 state is consistent with the simulations. The linear rise of the output frequency as a function of the stim- ulus strength above the multimodal transition was also confirmed. The excitation threshold in the antiresonant limit is higher by about a factor of two compared to the resonant regime. The rise of threshold for frequencies of current pulses exceeding the resonant frequency was ob- served experimentally by Kaplan et al.[26]. Further sup- port for the significance of the parity of the modes comes from the experiment of Paydarfar et al.[19], who found that the quiescent periods between highly regular bursts were always equal to even multiples of the resonant pe- riod. An ISI histogram with odd modes was obtained by Racicot and Longtin[27] in a chaotically forced FitzHugh- Nagumo (FHN) model. FHN equations are often used as a substitute for the full HH model. It would therefore be useful to investigate whether the main features of the odd-all ISI transition are reproduced in the FHN model with a deterministic and stochastic drive. Perturbing the system with noise changes significantly the f vs. I0 dependence. The local minima of this curve disappear already for D ≃ 10−3. In the regime below the multimodal transition the 3:1 plateau disappears rapidly for very small noise. The firing rate has a minimum and CV has a maximum as a function of the noise inten- sity. These predictions are expected to be valid for short stimuli of different shapes and can be tested experimen- tally. The multimodal transition and the accompanying stochastic antiresonance are important both for the un- derstanding of excitable systems and for potential neu- rological applications. Spike annihilation by determin- istic signals[28] is studied in the context of deep brain stimulation[29, 30], a therapeutic technique, in which synchronicity of certain parts of the brain is reduced by brief current pulses. [1] A. L. Hodgkin and A. F. Huxley, J. Physiol. (London) 117, 500 (1952). [2] P. H. E. Tiesinga, J. V. Jos´e, and T. J. Sejnowski, Phys. Rev. E 62, 8413 (2000). [3] B. Aguera y Arcas, A. L. Fairhall, W. Bialek, Neural Comput. 15, 1715 (2003). [4] D. Brown, J. Feng, and S. Feerick, Phys. Rev. Lett. 82, 4731 (1999). [5] S. Luccioli, T. Kreuz, and A. Torcini, Phys. Rev. E 73, 041902 (2006). [6] P. Rowat, Neural Comput. 19, 1215 (2007). [7] A. Longtin, A. Bulsara, and F. Moss, Phys. Rev. Lett. 67, 656 (1991). [8] J. R. Clay, J. Comput. Neurosci. 15, 43 (2003). [9] L. S. Borkowski, Nonlinear dynamics of Hodgkin-Huxley neurons (Adam Mickiewicz University Press, 2010). [10] L. S. Borkowski, Phys. Rev. E 80, 051914 (2009). [11] G. Matsumoto, K. Aihara, Y. Hanyu, N. Takahashi, S. Yoshizawa, and J. Nagumo, Phys. Lett. A 123, 162 (1987). 5 [12] N. Takahashi, Y. Hanyu, T. Musha, R. Kubo, and G. Matsumoto, Physica D 43, 318 (1990). [13] K. Wiesenfeld and F. Moss, Nature 373, 33 (1995). [14] B. Doiron, A. Longtin, N. Berman, and L. Maler, Neural Comput. 13, 227 (2000). [15] W. C. Stacey and D. M. Durand, J. Neurophysiol. 86, 1104 (2001). [16] M. Rudolph and A. Destexhe, J. Comput. Neurosci. 11, 19 (2001). [17] Hu Gang, T. Ditzinger, C. Z. Ning, and H. Haken, Phys. Rev. Lett. 71, 807 (1993). [18] A. S. Pikovsky and J. Kurths, Phys. Rev. Lett. 78, 775 (1997). [19] D. Paydarfar, D. B. Forger, and J. R. Clay, J. Neuro- physiol. 96, 3338 (2006). [20] C. K. Sim and D. B. Forger, J. Biol. Rhythms 22, 445 (2007). [21] B. S. Gutkin, J. Jost, and H. C. Tuckwell, Europhys. Lett. 81, 20005 (2008). [22] B. S. Gutkin, J. Jost, and H. C. Tuckwell, Naturwis- senschaften 96, 1091 (2009). [23] E. N. Best, Biophys. J. 27, 87 (1979). [24] R. L. Honeycutt, Phys. Rev. A 45, 600 (1992). [25] B. Lindner, L. Schimansky-Geier, and A. Longtin, Phys. Rev. E 66, 031916 (2002). [26] D. T. Kaplan, J. R. Clay, T. Manning, L. Glass, M. R. Guevara, and A. Shrier, Phys. Rev. Lett. 76, 4074 (1996). [27] D. M. Racicot and A. Longtin, Physica D 104, 184 (1997). [28] R. Guttman, S. Lewis, and J. Rinzel, J. Physiol. 305, 377 (1980). [29] P. A. Tass, Phys. Rev. E 66, 036226 (2002). [30] D. Calitoiu, B. J. Oommen, and D. Nussbaum, Biol. Cy- bern. 98, 239 (2007).
1012.3743	3	1012	2011-05-26T13:20:51	Bistability and resonance in the periodically stimulated Hodgkin-Huxley model with noise	[ "physics.bio-ph", "q-bio.NC" ]	We describe general characteristics of the Hodgkin-Huxley neuron's response to a periodic train of short current pulses with Gaussian noise. The deterministic neuron is bistable for antiresonant frequencies. When the stimuli arrive at the resonant frequency the firing rate is a continuous function of the current amplitude $I_0$ and scales as $(I_0-I_{th})^{1/2}$, where $I_{th}$ is an approximate threshold. Intervals of continuous irregular response alternate with integer mode-locked regions with bistable excitation edge. There is an even-all multimodal transition between the 2:1 and 3:1 states in the vicinity of the main resonance, which is analogous to the odd-all transition discovered earlier in the high-frequency regime. For $I_0<I_{th}$ and small noise the firing rate has a maximum at the resonant frequency. For larger noise and subthreshold stimulation the maximum firing rate initially shifts towards lower frequencies, then returns to higher frequencies in the limit of large noise. The stochastic coherence antiresonance, defined as the maximum of the coefficient of variation as a function of noise intensity, occurs over a wide range of parameter values, including monostable regions.	physics.bio-ph	physics	Bistability and resonance in the periodically stimulated Hodgkin-Huxley model with noise Department of Physics, Adam Mickiewicz University, Umultowska 85, 61-614 Poznan, Poland L. S. Borkowski We describe general characteristics of the Hodgkin-Huxley neuron's response to a periodic train of short current pulses with Gaussian noise. The deterministic neuron is bistable for antiresonant frequencies. When the stimuli arrive at the resonant frequency the firing rate is a continuous function of the current amplitude I0 and scales as (I0 − Ith)1/2, characteristic of a saddle-node bifurcation at the threshold Ith. Intervals of continuous irregular response alternate with integer mode-locked regions with bistable excitation edge. There is an even-all multimodal transition between the 2:1 and 3:1 states in the vicinity of the main resonance, which is analogous to the odd-all transition discovered earlier in the high-frequency regime. For I0 < Ith and small noise the firing rate has a maximum at the resonant frequency. For larger noise and subthreshold stimulation the maximum firing rate initially shifts toward lower frequencies, then returns to higher frequencies in the limit of large noise. The stochastic coherence antiresonance, defined as a simultaneous occurrence of (i) the maximum of the coefficient of variation, and (ii) the minimum of the firing rate vs. the noise intensity, occurs over a wide range of parameter values, including monostable regions. Results of this work can be verified experimentally. PACS numbers: 87.19.lb,87.19.lc,87.19.ln I. INTRODUCTION The Hodgkin-Huxley (HH) model [1] is a prime exam- ple of a resonant neuron. It was originally developed to explain experimental properties of the squid giant axon. Its behavior under the influence of constant, periodic [2 -- 6] and irregular external current [7] has been studied ex- tensively. It also served as a starting point for a number of reduced models [8 -- 10], designed to preserve the key features, while being more amenable to large-scale nu- merical simulations. However, a full understanding of all qualitative properties of its solutions has not been achieved yet. Deterministic HH equations can have both periodic and aperiodic, sometimes chaotic solutions [11]. Theo- retical [12 -- 14] and experimental [3, 13] analysis revealed that near the excitation threshold two solutions, the fixed point and the limit cycle, may coexist. A simple such ex- ample is the HH model driven by a constant current. As the current magnitude is increased the neuron starts re- sponding at a preferred frequency, which is close to 50 Hz for the original HH parameter set. The HH neuron in the presence of noise may display ei- ther resonant or antiresonant behavior, depending on the signal magnitude, frequency, and noise. The enhance- ment of weak signals by noise is known as stochastic res- onance [15 -- 17]. The opposite effect, stochastic antires- onance, in which a neuron's firing frequency is slowed down or even entirely stopped at some intermediate noise level, received some attention [18 -- 20] recently. Depend- ing on model parameters this behavior is associated with bistability [18, 19] or multimodality of the response [21]. A relation between bistability and antiresonance in inte- grator neurons was also studied recently [22]. In two previous papers on the HH model driven by pe- riodic sequence of short stimuli, we have shown that there is a transition between odd-only and all modes at high frequencies [21, 23]. This transition is located between the locked-in regions 2:1 and 3:1. The notation p : q means q output spikes for every p input current pulses. The edges of individual modes scale logarithmically in the vicinity of this singularity. This theoretical analysis agrees well with experimental data [24]. A natural ques- tion to ask is whether we can identify analogous parity transition, involving even-only modes on one side and all modes on the other side of the transition. A preliminary study [25] showed that the even and odd modes compete also near the main resonance. In the following we analyze the resonance regime in detail, looking for signatures of the even-all transition. Our aim is to map the response diagram of the HH neuron to a train of brief current pulses, rather than em- ulate typical in vivo scenarios. Stimuli in the form of brief pulses are better suited to reveal the internal dynamics of the neuron than signals varying on the time scale of the main resonance. In the case of the often used con- stant or sinusoidal input currents the neuron dynamics is obscured by the drive that is always different from zero. Gaussian noise is added to the model in order to reveal additional features of the model's dynamics. Although this particular type of randomness may not necessarily occur in a neuronal system we believe similar results will be obtained for any fast-varying irregular component of the stimulus. II. THE MODEL AND RESULTS We consider the stochastic HH model with the classic parameter set and rate constants [1], C dV dt = −IN a − IK − IL + Iapp + Cξ(t), (1) 2 der resonance is present at Ti ≃ 34ms and the third one at Ti ≃ 51ms (not shown). The dotted line separates the region with a single solution from the area where two solutions coexist. Bistable solutions appear between the resonant regions near Ti = 11ms and Ti = 23ms. In these regimes the transition to excitability occurs via the subcritical Hopf bifurcation. The picture is more complicated at the frequencies above 250 Hz, where the quiescent state often coexists not with a limit cycle but with a set of irregular orbits. The boundaries of bistable regions are determined with a simple continuation algo- rithm. The initial conditions of each run with a new value of a bifurcation parameter are equal to the end values from the previous iteration. The firing rate fo/fi depends continuously on I0 be- tween the tip of the resonance at Ti ≃ 17ms and the bistable area, which begins at Ti ≃ 20ms. Here fo, fi is output, input frequency, respectively. For 14ms < Ti < 17ms bistable intervals with integer p/q ratio alter- nate with irregular response, whose long-time average is a continuous function of I0 and Ti. Earlier, Clay reported [27] an irregular graded response near the excitation edge within a revised HH model [28] stimulated by 1-ms rect- angular current pulses and trains of half-sine waves. This variability was attributed to the deterministic nonlinear dynamics of the model. FIG. 1: Response diagram in the noiseless case for stimulation by rectangular current pulses of width τ = 0.6ms and height I0. The top (bottom) panel shows the high (intermediate) frequency regime. Bistable regions are marked "b". Some smaller bistable areas remain unlabeled. They are limited by a dotted line from above and a solid line from below. La- bels 1,2,3, and 4 mark regions with 1:1,2:1,3:1, and 4:1 mode locking, respectively. The resonance area near Ti = 17ms is dominated by the 3:1 state and higher order states. The sec- ond resonance near Ti = 34ms is dominated by the 2:1 mode. Higher modes appear near the threshold at Ti ≃ 34ms. where noise is given by the Gaussian process < ξi(t) >= 0, < ξ(t)ξ(t′) >= 2Dδ(t − t′), and D is expressed in mV2/ms. IN a, IK, IL, and Iapp, are the sodium, potas- sium, leak, and external current, respectively. C = 1µF/cm2 is the membrane capacitance. The input cur- rent is a periodic set of rectangular steps of height I0 and width 0.6ms, which is an order of magnitude be- low the resonant pulse width [20] and does not inter- fere with the neuron's internal dynamics. The deter- ministic (stochastic) HH equations are integrated us- ing the fourth-order Runge-Kutta algorithm (the second- order stochastic Runge-Kutta algorithm [26]), respec- tively. The simulations are carried out with the time step of 0.001ms and are run for 400s, discarding the initial 4s. Figure 1 shows the response diagram in the Ti − I0 plane without noise, where Ti is the stimulus period. The main resonance is located at Ti ≃ 17ms. The second or- FIG. 2: The firing rate as a function of the current pulse height at the resonance Ti = 17.5ms. Here τ = 0.6ms. Near the firing threshold fo is approximately a square root function of the pulse amplitude. Further away from the threshold the dependence of fo on I0 between the mode-locked states is irregular and nonmonotonic. For Ti ≥ 17ms the firing rate is approximately a square root function of the deviation from the threshold current amplitude (see Fig. 2). This dependence is characteris- tic of a saddle-node bifurcation [29]. The scaling fo/fi ∼ (I0 − Ith)β, with β = 1/2 is reminiscent of a mean-field second-order phase transition, with fo/fi playing the role of an order parameter. The same exponent is obtained for a relaxation time near Ic ≃ 6.264µA/cm2 for a con- stant current I [30], where Ic is the value of I at the saddle-node bifurcation. Below Ic the neuron returns 3 FIG. 4: Sample voltage trace in the chaotic regime near the transition between even-only and all modes. The applied cur- rent pulses are shown with dotted lines. There is only one odd multiple (7:1) of the input period in this sample. Here Ti = 13.9ms and I0 = 10.75µA/cm2. FIG. 5: The Poincare section of the membrane potential near the bistable region and irregular regime for current ampli- tude I0 = 10.65µA/cm2. The values of V (t) in each run are recorded at the points t = nTi, where n = 1, 2, .... The low- est and the highest flat branches on the left belong to the 2:1 states. The middle one belongs to the steady state. It undergoes a period doubling bifurcation on approach to the irregular regime above Ti = 13.3ms. ually transferred to the 4:1 mode, which dominates for Ti slightly below the even-all transition point Tea. On the other side of Tea, h5 = 1 over a narrow interval. The histogram of the dominant modes for D > 0 is shown in Fig. 6 (right). Some weight is now transferred to lower modes. The 4:1 and 5:1 modes are still well pronounced over a range of Ti but they are now mixed with 2:1 and 3:1 modes, respectively. Noise extends the range of presence of high order modes around Tea, as expected. The average firing rate between nearby n:1 and (n+1):1 states often has a narrow local minimum due to the pres- ence of slower modes. The minima are more pronounced near the excitation threshold along the left edge of the resonance tip in Fig. 3, where Ti < 17ms. One such valley is shown in Fig. 7. Here the 4:1 mode dominates in a narrow range of parameter values below the parity transition. The firing rate minimum is robust to small levels of FIG. 3: Details of the response diagram at the resonance. Borders of even (odd) mode-locked states are shown with bro- ken (continuous) lines respectively. The dotted lines mark the transitions to repetitive firing. Areas enclosed by the dotted line and continuous or broken lines are bistable. Numbers 2,...,5 indicate the states 2:1,...,5:1. The states beyond 7:1 are not shown. Filled squares indicate the parity multimodal transition between even-only and all modes. to quiescence after emitting a series of spikes. The re- laxation time, defined as the time from the first to the last spike, diverges as (Ic − I)∆, where ∆ ≃ 1/2. Sim- ilar values of ∆ were obtained for the Morris-Lecar and FitzHugh-Nagumo models [30]. Figure 3 shows the bifurcation diagram near the tip of the resonance. The synchronized states alternate with irregular firing. There are small bistable regions where the quiescent state coexists with a limit cycle. We have calculated their boundaries for states up to order 5:1. It is likely that they extend all the way to the tip of the resonance. The behavior in the intermediate regions is chaotic due to a competition between odd and even modes [23]. An example of V (t) dependence for a point located between states 2:1 and 3:1 is shown in Fig. 4. This sample contains only one odd multiple of the input period. Odd interspike intervals vanish in the vicinity of the 2:1 state. Careful analysis of the interspike interval histograms (ISIH) reveals a transition between the set of even and the set of all modes. This parity transition of ISIHs in the HH model could be tested experimentally in a squid axon experiment, similar to the odd-all transition discovered earlier [21, 24]. The transition between quiescence and chaotic firing marked by the dotted line in the intermediate zones be- tween the locked-in states occurs via period doubling (see Fig. 5). The lowest and the highest values of V below Ti ≃ 13.24ms belong to the 2:1 limit cycle. The line slightly below V = −65mV, splitting above Ti = 13ms, is associated with the steady state. The even, odd multiples of Ti dominate near the states with even, odd p/q ratio, respectively (see Fig. 6, left). This is a general property of regimes of irregular firing. Starting from the 2:1 mode, the histogram weight is grad- 4 FIG. 6: Histograms of the lowest order modes for D = 0 (left) and D = 10−5 (right) at I0 = 10.75µA/cm2. This is an irregular firing regime between the states 2:1 and 3:1. Data in the noisy case are Bezier-smoothed averages. FIG. 7: Minimum of the firing rate between the 2:1 and 3:1 states for the noiseless stimulus. The state 4:1 is located near the 2:1 mode. FIG. 8: The firing rate as a function of Ti in the intermediate region between the 2:1 and 3:1 states for D = 10−5 and Ti = 10.75ms. The competition of odd and even modes manifests itself as a minimum of the firing rate. noise, as can be seen in Fig. 8. The minimum of fo occurs for Ti & Tea, where modes of both parity are available. In the odd-all transition of the high-frequency regime [23] the minimum fo was similarly located close to the transition, on the side of all modes. It is well known that the HH neuron has a tendency to spike in bursts when subjected to a noisy stimulus in a bistable regime. If the deterministic system is prepared in one of high order bistable states from Fig. 3, the addition of noise results in slow bursts, where the ISI within a burst is a high integer multiple of Ti. An example is shown in Fig. 9, where each burst's ISI is equal to 3Ti. We expect slow bursts of order 6:1 and higher to be found in a more detailed calculation. For stimuli below the deterministic threshold noise ac- 5 FIG. 9: Sample voltage trace in the bistable regime of the state 3:1 under the influence of small noise. The most common form of response consists of 3:1 bursts separated by longer silent intervals. Here Ti = 13.9ms, I0 = 10.75µA/cm2, and D = 10−3. FIG. 10: The firing rate as a function of Ti for different noise levels and I = 10µA/cm2, about 0.1µA/cm2 below the deter- ministic threshold. tivates firing with maximum response at the resonant fre- quency (see Fig. 10). When noise is larger, excitations into the 2:1 state occur more frequently and the maxi- mum of the firing rate shifts toward Ti = 20ms. This maximum is a result of the interplay between the 1:1 mode and the 2:1 mode. The relative frequency of par- ticipation of the 2:1 mode grows sharply for intermediate noise levels in the neighborhood of the bistable regime for Ti ≃ 20ms. For signals with a periodic subthreshold component noise may play the role of a frequency selector. In a net- work with some elements firing in unison at the resonant frequency the remaining uncorrelated neurons also play an important role. The intensity of their background ac- tivity may select the firing rate of a network. The dependence of CV on Ti for fixed noise intensity is shown in Fig. 11. As expected, the most coherent response occurs near the resonance. The minimum near Ti = 20ms for suprathreshold inputs follows the left edge of the 1:1 region from Fig. 1. For stimuli below threshold there is a deep minimum at the resonance and a large maximum in the bistable regime for Ti > 20ms. For suprathreshold stimulation in the resonance regime the firing rate in general increases monotonically as a function of D with the exception of states having p/q FIG. 11: The coefficient of variation as a function of Ti for different pulse amplitudes. The noise intensity if fixed at D = 0.2. The local minimum at larger I0 follows the left edge of the 1:1 regime. values close to 1 (see Fig. 11). Starting in the 1:1 state at D = 0, increasing noise slows down the response by annihilating some of the action potential spikes. The minimum fo/fi is reached at an intermediate D. In the limit of large noise fo approaches the inverse of the refrac- tory period. The CV maxima are caused by redistribu- tion of histogram weight among several principal modes. This effect was described earlier as stochastic coherence antiresonance [21]. Maximization of spike train incoher- ence was shown earlier to occur in the FitzHugh-Nagumo [31] and leaky integrate-and-fire [32] models. However in those studies the maxima of CV were not accompanied by the minima of fo/fi. Figure 13 shows the response diagram in the pres- ence of noise. The tip of the resonance is broadened and shifted to higher frequencies. The 3:1 state vanishes completely for D slightly larger than 10−3. Higher-order states are more sensitive to noise and are quickly washed out. The bistable area at the edge of the 2:1 state shrinks more gradually. When bistability is finally eliminated, fo remains discontinuous over part of that border. For D = 10−2 the discontinuity occurs below Ti ≃ 10.5ms. Above Ti ≃ 10.5ms the firing rate is a continuous func- tion of I0. The behavior in the immediate vicinity of Ti = 10.5ms seems to be weakly irregular. A more de- tailed study would be needed for a proper description of this area. The loss of stability by the 2:1 state in favor of the quiescent state and further transition to bursting are illustrated in Fig. 14. The reaction to noise at high frequencies is qualita- tively different from behavior at near-resonant frequen- cies (see Fig. 15). In the top panel of Fig. 15 the firing rate drops almost everywhere for small D. This is easy to understand if we recall that the odd-all multimodal tran- sition [21, 23] occurs just below I0 = 20µA/cm2. The slow modes are easily available in this regime and small perturbations suffice to switch the system among differ- ent trajectories. The firing rate drops over the entire 3:1 plateau due to the appearance of the 5:1, 7:1 and other 6 FIG. 13: The response diagram in the presence of noise. Solid lines are the boundaries of the main mode-locked states with- out noise (see also Fig. 1). The broken and dotted lines are boundaries for D = 10−3 and 10−2, respectively. FIG. 12: The firing rate and CV as a function of D for several values of I0 and Ti = 17ms. Maximum of CV at intermediate values of D is due to the multimodal distribution of interspike intervals. odd modes. For larger D also the even modes are sam- pled by the system and fo/fi increases. At moderate, near-resonance frequencies, the central part of the 2:1 plateau in the lower part of Fig. 15 is more robust to noise. The average fo is preserved for D up to 0.1. Noise induces both the 1:1 mode as well as the 3:1 and slower modes. However the participation rate of the 1:1 mode is balanced by the slower modes in such a way that average fo/fi remains close to 2. This type of resilience to noise, characteristic of the resonance regime, can also be seen in other mode-locked states. Finally, let us briefly describe the role played by the current pulse width τ . Figure 16 shows the excita- tion threshold as a function of τ for a resonant drive, Ti = 17ms. Initially the threshold decreases as the in- verse of τ . For τ > 8ms the tip of the resonance (shown in Fig. 3) gradually becomes narrower and vanishes. The threshold rises again and reaches maximum near τ = 17ms, when I(t) = const. Bistability appears above τ ≃ 16ms. It is clear that stimulation by a constant cur- rent forces the neuron into an antiresonant regime. The bifurcation diagram in Fig. 16 implies that the addi- tion of any charge-unbalanced component to the constant stimulus takes the system away from this antiresonant limit. The precise nature of such a component, whether deterministic or stochastic, is not important. What mat- FIG. 14: Sample dependence of V (t) for three intensities of noise, from top to bottom: D = 0.001, 0.01, and 0.03. The spiking action is switched off for intermediate values of D. Here Ti = 11ms, I0 = 10.5µA/cm2. ters is the amount of charge delivered within approxi- mately 4ms from the stimulus onset. III. CONCLUSIONS We studied the response of the HH neuron to a periodic pulse current with a Gaussian noise. The global bifurca- 7 FIG. 16: Bifurcation diagram near the excitation threshold as a function of the current pulse width. The input period is set at Ti = 17ms. At τ = 17ms the input current is constant. The details of the antiresonant limit with bistable behavior are shown in the inset. Solid lines mark the excitation threshold. The dotted lines are the boundaries of the 1:1 state. It was shown in the HH [33, 34] and the Hindmarsh-Rose [35] models that this could be obtained by introducing an additional control function. We demonstrated that the threshold behavior can also be altered by selecting the frequency and amplitude of external current. For subthreshold stimuli noise enables spiking in the vicinity of the resonance. The stimulus frequency, for which the maximum firing rate is obtained, depends on the magnitude of noise. For large D, the maximum fo oc- curs for Ti > Tres. It would be interesting to investigate the same regime in a HH network where the connectivity pattern of neurons as well as their individual properties might lead to the emergence of subpopulations of neurons firing with different average frequencies in response to a correlated input with background noise. It was shown earlier by several authors that both the background noisy activity [36, 37] and correlated inputs [38] are important in explaining the neuronal response in vivo. Qualitative results of this work do not depend on the precise func- tional form of the current pulse provided the width of each pulse does not exceed 4 ms. This invariance of the bifurcation diagram in the Ti − I0 plane is not surpris- ing because for short pulses the HH neuron's threshold is determined by the amount of charge delivered per pulse [39]. We have also found a new even-all multimodal transi- tion occurring between the states 2:1 and 3:1 close to the main resonance. For input period Ti below this singular- ity only even response modes exist. Both even and odd modes appear above the transition. This effect is accom- panied by a minimum of the firing rate, located close to the parity transition, on the side of all modes. Similar transitions may exist in other excitable systems. Our results are also relevant to studies of auditory FIG. 15: Typical behavior of the average firing rate in the regime of high (top) and moderate frequency (bottom) for different noise levels. At high frequencies and small noise the response is initially slowed down over an entire plateau, starting from both edges. At the lower diagram fo is held steady in the central part of the plateau, decreasing at the left edge and increasing at the right edge. The average firing rate in the central part of the plateau is preserved up to D ≃ 0.1. The irregular nonmonotonic behavior in the deterministic case is smoothed out for very small levels of noise. tion diagram in the Ti − I0 plane has a rich structure near the excitation threshold, where resonant regimes al- ternate with antiresonant ones. The model is bistable be- tween the resonances and in the limit of high-frequency stimulation. The firing rate is a continuous function of I0 for Tres . Ti < 20ms. The scaling of the firing rate with (I0 − Ith)1/2 is a signature of a saddle-node bifurcation at the threshold. For Ti . Tres bistable regions are sepa- rated by areas of irregular response with no well defined threshold and approximately continuous dependence of fo/fi on input parameters. As Ti approaches Tres from below, the subcritical Hopf bifurcation gradually softens. Bistable regions occupy smaller portions of the parame- ter space and disappear before Ti = Tres. The change of the type of neuronal excitability is important in the context of preventing the so called dynamical diseases, such as epilepsy, Alzheimer's, and Parkinson's disease. nerve fiber responses to electric stimulation [40 -- 43]. At low stimulation rates auditory nerve fibers fire regularly and are locked in to applied stimulus. At high stimulation rates these fibers respond irregularly. Some researchers attributed this effect to physiological noise [44, 45]. How- ever, an analysis within the FitzHugh-Nagumo model showed that the firing irregularities at high frequencies may be caused by deterministic dynamical instability [42, 43]. We have found similar instabilities in the HH model. They appear at high stimulation frequencies and along the excitation threshold. Both of these regimes are relevant to studies of hearing sensitivity. 8 Acknowledgments Computations were performed in the Computer Center of the Tri-city Academic Computer Network in Gdansk. [1] A. L. Hodgkin and A. F. Huxley, J. Physiol. (London) 117, 500 (1952). [2] A. V. Holden, Biol. Cyber. 21, 1 (1976). [3] R. Guttman, L. Feldman, and E. Jakobsson, J. Mem- [23] L. S. Borkowski, Phys. Rev. E 80, 051914 (2009). [24] N. Takahashi, Y. Hanyu, T. Musha, R. Kubo, and G. Matsumoto, Physica D 43, 318 (1990). [25] L. S. Borkowski, BMC Neuroscience 10 (Suppl. 1), brane Biol. 56, 9 (1980). P250 (2009). [4] H. L. Read and R. M. Siegel, Neuroscience 75, 301 (1996). [5] D. T. W. Chik, Y. Wang, and Z. D. Wang, Phys. Rev. E 64, 021913 (2001). [6] S. G. Lee and S. Kim, Phys. Rev. E 73, 041924 (2006). [7] H. Hasegawa, Phys. Rev. E 61, 718 (2000). [8] R. FitzHugh, Biophys. J. 1, 445 (1961). [9] J. S. Nagumo, S. Arimoto, and S. Yochizawa, Proceed- [26] R. L. Honeycutt, Phys. Rev. A 45, 600 (1992). [27] J. R. Clay, J. Comput. Neurosci. 15, 43 (2003). [28] J. R. Clay, J. Neurophysiol. 80, 903 (1998). [29] S. H. Strogatz, Nonlinear Dynamics and Chaos (Perseus, 1994). [30] M. A. D. Roa, M. Copelli, O. Kinouchi, and N. Caticha, Phys. Rev. E 75, 021911 (2007). [31] A. M. Lacasta, F. Sagues, and J. M. Sancho, Phys. Rev. ings of the IRE 50, 2061 (1962). E 66, 045105(R) (2002). [10] E. M. Izhikevich, Dynamic Systems in Neuroscience [32] B. Lindner, L. Schimansky-Geier, and A. Longtin, Phys. (MIT Press, 2006). Rev. E 66, 031916 (2002). [11] J. Guckenheimer and R. O. Oliva, SIAM J. Appl. Dyn. [33] J. Wang, L. Chen, and X. Fei, Chaos, Solitons and Frac- Syst. 1, 105 (2002). tals 33, 217 (2007). [12] J. Rinzel, in Mathematical Apects of Physiology, edited by F. C. Hoppensteadt (American Mathematical Soci- ety, Providence, RI, 1981), vol. 19 of Lectures in Applied Math. [13] A. T. Winfree, When Time Breaks Down (Princeton Uni- versity Press, 1987). [34] Y. Xie, L. Chen, Y. M. Kang, and K. Aihara, Phys. Rev. E 77, 061921 (2008). [35] Y. Xie, K. Aihara, and Y. M. Kang, Phys. Rev. E 77, 021917 (2008). [36] N. Ho and A. Destexhe, J. Neurophysiol. 84, 1488 (2000). [37] F. S. Chance, L. F. Abbott, and A. D. Reyes, Neuron [14] J. Cronin, Mathematical Aspects of Hodgkin-Huxley Neu- 35, 773 (2002). ral Theory (Cambridge University Press, 1987). [38] C. F. Stevens and A. M. Zador, Nature Neuroscience 1, [15] H. Gang, T. Ditzinger, C. Z. Ning, and H. Haken, Phys. 210 (1998). Rev. Lett. 71, 807 (1993). [39] C. Koch, Biophysics of computation (Oxford University [16] A. S. Pikovsky and J. Kurths, Phys. Rev. Lett. 78, 775 Press, New York, 1999). (1997). [17] A. Longtin, Phys. Rev. E 55, 868 (1997). [18] B. S. Gutkin, J. Jost, and H. C. Tuckwell, Europhys. Lett. 81, 20005 (2008). [40] A. J. Matsuoka, P. J. Abbas, J. T. Rubinstein, and C. A. Miller, Hear. Res. 149, 129 (2000). [41] O. Macherey, R. P. Carlyon, A. van Wieringen, and J. Wouters, J. Assoc. Res. Otolaryngol. 8, 84 (2007). [19] B. S. Gutkin, J. Jost, and H. C. Tuckwell, Naturwis- [42] D. E. O'Gorman, J. A. White, and C. A. Shera, J. Assoc. senschaften 96, 1091 (2009). Res. Otolaryngol. 10, 251 (2009). [20] L. S. Borkowski, Nonlinear dynamics of Hodgkin-Huxley neurons (Adam Mickiewicz University Press, Poznan, 2010). [21] L. S. Borkowski, Phys. Rev. E 82, 041909 (2010). [22] R. Guantes and G. G. de Polavieja, Phys. Rev. E 71, 011911 (2005). [43] D. E. O'Gorman, H. S. Colburn, and C. A. Shera, J. Acoust. Soc. Am. 128, EL300 (2010). [44] R. P. Morse and E. F. Evans, Nat. Med. 2, 928 (1996). [45] F. Moss, F. Chiou-Tan, and R. Klinke, Nat. Med. 2, 860 (1996).
1203.2006	1	1203	2012-03-09T07:32:05	Semi-classical statistical approach to Fr\"ohlich condensation theory	[ "physics.bio-ph" ]	Fr\"ohlich model equations describing phonon condensation in open systems of biological relevance are here reinvestigated in a semi-classical non-equilibrium statistical context (with "semi-classical" it is meant that the evolution of the system is described by means of classical equations with the addition of energy quantization). In particular, the assumptions that are necessary to deduce Fr\"ohlich rate equations are highlighted and we show how these hypotheses led us to write an appropriate form for the master equation. As a comparison with known previous results, analytical relations with the Wu-Austin quantum Hamiltonian description are emphasized. Finally, we show how solutions of the master equation can be implemented numerically and outline some representative results of the condensation effect. Our approach thus provides more information with respect to the existing ones, in what we are concerned with the time evolution of the probability density functions instead of following average quantities.	physics.bio-ph	physics	Semi-classical statistical approach to Frohlich condensation theory Jordane Preto1, ∗ 1Aix-Marseille University, Campus de Luminy, case 907, CNRS Centre de Physique Th´eorique, UMR 7332, 13288 Marseille Cedex 09, France (Dated: October 15, 2018) Abstract Frohlich model equations describing phonon condensation in open systems of biological relevance are here reinvestigated in a semi-classical non-equilibrium statistical context (with "semi-classical" it is meant that the evolution of the system is described by means of classical equations with the addition of energy quantization). In particular, the assumptions that are necessary to deduce Frohlich rate equations are highlighted and we show how these hypotheses led us to write an ap- propriate form for the master equation. As a comparison with known previous results, analytical relations with the Wu-Austin quantum Hamiltonian description are emphasized. Finally, we show how solutions of the master equation can be implemented numerically and outline some representa- tive results of the condensation effect. Our approach thus provides more information with respect to the existing ones, in what we are concerned with the time evolution of the probability density functions instead of following average quantities. PACS numbers: 87.15.Zg; 43.20.Tb; 87.10.-e 2 1 0 2 r a M 9 ] h p - o i b . s c i s y h p [ 1 v 6 0 0 2 . 3 0 2 1 : v i X r a ∗ [email protected] 1 I. INTRODUCTION Some decades ago, the study of open systems far from thermodynamic equilibrium showed, under suitable conditions, the emergence of self-organization. Striking similarities were observed among very different physical systems having in common the fact of being composed of many non-linearly interacting subsystems. When a control parameter, typi- cally the energy input rate, exceeds a critical value then the subsystems act cooperatively to self-organize in what is commonly referred to as a non-equilibrium phase transition. This is at variance with equilibrium phase transitions, for which the transition from disorder to order is driven by a lowering of temperature -- the control parameter -- below a critical value. Paradigmatic examples of non-equilibrium phase transitions are provided by the laser tran- sition and by the Rayleigh-B´enard convective instability, a quantum and a classical system, respectively. In the early '80s of the last century, the emergence of collective properties in open systems was studied in a general and interdisciplinary framework referred to as "Synergetics" [1]. This fascinating topic was pioneered in the late '60s by H. Frohlich [2]. Frohlich considered a system consisting in z normal modes of frequency ωk (ω1 < ω2 < ... < ωz), each characterized by a discrete energy spectrum (phonons). Due to interactions with a surrounding heat bath, it was supposed that the dynamics of phonons in each mode was governed by several kinds of stochastic events : (i) linear events resulting in the absorp- tion or the emission of a single phonon from a particular normal mode, (ii) non-linear events resulting in the simultaneous absorption of one phonon from a normal mode, and emission of one phonon to another mode. Besides interactions with the heat bath, each normal mode is assumed to be fed energy by some external source. Denoting by (cid:104)Nk(cid:105) the average number of phonons in the mode of frequency ωk, Frohlich suggested that the dynamics of the system was described by the following rate equations = sk + ϕk(β)(cid:0)(cid:104)Nk(cid:105) + 1 − (cid:104)Nk(cid:105)eβωk(cid:1) + (cid:88) Λkj(β)(cid:2)(cid:104)Nk + 1(cid:105)(cid:104)Nj(cid:105) − (cid:104)Nk(cid:105)(cid:104)Nj + 1(cid:105)eβ(ωk−ωj )(cid:3) , k = 1, ..., z , (1) d(cid:104)Nk(cid:105) dt j(cid:54)=k where β = 1/kT with T the temperature of the heat bath. Here, the first r.h.s. sk denotes the rate of external energy supply to the mode k whereas the second and the third terms account for the rate of change due to linear (i) and non-linear (ii) events, respectively (with 2 solved analytically Eq. (1) in the stationary case appropriate temperature-dependent coupling constants ϕk and Λkj, respectively). Frohlich of phonons increased linearly by increasing the total rate of external energy supply(cid:80) k sk beyond a certain threshold, while non-linear events (ii) tend to redistribute the energy excess = 0, showing that the total number d(cid:104)Nk(cid:105) dt into the mode of lowest frequency. In other words, for sufficiently large value of the sk's, the stationary state is shown to satisfy z(cid:80) (cid:104)N1(cid:105) ∼ (cid:104)N(cid:105) where (cid:104)N(cid:105) = (cid:104)Nj(cid:105), i.e., it can be considered that all the components of the system oscillate in a collective way at the lowest frequency of the spectrum. This state of average j=1 excitation is better known as Frohlich condensation. Since Frohlich's first proposal [2], Frohlich condensation has been investigated by many authors [3]. In particular, a generic model of Hamiltonian was provided by Wu and Austin to derive Frohlich rate equations [Eqs. (1)] from a microscopic quantum basis [4, 5]. This type of Hamiltonian was then found to induce excitations propagating coherently in a similar way as Davydov's solitons [6, 7]. From a practical point of view, Frohlich condensation has been suggested to play a central role in the self-organization of biological polar structures (here the set of normal modes arises from polar oscillations). This is the case, for instance, of microtubules and cell membranes that fulfilled the main requirements of Frohlich systems [3, 8] (there the hydrolysis of adenosine triphosphate (ATP) or guanosine triphosphate (GTP) represents a significant source of external energy supply). Recently, Pokorn´y experimentally observed a strong excitation in the spectrum of vibration of microtubules localized in the 10 MHz range [9]. Among other applications of condensation, Frohlich suggested that the excitation of a particular mode of polar oscillations in macromolecular systems could lead to strong long-range dipole-dipole interactions. Applied to biological systems, it was expected that such forces would have a profound influence on the displacement of specific biological entities [10, 11], and thus on the initiation of a particular cascade of chemical events. While long-range interactions have been reported at the cellular level [12], their possible role at the biomolecular level still remains an open question [13]. Now, coming back to Frohlich theory, it should be stressed that most studies performed on the condensation -- including the derivation of Eqs. (1) -- are based on a quantum Hamiltonian description akin to the one proposed originally by Wu and Austin. Given 3 the semi-classical nature of the rate equations postulated by Frohlich, one can legitimately wonder how necessary a quantum description is to account for Frohlich condensation or even what are the ingredients needed to contemplate such a phenomenon. In the present paper, we propose to tackle this issue from a semi-classical point of view (here "semi-classical" means that the evolution of the system is described by means of classical equations with the addition of energy quantization). In section II, Frohlich rate equations (1) are derived from clear semi-classical arguments (II A and II B) and we show, in qualitative terms, how Frohlich condensation can be "deduced" from these arguments (II C). In section III, the classical master equation which is at the grounds of the rate equations, is given from the main results found in section II. Then, the same equation is found to arise from the Wu-Austin Hamiltonian description under appropriate decoherence assumptions. Finally, as another original approach, Frohlich condensation is statistically studied by estimating the solution of the master equation numerically [section IV]. A lot of informations is thus provided that could, in our opinion, be of particular relevance to identify the phenomenon experimentally. II. DERIVATION OF THE RATE EQUATIONS For the sake of readability, the derivation of Frohlich rate equations has been itself splitted into two subsections. The first one (A) deals specifically with the derivation of the second r.h.s. of Eq. (1), i.e. that part which accounts for linear events, whereas the second one (B) is about the derivation of the equations in their entirety. In the former case, it is enough to consider the presence of a single normal mode in the system. Thus, this will provide a good introduction to the second subsection where the dynamics of all normal modes must be taken into account simultaneously in order to describe non-linear events. Finally, a subsection (C) has been added as a complement to show how Frohlich condensation can be "predicted" qualitatively on the basis of the transition probabilities given in (A) and (B) (see below). A. One variable process Although that point is usually not explicitly mentioned in the literature, Frohlich equa- tions are primarily based on the assumption that the time dependence of the number of phonons in each normal mode can be described as a homogeneous Markov process. Here 4 one normal mode of frequency ωk is considered and we call Nk(t) the stochastic process that accounts for the number of phonons in this mode. Under homogeneous Markov assumption it is possible to work with time-translation invariant conditional probabilities p(nk, tn(cid:48) k) such that satisfies the following master equation [14] p(nk, tn(cid:48) k) ≡ Prob{Nk(t + t(cid:48)) = nkNk(t(cid:48)) = n(cid:48) (cid:88) W (nkmk)p(mk, tn(cid:48) k,∀t(cid:48) ≥ 0}, ∂tp(nk, tn(cid:48) k) = k) − W (mknk)p(nk, tn(cid:48) k). (2) Here, the transition probabilities mk W (nkmk) = lim ∆t→0 p(nk, ∆tmk) ∆t , are supposed to be well defined. Moreover, since Eq. (2) holds true irrespectively of the initial condition n(cid:48) ones p(..., t) in what follows. k, conditional probabilities p(..., tn(cid:48) k) will be substituted with standard As mentioned in the heading of the section, only interactions that lead to the absorption or the emission of one phonon at a time are considered here; thus W (nkmk) is zero except when mk = nk±1 or mk = nk (birth and death process). Then, the equation for the evolution of the average number of phonons in the mode k is easily deduced d(cid:104)Nk(cid:105) dt = = (cid:88) (cid:88) nk>0 nk>0 nk∂tp(nk, t) nk [W (nknk − 1)p(nk − 1, t) + W (nknk + 1)p(nk + 1, t) −(W (nk + 1nk) + W (nk − 1nk))p(nk, t)] . (3) With some appropriate substitutions in the sum index with suitable relabeling, one gets [W (nk + 1nk) − W (nk − 1nk)] p(nk, t), (4) (cid:88) nk>0 d(cid:104)Nk(cid:105) dt = where the boundary conditions 5 W (−10) = 0, and p(nk, t) = 0, if nk < 0, (5) have been used. Now, let us look for an appropriate expression for W (nk + 1nk) and W (nk − 1nk) in terms of nk. First, we suppose that all phonons are "independent" so that W (nk − 1nk) = αk(β)nk, (6) where αk(β) is the probability per time unit (which is temperature-dependent in general) that one phonon is emitted from the mode k. Second, W (nk + 1nk) can be obtained by noticing that a birth and death process following boundary conditions (5) has a single stationary solution ps(nk) through Eq. (2) (see the Appendix), and that this solution satisfies the detailed balance condition W (nk + 1nk)ps(nk) = W (nknk + 1)ps(nk + 1), ∀ nk ≥ 0. (7) Moreover, due to the contact between the system and the surrounding heat bath, it is required that the stationary solution is given by a Bose-Einstein distribution ps(nk) = e−βωknk/Z, where Z = e−βωknk (cid:88) nk≥0 (8) (9) is the one-normal mode partition function. Eq. (7) then leads to : W (nk + 1nk) = W (nknk + 1)e−βωk (6) = αk(β)(nk + 1)e−βωk. Substituting this last equation and Eq. (6) into Eq. (4), we finally get the second r.h.s. of Eq. (1) : = ϕk(β)(cid:0)(cid:104)Nk(cid:105) + 1 − (cid:104)Nk(cid:105)eβωk(cid:1) , (10) d(cid:104)Nk(cid:105) dt where we have let ϕk(β) ≡ αk(β)e−βωk. B. Many variables process We now want to consider a possible influence of all other normal modes on the mode of frequency ωk. In this way, let N (t) = (N1(t), N2(t), ..., Nz(t)) be the vector of all Marko- 6 s(cid:88) (cid:88) corresponds to (cid:80) where the sum (cid:80) d(cid:104)Nk(cid:105) dt µ=1 = n n (cid:80) ... (cid:80) n1≥0 n2≥0 nz≥0 nk [W (nn − rµ)p(n − rµ, t) − W (n − rµn)p(n, t)] , (12) . Besides, since events which do not vian processes describing the number of phonons in each normal mode, so that the con- ditional probabilities are noted as p(n, tn(cid:48)), with n, n(cid:48) ∈ Nz, n = (n1, n2, ..., nz) and n(cid:48) = (n(cid:48) 1, n(cid:48) 2, ..., n(cid:48) z). In anticipation of a large number of interactions between the normal modes and the heat bath, we call Rµ, µ = 1, ..., s, all possible events that account for the simultaneous absorption(s) and/or emission(s) of phonon(s) from particular modes. Then, each event is i (1 ≤ i ≤ z) is equal to the corresponding characterized by a vector rµ whose element rµ variation in the number of phonons in the mode of frequency ωi. Thus, the master equation takes the usual form [14] s(cid:88) ∂tp(n, tn(cid:48)) = W (nn − rµ)p(n − rµ, tn(cid:48)) − W (n − rµn)p(n, tn(cid:48)). (11) In this way, the evolution of the average number of phonons in the mode k is immediately µ=1 deduced (similarly to the previous subsection, initial conditions are omitted) result in a variation in the number of phonons in the mode k have zero contribution to the equation, Eq. (12) can be readily simplified. 1. Second r.h.s. of Eq. (1) : as mentioned above, each event Rµ corresponds to the absorption or the emission of one phonon from a particular mode. Thus, there are only j = ±δjk, two events involving the mode k (2z events overall) and they are given by rµ for j = 1, ..., z. This case corresponds to the case tackled in the previous subsection whereby equation (10) was finally found. 2. Third r.h.s. of Eq. (1) : here each event Rµ corresponds to the simultaneous absorption of one phonon from a normal mode, and emission of one phonon from another mode. In particular, with regard to the mode k, there are 2(z − 1) events (2z(z − 1) events overall), each characterized by rµ such that k ↓ j ↓ (13) rµ = (0, ...., 0,±1, 0, ...., 0,∓1, 0, ...., 0), ∀ j (cid:54)= k, j ∈ [[1, z]] , 7 where j ∈ [[1, z]] means j = 1, 2, ..., z. For the sake of clarity in what follows, we shall note Wn(nk; nj nk ± 1; nj ∓ 1) the transition probabilities W (nn − rµ) associated with these events. Likewise, p(n− rµ, t) will be written as pn(nk ± 1; nj ∓ 1, t) (in this way, p(n, t) may be sometimes noted as pn(nk; nj, t)). In this way, Eq. (12) becomes d(cid:104)Nk(cid:105) dt = j(cid:54)=k n (cid:88) (cid:88) nk [Wn(nk; njnk − 1; nj + 1)pn(nk − 1; nj + 1, t) + Wn(nk; njnk + 1; nj − 1)pn(nk + 1; nj − 1, t) − Wn(nk + 1; nj − 1nk; nj)pn(nk; nj, t) − Wn(nk − 1; nj + 1nk; nj)pn(nk; nj, t)] . (14) Again, some appropriate substitutions on the sum indexes with suitable relabeling lead to (cid:88) (cid:88) j(cid:54)=k n d(cid:104)Nk(cid:105) dt = [Wn(nk + 1; nj − 1nk; nj)− Wn(nk − 1; nj + 1nk; nj)] pn(nk; nj, t), where the boundary conditions Wn(−1; nj + 10; nj) = 0, Wn(nk + 1;−1nk; 0) = 0, and (15) (16) pn(nk; nj, t) = 0, if nk or nj < 0, have been used. As previously, we look for the expression of the transition probabilities. On each normal mode j (cid:54)= k, we suppose that the phonons have the same probability γkj(β, nk) of being emitted while a phonon is absorbed in the mode k (in general terms, this quantity depends on the temperature as well as on the current number nk of phonons in the mode k). We thus get Wn(nk + 1; nj − 1nk; nj) ≡ Wn(nj − 1; nk + 1nj; nk) = γkj(β, nk)nj, ∀ nk, nj ≥ 0 and ∀ j (cid:54)= k, j ∈ [[1, z]] . (17) 8 At this stage, it is worth mentioning that the transition probabilities considered thus far have been given in such a way that the graph of the master equation is connected. Therefore, according to general results on master equations [14], it can be found that a stationary solution of Eq (11) exists and is unique. Moreover, in so far as no other external source is present, a stationary solution for which detailed balanced condition is fulfilled, always exists. In our case, this solution is thus unique. Besides, since the system of normal modes interacts with the heat bath only, the stationary solution has to be given by a Bose-Einstein distribution. Thus Wn(nk + 1; nj − 1nk; nj)ps n(nk; nj) = Wn(nk; njnk + 1; nj − 1)ps n(nk + 1; nj − 1), with ps n(nk; nj) = (cid:89) i e−βωini/Z, where Z = ∀ nk, nj ≥ 0 and ∀ j (cid:54)= k, j ∈ [[1, z]] , e−βωini. (18) (cid:89) (cid:88) ni≥0 i Using Eq. (17), Eq. (18) becomes γkj(β, nk) nk + 1 = γjk(β, nj − 1) nj e−β(ωk−ωj ), which must depend on k and j only (i.e., not on nk nor on nj). We finally deduce the properties with γkj(β, nk) = Λkj(β)(nk + 1), Λjk(β) = Λkj(β)eβ(ωk−ωj ), ∀ j (cid:54)= k, j ∈ [[1, z]] . Using these equations and Eq. (17) in Eq. (15), we obtain (cid:88) j(cid:54)=k d(cid:104)Nk(cid:105) dt = Λkj(β)(cid:2)(cid:104)(Nk + 1)Nj(cid:105) − (cid:104)Nk(Nj + 1)(cid:105)eβ(ωk−ωj )(cid:3) , (19) (20) (21) which is exactly the third r.h.s. of Frohlich equations (1) in so far as correlations between the normal modes can be neglected, i.e. (cid:104)Nk(Nj + 1)(cid:105) (cid:39) (cid:104)Nk(cid:105)(cid:104)Nj + 1(cid:105), or more specifically p(n, t) (cid:39)(cid:89) i p(ni, t). 9 k(βs)(cid:0)(cid:104)Nk(cid:105) + 1 − (cid:104)Nk(cid:105)eβsωk(cid:1) = ϕs d(cid:104)Nk(cid:105) dt where βs → 0, so that 3. First r.h.s. of Eq. (1) (source term) : the constant rate of external energy supply may be worked out by considering the external source as a heat bath with infinite temperature, with which the system interacts linearly. Thus, the rate of change in the mode k due to the source is given by an equation similar to (10) d(cid:104)Nk(cid:105) dt = sk ≡ ϕs k(βs) (22) k(βs)e−βsωk k(βs) is the probability per time unit that a phonon is emitted from the mode Here, βs = 1/kTs with Ts the temperature of the source, and ϕs where αs k(βs) = αs k to the source. Finally, by gluing together all the events listed above (points 1. 2. and 3.), we deduce d(cid:104)Nk(cid:105) dt = sk + ϕk(β)(cid:0)(cid:104)Nk(cid:105) + 1 − (cid:104)Nk(cid:105)eβωk(cid:1) + (cid:88) Λkj(β)(cid:2)(cid:104)Nk + 1(cid:105)(cid:104)Nj(cid:105) − (cid:104)Nk(cid:105)(cid:104)Nj + 1(cid:105)eβ(ωk−ωj )(cid:3) , k = 1, ..., z j(cid:54)=k (23) which correspond exactly to Eqs. (1) with ϕk(β) = αk(β)e−βωk and Λjk(β) = Λkj(β)eβ(ωk−ωj ). C. Qualitative approach to Frohlich condensation As mentioned in the Introduction, Frohlich condensation state is a stationary state for the system described by Eqs. (23) achieved when the rates of energy supply sk exceed some threshold value. In this situation, the energy of the system is primarily located in the mode of lowest frequency, i.e., it can be found that (cid:104)N1(cid:105) ∼ (cid:104)N(cid:105) ≡(cid:88) (cid:104)Nj(cid:105). Although the existence of this state has been emphasized both analytically [2, 10] and j numerically [5] from Eqs. (23), Frohlich condensation may be highlighted more directly on the basis of the transition probabilities computed in the previous subsection. In that regard, it should be stressed that Frohlich condensation is obtained as a consequence of the energy redistribution due to the influence of non-linear interactions between the system and the 10 heat bath [10]. Now, we found in the previous subsection that the probability per time unit that one phonon is emitted from the mode j while one phonon is simultaneously absorbed in the mode k reads as (see Eqs. (17) and (19)) Wn(nk + 1; nj − 1nk; nj) = Λkj(β)(nk + 1)nj. (24) Conversely, the probability per time unit that one phonon is emitted from the mode k while one phonon is absorbed in the mode j is simply Wn(nk − 1; nj + 1nk; nj) = Λjk(β)nk(nj + 1). Then, using Eq. (20) Λjk, one gets Wn(nk − 1; nj + 1nk; nj) = Λkjeβ(ωk−ωj )nk(nj + 1). (25) In particular, assuming that nk, nj (cid:29) 1, it appears that the probability that one phonon is absorbed in the mode of lower frequency through non-linear events is always greater than the probability that one phonon is emitted from that mode. Indeed, when nk, nj (cid:29) 1, one gets (nk + 1)nj ∼ nk(nj + 1). Thus, the ratio between the transition probabilities (24) and (25) can be approximated as eβ(ωk−ωj ). If ωk < ωj, eβ(ωk−ωj ) < 1 so that the probability of a simultaneous absorption in the mode k and emission from the mode j will be the larger one at any time (Eq. (24)). On the contrary, if ωk > ωj, eβ(ωk−ωj ) > 1 and the larger probability will be that of a simultaneous absorption in the mode mode j and emission from the mode k (Eq. (25)). Here, we see how relevant the condition of detailed balance through Eq. (20) is in order to get Frohlich condensation. In particular, a wrong choice of the constants Λij could lead to a stationary state characterized by the excitation of modes of higher frequency in the spectrum of the system. Coming back to the number of phonons in each mode, the condition for large values of nk, nj can be fulfilled in so far as the rates of energy supply sk are large enough. More specifically, it was shown at the end of the last subsection, that the interactions between the system and the external source could be depicted similarly to the linear interactions between the system and the heat bath with a condition of infinite temperature. For the latter, let us recall from subsection II A that the probability that one phonon is emitted from the mode k (through linear events) reads as Wn(nk − 1nk) = αk(β)nk, (26) 11 while the probability of absorption is given by Wn(nk + 1nk) = αk(β)(nk + 1)e−βωk. (27) Here, in the absence of source, the condition for large values for the nk's can never be fulfilled because if the nk's become too large, emission will be favored against absorption. On the contrary, when only the external source is considered, one has β = βs → 0 so that according to (26) and (27) (here αk must be changed in sk) absorption is always favored. Thus, in considering the presence of both heat bath and external source, the possibility of working with large number of phonons in each normal mode will depend to some extent on whether the ratio sk/αk is large or not. In this way, supposing that sk (cid:29) αk for all k, one will naturally get nk (cid:29) 1 after a certain time so that non-linear interactions will redistribute the energy into the mode of lowest frequency. III. THE ORIGINAL FR OHLICH MASTER EQUATION AND THE RELATION TO THE WU-AUSTIN HAMILTONIAN DESCRIPTION In the previous section, we have been able to derive the Frohlich rate equations (23) on the basis of purely semi-classical arguments. To summarize, it was assumed that : 1. The number of phonons in each normal mode of the system (as a function of time) could be represented as a (homogeneous) Markov process. This property allowed us to describe the evolution of the system on the basis of a classical master equation (11). 2. Due to the discrete nature of the energy (phonons), each process could be described as a birth and death process. 3. On each normal mode, all phonons could be considered independent. In addition, according to the detailed balance condition and the requirement of Bose-Einstein sta- tionary distributions of phonons when only one of both the external source and the heat bath is present, the transition probabilities that take place in the master equation have been worked out. In the end, the previous calculations allowed us to derive not only the Frohlich rate equations but also the original classical master equation from which the former can be deduced 12 ∂tp(n, t) = ∂tps(n, t) + ∂tpα(n, t) + ∂tpΛ(n, t). (28) Here, ps(n, t), pα(n, t) and pΛ(n, t) are the probability density functions of the number of phonons in all normal modes of the system (let us emphasize again that n = (n1, n2, ..., nz)) whose evolutions are respectively due to : • (Linear) events resulting from interactions between the system and the external source; • Linear events (i) resulting from interactions between the system and the heat bath; • Non-linear events (ii) resulting from interactions between the system and the heat bath. By substituting Eqs. (26) and (27) into Eq. (11), we get the master equation for pα(n, t) (cid:88) i ∂tpα(n, t) = αi(β){[(ni + 1)pn(ni + 1, t) − nipn(ni, t)] + +e−βωi[nipn(ni − 1, t) − (ni + 1)pn(ni, t)](cid:9) . (29) The master equation for ps(n, t) is found in a similar way by letting β = βs → 0 and changing αi in si (cid:88) ∂tps(n, t) = si(βs){(ni + 1)pn(ni + 1, t) + nipn(ni − 1, t) − (2ni + 1)pn(ni, t)} . (30) i Let us recall from subsection II B that pn(mi, t) is the probability of getting n1 phonons in the mode of frequency ω1, n2 phonons in the mode of frequency ω2,..., nz phonons in the mode of frequency ωz but mi phonons (instead of ni phonons) in the mode of frequency ωi. A similar notation pn(mi; mj, t) is used when the exception involves two modes. Naturally, pn(ni, t) and pn(ni; nj, t) correspond to p(n, t). Finally, the master equation corresponding to non-linear events is given by substituting Eqs. (24) and (25) into Eq. (11) so that (cid:88) ∂tpΛ(n, t) = Λij(β)ni(nj + 1)pn(ni − 1; nj + 1, t) − i,j>i Λij(β)(ni + 1)njpn(ni; nj, t) + Λji(β)(ni + 1)njpn(ni + 1; nj − 1, t) − (31) Λji(β)ni(nj + 1)pn(ni; nj, t), 13 with Λji(β) = Λij(β)eβ(ωi−ωj ). Eq. (28) together with Eqs. (29), (30) and (31) are the original classical master equations from which the Frohlich rate equations (23) may be deduced. In the Introduction, it was mentioned that Eqs (1) could be derived from a quantum Hamiltonian that was originally introduced by Wu and Austin [4, 5]. More specifically, it can be found that the above master equations are no more than an intermediate step of that derivation. In that regard, let us also emphasized that the possibility of deducing classical master equations (also called Pauli master equations) from a microscopic Hamiltonian is well-known in the realm of quantum open systems [15, 16]. In the case of Frohlich theory, that correspondence has not been given much attention in the literature. From this point of view as well as to provide a consistency check of Eqs. (28) to (31), let us emphasize how the Frohlich master equations can be deduced from the quantum approach given by Wu and Austin. To that purpose, we recall that the Wu-Austin Hamiltonian [4] is given by where H0 stands for the Hamiltonian of the system of normal modes, the heat bath plus H = H0 + Hint, the external source in the absence of interactions, and has the following form : z(cid:88) H0 = zb(cid:88) zs(cid:88) ωia † i ai + Ωkb † kbk + Ω(cid:48) pc† pcp. i=1 k=1 p=1 Here, a † i , ai are the creation and annihilation operators associated with the normal mode † k,bk are the creation and annihilation operators associated p, cp are the creation and p of the source. In of frequency ωi of the system, b with the normal mode of frequency Ωk of the heat bath, and c† annihilation operator associated with the normal mode of frequency Ω(cid:48) parallel, Hint is the interaction Hamiltonian such that (cid:88) (cid:88) (cid:88) Hint = λi,ka † i bk + χi,j,kaia † jbp + † i cp + H.c., ξipa (32) i,k i,j,k i,p Since one is interested in the statistical properties of observables related to the system only, it is convenient to work with the reduced density operator S(t), which is defined as the usual density operator traced over the states related to the heat bath and the source [15]. Assuming that the dynamics of each normal mode verifies the Markov property, it may 14 be shown [15, 17] that second perturbation theory in the interaction picture applied to the interaction Hamiltonian (32) actually yields (cid:17) † i S + Saia i − 2a † † + i Sai i ai − 2aiSa † (cid:17)(cid:111) † i † i aiS + Sa dS dt = − 1 2 (φiri + σi) (cid:110) (cid:88) (cid:88) i i,j,i(cid:48),j(cid:48) j>i,j(cid:48)>i(cid:48) i−i(cid:48)=j−j(cid:48) (cid:16) aia (cid:16) a i(cid:48),j(cid:48),j(cid:48)−i(cid:48)× π [φi(ri + 1) + σi] 2 χi,j,j−iχ∗ (cid:104)(cid:16) (cid:16) † jaj(cid:48)a aia −1 2 where : † i(cid:48)S + Saia † jaj(cid:48)a i(cid:48) − 2aj(cid:48)a † † aj(cid:48)a i(cid:48)aia † jS + Saj(cid:48)a † i(cid:48)aia (cid:17) † i(cid:48)Saia j − 2aia † † j (rj−i + 1) + † jSaj(cid:48)a † i(cid:48) rj−i (33) (cid:105) , (cid:17) (cid:17) (cid:0)λi,i2 + ξi,i2(cid:1) , σi = π 2 2ξi,i2(cid:16)(cid:104)c π i ci(cid:105)s − (cid:104)b † i bi(cid:105)b † φi = (34) Here, (cid:104)...(cid:105)s and (cid:104)...(cid:105)b are averages performed over states related to the source and the heat bath respectively. Now, the probability of observing n1 phonons in the first mode, n2 phonons in the second, and so on, corresponds to the matrix element (cid:104)nS(t)n(cid:105) := p(n, t), where n(cid:105) = n1(cid:105)...nz(cid:105). Naturally, ni(cid:105) is the eigenvector of the operator ni = a † i ai. Considering only the first sum in Eq. (33) (that stems from the linear terms in the Hamiltonian (32)), . one can easily deduce the equation : ∂tp(n, t) = −(cid:88) (φiri + σi) [(ni + 1)pn(ni, t) − nipn(ni − 1, t)] + i +(φi(ri + 1) + σi) [nipn(ni, t) − (ni + 1)pn(ni + 1, t)] (35) In particular when no source is present, one simply gets π φiri + σi = eβωi eβωi − 1 as the heat bath is required to be at thermal equilibrium with a temperature β = 1/kT . and φi(ri + 1) + σi = 2λi,i2(cid:104)b 2λi,i2 2λi,i2 eβωi − 1 i bi(cid:105)b = † π π 1 , Using these two last expressions into Eq. (35), we obtain : (cid:88) i ∂tp(n, t) = π 2λi,i2 eβωi eβωi − 1 {[(ni + 1)pn(ni + 1, t) − nipn(ni, t)] + +e−βωi[nipn(ni − 1, t) − (ni + 1)pn(ni, t)](cid:9) (36) 15 which is exactly the classical master equation (29) with αi(β) = π 2λi,i2 eβωi eβωi − 1 . Nat- urally, the classical master equation (30) related to the source is found in a similar way by considering only the ξ-terms in Eq. (33) and by using the condition (cid:104)c eβsωi − 1 with βsωi (cid:28) 1. i ci(cid:105)s = † 1 Although no further assumption has been needed to get Eqs. (29) and (30) from Eq. (33), the master equation (31) related to non-linear interactions may only be deduced by first supposing that all quantum coherences have been destroyed in the system, i.e : (cid:104)nS(t)m(cid:105) = p(n, t)δn,m, ∀n, m (37) where δn,m = δn1,m1δn2,m2...δnz,mz . Surprisingly, although Eq. (37) is a quite usual assumption when deriving classical master equations from a microscopic Hamiltonian, it was never clearly put in evidence in the case of Frohlich theory. Finally, one gets (we recall that according to Eq. (33), the condition i − i(cid:48) = j − j(cid:48) is required) : (cid:104)naia † jaj(cid:48)a i(cid:48)S(t)n(cid:105) = δi,i(cid:48)δj,j(cid:48)(ni + 1)njpn(ni; nj, t) = (cid:104)nS(t)aia † † jaj(cid:48)a i(cid:48)n(cid:105), † (cid:104)naj(cid:48)a † i(cid:48)Saia jn(cid:105) = δi,i(cid:48)δj,j(cid:48)ni(nj + 1)pn(ni − 1; nj + 1, t), † (cid:104)naj(cid:48)a † i(cid:48)aia jS(t)n(cid:105) = δi,i(cid:48)δj,j(cid:48)ni(nj + 1)pn(ni; nj, t) = (cid:104)nS(t)aj(cid:48)a † † i(cid:48)aia jn(cid:105), † and (cid:104)naia i(cid:48)n(cid:105) = δi,i(cid:48)δj,j(cid:48)(ni + 1)njpn(ni + 1; nj − 1, t). † † jSaj(cid:48)a From the second sum in Eq. (33) (terms in χ) and the relations rj−i = rj−i + 1 = eβ(ωj−ωi) eβ(ωj−ωi) − 1 , one obtains : 1 eβ(ωj−ωi) − 1 and 16 (cid:88) i,j>i ∂tp(n, t) = π 2χi,j,j−i2 1 {(ni + 1)njpn(ni; nj, t) + eβ(ωi−ωj ) − 1 ni(nj + 1)pn(ni − 1; nj + 1, t) + eβ(ωi−ωj ) ((ni + 1)njpn(ni + 1; nj − 1, t) − (38) ni(nj + 1)pn(ni; nj, t))} , that is exactly the master equation (31) accounting for the non-linear coupling between the eβ(ωi−ωj ) eβ(ωi−ωj ) − 1 Therefore, we see that the condition for detailed balance [Eq. (20)] is adequately verified. normal modes with Λij(β) = 2χi,j,j−i2 2χi,j,j−i2 eβ(ωi−ωj ) − 1 π 1 and Λji(β) = π . As a final complement, let us mention that the Frohlich master equation can be deduced not only from the Hamiltonian (32) but also from a large family of Hamiltonians for which the operators describing the thermal bath and the source can be very complex objects as various combinations of bosonic and/or fermionic, creation and/or annihilation operators [18]. In the end, only the expressions of the parameters s, ϕ and χ will differ from the ones given in this section. We refer the reader interested in a further detailed investigation on the quantum Hamiltonian formulation of Frohlich theory to references [4 -- 7] . IV. NUMERICAL COMPUTATIONS It should be stressed that most theoretical studies on Frohlich condensation are based on the numerical integration of the rate equations (23) whereas the issue of the fluctuations of the number of phonons in each mode is largely unexplored. From this point of view, a study of the distributions of the occupation numbers could be of utmost importance to identify Frohlich condensation experimentally at least in qualitative terms. As emphasized above, the time evolution of these distributions is provided by Eqs. (28) to (31). These equations, being master equations with many variables, are extremely difficult to solve analytically as the number of involved processes (here the number of normal modes) increases. Nevertheless, several algorithms allow to estimate the solution of such equations numerically. The most popular one, proposed by Gillespie, consists in generating stochastic trajectories governed by the master equation on the basis of Monte-Carlo procedures [19]. Eventually, these trajectories can be used to estimate the expected distribution at any time t as well as the related moments. 17 In this context, we investigate numerically Frohlich condensation through two different approaches. The first one, as a consistency check, involves the integration of the rate equations (23) so that the evolution of the average number of phonons of each normal mode is followed for different values of the source parameters sk. The second approach consists in generating stochastic trajectories whose evolution is governed by the Frohlich master equation [Eqs. (28) to (31)], by means of the Gillespie algorithm. By collecting the results in the form of histograms, it is thus possible to estimate the distribution of phonons for any normal mode at any time t. Some representative results of Frohlich condensation using the first approach are reported in Figure 1. The system is initially at thermal equilibrium, i.e, the average number of phonons is given at t = 0 by the Planck formula : (cid:104)Nk(t = 0)(cid:105) = 1 eβωk − 1 , ∀k. (39) Then, each solution tends to a stationary state with a convergence rate that is mainly depending on the values of the coupling constants sk, αk, Λkj. Since we are interested in a qualitative description of the condensation, we are not committed to an exhaustive exploration of the parameter space (in particular, the coupling constants are given in (time arbitrary units)−1) [20]. Let us simply specify that we required Λkj (cid:28) αk ∀k, j because events related to linear interactions are more likely to arise than events related to non-linear ones. Also, the coefficients Λkj have been fixed according to the relation Λjk = Λkjeβ(ωk−ωj ) which follows a theoretical condition of detailed balance (see section II B). Now, we see from Figure 1 how Frohlich condensation takes place in average terms : a situation wherein sk < αk, for all k, leads to a stationary state relatively close to the initial (thermal equilibrium) condition (see Figure 1(a)). Alternatively, a situation where sk (cid:29) αk holds for the majority of normal modes leads to a stationary state characterized by a large number of phonons populating the mode of lowest frequency, i.e., (cid:104)N1(∞)(cid:105) (cid:29) (cid:104)Nk(∞)(cid:105) for all k > 1 (see Figure 1 (b)). Besides, the values of (cid:104)Nk(∞)(cid:105) with k > 1 still remain close to the thermal equilibrium values. Representative results of Frohlich condensation using the second approach are depicted in Figure 2, 3 and 4. Each figure (histogram) was obtained by generating 10000 stochastic trajectories and by storing the values of the number of phonons of a particular mode at a time t when the system is considered to have reached the stationary state. In that regard, t 18 FIG. 1. (Color online) Time evolutions of the average number of phonons (cid:104)Nk(t)(cid:105) in the three (23). The modes of lowest frequency of a system of four modes obtained by integrating Eqs. frequencies of the normal modes are ωi = i.1012 Hz [22], i = 1, ..., 4 whereas the coupling constants associated with the interactions between the system and the heat bath are given by αk = 10 for all k and Λkj = 1 for k < j (let us recall that the relation Λjk = Λkjeβ(ωk−ωj ) is required). Initially, the system is supposed to be at thermal equilibrium so that (cid:104)Nk(t = 0)(cid:105) is given by Planck formula (39). The main change between both figures is the rate of energy supply : (a) sk = 1, (b) sk = 20, for all k. All coupling constants α, Λ and s are given in (time arbitrary unit)−1, as discussed in the text. 19 FIG. 2. (Color online) Normalized histograms of the number of phonons in the first two modes of a system of four modes at t = 30 arbitrary units (= stationary state) : (a) mode of frequency ω1 (b) mode of frequency ω2. Both figures have been obtained by generating 10000 stochastic trajectories by means of the Gillespie algorithm applied on Eqs. (28) to (31). Initially, the system is supposed to be at thermal equilibrium (see text). The frequency of the normal modes as well as the coupling constants related to the system-bath interactions are the same as those used Figure 1. The rate of energy supply is sk = 5 (time arbitrary unit)−1 for all k (for (a) and (b)). 20 FIG. 3. (Color online) Same as Figure 2 with a rate of energy supply sk = 20 (time arbitrary unit)−1 for all k (for (a) and (b)). has been approximated beforehand using the first approach. Again, the system is supposed to be initially at thermal equilibrium which means that each initial condition is chosen at random following a Bose-Einstein distribution for all the modes 21 FIG. 4. (Color online) Same as Figure 2 with a rate of energy supply sk = 50 (time arbitrary unit)−1 for all k (for (a) and (b)). p(nk, t = 0) = 1 Zk e−βωknk, with Zk = 1 1 − e−βωk , for k = 1, ..., z (rejection sampling). From a general point of view, the averages of the number of phonons deduced from the stochastic trajectories are in excellent agreement with 22 those stemming from the integration of the rate equations (23). To this respect, let us recall that the rate equations (23) can be deduced from the Frohlich master equation only if the correlations between the normal modes can be neglected p(n, t) (cid:39) (cid:89) p(ni, t) (see section II B). So far, with our choice of parameters, this assumption is thus confirmed independently i of whether Frohlich condensation is reached or not. Now, we see from Figure 2, 3 and 4 how Frohlich condensation appears statistically : for weak values of sk (Figure 2) the distributions of the number of phonons in the first two modes of lowest frequency (four modes overall, see Figures) remain close to the Bose-Einstein distributions. Progressively, by increasing sk, we see in the first mode that not only is the average number of phonons increasing as expected but also does the variance. As a general comment concerning the numerical results displayed here, let us remark that there is an analogy with the change of the statistical distributions of photon occupation numbers below and above the laser transition. In fact, the photon statistics below the laser transition is a thermal equilibrium one peaked at zero, whereas, above the transition, the photon statistics gives a non-zero most probable value of the number of photons in the lasing mode with a well- known Poisson distribution. Loosely speaking something similar occurs in correspondence with the condensation transition here explored. Here, we are not interested in deepening this analogy beyond the simple remark that the Frohlich condensation transition displays a phenomenology typical of non-equilibrium phase transitions in open systems. V. CONCLUSIONS In the present work, the phenomenon of Frohlich condensation has been addressed in a semi-classical framework. With "semi-classical" it is meant that the evolution of the system is described by means of classical equations with the addition of energy quantization. In the framework of open systems, this leads to the classical master equation replacing the evolution equation for the quantum density matrix. The quantum approaches hitherto proposed are based on the microscopic Hamiltonian originally proposed by Wu and Austin or on slightly modified versions of it. Here, we have proposed a different approach to highlight what are the necessary hypotheses that are required to yield the Frohlich condensation. This is not an academic exercise because Frohlich phenomenon could be relevant in very different physical contexts, and, in particular, to biologically relevant systems undergoing self-organization at 23 different scales (from single macromolecules up to collections of cells). Thus, the approach proposed here on the one hand has the merit of a critical reviewing of the physics underlying Frohlich condensation and on the other hand has the advantage of providing the time evo- lution of the probability density function which allows the computation of higher moments of the number of phonons beside their average quantities. Actually, though for a system with a small number of normal modes, we have shown a clearcut change of the histograms of the number of phonons in correspondence with the condensation transition. Though in a somewhat loose way, an analogy with the change of the distribution of photons below and above the laser transition has been noted. ACKNOWLEDGMENTS I warmly thank my colleagues and friends Marco Pettini and Ricardo Lima for enlight- ening comments and discussions. Appendix A: Detailed Balance and birth and death process We consider the homogeneous Markov process Nk(t) of section II A. According to the birth and death nature of Nk(t), the stationary solution ps(nk) of equation (2) is given ∀ nk ≥ 0 by 0 = J(nk) − J(nk − 1), (A1) where J(nk) = W (nknk + 1)ps(nk + 1) − W (nk + 1nk)ps(nk). We now sum (A1) so nk(cid:88) 0 = [J(ni) − J(ni − 1)] = J(nk) − J(−1), (A2) According conditions (5), J(−1) = 0, so that the detailed balance condition is obtained ni=0 0 = J(nk) = W (nknk + 1)ps(nk + 1) − W (nk + 1nk)ps(nk), ∀ nk ≥ 0 (A3) [1] H. Haken, Synergetics, an Introduction : Nonequilibrium Phase Transitions and Self- organization in Physics, Chemistry and Biology, (Springer-Verlag, New York, 1983). 24 [2] H. Frohlich, Int. J. Quantum Chem. 2, 641 (1968). [3] M. Cifra, J. Z. Fields and A. Farhadi, Prog. Biophys. Mol. Biol. 105, 223 (2011). [4] T. M. Wu and S. Austin, Phys. Lett. A64, 151-152 (1977). [5] J. Pokorn´y and T. M. Wu, Biophysical Aspects of Coherence and Biological Order, (Springer, Berlin, 1998). [6] J. A. Tuszynski, H. Bolterauer and M. V. Satari´c, Nanobiol. 1, 177-190 (1992). [7] M. V. Mesquita, A. R. Vasconcellos, R. Luzzi and S. Mascarenhas, Int. J. Quantum Chem. 102, 1116-1130 (2005). [8] V. Salari, J. A. Tuszynski, I. Bokkon, M. Rahnama and M. Cifra, J. Phys.: Conf. Ser. 329, 012006 (2011). [9] J. Pokorn´y, Bioelectrochem. Bioenerg. 48 (2), 267-271 (1999); J. Pokorn´y, Electromagn. Biol. Med. 28 (2), 105-123 (2009). [10] H. Frohlich, Adv. Electron. Electron Phys. 53, 85 (1980). [11] J. Preto and M. Pettini (2012, manuscript submitted); arXiv:1201.5187. [12] S. Rowlands, L. S. Sewchand, R. E. Lovlin, J. S. Beck and E. G. Enns, Phys. Lett. A82, 436 (1981); S. Rowlands, L. S. Sewchand, and E. G. Enns, Phys. Lett. A87, 256 (1982). [13] J. Preto, E. Floriani, I. Nardecchia, P. Ferrier and M. Pettini, Phys. Rev. E (2012), in press; arXiv/1201.2607. [14] C. W. Gardiner, A Handbook of Stochastic Methods, (Springer, Berlin, Third Edition, 2004). [15] W.H. Louisell, Quantum Statistical Properties of Radiation, (Wiley, New-York, 1973), p. 104- 109 and p. 238-246. [16] R. Zwanzig, Nonequilibrium Statistical Mechanics, (Oxford University Press, New York, 2001). [17] J. Hirsch, Phys. Lett. A121, 447-450 (1987). [18] I. Turcu, Phys. Lett. A234, 181-186 (1997). [19] D. T. Gillespie, J. Phys. Chem. 81, 2340-2361 (1977). [20] In that regard, we refer the reader to the studies performed by Pokorn´y and Wu [5]. [21] L. Mandel and E. Wolf, Optical Coherence and Quantum Optics, (Cambridge University Press, New York, 1995). [22] The range 0.1 − 1 THz was given by Frohlich [10] as the range of frequencies of typical vibrational biomolecular structures. 25
1804.01761	1	1804	2018-04-05T10:24:49	Modelling of Dictyostelium Discoideum Movement in Linear Gradient of Chemoattractant	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	Chemotaxis is a ubiquitous biological phenomenon in which cells detect a spatial gradient of chemoattractant, and then move towards the source. Here we present a position-dependent advection-diffusion model that quantitatively describes the statistical features of the chemotactic motion of the social amoeba {\it Dictyostelium discoideum} in a linear gradient of cAMP (cyclic adenosine monophosphate). We fit the model to experimental trajectories that are recorded in a microfluidic setup with stationary cAMP gradients and extract the diffusion and drift coefficients in the gradient direction. Our analysis shows that for the majority of gradients, both coefficients decrease in time and become negative as the cells crawl up the gradient. The extracted model parameters also show that besides the expected drift in the direction of chemoattractant gradient, we observe a nonlinear dependency of the corresponding variance in time, which can be explained by the model. Furthermore, the results of the model show that the non-linear term in the mean squared displacement of the cell trajectories can dominate the linear term on large time scales.	physics.bio-ph	physics	Modelling of Dictyostelium discoideum movement in linear gradient of chemoattractant Zahra Eidi,1 Farshid Mohammad-Rafiee,1 Mohammad Khorrami,2 and Azam Gholami3 1Department of Physics, Institute for Advanced Studies in Basic Sciences (IASBS), Zanjan 45137-66731, Iran 2 Department of Physics, Alzahra University, Tehran, Iran 3 Max Planck Institute for Dynamics and Self-Organization, Gottingen, Germany (Dated: November 6, 2018) Chemotaxis is a ubiquitous biological phenomenon in which cells detect a spatial gradient of chemoattractant, and then move towards the source. Here we present a position-dependent advection-diffusion model that quantitatively describes the statistical features of the chemotactic motion of the social amoeba Dictyostelium discoideum in a linear gradient of cAMP (cyclic adenosine monophosphate). We fit the model to experimental trajectories that are recorded in a microfluidic setup with stationary cAMP gradients and extract the diffusion and drift coefficients in the gradient direction. Our analysis shows that for the majority of gradients, both coefficients decrease in time and become negative as the cells crawl up the gradient. The extracted model parameters also show that besides the expected drift in the direction of chemoattractant gradient, we observe a nonlinear dependency of the corresponding variance in time, which can be explained by the model. Further- more, the results of the model show that the non-linear term in the mean squared displacement of the cell trajectories can dominate the linear term on large time scales. I. INTRODUCTION Dictyostelium discoideum (D.d.) is a well-established model organism for cellular motility. Chemotactic com- petent D.d. cells are highly motile and exhibit fast amoe- boid movements with a velocity of 10 − 20 µm/min on glass substrates [1 -- 3]. The chemotactic cell motion is highly organized over a length scale significantly larger than the size of a single cell (∼10 µm). When nutri- ents are depleted, D.d. cells secret a chemical called cAMP (cyclic adenosine monophosphate) that has an at- tractive effect on the cells themselves. Cells sense gradi- ents of cAMP and direct their chemotactic movements to- wards regions of higher concentration of cAMP [4]. When chemotactic attraction prevails over diffusion, the chemo- taxis can trigger a self-accelerating process until aggre- gation takes place. As a result, 105 -- 106 cells stream to- wards the aggregation centers and eventually transform into millimeter long slugs and ultimately form fruiting bodies bearing spores for long-term survival and long- range dispersal [5]. Different mathematical models incorporate chemotaxis in different ways; however, a common mechanism is to as- sume that chemotaxis biases the otherwise random mo- tion of crawling cells along the concentration gradients of chemoattractants [6]. The random cell movement is commonly described as a diffusion and the directional movement along the chemical gradient is incorporated as a combination of diffusion and advection. In the simplest model, the diffusion coefficient and the drift velocity of the cells are assumed to be constant. However, in gen- eral, these coefficients depend on both the absolute con- centration and the gradient of the chemical [7 -- 10]. The advection-diffusion equation has been previously used to describe the aggregation phase of D.d. cells where the chemotactic force pulls the amoebas towards the aggrega- tion centers. For example, a model of slime mold aggrega- tion has been introduced by Patlak [6] and Keller [10] in the form of two coupled differential equations. The first equation is an advection-diffusion equation describing the evolution of the concentration of amobae and the second equation is a diffusion equation with terms of source and degradation describing the evolution of the concentra- tion of the signaling molecule. The original form of the Keller-Segel model, would allow the diffusion coefficient and the drift velocity to depend on the cAMP concentra- tion and on the concentration of the amoeba. The case that these coefficients depend on the chemical cencentra- tion but not on the cell density has been considered by Othmer and Stevens [11]. This leads to ordinary mean field Fokker-Planck equations for cell density with space and time dependent coefficients [12]. On the other hand, if we assume that the diffusion coefficient and the drift velocity of the cells depend on their concentration and on the concentration of the secreted chemical, the original Keller-Segel model takes the form of a generalized mean field Fokker-Planck equation. The statistical characteristics of trajectories of motile D.d. cells have been the subject of several recent studies. These include experiments to characterize chemotactic cell movement in homogenous and inhomogenous chem- ical cues [7, 8, 13 -- 18], and parallel theoretical modeling to reproduce statistical features of the experimental ob- servations [7, 16 -- 19]. Recently, Li et al. have presented an experimental study of the individual cells in a homo- geneous medium [18]. They have proposed a generalized Langevin equation for the velocity of individuals. Their data-driven modeling showed a "programmed" periodic motion around a persistent direction of motion on short time scales and ordinary diffusive behavior on long time 8 1 0 2 r p A 5 ] h p - o i b . s c i s y h p [ 1 v 1 6 7 1 0 . 4 0 8 1 : v i X r a scales. Moreover, it is also well known that a cAMP gradient induces a bias of the position where pseudopo- dia emerge [15]. The measured probabilities of pseudo- pod directions were used to obtain an analytical model for chemotaxis of cell populations [16]. The prediction of the model are similar to measured chemotactic index of wild-type cells as well as the mutants. Besides, al- though it is well-known that the directed movement of the D.d. cells in response to the chemoattractant cAMP depends both on the absolute value of the local concen- tration (chemokinesis) and its gradient (chemotaxis), the exact dependency is not well understood. In this study, we aim to extract the concentration de- pendencies of the diffusion and drift coefficients in the Fokker-Planck equation (with respect to cell density), by analyzing the experimental trajectories of motile D.d. cells in Ref. [7, 8, 13]. We assume that these coeffi- cients depend on both the local cAMP concentration (the so-called midpoint concentration) and its gradient. The experiments are performed in a microfluidic device (see Section II-B) that generate linear stable gradients be- tween the two inlet concentrations Cmax and Cmin. As the cells crawl up the gradient, the average background concentration they experience increases. These exper- iments systematically explore different steepnesses and cover a wide range of gradients, at which chemotactic behavior is observed. In these experiments, an external flow removes the naturally produced cAMP secreted by the cells to avoid cell-cell signaling. This is completely different from aggregation process where the cell density is much higher and the cells signal each other. We start our analysis by assuming linear dependencies for the dif- fusion coefficient and the drift velocity of the cells along the width of microfluidic setup, where a linear gradient is established. We then use the experimental cell trajecto- ries to deduce the coefficients of these linear dependencies at different cAMP gradients. II. EXPERIMENTS A. Cell Culture All experiments were performed by M. Theves [7, 8, 13] with Dictyostelium discoideum AX3 wild type cells. Cells were grown in HL5 medium (7 g/L yeast extract, 14 g/L peptone, 0.5 g/L potassium dihydrogen phosphate, 0.5 g/L disodium hydrogen phosphate, 13.5 g/L glucose, ForMedium Ltd., UK). Cells were starved in shaking sus- pension of phosphate buffer (pH 6.0, 15 mM KH2 PO4 , 1 mM Na2 HPO4 ) at a density of 2 × 106 cells/mL for 5:30 hours. After one hour of starvation, the cells were exposed to periodic pulses of cAMP for the remaining time of starvation. The pulses had a concentration of 50 nM and were delivered with a period of 6 minutes. 2 B. Microfluidics A microfluidic gradient mixer [17, 20] with given di- mensions (width=525 µm, height= 50 µm) was used to establish a stable linear gradient over a region of 350 µm × 50 µm × 3000 µm in size (see Fig. 1). The gradients were generated using a pyramidal microfluidic network that provides well-defined concentration profiles with high temporal stability. Throughout the experi- ment, a constant flow is provided by syringe pumps. The flow provides a constant supply of oxygen and removes all substances released by the cells. This prevents cells from signaling each other, which would perturb the concentra- tion gradient and bias the chemotactic motion. Running at an adjustable average flow velocity of ¯v = 320 µm/s, the gradient is linear and stable within d = 350 µm in the middle of the channel. Above a lower threshold of ∇Cthresh ∼ 10−3 nM/µm cells started to show a direc- tional response. It is important to note, that all gradients have been established by mixing a phosphate buffer so- lution at one inlet, Cmin = 0, together with a solution of cAMP and phosphate buffer Cmax on the opposing inlet. Therefore the gradient ∇C = Cmax − Cmin = ∆C/d (1) always ranges from zero to this maximum concentration. Due to boundary effects, the profile is distorted near the walls. All cell trajectories within this non-linear area were excluded from statistics. Moreover, given the di- mension of the channel and the dynamics viscosity of the flowing phosphate buffer (η = 10−3 Pa s), one can calcu- late the shear stress applied on the cells at the imposed mean flow velocity of ¯v = 320 µm/s to be σ = 0.038 Pa. According to the literature, mechanosensing in D.d. has been observed above a threshold of σ = 0.5 Pa [21]. We are thus approximately one order of magnitude below the regime where flow induced shear stress would bias the motion of chemotactic cells. C. Cell Tracking resolution of 1024x1024 pixel Differential Interference Contrast (DIC) images were recorded for 180 min, with time resolution of 10 sec and spatial (1 pixel=0.6409 µm), and processed using Mathworks MATLAB 7.5 with the Image Processing Toolbox [7, 8, 13]. All the image processing steps are done by M. Theves et al. The images were binarized to distinguish the cells from the background and possible optical artifacts. The cell centroids in each binarized frame were identified. To produce cell trajectories one had to link these locations together in time and space. To achieve this, a customized version of the MATLAB cell tracking algorithm written by Crocker and Grier [22] was used. This tracking pro- 3 FIG. 1: (a) Microfludic Gradient Mixer for Chemotaxis Experiments: two different concentrations flown into the channel inlets undergo steps of diffusive mixing at each branch to form a linear stable gradient in the area of observation. (b,c) Line Profile of fluorescein intensity inside the gradient mixer perpendicular to the flow direction, (d) Differential Interference Contrast (DIC) image, showing the cell population being exposed to the gradient. Only cell trajectories within the region of interest (blue box) are considered for statistics. Moreover, Bin 1 corresponds to the area, with the highest, Bin 3 to the one with the lowest average midpoint concentration experienced by the cells. This figure is used by the permission of M. Theves from his master thesis [13]. cess is consisted of calculating and minimizing the sum over the squared displacements of all possible links be- tween the cell positions in two subsequent frames. In- terestingly, an analysis of the broken trajectories have shown that more than 90 percent of the cells were lost due to a sudden jump in the cell location or because two cells ran into each other and their center of mass in the binarized image became indistinguishable. In this case, the tracks will end and new ones will start, once the cells separate again. Tracks may also end when the segmen- tation program loses a cell due to image quality prob- lems. Once the cell is detected again, a new track will start. These different scenarios result in a distribution of tracks of different length with most of them shorter than the total measurement time. They also result that the number of trajectories (e.g. 582 trajectories in Fig. 2) is much greater than the number of cells (∼ 40 cells at t = 0) in the experiment. Note that during an experi- mental recording, the number of cells is not constant with time as most of the fast cells move out of the region of interest or new cells enter the field of view. Finally, it is important to note that since the cells begin responding to the cAMP at different time points, or as the cells col- lide and new tracks start, the starting time-points of all trajectories are set to zero. D. Selection of Trajectories In our analysis, to have a reliable statistics, we keep the number of trajectories during the averaging process constant. Trajectories are selected based on two criteria: (i) they should persist at least 20 min and (ii) within this time interval, the cells should migrate more than 20 µm in −y direction. The minimum displacement of 20 µm in the direction of gradient, for t = 20 min, gives an average motility of ¯vy > 1 µm/min. Cells with ¯vy < 1 µm/min are neglected to exclude dead or immobile cells from statistics. As previously mentioned, we lose track of the cells once they collide. Therefore, it is important to note that based on this criteria, if a cell collides with another cell and the time interval between two successive collisions is less than 20 min, this trajectory is excluded from statistics although the cell was crawling with ¯vy > 1 µm/min. Eventually, to improve our statistics, long trajectories, are truncated at 20, 40, 60,... min and, if the conditions above are satisfied, trajectories between 20 to 40 min, 40 to 60 min,etc. are considered as new trajectories and the starting time point of each trajectory is set to zero (see Fig. 2). 4 where ∂t denotes differentiation with respect to t. Using Eqs. (2) and (3), one can find the diffusion-advection equation for the problem as ∂tP = ∂y (D∂yP ) − ∂y(vP ). (4) The chemotactic motion of the cells depends on both the absolute local concentration (chemo-kinesis) and its gra- dients (chemotaxis) [7, 17]. Based on the experiments of Ref. [7], here we consider a constant spatial gradient of cAMP in the direction of −y. Therefore, one can ex- pect that both the diffusion coefficient, D, and the drift velocity, v, depend on the y component of the position vector. Since there is no direct experimental method to determine this dependency, it is plausible to expand the mentioned coefficients in terms of y as III. MODEL and Nonlinear mean field Fokker-Planck equations can find important applications in the context of chemotaxis [23]. Here, we attempt to implement an advection-diffusion approach to describe the chemotactic movement of the D.d. cells experiencing a linear stationary gradient [7, 8]. The statistical properties of the system are characterized by the values of the model parameters returned after fit- ting the model to the experimental trajectories. To study the chemotactic movement of the D.d. cells, we consider an advection-diffusion model in which the centroid of the cell's perimeter is represented as the po- sition of a particle. We define an orthonormal basis with the unit vectors x and y, where x is the flow direction and −y is the direction of the spatial gradient of cAMP (see Fig. 2). The position of each cell is given by (cid:126)r = xx + y y. The concentration of the D.d cells is low enough, so we can assume that each cell does not sense the presence of the other cells. As stated in Experiment section, mi- crofluidic gradient along y direction is generated by a continuous flow along x. To avoid mixing up the issues of chemotaxis in response to the chemoattractant and mechanotaxis under the influence of the shear stress due to viscous forces, we limit our model to the chemotac- tic movement of the cells along y. Let us assume that p(x, y, t) denotes the number density of cells at position (x, y) at time t. Then, we have the probability density P (y, t), which is the original density p(x, y, t) integrated over x. The current density along y, J, reads as J = −D∂yP + vP. (2) where v and D are drift velocity and diffusion coefficient, respectively, and ∂y means differentiation with respect to y. Now, the continuity equation for P and J then reads as ∂tP = −∂yJ, (3) v = v0 + v1y + ··· D = D0 + D1y + ··· (5) (6) We keep the terms only up to the first order of y, and treat v1, and D1 as perturbation coefficients. Here, we assume that the current in the y direction does not de- pend on x. Using equations (4) to (6), one finds ∂tP = (D0 + D1y)∂2 yP + (D1 − v0 − v1y)∂yP − v1P. (7) (cid:90) (cid:90) The mean value of y-component of the cells' positions is obtained by (cid:104)y(t)(cid:105) = y P (y, t)dy. (8) Differentiating the above expression with respect to time results (cid:104)y(t)(cid:105) = d dt dy y ∂tP (y, t). (9) After substituting ∂tP (y, t) from Eq. (7) and integrat- ing, one can find (cid:104)y(t)(cid:105) = v0 + D1 + v1(cid:104)y(t)(cid:105). d dt (10) By solving this simple ordinary differential equation we find (cid:104)y(t)(cid:105) = ev1t + (cid:104)y(cid:105)0 − v0 + D1 v1 , (11) where (cid:104)y(cid:105)0 ≡ (cid:104)y(cid:105)t=0 denotes the mean initial y-position of the cells. As it has been mentioned above, v1 and D1 are the small parameters and in our model, they have (cid:20) v0 + D1 v1 (cid:21) 5 FIG. 2: (Color online) (a) 582 trajectories tracked in a microfluidic channel with Cmax = 50 nM (∇C = 0.14 nM/µm). Cells migrate on average upwards from the bottom of the channel to the top areas with higher cAMP concentration. (b) 88 trajectories selected (out of 582) based on the two conditions explained in Section II.D. The stars mark the cell positions exactly at 20 min and (if the trajectory is long enough) at 40 min, 60 min and etc. (c) The same trajectories in (b) truncated at 20 min to keep the number of cells during the averaging process constant. For long trajectories, if the two conditions are satisfied, the tracks between 20 to 40 min or from 40 to 60 min, etc., are considered as new independent trajectories to improve the statistics. (d,e) The comparison between experimental data (red lines) and the fitted model (blue line) for (cid:104)y(cid:105) and σ2 y. been considered as perturbation parameters. After ex- panding the exponential factor and keeping the terms up to the first order of perturbation parameters, v1 and D1, one can find (cid:104)y(cid:105)(t) = (cid:104)y(cid:105)0 + (v0 + v1(cid:104)y(cid:105)0 + D1) t + 1 2 v0v1t2. (12) It is worth mentioning that since terms like v1D1 are the second order of perturbation parameters, these terms are dropped. The variance of the cells' positions along y is defined y(t) ≡ (cid:104)y(t)2(cid:105) − (cid:104)y(t)(cid:105)2. Using similar method (see as σ2 Appendix I for details), one can find σ2 y(t) as (cid:3) t y(0) + 2(cid:2)σ2 σ2 y(t) = σ2 y(0) v1 + D0 + D1(cid:104)y(cid:105)0 + (2D0v1 + D1v0) t2, (13) where σ2 along y. We note that in Eq. y(0) is the initial variance of the cells' positions (13), we have kept the terms up to the first order of perturbation parameters as well. IV. RESULTS Now we are in a position to determine the perturba- tion parameters of our model based on the experimental trajectories. The mean displacement of chemotactic cells and the corresponding variance can be calculated from the experimental trajectories as defined in Appendix I (Eqs. 23 -- 26). To characterize the chemotactic behavior of D.d. cells, based on Eqs. (12) and (13), we need to de- termine the values of v0, v1, D0 and D1. We treat these factors as tuning parameters and find their values simul- taneously by fitting (MATLAB, R2016b) the model to the experimental values of (cid:104)y(cid:105)exp and σ2 y,exp. Tables I-III include the best fit values of the above mentioned param- eters at different cAMP gradients. In Table I, (cid:104)y(cid:105)0 and σ2 y(0) denote the fitted mean and standard deviations at time zero. Fig. 2 and Figs. 6 -- 11 (see Appendix III) show the com- parison between the model and the experiments at differ- ent cAMP concentrations. The initial number of trajec- tories before the selection procedure are presented in part (a) of each figure. Trajectories for our analysis are then selected and truncated based on the criteria explained in Section II.D. Selected and truncated trajectories are shown in parts (b) and (c) of each figure, respectively. The initial number of trajectories as well as the number of selected trajectories are different for different cAMP gradients. In parts (d-e) of each figure, the red lines correspond to the experimental data and the blue lines correspond to the results of our model using the fitting parameters of Tables I-III. The important features of the figures and the tables are summarized below: • The mean position of the chemotactic D.d. cells, (cid:104)y(cid:105), decreases almost linearly in time, which shows that the chemotactic cells migrate towards areas with higher cAMP concentration (top areas of the channel). The nonlinear term v0v1/2 in Eq. 12 is small for all concentrations (see the last column of Table. II). • The mean square displacement function σ2 y(t) shows decreasing behavior at Cmax = 10, 50, 316 and 10000 nM . This trend is an experimental ob- servation, independent of the introduced model, and has to do with the fact that the cells tend to migrate to areas with high cAMP concentration (top part of the channel). Since in these areas mid- point cAMP concentration is high and most of the cAMP receptors are saturated, therefore the cells slow down and accumulate at the top of the chan- nel. This "accumulation" can give rise to a decreas- ing σ2 y(t). In the other word, the possible decrease in the variance is due to a drift towards the top areas of the channel. • The diffusion coefficient in y direction D0 + D1(cid:104)y(cid:105) is initially positive for all concentrations but as the cells migrate upwards and the value of (cid:104)y(cid:105) de- creases, it becomes negative for all concentrations except for Cmax = 1 nM . We think that this neg- ative diffusion coefficient extracted from the data is an artifact of the perturbative approximation. To be more exact, the diffusion coefficient depends on the position, and we have Taylor-expanded it and kept only the zeroth and the first order terms (the latter as a perturbative term). While the full position-dependent diffusion coefficient should probably be non-negative, there is no such restric- tion on its truncated form (which contains only the first two terms): it is just a parameter which is determined through a best fit. Of course the pa- 6 rameters should respect the non-negativity of the variance, and they do, as it is seen that the fitted variance does not become negative. • We define the mean drift velocity of chemotactic D.d. cells as d(cid:104)y(cid:105) dt vdrift = = v0 + v1(cid:104)y(cid:105)0 + D1 + v0v1t, (14) which shows that the drift velocity depends not only on v0 and v1 but also on D1. The coefficient v0v1 is a small number for different cAMP gradi- ents (see Table. II), therefore vdrift is essentially constant in time. The extracted values of drift ve- locity at time zero are listed in the fifth column of Table. II. It is interesting that the drift velocity in y direction doesn't depend significantly on the cAMP gradient and fluctuates around 4 µm/min. This is consistent with an independent data anal- ysis performed by M. Theves [13] (see Fig. 3): within a plateau, ranging from 10−2 nM/µm ≤ ∇C ≤ 1 nM/µm over two orders of magnitude, the chemotactic velocity is almost constant. For gradients above ∇C = 1 nM/µm the directional- ity of movement is decreased. Exceeding an upper threshold of ∇Cup ∼ 102 nM/µm, the cell motion becomes isotropic again. • For all gradients, while the cells crawl up the gra- dient, the magnitude of (cid:104)vy(cid:105) = v0 + v1(cid:104)y(cid:105) decreases as the midpoint concentration increases. However, the independent data analysis of M. Theves [13] shows a transition: for shallow gradients, right af- ter the onset of chemotaxis ∇C = 0.003 nM/µm, vy increases as the background concentration rises. This effect reverses for steep gradients ∇C > 0.3 nM/µm (see right panel of Fig. 3). V. DISCUSSION We have analyzed large data sets of D.d. chemotaxis in linear gradients of cAMP recorded by Theves et al. in a microfluidic setup [7, 8, 13]. Data sets with different steepnesses of cAMP gradient were included in our anal- ysis, covering a large range of gradients, in which chemo- tactic behavior was observed. Inspired by the experi- mental conditions of Ref. [7], we introduced a minimal phenomenological model that explicitly incorporated the dependency of diffusion matrix and velocity of the cells on their positions which corresponds to the position de- pendence of the local concentration of chemotactic cues. Based on this model, we extracted the physical proper- ties of the chemotactic D.d. cells using the mean and variance of the experimental cell tracks. What is the benefit of this phenomenological model? As highlighted 7 FIG. 3: (Color online) Independent data analysis done by M. Theves [13] showing (left) chemotaxis as a function of gradient steepness ∇C: above a threshold at ∇C (cid:39) 10−3 nM/µm cells show a positive (in our coordinate system negative) average velocity in gradient direction (vy) as well as an increased total motility v, while the perpendicular velocity component in flow direction (vx) remains random and averages to zero within error bars. For gradients ranging over two orders of magnitude, 10−2nM/µm ≤ ∇C ≤ 1 nM/µm, the chemotactic speed is constant. (right) Average chemotactic velocity vy as a function of gradient steepness evaluated separately for three different areas, subdividing the region of interest (see Fig. 1d). The midpoint concentration decreases from bin 1 to bin 3. For shallow gradients, the chemotactic velocity increases with a raise in the average midpoint concentration. This effect seems to reverse for steep gradients above 1 nM/µm, where vy decreases slightly in higher concentration backgrounds. The figures are used by the courtesy of M. Theves from his master thesis [13]. TABLE I: The mean initial positions of the cells and the corresponding variances at time zero for different cAMP concentrations. Cmax(nM ) ∇C (nM/µm) (cid:104)y(cid:105)0 (µm) σ2 x(0)(µm2) σ2 1 10 31.6 50 100 316 10000 0.003 0.03 0.09 0.14 0.29 0.9 28.6 373.39 347.22 417.88 365.99 410.24 371.13 382.33 10880 25893 24136 27253 32453 30854 27665 y(0)(µm2) 5454.2 6857.1 7608.9 6707.3 4719.2 7033.1 6522.6 previously, chemotactic movement of the cells depend on both the chemoattractant gradient and the average ambi- ent chemoattractant concentration (midpoint concentra- tion). In the microfluidic setup of Theves et al. [7], the cells are exposed to a constant gradient, while the mid- point concentration increases when the cells are moving up the gradient. Traditionally, chemotactic cell motion is described by Langevin-type equation where for each cell track, the velocity and acceleration of the cells are calculated at each point by finite differences from the cell positions [7, 8, 24]. Therefore in these types of analysis midpoint concentration is globally averaged out. Other quantities such as chemotactic index, defined as the dis- tance moved in gradient direction divided by the total distance, are also averaged quantities where information on mid-point concentration is lost. However in our analy- sis, instead of velocity and acceleration, we work directly with spatial position of the cells and explicitly include the dependence of the diffusion coefficient and the drift velocity on the midpoint concentration. Taylor expan- sion of these coefficients up to the first order in y leads to a closed set of equations that can be solved to obtain the fitting parameters. It is worth to check the effect of the dependency of the diffusion and velocity on the local concentration. To do this, let us assume that the drift velocity and the diffusion coefficient were constant. We denote the constant drift velocity and diffusion constant by v0 and D0, respectively, to obtain (cid:104)y(cid:105)(t) = (cid:104)y(cid:105)0 + v0t, σ2 y(t) = σ2 y(0) + 2 D0t. (15) (16) These equations predict a linear dependency on t, in both the mean and the variance of the position, which is not 8 FIG. 4: (Color online) (a) Trajectories of three different experiments recorded at the same cAMP gradient of 0.14 nM/µm are combined to improve the statistics. (b) 207 (out of 1097) trajectories, are selected and (c) truncated based on the criteria in Section II.D. These trajectories participate in our analysis which is more than two times the number of selected trajectories in Fig. 2. (d,e) The comparison between experimental data (red lines) and the fitted model (blue line) for (cid:104)y(cid:105) and σ2 y. TABLE II: Drift coefficients for different cAMP gradients. The mean drift velocity of the cells at time zero is presented in the fifth column. Cmax(nM ) v0(µm/min) v1(1/min) v0 + v1(cid:104)y(cid:105)0 v0 + v1(cid:104)y(cid:105)0 + D1 v0v1/2 −0.024 0.045 0.033 −0.017 −0.027 0.058 0.079 −0.017 −0.012 −0.010 −0.016 −0.016 −0.013 −0.015 −3.32 −11.69 −11.04 −3.95 −3.16 −13.77 −16.36 1 10 31.6 50 100 316 10000 2.87 −7.62 −7.27 2.04 3.36 −8.98 −10.71 −2.94 −4.06 −4.66 −3.45 −3.08 −4.83 −3.97 y (see Fig. 2 and Figs. 6 -- 11). consistent with the experimental data especially in the variance of σ2 In fact it has been shown in Ref. [25], that any linear diffusion model (even anomalous), which enjoys both time transla- tional invariance and space translational invariance leads to means and variances which are at most linear in time. This is a motivation to use drift and diffusion parame- ters which do depend on the position. There is an obvious position-dependence in our system: C (the concentration of cAMP) does depend on the coordinate y. Assuming that the drift and diffusion parameters do depend on C, one is left with a y-dependence in the drift and diffusion parameters. A simple manageable model is to Taylor- expand this y-dependence and keep only terms which are up to first order in y. The result is a first order perturba- tion model, which has been studied here. We emphasize that the experimental conditions of Ref. [7] fulfill the nec- essary conditions of the mentioned study. In previous studies, wild-type and mutated epithe- lial canine kidney cells, it has been shown that the cell dynamics can be characterized by an anomalous dif- fusion [26]. In particular, mean squared displacement TABLE III: Diffusion coefficients in y direction for different cAMP gradients. 9 Cmax(nM ) D0 D1 D0 + D1(cid:104)y(cid:105)0 −49.10 1 0.37 −2612.60 7.63 10 −2581.40 6.38 31.6 −148.60 0.50 50 100 129.80 0.09 −3270.57 8.90 316 10000 −4690.80 12.39 89.50 38.23 86.67 32.52 164.54 45.22 46.59 2D0v1 + D1v0 2.69 3.09 0.27 5.87 −3.84 4.15 6.00 2(σy(0)2v1 + D0 + D1(cid:104)y(cid:105)0) −1.81 −84.38 35.76 −154.38 179.00 −91.00 −99.65 FIG. 5: (Color online) (a) The same trajectories as in Fig. 2 which are shifted to x = 0, (b) selected and (c) truncated based on the criteria in Section II. D. The y dependency of all trajectories are kept as before because the mid-point concentration is different along the width of the channel. (d) and (e) show the behavior of σ2 x as a function of t, before and after shifting all the tracks to x = 0, respectively. Red lines correspond to the experimental data and blue lines correspond to fitted quadratic polynomials, respectively. shows a super-diffusive behavior. This super-diffusive behaviour was also observed in the mean square dis- placement of Hydra cells [27]. However, experimental trajectories of chemotactic D.d. cells in Ref. [18], were interpreted by a data-driven model with purely diffusive behavior. As we mentioned above, a pure diffusive model can not explain the non-linear behavior observed in the experimental data of Ref. [7]. In our analysis, we observed that at all concentrations D0 +D1(cid:104)y(cid:105) decreases with time and becomes negative for concentrations of 10, 31.6, 316, and 10000 nM . In or- der to make sure that negative diffusion coefficients are not due to our low statistics after the selection proce- dure, we combined trajectories of three different exper- iments performed at the same cAMP gradient, namely ∇C = 0.14 nM/µm (Cmin = 0, Cmax = 50 nM ). Com- parison between Fig. 2 and Fig. 4 shows a similar de- creasing behavior in σ2 y(t). Indeed, an absorbing point on top of the channel can produce a decreasing vari- ance, not through the diffusion but through the upstream drift. Let us suppose that D0 = D1 = 0. Then, accord- ing to Eq. 21, σ2 If everything is expanded up to first order in v1, then the result is σ2 y(t) = σ2 y(0) could become negative after a while. But that is an arti- fact of the approximation. We intended to find position- dependence of diffusion and drift coefficients. Since the exact position-dependence is not known, even if the in- homogeneity of the surface is known, we expanded the diffusion and drift coefficients in power of the position y(0)(1 + 2v1t). As v1 ≤ 0, it seems that σ2 y(0) exp(2v1t). y(t) = σ2 y. That the time dependence of the variance does match the experiments, means that the method works. But the perturbative parameters should not be misleading. Furthermore, with our model, we can test directly the space-time symmetries of the cell movement. Based on the reports of the experiments the gradient in the y di- rection is homogeneous in x. However, the cell tracks shown in the panel (a) of Figs. 2, 6, 7, 9, 10, and 11 seem to show a drift in positive x direction (in addition to the chemotactic drift in -y direction). To check the spatial homogeneity in the x direction, we shifted all the tracks of Fig. 2 to x = 0 (see Fig. 5). It is interest- ing that in this case σ2 x(t) is not a pure translation of the same function for unshifted trajectories (see Fig. 5). It seems that the behavior of the function depends on the initial condition. This non-pure shift in σ2 x(t) could be a hallmark of correlation between the displacement of individual cells along x direction and their initial x- positions (see Appendix II). Apparently, the system does not have the translational symmetry along the x direc- tion. This is surprising, since analysis in Section II.B. shows that with flow speed of 320 µm/s we are far be- low the regime, where mechanotactic effects have been observed in D.d. cells. However, the authors of Ref. [21] conducted their experiments with vegetative cells. This suggests that starvation may increase the mechanosensi- tivity of D.d. cells. We emphasize that with this cor- rection all of our analysis in the y direction is still valid, given that the current in the y direction does not de- pend on x. This assumption is nothing but a mean-field approximation. To improve our statistics, we have divided long mother trajectories to shorter ones and if the criteria in Section. II.D are satisfied, we have included daughter trajectories as completely independent tracks in our analysis. The main difference between these new daughter trajectories is the average midpoint concentration that the cells ex- perience as they crawl up the gradient. This corresponds to moving up from Bin 3 to Bin 1 in Fig. 1d, where in each Bin cells are exposed to a different average midpoint concentration. Detailed analysis by Theves et al. have shown that (see the right panel of Fig. 3) with a raise in the average midpoint concentration the average chemo- tactic velocity vy doesn't show any clear trend for inter- mediate gradients, 10−2nM/µm ≤ ∇C ≤ 0.3 nM/µm. However, for shallow gradients vy increases with mid- point concentration and for steep gradients it decreases. Most of our analysis are done at intermediate gradients where the variation in chemotactic velocity vy between three different Bins is less than 25 percent. Exemplary, at ∇C = 0.3 nM/µm, average chemotactic velocity changes from ∼ 3 µm/min to ∼ 4 µm/min for three bins with different midpoint concentrations of 17 nM, 50 nM and 83 nM . At steep gradients larger than 0.3 nM/µm, the variations in vy is even less than 10 percent. Thereby we believe that shorter daughter trajectories which be- 10 long to one mother long trajectory, do not significantly differ in their chemotactic properties. In other words, by dividing long mother trajectories to shorter daughter tracks, we don't introduce new types of trajectories with completely different statistical properties. In the present work, even though we have analyzed a substantial amount of data, much larger data sets with longer trajectories would be required in order to improve our statistics. Possible future experiments with wider microfluidic channels can be helpful to obtain longer tra- jectories. Experiments with lower cell density (to avoid cell-cell collision) can also help us to obtain longer tra- jectories, as the cells after collision are indistinguishable from each other and two new trajectories are detected by the cell tracking algorithms. In summary, we have analyzed the experimental data of chemotactic D.d. cells in the linear gradient of cAMP. In order to have a reliable statistics, we kept the number of trajectories during our analysis constant. Trajectories were selected based on two criteria: (i) they should per- sist at least 20 min, and (ii) within this time interval, the cells should have migrated more than 20 µm in the direc- tion of the the gradient of cAMP. We have shown that by introducing an advection-diffusion model that includes the position dependence on the cAMP concentration, a quantitative description of experimental cell tracks of the amoeba D.d. is achieved. Our analysis goes beyond a pure diffusive model and shows that the super-diffusive behavior can dominate at larger time scales. Specifically, while in a conventional advection-diffusion model both the mean and the variance are linear in time, here in both cases terms arise which are quadratic in time. In future study, we aim to apply our analysis to the tra- jectories of cells migrating on surfaces of differing com- position. In a recent study, it has been shown that D.d. cells migrate similarly on surfaces with various chemical composition [28]. As the substrate composition changes, the cells regulate forces generated by actomyosin network to maintain optimal cell-surface contact area and adhe- sion. We will assess migration trajectories of the cells on different surfaces and investigate the variations in the fitting parameters of our model. Furthermore, we aim to extend our analysis to the trajectories of mutant cell lines that single or multiple components of the chemo- tactic signaling pathway are deficient and consequently the character of the cell trajectories may change consid- erably. Structural differences between the trajectories of wild-type and mutant cells may reflect important in- formation about the role of the various proteins in the signaling pathway of D.d. cells, which possibly can not be resolved in the models that mid-point concentration informations are averaged out. The objective is to corre- late various parameters of our model to the key molecular players involved in chemotaxis. This can provide a link between the observed macroscopic dynamics and the un- derlying microscopic mechanism which is an important 11 (cid:20) σ2 y(0) + σ2 y(t) = − D1v0 v1 t − D0 + D1(cid:104)y(0)(cid:105) (cid:20) D0 + D1(cid:104)y(0)(cid:105) v1 + v1 D1v0 2v2 1 D1v0 2v2 1 + (cid:21) (cid:21) e2v1t , (21) where σ2 y(0) denotes the initial variance of the cells' po- sitions along y. After expanding the above equation and keeping the terms up to the first order of v1 and D1, one has σ2 y(t) = σ2 y(0) v1 + D0 + D1(cid:104)y(cid:105)0 + (2D0v1 + D1v0) t2. (22) y(0) + 2(cid:2)σ2 (cid:3) t The mean displacement of chemotactic cells, in both x and y directions, and the corresponding variances can be calculated from the experimental data as follows (cid:104)x(cid:105)exp(t) = (cid:104)y(cid:105)exp(t) = σ2 x,exp(t) = σ2 y,exp(t) = 1 N 1 N 1 N 1 N i=1 N(cid:88) N(cid:88) N(cid:88) N(cid:88) i=1 i=1 i=1 xi(t), yi(t), [xi(t) − (cid:104)x(cid:105)exp(t)]2 , [yi(t) − (cid:104)y(cid:105)exp(t)]2 , (23) (24) (25) (26) Appendix II Here we show how shifting the cells' tracks to x = 0 can affect on the variance of the x-component through the time, σ2 x(t). Let xi(t) be the position of the i'th particle in x-direction at time t. The displacement of the i'th particle in x-direction through the time is zi(t) = xi(t) − xi(0). Simply, one has (cid:104)z(t)(cid:105) = (cid:104)x(t)(cid:105) − (cid:104)x(0)(cid:105) where (cid:104)x(t)(cid:105) and (cid:104)x(0)(cid:105) denote the mean values of x-component of the particles at time t and t = 0, respec- tively, and (cid:104)z(t)(cid:105) is the mean displacement of the parti- cles. It is worth mentioning that for example (cid:104)x(t)(cid:105) ≡ i xi(t), where N denotes the number of cells. In 1/N(cid:80) order to find the variance, first we note that xi(t) − (cid:104)x(t)(cid:105) = [zi(t) − (cid:104)z(t)(cid:105)] + [xi(0) − (cid:104)x(0)(cid:105)]. (27) (cid:104)y(t)2(cid:105) = y2P (y, t)dy. (17) where N denotes the number of cells. goal in the field of eukaryotic chemotaxis. as ACKNOWLEDGMENT We are deeply grateful to M. Theves, E. Bodenschatz, G. Amselem and C. Beta for sharing the experimental data of Ref. [7] and A. Bae for critical reading of the manuscript. Z.E. and F.M.-R are grateful to A. Celani, R. Golestanian and L. Mollazadeh-Beidokhti for helpful discussions. Z.E. and F.M.-R. acknowledge the hospital- ity of ICTP in Trieste, where some parts of this work is done. F.M.-R. acknowledges the hospitality of MPIDS- LFPN Group in Gottingen, where this work was initi- ated. M.K. acknowledges the support of the research council of the Alzahra University. A. G. acknowledges the support of the MaxSynBio Consortium which is jointly funded by the Federal Ministry of Education and Re- search of Germany and the Max Planck Society. Appendix I Here we show how to derive Eq. (13). Indeed, for what follows we do not need the exact form of P (y, t) itself, but just the time dependence of its moments. (cid:104)y(t)2(cid:105) is defined as (cid:90) ∂ ∂t dt d dt We can directly obtain an equation for the time evolution of (cid:104)y(t)2(cid:105) by multiplying the master equation, Eq. (7), by y2 and integrate over y. This results in dy y2(cid:2)(D0 + D1y) ∂2 yP + (D1 − v0 − v1y)∂yP − v1P ] .(18) dy y2P (y, t) = The left-hand side of this equation is simply equal to d(cid:104)y(t)2(cid:105) . We apply partial integration to the right-hand side and obtain (cid:104)y(t)2(cid:105) = 2D0 + 2(v0 + 2D1)(cid:104)y(t)(cid:105) + 2v1(cid:104)y(t)2(cid:105). (19) dt(cid:104)y(t)2(cid:105)− y(t) ≡ (cid:104)y(t)2(cid:105)−(cid:104)y(t)(cid:105)2, and d dt(cid:104)y(t)(cid:105), one finds y(t) = d dt σ2 Since σ2 2(cid:104)y(t)(cid:105) d d dt y(t) = 2D0 + 2D1(cid:104)y(t)(cid:105) + 2v1σ2 σ2 (20) where according to Eq. (10), d(cid:104)y(cid:105)/dt has been replaced by v0 + D1 + v1(cid:104)y(t)(cid:105). The solution of Eq. (20) is found y(t), (cid:90) (cid:90) (cid:68) [z(t) − (cid:104)z(t)(cid:105)]2(cid:69) (cid:68) [x(0) − (cid:104)x(0)(cid:105)]2(cid:69) = + 2(cid:104)[x(0) − (cid:104)x(0)(cid:105)] [z(t) − (cid:104)z(t)(cid:105)](cid:105) . + 12 [5] R.H. Kessin, Dictyostelium: evolution, cell biology, and the development of multicellularity (Cambridge Univer- sity Press, 2001). [6] C.S. Patlak, Bull. of Math. Biophys.15, 311 (1953) [7] G. Amselem, M. Theves, A. Bae, E. Bodenschatz, and C. Beta, PLoS ONE 7, e37213 (2012). (28) [8] G. Amselem, M. Theves, A. Bae, C. Beta, and E. Boden- schatz, Phys. Rev. Lett. 109, 108103 (2012). [9] A. Friedman, and C.-Y. Kao, Mathematical Modeling of Biological Processes (Springer, 2014). [10] E. F. Keller and L. A. Segel, J. Theor. Biol. 26, 399 (1970). [11] H. Othmer, A. Stevens, SIAM J. Appl. Math.57, 1044 (1997) [12] P. Chavanis, Eur. Phys. J. B 62, 179 (2008) [13] M. Theves, Quantitative Study of Eukaryotic Chemo- taxis with Microfluidic Devices (Master's thesis), Georg- August Universitat Gottingen, and Max Planck Institut fur Dynamik und Selbstoganisation (2009). [14] N. Andrew, and R.H. Insall, Nat. Cell. Biol. 9, 193 (2007). [15] L. Bosgraaf, and P.J.M. Van Haastert, PLoS ONE 4, e6842 (2009). [16] P.J.M. Van Haastert, Biophys. J. 99, 3345 (2010). [17] L. Song, S.M. Nadkarni, H.U. Bodeker, C. Beta, A. Bae, et al., Eur. J. Cell Biol. 85 981 (2006). [18] L. Li, E.C. Cox, and H. Flyvbjerg, Phys. Biol. 8, 046006 (2011). [19] N. Makarava, S. Menz, M. Theves, W. Huisinga, C. Beta, and M. Holschneider, Phys. Rev. E 90, 042703 (2014). [20] N. L. Jeon, S. K. W. Dertinger, D. T. Chiu, I. S. Choi, A. D. Stroock, and G. M. Whitesides. Langmuir, 16 168311 (2000). [21] E. Decave, D. Rieu, J. Dalous, S. Fache, Y. Brechet, B. Fourcade, M. Satre, and F. Bruckert. J. Cell Sci. 116 4331 (2003). [22] J. C. Crocker and D. G. Grier, J. of Colloid and Interface Sci., 179, 298 (1996). [23] J.D. Murray, Mathematical Biology (Springer, Berlin, 1991). [24] D. Selmeczi, S. Mosler, P.H. Hagedorn, N.B. Larsen NB, H. Flyvbjerg, Biophys. J. 89, 912 (2005). [25] M. Khorrami, A. Shariati, A. Aghamohammadi, and A. H. Fatollahi, Phys. Lett. A 376, 687 (2012). [26] P. Dieterich, R. Klages, R. Preuss, and A. Schwab, Proc. Natl. Acad. Sci. USA 105, 459 (2008). [27] A. Upadhyaya, J.-P. Rieu, J.A. Glazier, and Y. Sawada, Physica A 293, 549 (2001). [28] C.P. McCann, E.C. Rericha, C. Wang, W. Losert, and C. A. Parent, PLoS ONE 9, 87981 (2014). After squaring both sides of the above equation and av- eraging, one finds (cid:68) [x(t) − (cid:104)x(t)(cid:105)]2(cid:69) covariance of z(t) and x(0) The is defined as cov[z(t), x(0)] ≡ (cid:104)(z(t) − (cid:104)z(t)(cid:105))(x(0) − (cid:104)x(0)(cid:105))(cid:105). This quantity provides a measure for the strength of the cor- relation between two stochastic variables. Using the def- inition of the variance and covariance, Eq. (28) can be written as σ2 x(t) = σ2 z (t) + σ2 x(0) + 2 cov[z(t), x(0)]. (29) We see that when z(t) and x(0) are independent, one has (cid:104)z(t)x(0)(cid:105) = (cid:104)z(t)(cid:105)(cid:104)x(0)(cid:105) and cov[z(t), x(0)] becomes zero. In this case, σ2 z (t) by just a con- stant shift. In other words, the necessary and sufficient condition for a pure shift in the variances of σ2 x(t) and σ2 z (t) is vanishing of the covariance of z(t) and x(0). x(t) differs from σ2 Appendix III As we discussed in the main text, the experiments had been done in different cAMP concentrations. Here we present the trajectories, before and after selection proce- dure, as well as the corresponding analysis for the cAMP concentrations of Cmax = 1, 10, 31.6, 100, 316, and 10000 nM. References [1] H. Takagi, M.J. Sato, T. Yanagida, and M. Ueda, PLoS ONE 3, e2648 (2008). [2] W. F. Loomis, D. Fuller, E. Gutierrez,A. Groisman,W. J. Rappel, PLoS ONE 7, 42033 (2012). [3] M. Buenemann, H. Levine, W.J. Rappel, L.M. Sander, Biophys. J. 99, 50 (2010). [4] K.F. Swaney, C.H. Huang, and P.N. Devreotes, Annu. Rev. Biophys. 39, 265 (2010). 13 FIG. 6: (Color online) In an experiment with Cmax = 1 nM (∇C = 0.003 nM/µm), out of 282 trajectories shown in panel (a), after applying the selection conditions of Section II.D, only 41 trajectories are selected in panel (b) and truncated in panel (c). The comparisons between the experimental data (red) and the model (blue) are presented in panels (d) and (e). FIG. 7: (Color online) (a-c) Based on our selection criteria in Section.II.D, only 27 (out of 815) trajectories participate in our analysis for Cmax = 10 nM (∇C = 0.028 nM/µm). (d) and (e) show (cid:104)y(cid:105) and σ2 y plotted for both experimental data (red) and the model (blue). 14 FIG. 8: (Color online) (a-c) 61 (out of 650) trajectories satisfy the selection conditions of Section II.D for gradient of ∇C = 0.09 nM/µm (Cmax = 31.6 nM ). In panels (d) and (e) the outcome of experimental data and the model are compared. FIG. 9: (Color online) Out of 2321 trajectories in panel (a), only 25 are selected in panel (b) and truncated in panel (c) for Cmax = 100 nM (∇C = 0.28 nM/µm). A large number of trajectories, either show immobile cells or become discontinuous as the cells collide. Again panels (d) and (c) are the comparison between the data and the model. 15 FIG. 10: (Color online) (a-c) 39 trajectories (out of 254) are participating in our analysis for Cmax = 316 nM (∇C = 0.9 nM/µm). y and σ2 y are calculated from the truncated trajectories (red lines) and compared with the fitted model (blue lines) in panels (d) and (e). FIG. 11: (Color online) (a-c) In an experiment with Cmax = 10000 nM (∇C = 28.6 nM/µm), out of 401 trajectories, 41 are selected and truncated. Comparisons between the experimental measured quantities (red) and the fitted model (blue) are shown in panels (d) and (e).
1808.00314	2	1808	2018-08-03T11:27:34	Morphology and Motility of Cells on Soft Substrates	[ "physics.bio-ph", "q-bio.CB" ]	Recent experiments suggest that the interplay between cells and the mechanics of their substrate gives rise to a diversity of morphological and migrational behaviors. Here, we develop a Cellular Potts Model of polarizing cells on a visco-elastic substrate. We compare our model with experiments on endothelial cells plated on polyacrylamide hydrogels to constrain model parameters and test predictions. Our analysis reveals that morphology and migratory behavior are determined by an intricate interplay between cellular polarization and substrate strain gradients generated by traction forces exerted by cells (self-haptotaxis).	physics.bio-ph	physics	Morphology and Motility of Cells on Soft Substrates Andriy Goychuk,1 David B. Bruckner,1 Andrew W. Holle,2 Joachim P. Spatz,2, 3 Chase P. Broedersz,1 and Erwin Frey1 1Arnold Sommerfeld Center for Theoretical Physics and Center for NanoScience, Department of Physics, Ludwig-Maximilians-Universitat Munchen, Theresienstr. 37, D-80333 Munich, Germany 2Department of Cellular Biophysics, Max-Planck-Institute for Medical Research, D-69120 Heidelberg, Germany 3Department of Biophysical Chemistry, University of Heidelberg, D-69120 Heidelberg, Germany Recent experiments suggest that the interplay between cells and the mechanics of their substrate gives rise to a diversity of morphological and migrational behaviors. Here, we develop a Cellular Potts Model of polarizing cells on a visco-elastic substrate. We compare our model with experiments on endothelial cells plated on polyacrylamide hydrogels to constrain model parameters and test predictions. Our analysis reveals that morphology and migratory behavior are determined by an intricate interplay between cellular polarization and substrate strain gradients generated by traction forces exerted by cells (self-haptotaxis). Cell migration is a highly complex process determined by internal chemo-mechanical processes and the interac- tion of the cell with its environment [1 -- 4]. Indeed, cells respond to the mechanical properties of the substrate to which they adhere [6 -- 13, 18]. Interestingly, with increas- ing substrate rigidity, different cell types show qualita- tively distinct migratory behavior. For example, glioma cells [14], glioblastoma cells [15], and human adipose- derived stem cells [12] plated on polyacrylamide (PA) hydrogels, as well as fish keratocytes on PA and poly- dimethylsiloxane (PDMS) hydrogels [20], move faster and more persistently with increasing elastic modulus. In contrast, rat fibroblasts plated on polyethylene glycol- based (PEG) hydrogels [17], as well as 3T3 fibroblasts on PA hydrogels [18], show the opposite behavior and slow down, while still increasing their persistence of mi- gration on stiffer substrates. What then are the physical principles that lead to such diverse cell behaviors? [19, 20]. Substrates like PA and PEG hydrogels are widely re- garded as almost ideally elastic materials In general, however, substrate viscosity may also affect cell migration. For example, correlations in the movement of epithelial sheets have been shown to increase with sub- strate viscosity [21], and a recent computational study has demonstrated the relevance of viscous substrate re- modelling for cell spreading [22]. These studies suggest an intricate interplay between cell migration and both the elastic and viscous properties of the environment. It remains to be resolved, however, whether and how these cell-substrate interactions can reconcile the apparently contradictory migratory responses of various cell types on different substrates. Previous computational approaches, including phase field models [23 -- 29], cellular Potts models (CPM) [6, 7, 12, 31, 32, 35, 36], particle-based models [37 -- 45], and various continuum models [7, 8, 11, 46 -- 53], have led to important advances in understanding cell traction force generation and cell migration. In particular, these stud- ies have helped to rationalize the coupling between single- cell motion and substrate deformation [11, 25, 32, 36, 45 -- 47, 51, 52]. However, these models neglect spatial cou- pling of substrate deformations [25], cannot capture cell shape [11, 45 -- 47], do not include a cell polarization mech- anism [11, 32, 36, 46, 47], and mostly exclude persistent cell migration. Here, we study the morphology and migratory behav- ior of actively polarizing cells on visco-elastic substrates of varying elastic stiffness and different degrees of viscous friction. To this end, we develop a CPM of actively polar- izing motile cells [6, 7] that mechanically interact with a simple visco-elastic substrate [Fig. 1], using experimen- tal measurements on human umbilical vein endothelial cells (HUVECs) plated on PA gels to constrain the model parameters. Our combined experimental and theoreti- cal investigations suggest that a cell's response to the physical properties of the substrate can be understood in a relatively simple way, without explicitly taking into account additional effects like stiffness-dependent adhe- sions. Within our picture, cells generate substrate strain gradients, which guide shape changes and cell migration (self-haptotaxis). The interaction with the substrate can in turn interfere with, and even override, internal feed- FIG. 1. Sketch of the computational model. The substrate is represented by nodes i at positions xi, each connected to six nearest neighbors j ∈Ni by loaded springs. A cell C is comprised of a set of hexagons with respective areas a(xi, t) and local protrusion energies (xi, t)∈ [q, Q] (color scale). As the cell exerts traction forces T on the nodes, it compresses the substrate beneath, while stretching the surrounding sub- strate. The cell protrudes or retracts over an effective distance d in the direction ±d, where d = xj−xi. nodeinodejkζijaprotrusionfieldqQTdback mechanisms that would under normal circumstances lead to cell polarization. Experimental observations. We started our analysis by experimentally investigating HUVECs plated on PA gels. Depending on the substrate stiffness, we observed three distinct migratory cell behaviors [Fig. 2(a)-(e)]. At low stiffness, cells are elongated and localized (elongation): Even though they locally move at some slow speed v in random directions, they remain localized within a cer- tain substrate area and do not show persistent motion. As substrate stiffness is increased, cells first round up and increase their local speed, but remain localized (round- ing). Only when the substrate stiffness is increased above some threshold value, do cells begin to show persistent cell migration (running), which can be described as a persistent random walk with ballistic motion on short timescales and diffusive motion on long timescales. Generalized CPM. To rationalize these diverse cell be- haviors we build on and extend a recently introduced generalization of the CPM [6, 7], which includes the fol- lowing basic features of cellular dynamics: Elasticity of the cell membrane and cortex, dynamic cell polarization through a chemo-mechanical feedback mechanism, and force generation driven by the interplay between actin polymerization and contraction of acto-myosin networks. As described below, we add as a new feature the visco- elastic coupling of cell and substrate deformations. We consider a cell as a connected set C of occupied grid sites (hexagons) i, with positions xi(t) and areas a(xi, t) that change with time t [Fig. 1]. Cell motion and cell shape changes are implemented as elementary protrusion and retraction events, corresponding to an in- crease and decrease, respectively, in the number of occu- pied grid sites. Moreover, through coupling with a visco- elastic substrate, the individual grid areas may change dynamically. The dynamics of each cell is determined by a Monte Carlo update scheme with the 'statistical' weight being given by a Boltzmann factor with a Hamil- tonian H =HP +HM , which describes the balance be- tween a cell's tendency to protrude and migrate and the constraints imposed by membrane elasticity. FIG. 2. Migratory cell states. Depending on substrate stiff- ness, HUVECs show distinct migratory cell states: (a) elonga- tion, (b) rounding, and (c) running, as quantified by the his- tograms for the cell extension α = 1− 4πA/P 2 (d), and local cell speed v (e) in [nm/s]). (f)-(j) Analogous results from the computational model for viscous friction ζ = 17.5 s nN/µm. 2 As in the original CPM [12], deformations of a cell's membrane and cortex are assumed to be constrained by the elastic energy HM = κAA(t)2+κP P (t)2, with κA and κP denoting the stiffnesses corresponding to the area A(t) and perimeter P (t) of the cell, respectively. The ensuing contractile forces are counteracted by outwardly directed forces generated by cytoskeletal structures anchored to the substrate at focal adhesion sites [13, 14]. In our model, the local energetic contribution from this cellular i∈C (xi, t), with the activity is described by HP =− scalar protrusion field (xi, t)∈ [q, Q] [6, 7]. The protru- sion field is dynamic, reflecting the response of cytoskele- tal structures to external mechanical stimuli through feedback mechanisms involving regulatory cytoskeletal proteins [15, 16]. In the generalized CPM these complex biochemical processes are accounted for in a simplified way by regulatory factors that reinforce the protrusion field in a positive feedback loop that can lead to sponta- neous cell polarization [6, 7]; for details see the Supple- mental Material (S.M.) [58]. (cid:80) Cell-substrate coupling. How can one account for sub- strate deformations and their coupling to cell deforma- tion in a CPM? As the cell's cytoskeleton is anchored to the substrate via focal adhesion sites while the cell is exerting force on the substrate, we will assume that each hexagon area a(xi, t) deforms in an affine way with the substrate. In the continuum limit, this implies that the protrusion energy density is given by (x, t)/a(x, t), and the total protrusion energy can thus be written as an integral over the cell area A: HP = − A d2x a0 (x, t) σ(x, t) , (1) where σ(x, t) = a0/a(x, t) represents the local compres- sion or dilatation with respect to the area of an unde- formed hexagon a0. Hence HP favors high protrusion energy density (x, t) σ(x, t). When a cell attempts to protrude/retract in the direc- tion ±d [Fig. 1], the forces F facilitating this effort are balanced (on the scale of the grid sites) by traction forces T. For instance, during a protrusion, the actin cytoskele- ton exerts a pushing force FP (xi, t) over the distance d(xi, t), which is determined by a change in polariza- tion energy accounted for by Eq. (1). This pushing force is transmitted to the substrate by focal adhesions and balanced locally by a traction force TP =−FP , which is directed towards the cell interior: (cid:90) TP (xi, t) = −∆HP (xi, t) d(xi, t)2 d(xi, t) . Similarly, the change in cell morphological energy asso- ciated with a protrusion or retraction can be related to an effective contractile force on the cell membrane: FM (xi, t) = −∆HM (xi, t) d(xi, t)2 d(xi, t) . We assume that the cytoskeleton facilitates this con- tractility by transmitting forces instantaneously through- out the cell [59]. Then, the contractile force FM is (2) (3) (a)0.2kPa(b)2kPa(c)34kPa(d)0α1(e)0v75(f)0.5nN/µm(g)1nN/µm(h)8.75nN/µm(i)0α1(j)0v250distributed homogeneously over all hexagons j ∈ C occupied by the cell and balanced by traction forces TM (xj) =−FM (xi)/C. In the course of spreading and migration, the cell ex- erts the traction forces T(xi) = TP (xi) + TM (xi) on the nodes xi of the substrate, which is described as a discrete network of beads [Fig. 1] subject to viscous damping with viscous friction coefficient ζ [5], and connected by loaded springs with spring coefficient k; for details see S.M. [58]. Force balance then determines the overdamped dynamics for each node: ζ xi(t) = T(xi) + k(cid:80) j∈Ni (xj − xi). Parameter estimation. To compare the experimen- tal results with our computational model, we chose the model parameters to ensure physiological values for the cell speed v, spreading area A and traction forces on the substrate [58]. We determined the range of studied spring coefficients k to match the elastic properties of the sub- strate. Specifically, a spring coefficient of k = 0.5 nN/µm corresponds to a substrate modulus of E ≈ 0.6 kPa [58]. For the choice of the friction coefficient ζ we distinguish between two representative cases, depending on the rela- tive timescales for relaxation of the visco-elastic network (τR = ζ/k) and cell migration (τC). Since a lower bound for τC is given by the inverse update rate of internal cell polarization, τC ≥ 1/g, we expect viscous friction effects to become significant at ζ (cid:63) ≈ 35 s nN/µm. This motivates our choice of the representative values ζ = 121 s nN/µm and ζ = 17.5 s nN/µm for what we call high and low sub- strate viscosity, respectively, in the following. A table of the parameter values of the CPM is given in the S.M. [58]. Low viscous friction. Because PA gels are dominated by their elastic and not their viscous properties [19], we used a low value ζ = 17.5 s nN/µm, below the threshold value ζ (cid:63), and find the same phenomenology as in our ex- periments. Not only does our model capture the distinct morphologies at different substrate stiffness [Fig. 2 (f)- (j)], it also accounts for the onset of motility, i.e. the transition towards the running state beyond a thresh- old in substrate stiffness [Fig. 3(a)]. Indeed, our simu- lations are consistent with experiments which show that the cell's speed increases with substrate stiffness [inset of Fig. 3(a)]. Previous experiments [20] show a mono- tonic increase in cell elongation for high substrate rigid- ity (>1.5 kPa). We observe the same monotonic trend in our experiments with HUVECs plated on PA gels of com- parable rigidity (>1 kPa). Interestingly, extending these measurements to low substrate rigidity (less than 1 kPa), we observe pronounced cell elongation [inset of Fig. 3(b)]. This biphasic behavior is fully in accordance with our computer simulations without any further adjustment of parameters [Fig. 3(b)]. Moreover, the computational model predicts that the persistence time τ , as determined from fitting a persistent random walk to the cell trajec- tories, increases with substrate stiffness [Fig. 3(c)], in full agreement with previous experimental results [14, 15, 20]. High viscous friction. We then looked at the effects of substrate viscosity on the migratory behavior of cells. On raising the viscous friction coefficient ζ above the 3 FIG. 3. Characterization of cell migration and morphology. For a reference, horizontal dashed lines indicate the cell be- havior for rigid substrates. All lines are guides to the eye. (a) For ζ = 121 s nN/µm (high ζ, filled black circles), the lo- cal cell speed v decreases with increasing substrate stiffness k. In contrast, and in accordance with experimental results for HUVEC cells on PA gels (Inset), v increases with stiffness for ζ = 17.5 s nN/µm (low ζ). There are three distinct mi- gratory cell states: elongation (red triangles), rounding (blue squares), and running (green circles). (b) For low ζ, both experiment (Inset) and the theoretical model show biphasic behavior in the cell extension α with pronounced elongation for low k. (c) The persistence time τ of directed cell migra- tion, as obtained from fitting a persistent random walk to the cell trajectories, shows a positive correlation with k. (d) Cell polarization p as a function of substrate stiffness for low and high substrate viscosity. Inset: Correlation plot of persistence time τ versus cell polarization p. threshold value ζ (cid:63), we find a considerable change in phe- nomenology [Fig. 3(a)-(c)]: Cells now only exhibit run- ning states, with cell speed decreasing, and both persis- tence time and elongation monotonically increasing with substrate stiffness. Qualitatively, these trends are similar to measurements of fibroblast motility on polyethylene glycol-based hydrogels [17] and on PA gels [18]. Cell phenotypes and substrate properties. How can one rationalize the dependence of the observed cellular phe- notypes (morphology and motility) on substrate stiffness and viscosity [Fig. 3] in terms of the interplay between substrate dynamics and cell polarization? Since by con- struction of the generalized CPM, a cell has highest prob- ability to migrate in the direction of maximal protrusion energy density σ , we analyzed correlations between this quantity and the persistence time. We define the strength of cell polarization as p = 1 0 dθ cos θ b(θ, θ) σb(θ, θ) π with θ being the angle relative to the average polariza- (cid:82) π 2468stiffnessk[nN/µm]0510kPa102030406080speedv[nm/s]highζlowζrunningroundingelongation(a)2468stiffnessk[nN/µm]0510kPa0.60.70.8extensionα0.40.60.8highζlowζelongationroundingrunning(b)2468stiffnessk[nN/µm]0246persistenceτ[103s]highζlowζ(running)(c)2468stiffnessk[nN/µm]012p[102pNnm]012p[102pNnm]0246τ[103s](d)4 dividual protrusion and retraction rates is small, lead- ing to a broad velocity distribution with respect to the (transient) axis of polarization and thereby to a low lo- cal cell speed. Conversely, for high values of substrate stiffness, substrate deformations are small and cells can polarize strongly [Fig. S8b]. Due to this strong polar- ization, cells migrate persistently [Fig. 3(c)], and also at relatively high speeds [Fig. 3(a)], as cell velocities are nar- rowly distributed with respect to the cell's polarization axis. For high ζ, the response of the substrate is slow com- pared to the intracellular dynamics. Thus, to a first ap- proximation, a cell behaves as if it were migrating on a completely rigid substrate, and hence can easily polarize even for low substrate stiffnesses [Fig. S8(c)]. The slow response of the substrate to cellular forces leads to a trail of increased substrate density σ behind the cell, and to a decrease in substrate density at the sides of the cell. This leads to a lensing effect, which decreases the probability that the cell will deviate from a straight path. This also explains why cell speed is enhanced at low substrate stiff- ness. Moreover, the particular substrate density profile effectively reduces the cell polarization strength p and with it the persistence time. With increasing substrate stiffness all of these effects are attenuated as substrate de- formations become smaller. As a consequence, cell speed decreases and persistence time increases, asymptotically approaching the corresponding values for low viscous fric- tion of the substrate. Conclusion. Though we cannot exclude gene regula- tion as a possible cause for distinct cellular responses to substrate stiffness and viscosity, our study shows that variability in cell behaviors can also be explained sim- ply in terms of the physical properties of the substrate and its interplay with cell polarization. This has po- tentially far-reaching consequences, as the mechanics of the physiological environment of cells varies depending on the tissue they are embedded in -- and this not only determines cell migration [4] but also stem cell differen- tiation and fate [61, 62]. Based on our results, one may speculate that the typical shape of cells (e.g. elongated 'neurons' at low stiffnesses, round 'adipocytes' at inter- mediate stiffnesses, 'keratocytes' at high stiffnesses) is not only pre-determined by gene regulation, but strongly affected by mechanical cross-talk with the extracellular matrix. ACKNOWLEDGMENTS E.F. and C.B. acknowledge support by the German Excellence Initiative via the program 'NanoSystems Ini- tiative Munich' (NIM) and by the Deutsche Forschungs- gemeinschaft (DFG) via Collaborative Research Center (SFB) 1032 (projects B02 and B12). A.G. and D.B.B. are supported by a DFG fellowship through the Gradu- ate School of Quantitative Biosciences Munich (QBM). J.P.S. is the Weston Visiting Professor at the Weizmann FIG. 4. Cell polarization and substrate deformation. Mir- rored halves of a cell, with the local protrusion energy per hexagon (x, y) [102 pN nm] shown in the top halves and the substrate density (number of hexagons per unit area) σ(x,−y) shown in the bottom halves. These quantities are obtained by collapsing the data for many cells in their center of mass frame with the polarization axis oriented along the x-axis (θ = 0). Note the differences in the scales of the substrate density σ. tion axis θ, and b(θ, θ) and σb(θ, θ) the protrusion energy and substrate density for hexagons along the in- terior boundary of a cell, respectively. For high ζ, a cell's persistence time τ remains finite even for very soft substrates [Fig. 3(c)]. In contrast, for low ζ, there is a threshold value k(cid:63) ≈ 1.58 nN/µm below which cells lose their persistence (τ = 0) and become self- trapped. In this state, cells still show a finite average polarization [Fig. 3(d)], but repolarize frequently, indi- cating that a threshold polarization strength is needed to sustain persistent cell migration against the substrate strain that tends to pull the cell back. Interestingly, we find that, regardless of the substrate properties, there is a universal increase of a cell's persistence time τ with cell polarization p, identifying it as the main determinant of the migratory persistence [inset of Fig. 3(d)]. Finally, we would like to illustrate how the interplay between cell polarization and substrate deformation leads to the different migratory states of a cell [Fig. S8]. Our simulations show that for low ζ and low substrate stiff- ness, cellular protrusion forces induce a strong compres- sion of the substrate beneath the cell [Fig. S8(a)]. For a cell to move, it needs to protrude on one side and retract on the opposite side. However, because substrate den- sity is strongly increased below the cell, all retractions are energetically penalized. Even in the event that a cell should manage to move, it would be energetically advan- tageous to simply move back to its previous position due to the local strain gradient (self-haptotaxis). Due to the feedback between internal cell polarization and cell pro- trusions or retractions, inhibiting retractions effectively hampers cell polarization. As a consequence, a cell stops performing a persistent random walk and becomes self- trapped on substrates with low stiffness. Moreover, as the cell is only transiently polarized, the bias in the in- k=1nN/µmk=8.75nN/µmlowζ17.5snN/µm(a)11.21.4σ5678915µm(b).981.02σ56789highζ121snN/µm(c).981.02σ56789compress.dilatation(d).981.02σ56789Institute of Science and member of the cluster of excel- lence CellNetworks at Heidelberg University. J.P.S. and A.H. acknowledge support from the Max Planck Society. Parts of this work was performed at the Aspen Center for Physics, which is supported by National Science Founda- tion grant PHY-1607611. 5 [1] G. Charras and E. Sahai, Nat. Rev. Mol. Cell Biol. 15, Sci. U.S.A. 109, 6851 (2012). 813 (2014). [25] F. Ziebert and I. S. Aranson, PLoS ONE 8, e64511 [2] C. De Pascalis and S. Etienne-Manneville, Mol. Biol. Cell (2013). 28, 1833 (2017). [26] B. A. Camley, Y. Zhao, B. Li, H. Levine, and W.-J. [3] V. Hakim and P. Silberzan, Rep. Prog. Phys. 80, 076601 Rappel, Phys. Rev. Lett. 111, 158102 (2013). (2017). [4] S. van Helvert, C. Storm, and P. Friedl, Nat. Cell Biol. 20, 8 (2018). [27] B. A. Camley, Y. Zhang, Y. Zhao, B. Li, E. Ben-Jacob, and W.-J. Rappel, Proc. Natl. Acad. Sci. H. Levine, U.S.A. 111, 14770 (2014). [18] C.-M. Lo, H.-B. Wang, M. Dembo, and Y.-L. Wang, [28] B. A. Camley and W.-J. Rappel, Phys. Rev. E 89, 062705 Biophys. J. 79, 144 (2000). (2014). [6] B. C. Isenberg, P. A. DiMilla, M. Walker, S. Kim, and [29] J. Lober, F. Ziebert, and I. S. Aranson, Sci. Rep. 5, 9172 J. Y. Wong, Biophys. J. 97, 1313 (2009). (2015). [7] A. Zemel, F. Rehfeldt, A. E. X. Brown, D. E. Discher, and S. A. Safran, J. Phys. Condens. Matter 22, 194110 (2010). [12] F. Graner and J. A. Glazier, Phys. Rev. Lett. 69, 2013 (1992). [31] P. J. Albert and U. S. Schwarz, Biophys. J. 106, 2340 [8] A. Zemel, F. Rehfeldt, A. E. X. Brown, D. E. Discher, (2014). and S. A. Safran, Nat. Phys. 6, 468 (2010). [9] L. G. Vincent, Y. S. Choi, B. Alonso-Latorre, J. C. del ´Alamo, and A. J. Engler, Biotechnol. J. 8, 472 (2013). [10] C. D. Hartman, B. C. Isenberg, S. G. Chua, and J. Y. Wong, Proc. Natl. Acad. Sci. U.S.A. 113, 11190 (2016). [11] R. Sunyer, V. Conte, J. Escribano, A. Elosegui-Artola, A. Labernadie, L. Valon, D. Navajas, J. M. Garc´ıa- Aznar, J. J. Munoz, P. Roca-Cusachs, and X. Trepat, Science 353, 1157 (2016). [12] W. J. Hadden, J. L. Young, A. W. Holle, M. L. McFetridge, D. Y. Kim, P. Wijesinghe, H. Taylor-Weiner, J. H. Wen, A. R. Lee, K. Bieback, B.-N. Vo, D. D. Samp- son, B. F. Kennedy, J. P. Spatz, A. J. Engler, and Y. S. Choi, Proc. Natl. Acad. Sci. U.S.A. 114, 5647 (2017). [13] D. Lachowski, E. Cortes, D. Pink, A. Chronopoulos, S. A. Karim, J. P. Morton, and A. E. del R´ıo Hern´andez, Sci. Rep. 7, 2506 (2017). [14] T. A. Ulrich, E. M. de Juan Pardo, and S. Kumar, Can- cer Res. 69, 4167 (2009). [15] T. J. Grundy, E. De Leon, K. R. Griffin, B. W. Stringer, and G. M. B. W. Day, B. Fabry, J. Cooper-White, O'Neill, Sci. Rep. 6, 23353 (2016). [20] M. Riaz, M. Versaevel, D. Mohammed, K. Glinel, and S. Gabriele, Sci. Rep. 6, 34141 (2016). [17] D. Missirlis and J. P. Spatz, Biomacromolecules 15, 195 (2014). [18] R. J. Pelham and Y.-L. Wang, Proc. Natl. Acad. Sci. U.S.A. 94, 13661 (1997). [19] D. Calvet, J. Y. Wong, and S. Giasson, Macromolecules 37, 7762 (2004). [32] R. F. M. van Oers, E. G. Rens, D. J. LaValley, C. A. Reinhart-King, and R. M. H. Merks, PLOS Comput. Biol. 10, e1003774 (2014). [7] F. Thuroff, M. Reiter, A. Goychuk, and E. Frey, "Bridg- ing the gap between single cell migration and collective dynamics," Unpublished. [6] F. J. Segerer, F. Thuroff, A. Piera Alberola, E. Frey, and J. O. Radler, Phys. Rev. Lett. 114, 228102 (2015). [35] P. J. Albert and U. S. Schwarz, PLOS Comput. Biol. 12, e1004863 (2016). [36] E. G. Rens and R. M. H. Merks, Biophys. J. 112, 755 (2017). [37] T. Vicsek, A. Czir´ok, E. Ben-Jacob, I. Cohen, and O. Shochet, Phys. Rev. Lett. 75, 1226 (1995). [38] M. H. Zaman, R. D. Kamm, P. Matsudaira, and D. A. Lauffenburger, Biophys. J. 89, 1389 (2005). [39] N. Sep´ulveda, L. Petitjean, O. Cochet, E. Grasland- Mongrain, P. Silberzan, and V. Hakim, PLOS Comput. Biol. 9, e1002944 (2013). [40] E. L. Barnhart, K.-C. Lee, K. Keren, A. Mogilner, and J. A. Theriot, PLOS Biol. 9, e1001059 (2011). [41] S. Garcia, E. Hannezo, J. Elgeti, J.-F. Joanny, P. Sil- berzan, and N. S. Gov, Proc. Natl. Acad. Sci. U.S.A. 112, 15314 (2015). [42] O. Chepizhko, C. Giampietro, E. Mastrapasqua, M. Nourazar, M. Ascagni, M. Sugni, U. Fascio, L. Leg- gio, C. Malinverno, G. Scita, S. Santucci, M. J. Alava, S. Zapperi, and C. A. M. La Porta, Proc. Natl. Acad. Sci. U.S.A. 113, 11408 (2016). [43] E. A. Novikova, M. Raab, D. E. Discher, and C. Storm, [20] G. P. Raeber, M. P. Lutolf, and J. A. Hubbell, Biophys. Phys. Rev. Lett. 118, 078103 (2017). J. 89, 1374 (2005). [44] S. K. Schnyder, J. J. Molina, Y. Tanaka, and R. Ya- [21] M. Murrell, R. Kamm, and P. Matsudaira, Biophys. J. mamoto, Sci. Rep. 7, 5163 (2017). 101, 297 (2011). [22] O. Chaudhuri, L. Gu, M. Darnell, D. Klumpers, S. A. Bencherif, J. C. Weaver, N. Huebsch, and D. J. Mooney, Nat. Commun. 6, 6364 (2015). [23] F. Ziebert, S. Swaminathan, and I. S. Aranson, J. Royal Soc. Interface 9, 1084 (2012). [45] M. Dietrich, H. Le Roy, D. B. Bruckner, H. Engelke, R. Zantl, J. O. Radler, and C. P. Broedersz, Soft Matter 14, 2816 (2018). [46] G. F. Oster, J. D. Murray, and A. K. Harris, J. Embryol. Exp. Morphol. 78, 83 (1983). [47] J. D. Murray, G. F. Oster, and A. K. Harris, J. Math. [24] D. Shao, H. Levine, and W.-J. Rappel, Proc. Natl. Acad. Biol. 17, 125 (1983). 6 [48] A. Besser and U. S. Schwarz, Biophys. J. 99, L10 (2010). [49] B. Harland, S. Walcott, and S. X. Sun, Phys. Biol. 8, 015011 (2011). [50] B. M. Friedrich and S. A. Safran, Soft Matter 8, 3223 (2012). [51] A. Elosegui-Artola, E. Bazelli`eres, M. D. Allen, I. An- dreu, R. Oria, R. Sunyer, J. J. Gomm, J. F. Marshall, J. L. Jones, X. Trepat, and P. Roca-Cusachs, Nat. Mat. 13, 631 (2014). [52] A. Elosegui-Artola, R. Oria, Y. Chen, A. Kosmalska, C. P´erez-Gonz´alez, N. Castro, C. Zhu, X. Trepat, and P. Roca-Cusachs, Nat. Cell Biol. 18, 540 (2016). [53] M. Bennett, M. Cantini, J. Reboud, J. M. Cooper, P. Roca-Cusachs, and M. Salmeron-Sanchez, Proc. Natl. Acad. Sci. U.S.A. 115, 1192 (2018). [13] T. D. Pollard and G. G. Borisy, Cell 112, 453 (2003). [14] A. Mogilner and K. Keren, Curr. Biology 19, R762 (2009). [15] A. F. M. Mar´ee, A. Jilkine, A. Dawes, V. A. Grieneisen, and L. Edelstein-Keshet, Bull. Math. Biol. 68, 1169 (2006). [16] A. F. M. Mar´ee, V. A. Grieneisen, and L. Edelstein- Keshet, PLOS Comput. Biol. 8, e1002402 (2012). [58] See Supplemental Material at [URL will be inserted by publisher] for further details on the model as well as ad- ditional data. [59] N. Wang, J. D. Tytell, and D. E. Ingber, Nat. Rev. Mol. Cell Biol. 10, 75 (2009). [5] M. G. Yucht, M. Sheinman, and C. P. Broedersz, Soft Matter 9, 7000 (2013). [61] A. J. Engler, S. Sen, H. L. Sweeney, and D. E. Discher, Cell 126, 677 (2006). [62] M. Akhmanova, E. Osidak, S. Domogatsky, S. Rodin, and A. Domogatskaya, Stem Cells Int. 2015, 167025 (2015). MORPHOLOGY AND MOTILITY OF CELLS ON SOFT SUBSTRATES -- SUPPORTING INFORMATION CONTENTS A. Numerical methods 1. Mathematical description of the substrate 2. Mathematical description of the cell 3. Observable definitions 4. Principal component analysis of the cell 5. Cell persistence measurement 6. Substrate model 7. Cell model a. Metropolis algorithm b. Contractility of the cell membrane and cortex c. Actin network of the cell d. Mechanochemical positive feedback 8. Simulation parameters B. Supplemental discussion 1. Measuring the persistence time of directed migration of the cell a. Mean-square displacement b. Normalized velocity auto-correlation c. Reorientation of the protrusion energy profile d. Reorientation of the cell body e. Robust measurement of cell orientation 2. Cell trapping 3. Cell migration on a deformable substrate a. Cell migration on a stiff substrate b. Can the cell shape serve as memory? c. How are substrate deformations induced? d. What determines cell speed? e. What determines the persistence time of directed migration of the cell? f. Low viscous friction g. High viscous friction 4. Modulation of cell morphology by substrate interactions C. Experimental methods 1. Preparation of polyacrylamide substrate 2. Cell culture 3. Microscopy 4. Cell shape analysis 5. Cell speed analysis References 2 2 3 3 5 5 5 5 6 6 7 7 8 9 9 9 9 9 11 11 11 13 13 14 14 14 15 16 16 17 18 18 18 18 18 18 19 2 Appendix A: Numerical methods In the following sections, we describe the numerical methods employed in this paper and the detailed imple- mentation of the model. We first provide mathematical definitions for the substrate and the cell, recapitulate the Cellular Potts model and introduce our proposed exten- sion to take substrate strains into account. Furthermore, we give give concise definitions for our observables and an overview over all model parameters and their values. 1. Mathematical description of the substrate The substrate (or grid ) is represented by a space-filling triangular lattice with time-dependent lattice vectors (or nodes) {xi(t)}i=1,...,N . This results in a hexagonal tesse- lation [Fig. S1] of the substrate consisting of N hexagons with indices i ∈ (1 . . . N ). Each hexagonal tile i is sur- rounded by six nearest neighbors that define the neigh- borhood N i: (cid:110) j(cid:12)(cid:12) xj(0) is nearest neighbor of xi(0) (cid:111) (S1) N i = Strains in the substrate are modeled by deviations of the lattice vectors xi(t) from the unstrained state. In this unstrained state (at t = 0 or for an infinitely stiff and undeformable substrate k →∞), nearest neighbors have a fixed distance from each other: xj(0) − xi(0) = d0 ⇐⇒ j ∈ N i . (S2) Furthermore, we impose a (clockwise) cyclic order on the set N i with respect to the center tile i: FIG. S1. Sketch of the cell and substrate morphology. The substrate consists of hexagons with indices i at positions xi. The vertices of the hexagons vi(k) are obtained by in- terpolation, and the areas a(xi, t) by the shoelace formula. The cell bulk is given by a morphologically connected set of hexagons i ∈ C. The cell membrane B is the set of membrane segments ei(k) lining the border of the cell bulk. Here the hexagons i, j and l are pairwise nearest k+4 and k+1, j =N i k−1 =N l k =N j neighbors such that l =N i i =N l k+3 =N j k+2. We assume that the hexagons are at all times (in- cluding strained states of the substrate) simple poly- gons. The area a of a hexagon spanned by the vertices vi k = (X i k ) is then given by Gauss area formula k, Y i (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 6(cid:88) k=1 (cid:0)Y i (cid:1)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . a(xi, t) = 1 2 X i k k+1 − Y i k−1 (S4) We complete the morphological description of the sub- strate by defining the six edges ei k of a hexagon i as Note that we do not perform a Voronoi-Tesselation here, because our current implementation of the CPM requires each tile to strictly have six neighbors and thus have a hexagonal shape. The shape of a hexagon i is defined by its six vertices vi k, which are obtained by in- terpolating between the positions of hexagon i and two mutually connected nearest neighbors: (cid:105) (cid:104) vi k = 1 3 xi + xN i k + xN i k−1 . (S3) ei k = (vi k, vi k+1) , (S5) This ensures a circular order in the set of vertices of a hexagon V = {vi k} and can be graphically represented as follows: k =vi with lengths ei k can also be understood as the border between the hexagons i and N i k. k+1. Thus, the edge ei k − vi iNi1Ni2Ni3Ni4Ni5Ni6anodeiatxinodeNi5vertexvi1membraneBedgeei1cellCxixjxlvik2. Mathematical description of the cell The bulk of the cell is described by a set C of simply connected hexagons [Fig. S1]: (cid:110) i(cid:12)(cid:12) i is occupied by cell (cid:111) The cell membrane is the set of hexagon edges ei the border of the cell bulk C [Fig. S1]: k lining (cid:41) (cid:40) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) B = ei k i ∈ C , N i k /∈ C . (S7) . (S6) 3. Observable definitions C = Observable Cell area Cell perimeter In this section we summarise the definition of all observables in Tables. TABLE S.I: Cell shape descriptors Description and remarks The area of the cell is the sum of the areas of all hexagons occupied by the cell. Definition The perimeter of the cell is the sum of the lengths of all edges lining the boundary of the cell. a(xi, t) A =(cid:88)i∈C P = (cid:88)ei k∈B ei k Cell extension The cell shape factor ranges from 0 (circular cells) to 1 (infinitely elongated cells). α = 1 − 4πA P 2 TABLE S.II: Cell position and orientation descriptors Observable Cell coordinates (center of mass) Description and remarks The center of mass of the cell body is deter- mined under the assumption that each hexagon has the same mass density. Definition 3 (S8) (S9) (S10) (S11) (S12) (S13) (S14) Cell coordinates (center of protrusion energy) Cell velocity Instantaneous cell polar- ization vector Average cell polarization vector Principal axes The center of protrusion energy of the cell body is similar to the center of mass. However, here each hexagon is weighted with its respective protrusion energy. The cell velocity is obtained from the differ- ence in the center of mass coordinates after ∆t = 1 MCS. The overall direction of the instantaneous cell polarization always points in the direction of the leading edge of the cell. The superscript (cid:126) indicates the usage of the non-averaged (in- stantaneous) polarization vector The overall average direction of the cell polar- ization always points in the direction of the leading edge of the cell. Compared to the in- stantaneous cell polarization vector n(cid:126) , it ex- hibits less fluctuations. The vectors n± corresponding to the two prin- cipal axes of the cell are the eigenvectors of the cell shape covariance matrix Cov(C); see below for a detailed description. xC = (cid:80)i∈C (cid:80)i∈C x = (cid:80)i∈C (cid:80)i∈C a(xi) xi a(xi) (xi) xi (xi) v(t) = xC (t + ∆t) − xC (t) ∆t n(cid:126) = (n(cid:126) , θ(cid:126) ) = x − xC n(t) = (n, θ) = n(cid:126) (t + t(cid:48)) (S15) 1 50 49(cid:88)t(cid:48)=0 Cov(C) n± = λ± n± (S16) Observable MSD Description and remarks mean-square Displacement of the cell. Definition TABLE S.III: Cell trajectory descriptors VACF PACF SAACF Normalized Velocity Auto-Correlation Func- tion of the cell. Normalized Correlation Function of the cell. Polarization Vector Auto- Normalized Short Axis Auto-Correlation Func- tion of the cell. (cid:10)R(t)2(cid:11) =(cid:10)x(t0 + t) − x(t0)2(cid:11)t0 v(t0 + t)v(t0)(cid:29)t0 CV(t) =(cid:28) v(t0 + t) v(t0) CP(t) =(cid:28) n(cid:126) (t0)(cid:29)t0 (t0 + t) n(cid:126) n(cid:126) (t0 + t)n(cid:126) (t0) CSA(t) =(cid:28) n−(t0 + t) n−(t0) n−(t0 + t)n−(t0)(cid:29)t0 TABLE S.IV: Angular profiles in relative coordinates Substrate density (outer cell boundary) Observable Substrate density (inner cell boundary) Description and remarks Substrate density at the cell boundary and in- side of the cell, at the angle θ relative to the av- erage direction of cell polarization θ. The rela- tive coordinates are defined as x = (r, θ) = x − xC . Substrate density at the cell boundary and outside of the cell, at the angle θ relative to the average direction of cell polarization θ. The relative coordinates are defined as x = (r, θ) = x − xC . Cell protrusion energy at the cell boundary, at the angle θ relative to the average direction of cell polarization θ. The relative coordinates are defined as x = (r, θ) = x − xC . Cell polarization strength Measure for the strength of the cell polariza- tion, i.e. the distinctness of the cell's edhesion energy profile. Cell (cell boundary) protrusion energy Definition σb,I(θ) =(cid:28) a0 a(xi)(cid:29)ei k∈B, φi≈θ±θ σb,O(θ) =(cid:42) a0 a(xN i k )(cid:43)ei k∈B, φN i k ≈θ±θ (θ) =(cid:10)(xi)(cid:11)ei k∈B, φi≈θ±θ 4 (S17) (S18) (S19) (S20) (S21) (S22) (S23) p = 1 π(cid:90) π 0 dθ cos(θ) b(θ, θ) σb,I(θ, θ) (S24) TABLE S.V: Two-dimensional profiles in relative coordinates Observable Substrate density profile Cell occupation probabil- ity Description and remarks The spatial profile of the average substrate density around the average cell polarization axis is obtained by radial and angular bin- ning. The relative coordinates are defined as x = (r, θ) = x − xC . The probability of substrate occupation around the average cell polarization axis is obtained by radial and angular binning. Here, Θ is the Heaviside step function. The relative coordi- nates are defined as x = (r, θ) = x − xC . Definition σ(r, θ) =(cid:28) a0 a(xi)(cid:29)xi≈(r, θ±θ) (S25) Prob(r, θ) = (cid:80)xi≈(r, θ±θ) Θ((xi) − q) (cid:80)xi≈(r, θ±θ) 1 (S26) TABLE S.V: Two-dimensional profiles in relative coordinates Description and remarks Observable Protrusion energy profile The spatial profile of the average local pro- trusion energy around the average cell polar- ization axis is obtained by radial and angular binning. The relative coordinates are defined as x = (r, θ) = x − xC . The spatial profile of the average local protru- sion energy around the cell center and the av- erage cell polarization axis under the condition that the substrate is occupied. Protrusion energy profile (occupied) Definition (r, θ) =(cid:10)(xi)(cid:11)xi≈(r, θ±θ) 5 (S27) (r, θ) = (r, θ) Prob(r, θ) (S28) 4. Principal component analysis of the cell We perform a principal components analysis of the cell shape to obtain data on its orientation in the form of its long and short axes n± = (n± , θ±). Consider the covariance matrix of the cell, which is defined as Cov(C) = . (S29) With the coordinates of each substrate hexagon relative to the cell center xi = xi − xC = (xi , yi), the elements of the covariance matrix are given by (cid:18)AXX AXY AXY AY Y (cid:19) (cid:80) (cid:80) (cid:80) (cid:80) (cid:80) (cid:80) i∈C a(xi) xi xi i∈C a(xi) i∈C a(xi) xi yi i∈C a(xi) i∈C a(xi) yi yi i∈C a(xi) AXX = AXY = AY Y = , , . (S30) (S31) (S32) Then, the short or long axis of the cell is defined as the eigenvector n± of Cov(C) corresponding to the smaller or larger eigenvalue λ±, respectively: Cov(C) n± = λ± n± . (S33) Both eigenvectors are chosen such that they point in the direction of the polarization vector: n±· n > 0. Since we are only interested in the direction of the eigenvectors, no particular normalization is needed. 5. Cell persistence measurement The persistence time of directed migration τ referenced in the main text denotes a typical timescale on which the cell reorients its direction of migration. It is obtained by fitting a persistent random walk model to the mean- square displacement of the cell in the simulations: (cid:10)R(t)2(cid:11) = 2v2τ 2(cid:104) t/τ + e , (S34) (cid:105) −t/τ − 1 with two fit parameters: Simulations on a deformable substrate are fitted using the Interior v and τ . Point method [1, 2], while reference simulations on a rigid substrate are fitted using the Levenberg-Marquardt method [1, 3, 4]. 6. Substrate model As discussed in section A 1, the substrate is de- scribed by a triangular lattice with time-dependent nodes {xi(t)}i=1,...,N . These nodes are elastically coupled with their nearest neighbors by loaded springs of zero rest length and are furthermore subject to a viscous dampen- ing and a traction force T: ζ xi = T(xi, t) + k (xj − xi). (S35) (cid:88) j∈Ni By assuming the rest length of the springs to be zero, we enforce a strictly linear response of the substrate to stresses. For a different approach of linearizing the full equation of motion including a non-zero spring rest length, we refer the reader to [5]. We have checked that both approaches yield the same phenomenology. A sec- ond alternative approach would be to use a continuum elastic theory to compute substrate strains. Note that in the absence of traction forces T, the lattice returns to its 'rest state' (all neighbors i, j have the same distance from each other) due to periodic boundary conditions. To compute the time-dependent node positions, we use an Euler forward method. 7. Cell model For the sake of completeness, we briefly recapitulate the cell model ultilized in this study, which has been previously introduced in [6, 7]. Please refer to [6, 7] for a detailed discussion and biological motivation of the core model for cell polarity and migration. For a quick overview over a single Monte Carlo Step, we refer the reader to to Fig. S2. 6 FIG. S2. Overview of a single Monte Carlo Step. An attempted protrusion or retraction event is accompanied by prospective changes in protrusion HP and morphological energy HM . These energy changes can be related to effective protrusive FP and contractile FM forces (illustrated in red for several simultaneously attempted events). The acceptance probability of such an event is calculated from the total energy difference ∆H = ∆HP + ∆HM [See Eq. S40]. Successful protrusions are followed by a secretion of internal signals within a radius R. Similarly, retractions lead to depletion of the mentioned internal signals. Over the course of a Monte Carlo Step, many such signals accumulate. Then, positive signalling increases the effective local cell protrusion energy , while negative signalling decreases it. Assuming force balance, the protrusive FP and contractile forces FM can be related to effective traction forces T on the substrate, leading to deformation. a. Metropolis algorithm incorporated into the cell bulk A single Monte Carlo Step in our simulations consists of many individual protrusion or retraction events, where the cell attempts to change its configuration. By ap- propriately defining of the total number of protrusion and retraction attempts B, we make sure that dur- ing a Monte Carlo Step on average each membrane seg- ment will experience contractile [Sec. A 7 b] and protru- sive forces [Sec. A 7 c], and as a result attempt to protrude or retract. During a Monte Carlo Step, a random membrane seg- k ∈ B is selected with a probability proportional ment ei to its length: Prob(ei k) = ei k P . (S36) With equal probability, the cell attempts to either pro- trude or retract along the normal vector of the chosen membrane segment ei k. The effective distance vector of such an attempted protrusion is given by +d, while for an attempted retraction it is given by −d with d = xN i k − xi , (S37) where i denotes the hexagon inside of the cell that shares the edge ei k [See Sec. A 1]. If a protrusion was successful, the conquered hexagon N i k is k with its k-th neighbor N i E+(ei k) : C (cid:55)→ C ∪ N i k . (S38) Similarly, in the case of a retraction the hexagon i is removed from the cell bulk E−(ei k) : C (cid:55)→ C \ i . (S39) Each cell configuration is associated with a Hamiltonian H. Thus, changes in the configuration are reflected by the energy state of the cell. The probability for an event E±(ei k) to be successful is then determined by the energy difference ∆H between the initial and the attempted cell state p(∆H) = min(cid:0)e , 1(cid:1) . −β∆H (S40) The inverse effective temperature β is a measure for the fluctuations and activity of the cytoskeletal dynamics on a cellular scale. Thus, in general β does not corre- spond to the room temperature. The energy difference ∆H = ∆HM + ∆HP is determined from the Hamiltonian modelling the contractility of the cell membrane and its cortex (HM , see Sec. A 7 b), and the Hamiltonian mod- elling the protrusive actin network (HP , see Sec. A 7 c). b. Contractility of the cell membrane and cortex The geometry of the cell is constrained by its elas- tic membrane and the contractile cytoskeleton, which is ++++++++------+++++---------forceFPforceFMeventsignalling1MCSaccumulatesignalsupdatetractionTstrainnextMCSqprotrusionfieldQadhered to the substrate [8 -- 11]. Thus, it is reasonable to assume in a first approximation that -- similar as in the original CPM [12] -- deformations of a cell's mem- brane and cortex are constrained by the elastic energy HM = κAA(t)2+κP P (t)2 with κA and κP denoting the stiffnesses corresponding to the area A(t) and perimeter P (t) of the cell, respectively. A change in cell morphology is accompanied by a change in the morphological energy ∆HM . This can be related to an effective contractile force always pointing inwards of the cell and acting on the membrane at each attempt to protrude over an effec- tive distance +d or retract over an effective distance −d [Eq. S37]: FM (xi, t) = −∆HM (xi, t) d(xi, t)2 d(xi, t) . (S41) We assume that the cytoskeleton facilitating this contrac- tility transmits forces instantaneously throughout the cell. Then, the contractile force FM is distributed ho- mogeneously over all hexagons j ∈ C occupied by the cell and balanced by traction forces TM (xj) =−FM (xi)/C. This denotes the traction force contribution stemming from the contractile force on a membrane segment dur- ing a single protrusion or retraction event. Note that the contraction is isotropic throughout the cell and as a result the average contractile forces along the cell membrane B ={ei k} vanish (cid:104)FM (xi, t)(cid:105)ei k∈B, t = 0 . (S42) Contractile forces and thus also traction forces result- ing from cell contractility are distributed homogeneously over all hexagons occupied by the cell j ∈ C: (cid:104)TM (xj)(cid:105)t = (cid:104)TM (xj)(cid:105)j∈C, t . (S43) Hence, the average traction force contribution on occu- pied hexagons j ∈ C resulting from contractile forces is negligible (cid:104)TM (xj)(cid:105)t = −(cid:104)FM (xi, t)/C(cid:105)ei k∈B, t = 0 . (S44) c. Actin network of the cell The homogeneous contractile forces facilitated by the contractile cytoskeleton are counteracted by local and inhomogeneously distributed outwardly directed push- ing forces generated by cytoskeletal structures. These pushing cytoskeletal structures are locally anchored to the substrate at focal adhesion sites [13, 14]. Because of this anchoring, they will behave in an affine way to the substrate, and the local amount of cytoskeleton per hexagon will remain constant under substrate deforma- tions. Thus, we describe the local energetic contribution from this cellular activity with (cid:88) i∈C HP =− (xi, t) , (S45) 7 per the scalar field [6, 7]. protrusion hexagon with (xi, t)∈ [q, Q] The protrusion field is dy- namic, reflecting the response of cytoskeletal structures to external mechanical stimuli through feedback mecha- nisms involving regulatory cytoskeletal proteins [15, 16], as will be described in the next section. Note that this protrusion field could as well be interpreted as a local, inhomogeneously distributed adhesion energy to the substrate. In this picture, the adhesion energy per hexagon would also remain constant under substrate deformations, as adhesions sites per definition deform affinely with the substrate. If the cell acquires a new hexagon, the conquered hexagon (target) will have the same protrusion field as the hexagon pushing the membrane (conqueror), and the overall polarization energy increases by the local protru- sion field of the conquering hexagon. This can be in- terpreted as the pushing cytoskeletal structures moving into the acquired hexagon. Similarly, in the case of a retraction the overall polarization energy decreases by the local protrusion energy of the lost hexagon. These energy changes can be related to an effective outward pushing force locally exerted by the cytoskeleton on the cell membrane at each attempt to protrude over an ef- fective distance +d or retract over an effective distance −d: FP (xi, t) = ∆HP (xi, t) d(xi, t)2 d(xi, t) . (S46) This pushing force is transmitted to the substrate by fo- cal adhesions and balanced locally by a traction force TP = − FP , which points towards the cell interior. The total traction force stemming from a single pro- trusion or retraction event that is locally exerted on the substrate T = TM + TP consists of a contribution from the contractility of the cell and and a contribution from the protrusive cytoskeleton. d. Mechanochemical positive feedback Cell migration is assumed to be driven mainly by a positive feedback loop involving the actin cytoskeleton and some -- a priori unknown -- signalling molecule. The relative amount of signalling molecules is coarse grained into an integer field m(xi), which can also take negative values. Because the internal dynamics of the cell is as- sumed to be fast, the amount of signalling molecules is reset after each Monte-Carlo Step. Consider a successful protrusion event E+(ei k), with the position of the acquired hexagon given by y = xN i . Analogously to the protrusion energy, in the case of a successful protrusion the signalling field m(xi) of the hexagon facilitating the protrusion is copied unto the acquired hexagon m(y) (cid:55)→ m(xi). Then, signalling molecules are secreted and diffuse within a signalling ra- k TABLE S.VI. Simulation parameters 8 energy energy 500 pN µm 1000 pN µm Value(s) effective temperature 100 pN µm Cell protrusion (lower bound) protrusion (upper bound) area stiffness perimeter stiffness signalling radius of in- ternal cell dynamics update rate of inter- nal cell dynamics Substrate stiffness viscous friction 0.5 pN/µm3 0.75 pN/µm 7.07 µm 0.014 s−1 0.5 to 8.75 nN/µm 17.5 to 121 s nN/µm q Q κA κP R g k ζ umbilical vein endothelial cells have been measured to exert physiological traction stresses up to 600 Pa [17]. On average, similar traction stresses have been mea- sured for fibroblasts, though also reaching up to sev- eral kPa [18]. Recent measurements have found simi- lar values for the stresses exerted by MDA-MB-231 cells on their three-dimensional environment [19]. To obtain traction forces on the correct order of magnitude, we set q = 500 pN µm and Q = 1000 pN µm for the lower and up- per protrusion energy bounds, respectively. Assuming the substrate depth to be on the order of the lattice con- stant 1.41 µm, the protrusion energy bounds correspond to traction stresses ranging from 306.19 Pa to 612.37 Pa. Similarly, the studied substrate stiffness of 0.5 nN/µm to 8.75 nN/µm can be related to an effective elastic modulus ranging from 0.61 kPa to 10.72 kPa. We study the influence of the viscous friction of the substrate on the cell behavior within the range 17.5 s nN/µm to 121 s nN/µm. To obtain a cell size of approximately 430 µm2 [20], the area stiffness is chosen as κA = 0.5 pN/µm3. The low perimeter stiffness m = 0.75 pN/µm allows for significant membrane fluctuations. For the remaining parameters, we set the signalling radius to R = 7.07 µm and the po- larization update rate g = g ∆t = 0.1. To achieve a cell speed of approximately 0.1 µm/s [20], the duration of a single Monte Carlo Step is set to ∆t = 7 s. dius R of the conquered hexagon: m(xj) (cid:55)→ m(xj) + 1, ∀j ∈ C : xj − y < R m(xj), else. (S47) Parameter Description β−1 (cid:40) (cid:40) Similarly, in the case of a retraction event E−(ei k) sig- nalling molecules are depleted within the signalling ra- dius R of the lost hexagon y = xi: m(xi) (cid:55)→ m(xi) − 1, ∀i ∈ C : xi − y < R m(xi), else. (S48) Because the lost hexagon removed from the cell, its corre- sponding signalling molecules are reset to zero m(y) (cid:55)→ 0. These protrusion or retraction events are driven by the actin cytoskeleton, which is modelled by the scalar protrusion energy . Protrusion events are more likely to occur in regions of high local protrusion energy , while retractions are more numerous in regions of low . Throughout a single Monte Carlo Step, many such protrusion and retraction events occur, and the corre- sponding signals overlap. Finally, at the end of a Monte Carlo Step with duration ∆t, the actin cytoskeleton is assumed to be reinforced in regions of high protrusive activity, and disassembled in regions of low protrusive activity with a rate g = g ∆t (xi, t) + g(Q − (xi, t)), m(xi) > 0, (xi, t) + g(q − (xi, t)), m(xi) < 0, (xi, t) + g(¯ − (xi, t)), else. (S49) Thus, a positive feedback loop is incorporated into the Cellular Potts model. Here, g is a measure for the speed of the cytoskeletal remodelling, and ¯ = (Q+q)/2. Before performing the actual simulations, we pre-equilibrate the cell by letting it grow on a non-deformable substrate for 1000 MCS with the positive feedback switched off and the protrusion field fixed at = ¯. There, the cell starts off as a single hexagon and grows until it reaches equilibrium. (xi, t + ∆t) = 8. Simulation parameters To allow for sufficient ruffling of the cell membrane, we choose the effective temperature β−1 = 100 pN µm, which corresponds to an effective temperature much larger than room temperature. A single cell is simulated over the course of 104 Monte Carlo Steps, each divided into 103 substrate update steps. We choose the initial distance between adjacent hexagons d0 = 1.41 µm (lattice con- stant). Hence, a cell spreading over an area of 400 µm2 consists of roughly 2.3 × 102 hexagons. The substrate is 283 µm wide and 245 µm high, with periodic boundary conditions. The lower and upper protrusion energy bounds repre- sent the ability of the cell to exert protrusive forces on the membrane and traction on the substrate. Human 9 Appendix B: Supplemental discussion In our model, the cell can exhibit different migratory states, depending on the mechanical properties of the substrate: running, rounding and elongation. Comple- mentary to the discussion in the main text, a more exten- sive explanation of the phenomenology is provided in the sections below. The additionally provided data serves to improve the intuition for the cell behavior across a wide range of parameters. 1. Measuring the persistence time of directed migration of the cell a. Mean-square displacement The persistence time of directed migration τ repre- sents the typical time over which the cell decorrelates (in other words reorients) its direction of motion. It has been shown previously that the migration of a polarized cell in the Cellular Potts model [7] can be approximated by a persistent random walk model with an exponen- tially decaying velocity auto-correlation function. Here we obtain the persistence time of directed migration τ by fitting the mean-square displacement with the corre- sponding expression for a persistent random walk model [Section A 5]:(cid:10)R(t)2(cid:11) = 2v2τ 2(cid:16) (cid:17) −t/τ − 1 t/τ + e . (S1) In this model, the cell migrates on an almost straight path on short timescales (t (cid:28) τ ) and performs a random walk on long timescales (t (cid:29) τ ). b. Normalized velocity auto-correlation In, we also investigated the normalized velocity auto- correlation function (VACF) CV. Empirically, we find that the behavior is well described by a bi-exponentional decay [Fig. S3(a)]: −t/τ −t/τ + − V + b e CV(t) = a e V + (1 − a − b) δt,0 , (S2) V > τ which is fitted to simulated data using the Interior Point − method and four fit parameters τ + V and b > a. We include the spike at t = 0 to capture the rapid decay of CV in the very first time step and to ensure that CV(0) = 1. We identify 1 − a − b as an effective 'noise strength' because the spike at t = 0 directly originates from the stochasticity of our simulations. It reflects a 'randomness' in the acceptance of protrusion and re- traction events: Though they are biased by the cell's non-uniform protrusion energy field , all protrusion and retraction events are essentially stochastic. In partic- ular, their randomness can be increased either (a) by decreasing the bias resulting from or alternatively (b) FIG. S3. Correlation functions and persistence times. (a) Exemplary correlation functions for a stiff substrate k = 8.75 nN/µm and low viscous friction ζ = 17.5 s nN/µm. The normalized velocity auto-correlation function (VACF) CV exhibits a bi-exponential decay with a long and a short timescale afterwards and a peak at t = 0. The peak at t = 0 can be attributed to the 'randomness' in the protrusion and retraction process (noise). In the normalized polarization vec- tor auto-correlation function (PACF) CP, noise is integrated out, and only the bi-exponential decay remains. The nor- malized short axis auto-correlation function (SAACF) CSA, which measures the actual reorientation of the cell body, shows a mono-exponential decay (all shorter timescales are integrated out). (b)-(d) The timescales obtained by fitting the mean-square displacement (MSD) to a persistent random walk model coincide with the (long) timescales of the VACF, PACF and the SAACF both for low (ζ = 17.5 s nN/µm) and for high (ζ = 121 s nN/µm) viscous friction coefficients. The error bars correspond to the estimated fitting errors. Thus, these are all equivalent measures for the persistence time of directed migration. by increasing the effective temperature β−1 [Eq. S40]. We observe that the long (dominant) timescale τ + V coin- cides with the persistence time of directed migration τ [Fig. S3(b)] and thus determines long-term cell behavior. c. Reorientation of the protrusion energy profile In addition to the mean-square displacement and the normalized velocity auto-correlation function we have also investigated the dynamics of the cell protrusion en- ergy profile and of cell repolarization. The current orien- tation of the protrusion energy profile at a given Monte Carlo Step (MCS) is captured by the polarization vec- tor n(cid:126) ) = x − xC pointing from the center = (n(cid:126) , θ(cid:126) 0123timet[103s]0.40.60.81.0auto-correlationCVCPCSA(a)0246τ[103s]0246τ+V[103s]highζlowζ(running)(b)0246τ[103s]0246τ+P[103s]highζlowζ(running)(c)0246τ[103s]0246τSA[103s]highζlowζ(running)(d)of mass xC towards the center of protrusion energy x of the cell [Section A 3]. Here and in later sections, the superscript (cid:126) indicates the usage of the non-averaged (in- stantaneous) polarization vector n(cid:126) ; for the definition of the averaged polarization vector n we refer the reader to Sec. A 3 and B 1 e. Because protrusions/retractions form preferably in regions of high/low respectively, the cell will on average migrate along the gradient of its protrusion energy field . Hence the polarization vec- tor pointing towards that side of the cell with a higher local protrusion energy will determine the cell's direc- tion of motion and its leading (protruding) edge. The change in the instantaneous cell polarization is captured by the normalized polarization vector auto-correlation function (PACF) CP. We observe that CP exhibits a bi-exponential decay [Fig. S3(a)]: CP(t) = a e −t/τ − P + (1 − a) e −t/τ + P , (S3) P > τ which is fitted using the Interior Point method with three − fit parameters τ + P and a < 0.5. The second pref- actor (1 − a) is determined naturally from the condi- tion CP(0) = 1. In contrast to CV, we do not observe a spike at t = 0, because the randomness of the protru- sion/retraction process is filtered out ('integrated out') by the internal polarization mechanism of the cell. Be- cause the direction of cell migration is slaved to the di- rection of instantaneous cell polarization, it is reason- able that both CP and CV show a similar time evolution and specifically the same decay rates at long timescales [Fig. S3(b),(c)]. FIG. S4. Polarization vector auto-correlation func- tions and their short timescale dynamics. (a) Semi- Log plot of exemplary polarization vector auto-correlation functions for different substrate properties as indicated by the graph color. At short times t < 0.4 × 103 s, all correlation functions decay identically. (b) The short timescale in the po- larization vector auto-correlation function (PACF) CP is on the order of 100 s for different substrate stiffnesses and viscous frictions. For low viscous friction (ζ = 17.5 s nN/µm), the hori- zontal and vertical dashed lines respectively indicate the lower bound of measured cell persistence times τ and the corre- sponding substrate stiffness k [Fig. 3(c)]. If the long timescale P ≈ τ and the short timescale τ−P are of the same order of τ + magnitude, the timescale separation in the bi-exponential fit fails. The error bars denote the estimated fitting errors. 10 We will now illustrate the origin of the short timescale observed in the normalized polarization vector auto- correlation function (PACF). Let us consider a scenario where the cell is polarized at a given time, that means the cell has a pronounced protrusion energy profile. Fur- thermore, we have argued before [Sec. B 1 b] that there is a certain 'randomness' in the protrusion and retraction processes. For now let's assume that this 'randomness' dominates and thus the bias of each individual protru- sion or retraction event by the local protrusion energy is negligible. Protrusions and retractions are then in good approximation equally likely everywhere at the cell edge, as would be the case in the limit of high effective tem- peratures. In such a scenario, all long-time correlations will be lost to the stochasticity of the cell. What is then the typical timescale on which a polarized cell will depo- larize and change the direction of its polarization vector in response to random protrusion and retraction events? Such random protrusion and retraction events are filtered by the internal cell dynamics, which is responsible for the formation and maintenance of the cell's protrusion energy profile. While a stable asymmetric protrusion en- ergy profile can in general not be maintained in the ab- sence of a bias in the protrusion/retraction process (as for e.g. high effective temperatures), it still allows for the formation of a highly volatile transient polarization pro- file due to the stochasticity of the system. This transient profile will then decay and its corresponding polarization vector will reorient with a typical timescale set by the in- − P ≈ 1/g = 70 s. ternal dynamics of the cell [Section A 8] τ However, in general the protrusion/retraction process is biased by the protrusion field , allowing stable polariza- tion profiles. According to our argumentation, the 'ran- domness' in the protrusion/retraction process [Sec. B 1 b] will then typically lead to a small decay and reorienta- tion of the cell's polarization profile on a short timescale set by the internal dynamics of the cell [Section A 8] − P ≈ 1/g = 70 s. This is in good agreement with the τ − P ≈ 100 s that was typically observed in short timescale τ our simulations [Fig. S4]. Thus we can conclude that the − short timescale τ P generally observed in our simulations originates from the stochastic behavior of the cell and the resulting decorrelation of the polarization vector at short timescales. Next let us turn to the origin of the long timescale observed in the normalized polarization vector auto- correlation function (PACF). For finite effective temper- atures cell protrusions and retractions are biased by the cell's protrusion energy field . The preference of the cell to protrude at its leading edge and to retract at its trailing edge leads to the reinforcement of the protrusion energy at the leading edge and its weakening at the trail- ing edge of the cell. This in turn sustains the protrusion energy profile over long periods of time and leads to the emergence of a long timescale τ + P . Note that with decreasing substrate stiffness, the per- sistence time of a cell and thus also the long timescale τ + P of its PACF decreases. Once the long timescale 0246timet[103s]1.8.6.4.2CP(t)highζ;k=0.50nN/µmhighζ;k=8.75nN/µmlowζ;k=4.00nN/µmlowζ;k=8.75nN/µm(a)2468stiffnessk[nN/µm]1234τ−P[102s]highζlowζ(running)lowboundτ(b)becomes small enough to be of the same order as the short timescale, the timescale separation in the bi- exponential fit fails [Fig. S4(b): for low viscous friction ζ = 17.5 s nN/µm; the corresponding stiffness and persis- tence time are respectively indicated by the vertical and horizontal dashed lines]. This explains the large spread − P for low substrate stiffnesses and low viscous friction of τ [Fig. S4(b)]. d. Reorientation of the cell body In the absence of external guiding cues, there are dif- ferent ways in which a cell can orient itself relative to its direction of migration and vice versa: a symmetrical cell can preferably move along its short axis, along its long axis, or the cell has no particular shape (symmetry) at all and performs a random walk. In our model the short axis of the cell aligns with the direction of motion and the polarization axis. Because the internal cell dy- namics is much faster than the motion of the cell, the short-timescale decorrelation in the polarization vector auto-correlation function (which arises from random, un- correlated protrusions or retractions) does not matter for the long-term orientation of the cell and is integrated out. Thus, if one performs a principal components anal- ysis of the cell shape and considers the short axis auto- correlation function (SAACF) CSA, it exhibits a mono- exponential decay [Fig. S3(a)]: CSA(t) = e −t/τSA , (S4) which we fit using the Interior Point method. In accor- dance with our arguments, this timescale τSA coincides with the dominant timescales of the velocity and polar- ization vector auto-correlation functions, as well as the persistence time τ [Fig. S3]. We have thus shown that the long-term behavior of the cell (e.g. its persistence time of directed migration) can be determined either from the mean-square displacement of the cell, or the velocity, polarization vector or short axis auto-correlation func- tions [Secs. B 1 a, B 1 b, B 1 c and B 1 d]. Each approach gives quantitatively identical results in a consistent way [Fig. S3(b)-(d)]. Note that there is a caveat for the measurement of the SAACF: the principal component analysis fails for round cells. For high viscous friction (ζ = 121 s nN/µm) cells are rounder than for low viscous friction (ζ = 17.5 s nN/µm) [Fig. 3] and migrating cells are both rounder and less per- sistent with decreasing substrate stiffness [Fig. 3]. This explains the deviation of τSA from τ for less persistent cells [Fig. S3(d)]. e. Robust measurement of cell orientation We have seen that the cell orientation can be captured by using a principal components analysis. However, this 11 approach fails for round cells because one can then obvi- ously not discern the long from the short axis of the cell. Hence, the long-term cell orientation (and thus direction of migration) can be robustly measured for all cell shapes only by either using the velocity or the polarization vec- tor, as we know that they both capture long-term cell behavior. However, the instantaneous polarization vec- tor exhibits short-timescale decorrelations on the order of 100 s which stem from the intrinsic noise of the Monte- Carlo simulation. Additionally, the velocity vector is not only slaved to those decorrelations, but furthermore also directly shows the mentioned intrinsic noise. We there- fore choose to utilize the polarization vector for measur- ing cell orientation, and average it over 50 Monte Carlo Steps (350 s): n(t) = (n, θ) = 1 (t + t(cid:48)). (cid:80)49 t(cid:48)=0 n(cid:126) 50 2. Cell trapping In this section, we will briefly discuss the conditions on the mechanical properties of its substrate for the cell to stop performing a persistent random walk and to become self-trapped. For quickly responding substrates, e.g. low substrate viscous friction, the normalized velocity auto- correlation function (VACF) oscillates if the stiffness falls below a threshold stiffness k(cid:63) = 1.58 nN/µm [Fig. S5(a)]. Analogously, the mean-square displacement also deviates from that of a persistent random walk model [Fig. S5(b)], and a typical cell persistence time cannot be determined anymore. We identify this behavior leading to a decrease in overall cell motility as cell trapping, or, as cell rounding because of the corresponding cell shape. To test our computational results, we have measured the cell trajectories from experiments on HUVECs plated on polyacrylamide gels [Sec. C]. We have already seen in the main text that the cell speed decreases with sub- strate stiffness in qualitative accordance with our model [Fig. 3(a)]. This is complemented by Fig. S6, where we can see from the cell trajectories that cell motility in- creases with substrate stiffness. Additionally, Fig. S5(c) shows anti-correlations in the VACF for E < E(cid:63) ≈ 7 kPa, indicating cell trapping. We have seen in the main text that for low stiffness and low viscous friction of the substrate, cells have marginal motility due to trapping, and that motility can be re- stored if the substrate stiffness is high enough [Fig. 3]. Can a similar effect be induced by tuning the viscous friction of the substrate? To answer this question, we have performed simulations at a fixed low substrate stiff- ness k = 0.5 nN/µm < k(cid:63), where cells are trapped and in the elongation state for low viscous friction ζ, and have varied ζ. Our measurements clearly indicate that with increasing viscous friction of the substrate an in- creasing amount of cells exhibit persistent cell migra- tion [Fig. S7(a),(b)]. Note that an individual trapped cell is characterized by an oscillating normalized veloc- ity auto-correlation function (VACF) and a saturating mean-square displacement (MSD), while a migrating cell 12 FIG. S6. Cell trajectories in the experiments. The substrate stiffness is indicated at the bottom of the corre- sponding frames. The amount of measured trajectories NT and the amount of measured data points NP is indicated by (NT NP ) at the top of the corresponding frames. The color code corresponds to the elapsed time in the respective trajec- tory (colorbar). The duration of the longest measured trajec- tory is 21.8 h. Note that the amount of measurements varies for different frames, influencing the visual perception to some extent. At low substrate stiffnesses, cells migrate less than at high stiffnesses. trapped and migrating cells. We expect that with in- creasing viscous friction of the substrate the contribu- tion of migrating cells to the VACF and to the MSD increases. Our expectation is confirmed by the increase in correlated cell movement [Fig. S7(c)] and by the MSD desaturation [Fig. S7(d)] for high viscous friction coeffi- cients ζ > ζ (cid:63) = 75 s nN/µm. Why can, for high substrate viscous friction, cell mi- gration occur even at low substrate stiffnesses? For high enough viscous friction, the response of the sub- strate (τR = ζ/k) is slow compared to the cell dynam- ics (τC ≥ 1/g), and the cell will approximately behave as on a locally non-deformable substrate. Then, the cell will polarize strongly even on low substrate stiff- nesses, before the positive feedback loop can be signif- icantly inhibited by substrate deformations. By compar- ing the corresponding timescales of substrate response and cell response (τR = ζ/k ≥ τC ≥ 1/g) we can make a simple estimation that viscous effects become dom- inant above a lower bound of viscous friction of at least ζ (cid:63) ≥ 35 s nN/µm. Indeed, it can be observed that ζ (cid:63) = 75 s nN/µm [Fig. S7]. This indicates a parameter regime where the cell is fast enough to 'outrun' substrate FIG. S5. Persistent motion and self-trapping. (a) Velocity auto-correlation functions CV in the simulations for different substrate stiffnesses, as indicated by the graph colors and the legend. For k < k(cid:63) = 1.58 nN/µm we observe oszilla- tions in CV. The frequency of these oscillations increases with k. Note that with increasing k cell migration becomes increas- ingly uncorrelated. (b) Mean-square displacements(cid:10)R2(cid:11) cor- responding to the velocity auto-correlation functions shown in (a). For k < k(cid:63) = 1.58 nN/µm we observe a saturation of the mean-square displacement. In the simulations, the oszilla- tions in CV and the saturation of the mean-square displace- ment strongly indicate cell trapping (cells change from the running to the rounding state). (c) Velocity auto-correlation functions CV of HUVECs plated on polyacrylamide (PA) gels for different substrate stiffnesses, as indicated by the graph colors and the legend. For E < E(cid:63) ≈ 7 kPa we observe anti- correlations in CV (CV < 0). Note: In the plot we excluded the data point at 0.2 kPa because there the cells dramati- cally changed their mode of migration and ceased the for- mation of lamellipodia. Specifically, for 0.2 kPa we observed that cells now migrated preferably along their long axis. (d) Mean-square displacements (cid:10)R2(cid:11) of HUVECs plated on PA gels corresponding to the velocity auto-correlation functions shown in (c). In our experiments, the anti-correlations in CV suggest cell trapping. is characterized by a bi-exponentially decaying VACF and no such saturation in the MSD. In particular, note that the VACF decays faster with decreasing substrate stiffness [Fig. S5]. Furthermore, we cannot assume that at ζ ≈ ζ (cid:63) all cells will be able to outrun the substrate deformations and elude trapping at all times; due to the stochastic nature of cell migration a cell might just reori- ent itself and become trapped. In general the normalized velocity auto-correlation function [Fig. S7(c)] and the mean-squared displacement [Fig. S7(d)] averaged over all measured cells will then be a weighted average of both 0123timet[103s]00.51CVsubstratestiffnessk1.33nN/µm1.42nN/µm1.50nN/µm1.58nN/µm1.67nN/µm1.75nN/µm(a)0123timet[103s]0510(cid:10)R2(cid:11)[103µm2](b)0246timet[103s]00.51CVsubstraterigidityE1kPa2kPa3kPa7kPa9kPa11kPa15kPa34kPa(c)0246timet[103s]01234(cid:10)R2(cid:11)[103µm2](d)0.2kPa(491370)1kPa(431079)2kPa(934675)3kPa(492214)7kPa(823570)9kPa(914120)11kPa(1115001)15kPa(883798)34kPa(723321)05101520timet[h]13 density [Fig. S8(e)-(h)]. Thus, we expect that even for high viscous friction, we will sporadically observe trapped cells if we wait long enough [Fig. S7(a),(b)]. 3. Cell migration on a deformable substrate In this section, we present a few exemplary cases for both low viscous friction (ζ = 17.5 s nN/µm) and high vis- cous friction (ζ = 121 s nN/µm) with the intent to give the reader a better intuitive understanding of the dynamics underlying both cell migration and substrate deforma- tion in our Cellular Potts model. Figure S8 shows the two-dimensional spatial profiles of protrusion energy and substrate density. The corresponding angular profiles of protrusion energy, substrate density and protrusion en- ergy density are shown in Fig. S9. a. Cell migration on a stiff substrate To obtain a fitting explanation for a new phenomenol- ogy, it is often fruitful to begin by considering a sim- plified and particular example, before advancing to the more general and complex case. Hence, we will first de- scribe cell migration for vanishingly small substrate de- formations in the limit of large substrate stiffness k →∞ or large viscous friction ζ →∞. Then, all hexagonal substrate tiles have the same size and shape, result- ing in a uniform substrate density σ ≡ 1. This line of argument can as a first approximation also be applied to simulations showing negligible substrate deformations [Fig. S8(d)-(h)]. The cell makes random protrusions or retractions at its boundary, as described in detail in section A 7. A high local protrusion energy increases the rate of making pro- trusions relative to the rate of making retractions, by in- creasing the energy gain/loss for protrusions/retractions, respectively. Although the direction of cell migration on average aligns with the cell's protrusion energy field, the core of the model is still stochastic. Because of this stochastic nature of the computational model, the out- come of an event is a priori unknown, and only the rela- tive probabilities of protrusions or retractions are biased by the local protrusion energy . This means that, though less likely, retractions can also occur at the leading edge, and protrusions can also occur at the trailing edge of the cell. Now, let us consider a scenario with a more pronounced protrusion energy profile, e.g. the gradient in protrusion energy is steeper than before. This will lead to an in- creased bias for protrusions to form at the leading edge and retractions at the trailing edge, and thereby to a better alignment of the cell's individual protrusions and retractions with its protrusion field. Hence, we can con- clude that the alignment of the cell velocity v = (v, θv) with its instantaneous polarization axis n(cid:126) , θ(cid:126) ) will also improve, and the overall effect of stochastic- = (n(cid:126) FIG. S7. Dependence of the cell behavior on substrate viscous friction. (a) Cell speed depends on the viscous friction ζ of the substrate at low stiffness k = 0.5 nN/µm < k(cid:63) [Fig. S5]. The color code represents the current elapsed time of a given data point in the simulation (color bar). The solid black line corresponds to the averaged behavior of the cells. For viscous friction coefficients ζ > ζ (cid:63) = 75 s nN/µm indicated by the vertical dashed line, we observe that cells can (at least transiently) migrate and elude trapping. Migrating cells then have a typical velocity v ≈ v∞, where the horizontal red line denotes the velocity v∞ of a cell on a non-deformable sub- strate. (b) Cell extension depends on the viscous friction ζ of the substrate at low stiffness. The color code represents the current elapsed time of a given data point in the simulation (color bar). The solid black line corresponds to the averaged behavior of the cells. For ζ > ζ (cid:63) indicated by the vertical dashed line we observe that cells can (at least transiently) drastically decrease their elongation. Together with (a) this suggests that the cells switch from the elongation to the run- ning state. (c) Velocity auto-correlation functions CV in the simulations for different substrate viscous friction ζ, as indi- cated by the graph colors and the legend. Note the dramatic decay of CV from CV = 1 to CV ≈ 0.05 in the first time step. For ζ > ζ (cid:63) = 75 s nN/µm, long range correlations in the ve- locity auto-correlation appear and the functional form of the mean-square displacement of the cell approaches that of a per- sistent random walk model. (d) Mean-square displacements (cid:10)R2(cid:11) corresponding to the velocity auto-correlation functions shown in (c). deformations. However, due to the stochastic nature of our simulations, it cannot be expected that the cell continues to do so indefinitely. Specifically, the cell can still become trapped if it runs into a localized region of strongly increased substrate density, e.g. by making a u-turn and running into its trail of increased substrate 5101520frictionζ[snN/µm]50100speedv[nm/s]v∞ζ?=75snN/µm(a)5101520frictionζ[snN/µm]0.50.60.70.8extensionα(b)0246timet[104s]0246timet[104s]0123timet[103s]00.050.1CVviscousfrictionζ71snN/µm72snN/µm73snN/µm74snN/µm75snN/µm76snN/µm(c)0123timet[103s]012(cid:10)R2(cid:11)[103µm2](d)14 FIG. S8. Cell polarisation and substrate deformation. The top halves and bottom halves of each panel represent the two mirrored halves of a cell. The bottom half of each panel depicts the local protrusion energy of the cell per hexagon (x, y). The bottom half of each panel depicts the substrate density below the cell (number of hexagons per unit area) σ(x,−y). These quantities are obtained by collapsing the data of many cells in their center of mass frame with the polarization axis oriented along the x-axis (θ = 0). In general substrate is dilated outside of the cell and compressed inside of the cell. Note the differences in the scales of the substrate density σ. The cell will preferably migrate along the gradient ∇( σ). (a),(b) For large substrate deformations, the migration of the cell is in the leading order dominated by the gradient in substrate density ∇σ, thus trapping the cell in the region of high substrate density. (c)-(h) For small substrate deformations, cell migration is in the leading order dominated by the gradient in protrusion energy ∇ and will migrate in the direction indicated by the black arrows. ity on cell behavior will decrease. Furthermore, as it takes more effort to rotate a pronounced protrusion en- ergy profile than to rotate a flat protrusion energy pro- file, we would expect an increased persistence time of directed migration. Observations in agreement with our qualitative assessment have been made in the preceding study [7] by increasing the maximal polarizability of the cells ∆Q = Q − q. b. Can the cell shape serve as memory? We cannot exclude the hypothesis that cell shape con- tributes to the memory of the cell, in addition to its pro- trusion energy field. Because we have assumed that the internal cell dynamics of the cell is much faster than indi- vidual protrusions/retractions, reorientations of the cell's protrusion energy field should in principal be faster than rotations of the whole cell body. However, we have ob- served that cells on substrates with stiffnesses 8 nN/µm and 4 nN/µm have different persistence times of directed migration in spite of having similar shapes [Fig. 3(b),(c)]. Thus, we conclude that cell shape can not be the sole originator of the cell memory. c. How are substrate deformations induced? Because the total mass of the substrate is conserved under deformations, a substrate dilatation (decrease in substrate density) at any given location has to be accom- panied by a substrate compression (increase in substrate density) elsewhere. Furthermore, cell traction forces are always pointed inwards of the cell. Together, this imposes substrate density gradients with substrate dilatation at the cell rim and substrate compression at the cell center, both proportional to the applied traction force. At the leading edge of the cell, traction forces are larger than at the trailing edge, due to a higher local protrusion energy [Figs. S8 and S9(a),(e)]. This leads to a stronger sub- strate dilatation at the leading edge than at the trailing edge of the cell, and a net increase of substrate density at the trailing edge [Figs. S8 and S9(b),(c),(f),(g)]. d. What determines cell speed? As we have discussed in sections B 1 b and B 1 c the 'noise strength' is a measure for the randomness of cell protrusions and retractions at the cell boundary. This randomness can be increased by increasing the effective temperature β or analogously by decreasing the bias in- troduced by the local protrusion energy . In general, un- k=0.5nN/µmk=1nN/µmk=2nN/µmk=8.75nN/µmlowζ=17.5snN/µm(a)123σ56789[102nNµm]15µm(b)11.21.4σ56789[102nNµm]15µm(c)11.1σ56789[102nNµm]15µm(d).981.02σ56789[102nNµm]15µmhighζ=121snN/µm(e).981.02σ56789[102nNµm]compress.dilatation15µm(f).981.02σ56789[102nNµm]15µm(g).981.02σ56789[102nNµm]15µm(h).981.02σ56789[102nNµm]15µm15 FIG. S9. Angular profiles of cell polarisation and substrate deformation. (a),(e) Profile of protrusion energy per hexagon at the boundary of a cell b for low (ζ = 17.5 s nN/µm) and high (ζ = 121 s nN/µm) viscous friction, respectively. (b),(f ) Profile of substrate density (number of hexagons per unit area) along the interior boundary of a cell σb,I for low and high viscous friction, respectively. (c),(g) Profile of substrate density (number of hexagons per unit area) along the exterior boundary of a cell σb,O for low and high viscous friction, respectively. (d),(h) Profile of protrusion energy density at the boundary of the cell b σb,I for low and high viscous friction, respectively. The dashed lines correspond to cells in the rounding state [Fig. S5: cells are trapped and have an oscillating VACF], while solid lines denote cells in the running state. like in section B 3 a, the substrate can be deformed. Thus, the probability of locally gaining or losing an infinitesi- mal area dA is determined by the local protrusion energy density σ. A stronger random contribution in the pro- trusion/retraction process leads directly to a broader cell velocity distribution around the instantaneous cell polar- ization vector and consequently a lower cell speed. This can easily be illustrated as follows: Consider a cell start- ing with a pronounced polarization profile (i.e. no initial symmetry breaking required). Then, in the low temper- ature limit (β →∞), the cell will always protrude at the location of highest polarization energy density and always retract at the location of the lowest polarization energy density, leading to a narrow cell velocity distribution. Be- cause each individual protrusion and retraction event dis- places the cell in the same direction and the total amount of such events per Monte Carlo Step is limited, this will lead to a high cell translocation speed. Conversely, in the high temperature limit (β → 0), cell protrusions are completely unbiased by the polarization profile, leading to a flat cell velocity distribution. Because all individual protrusion and retraction events displace the cell in dif- ferent directions and the total amount of such events per Monte Carlo Step is limited, this will lead to a negligible cell translocation speed. We conclude that the cell will migrate faster if the bias for individual protrusion/retraction events to align with the cell's protrusion energy field is increased. Fur- thermore, this alignment can be measured by the width of the symmetrical velocity distribution around the po- larization axis, or the overall randomness of the protru- sion/retraction process ('noise strength') inferred from Eq. S2. We consistently find that the cell speed increases with both decreasing cell velocity distribution width and with decreasing 'noise strength' [Fig. S10]. e. What determines the persistence time of directed migration of the cell? To answer this question, we introduce a quantity that characterizes the polarization strength of a cell (cid:90) π p = 1 π dθ cos(θ) b(θ, θ) σb,I(θ, θ) , (S5) 0 with the angle θ of the average polarization axis rela- tive to the x-axis. The choice of this quantity is based on the idea that the cell is strongly polarized, if there are many tiles (high substrate density) with a high pro- trusion energy at the leading edge of the cell, and few tiles (low substrate density) with low protrusion energy at the trailing edge of the cell. The persistence time of directed migration increases exponentially with the cell polarization strength, because it is more costly to rotate the polarization vector of a cell with a pronounced pro- trusion energy profile than that of an unpolarized cell lowζ=17.5snN/µm0πθ−θ6789b[102pNnm]5nN/µm4nN/µm3nN/µm2nN/µm1.5nN/µm1nN/µm(a)0πθ−θ11.11.2σb,I(b)0πθ−θ1.01.1σb,O(c)0πθ−θ789bσb,I[102pNnm](d)highζ=121snN/µm0πθ−θ789b[102pNnm]5nN/µm4nN/µm3nN/µm2nN/µm1nN/µm0.5nN/µm(e)0πθ−θ11.05σb,I(f)0πθ−θ0.951.0σb,O(g)0πθ−θ789bσb,I[102pNnm](h)16 ing forces. To effectively translocate, the cell needs to protrude at one side and retract at the opposite side. However, because the substrate density is dramatically increased at the position of the cell, all retractions are energetically penalized, and it becomes energetically dis- advantageous for the cell to translocate. Analogously, one might consider the event that the cell, due to the stochastic nature of the model, manages to move in some random direction. In that case, it would be energeti- cally advantageous to simply move back to its previous location. Because the positive feedback mechanism is fu- eled by protrusions and retractions alike [Section A 7], inhibiting retractions effectively inhibits the formation of a pronounced cell protrusion energy density profile [Fig. S9(d)]. As a result, there is no notable bias towards protrusions or retractions throughout the cell, leading to a broad velocity distribution and consequently a low cell speed. Furthermore, the flat energy density profile can be related to a marginal cell polarization strength, thus making cell reorientations cheap and frequent. Alto- gether, this leads to an effective cell trapping, where the cell initially attempts to polarize and is then prompted to turn around in its attempt to occupy areas of high substrate density. This is evidenced by the oscillating VACF [Section B 2] and leads to a vanishing cell persis- tence time of directed migration. The dynamics of cells seeded at intermediate substrate stiffnesses can be understood as an interpolation be- tween the low and high stiffness cases: in Fig. S8(c), the substrate stiffness is slightly increased compared to Fig. S8(b). This reduces substrate deformations and leads to less inhibition of the positive feedback mech- anism as compared to Fig. S8(b). Thus, here the cell can establish a more pronounced protrusion energy den- sity profile than in Fig. S8(b) and migrate persistently, i.e. no oscillations occur in the VACF. The width of the protrusion energy density profile, and consequently also the cell speed, lies in between Fig. S8(b) and Fig. S8(d). g. High viscous friction For high enough viscous friction [Fig. S8(e)-(h)], sub- strate deformations are too small to completely in- hibit the positive feedback mechanism. Hence, the cell can always polarize and migrate persistently, similar to Fig. S8(d). The profile of protrusion energy per hexagon does not particularly change for different stiff- nesses [Fig. S9(e)], and can thus by itself not explain the observed noticeable change in cell persistence time of directed migration. This can be corrected by taking the substrate density σ into account, which represents the number of hexagons per area, and thus considering the protrusion energy density σ [Fig. S9(h)]. The total polarization strength is represented by p [Eq. S5]. Let us first compare the behavior of a cell on high stiffness [Fig. S8(h)], high viscous friction substrate to that on a high stiffness, low viscous friction substrate FIG. S10. Stochasticity of the protrusion/retraction process in the simulations. (a) The cell speed decreases with the distribution width of the velocities around the polar- ization axis of the cell. Here, θv − θ(cid:126) is the angle between the cell velocity vector v and the instantaneous cell polarization vector n(cid:126) . (b) The cell speed decreases with increasing 'noise strength' (stochasticity of the protrusion/retraction process) inferred from Eq. S2. (c) Dependence of the distribution width of the velocities around the polarization axis of the cell on the substrate stiffness k. (d) Dependence of the 'noise strength' inferred from Eq. S2 on the substrate stiffness k. [Fig. 3(d)]. f. Low viscous friction For low viscous friction, substrate relaxation and re- sponse to the traction forces of the cell occur on short time scales τR = ζ/k compared to the typical timescale of the cell dynamics τC. Thus substrate deformations keep up with the cell and -- depending on the substrate stiffness k -- can become large enough to impair cell mo- tion. For high values of substrate stiffness [Fig. S8(d)], sub- strate deformations are small due to high restoring forces. Thus, the influence of the substrate on the cell behavior becomes negligible and the cell behaves similarly as on a non-deformable substrate (k →∞), see section B 3 a. The cell polarizes strongly and has a long persistence time of directed migration. Moreover, the cell veloci- ties are narrowly distributed around the cell polarization axis, in correspondence with the high cell speed. In contrast, low substrate stiffness [Fig. S8(b)] leads to a quick and profound compression of substrate at the position of the cell [Fig. S9(b)] due to the low restor- 246h(θv−θ~)2i−1406080speedv[nm/s]highζlowζ(running)(a)0.150.20.25noisestrength1−a−b0.40.60.8speedv[nm/s]highζlowζ(running)(b)2468stiffnessk[nN/µm]246h(θv−θ~)2i−1highζlowζ(running)(c)2468stiffnessk[nN/µm]0.150.20.25noisestrengthhighζlowζ(running)(d)17 FIG. S11. Cell extension. (a) Strong cell elongation oc- curs for low substrate stiffnesses and within a broad range of viscous friction coefficients. The red vertical dashed line corresponds to the stiffness k ≈ 1.58 nN/µm, below which no persistent cell migration occurs. The red horizontal dashed line corresponds to the viscous friction ζ (cid:63) = 75 s nN/µm, above which we on average observe only running states [Fig. S7(d)]. The black dashed line corresponds to the lower limit of ζ (cid:63), where we from a simple estimation would expect viscous ef- fects to dominate. The color code represents the extension of the cell for a given parameter combination. In the color code, the elongation of a cell on a non-deformable substrate is indi- cated by the dashed line. (b) Cell elongation at low substrate stiffnesses does not depend on cell polarizability ∆Q = Q − q. The color code represents the current elapsed time of a given data point in the simulation (color bar). cell. Hence, the cell has an increased protrusion activity at its short sides and subsequently stretches, as it tries to occupy areas of high substrate density. [Fig. S8(d)]. The cell speeds have similar magnitude due to a similar width of the velocity distribution. Cell persis- tence, however, is lower in Fig. S8(h), because the partic- ular substrate density profile effectively reduces the cell polarization strength p in a corresponding way. In par- ticular, note that the substrate density is reduced at the leading edge, and increased at the trailing edge. For low substrate stiffness [Figs. S8(e) and S9(g)], the lowest substrate density is not encountered at the leading edge, but at the side of the cell. This particular substrate density profile decreases the probability to protrude at the side of the cell, compared to protruding at the lead- ing edge. Because the cell has a strong bias to protrude at the leading edge, and the particular substrate den- sity profile slightly discourages motion to the side, the cell velocity distribution around the polarization vector is focussed. This effectively increases the average cell speed compared to Fig. S8(d), where no such focussing takes place. Compared to Fig. S8(d), the particular sub- strate density profile further reduces the cell polarization strength p, and thus the persistence time of directed mi- gration, because substrate deformations are larger on a soft substrate than on a stiff substrate. As before, intermediate substrate stiffnesses can be un- derstood as an interpolation between the low and the high stiffness cases [Fig. S8(f),(g)]. 4. Modulation of cell morphology by substrate interactions The change in cell shape due to substrate deforma- tions can be explained in a simple way by looking at the substrate density. In general, the shape of persistently migrating cells in the CPM [6, 7] tends to be fan-like and elongated perpendicular to the direction of motion. In these previous studies, the elongation increases with the polarizability ∆Q = Q − q of the cell. This maximal po- larizability translates to a realized polarization strength p of the cells in question. Therefore, decreasing the po- larization strength of the cell will effectively also lead to a rounder cell shape, as happens with decreasing substrate stiffness. Additionally, one can argue for high substrate viscous friction that the substrate density profile leading to the focussing effect of the cell velocities 'squeezes' the cell together because fewer protrusions occur at the sides of the cell. For low viscous friction and low stiffness of the sub- strate, we observe a profound elongation of immotile cells. This observation can be explained in the follow- ing way: Due to the inhibition of the positive feedback mechanism, the cell is unpolarized. Hence, its behavior is dominated by the substrate density profile alone, and is indeed the same for cells of a wide range of different po- larizabilities ∆Q = Q− q [Fig. S11], but identical average traction force. This particular substrate density profile is characterized by an increased substrate density at the short sides of the cell compared to the long sides of the 24stiffnessk[nN/µm]50100frictionζ[snN/µm](a)0200400polarizability∆Q[pNµm]0.60.70.8extensionα(b)0.60.70.8extensionα0246timet[104s]18 Appendix C: Experimental methods 1. Preparation of polyacrylamide substrate Acrylamide solutions corresponding to 0.2 kPa, 1 kPa, 2 kPa, 3 kPa, 7 kPa, 15 kPa, 34 kPa and 100 kPa polyacry- lamide hydrogels were prepared according to previous publications [21]. Briefly, an acrylamide/bis-acrylamide solution (Bio-Rad) was degassed and mixed with 1/100 volume 10% ammonium persulfate and 1/1000 volume tetramethylethylenediamine (Sigma Aldrich). 20 µL of solution was pipetted into a single well of an un- treated 12 well glass bottom plate (In Vitro Scien- tific). A 12 mm glass coverslip chlorosilanized with dichlorodimethylsilane (Sigma Aldrich) was placed above the acrylamide solution. Upon polymerization, hydro- gels were rinsed and functionalized with photoactivatable Sulfo-SANPAH (Thermo Fisher) before overnight conju- gation with 100 µg/mL Collagen type I (Gibco). Prior to cell culture all hydrogels were UV sterilized. FIG. S12. Manual image analysis procedure. Cells are selected and measured over the course of several frames to average out fluctuations in cell shape. Each cell satisfies the following conditions: (i) no alignment with the cracks on the substrate; (ii) no cell division within this timeframe; (iii) distinctness from background and neighboring cells. • good distinguishability of cells from the background and neighboring cells. 2. Cell culture 5. Cell speed analysis Human umbilical vein endothelial cells (HUVECs) were cultured in ready-to-use Endothelial Cell Growth Media (PromoCell) with 100 units/mL penicillin and 100 µg/mL streptomycin (Gibco). For live cell imag- ing, cells were plated at a density of approximately 2500 cells/cm2, or 10 000 cells/well. The cell positions were extracted manually from a subset of the phase contrast images using the software Fiji [22] and the Plug-In MTrackJ [23] [Fig. S12]. Math- ematica was used to perform a statistical analysis of the cell speeds [1]. 3. Microscopy Cells on hydrogels were maintained in a microscope- mounted incubator at 37 ◦C and 5% CO2. An AxioVert 200M with Axiovision software (Zeiss) and a PerkinElmer UltraVIEW ERS with Volocity software (PerkinElmer) were used to capture phase contrast images every ten minutes for a period of 48 h. 4. Cell shape analysis The cell shapes were extracted manually from a subset of the phase contrast images using the software Fiji [22] [Fig. S12]. Mathematica was used to perform a statistical analysis of the cell shapes [1]. Cells were selected and measured at various times to mitigate fluctuations in cell shape. Each cell should satisfy three conditions: • no alignment with the cracks on the substrate, as this is a strong bias for the measurement towards elongated cells, • no divisions at the time of measurement, as this is a strong bias for the measurement towards round cells, (a)0.2kPa(b)1kPa(c)2kPa(d)3kPa(e)7kPa(f)9kPa19 [1] Wolfram Research, Inc., "Mathematica, Version 11.3," [14] A. Mogilner and K. Keren, Curr. Biology 19, R762 Champaign, IL, 2018. (2009). [2] A. Forsgren, P. Gill, and M. Wright, SIAM Review 44, 525 (2002). [3] D. W. Marquardt, J. Soc. Indust. Appl. Math. 11, 431 (1963). [4] K. Levenberg, Quart. Appl. Math. 2, 164 (1944). [5] M. G. Yucht, M. Sheinman, and C. P. Broedersz, Soft Matter 9, 7000 (2013). [6] F. J. Segerer, F. Thuroff, A. Piera Alberola, E. Frey, and [15] A. F. M. Mar´ee, A. Jilkine, A. Dawes, V. A. Grieneisen, and L. Edelstein-Keshet, Bull. Math. Biol. 68, 1169 (2006). [16] A. F. M. Mar´ee, V. A. Grieneisen, and L. Edelstein- Keshet, PLOS Comput. Biol. 8, e1002402 (2012). [17] T. Pompe, M. Kaufmann, M. Kasimir, S. Johne, S. Glo- rius, L. Renner, M. Bobeth, W. Pompe, and C. Werner, Biophys. J. 101, 1863 (2011). J. O. Radler, Phys. Rev. Lett. 114, 228102 (2015). [18] C.-M. Lo, H.-B. Wang, M. Dembo, and Y.-L. Wang, [7] F. Thuroff, M. Reiter, A. Goychuk, and E. Frey, "Bridg- ing the gap between single cell migration and collective dynamics," Unpublished. [8] S. Tojkander, G. Gateva, and P. Lappalainen, J. Cell Sci. 125, 1855 (2012). Biophys. J. 79, 144 (2000). [19] Y. L. Han, P. Ronceray, G. Xu, A. Malandrino, and (2018), R. D. Kamm, M. Lenz, C. P. Broedersz, M. Guo, Proc. Natl. Acad. Sci. U.S.A. 10.1073/pnas.1722619115. [9] H. Herrmann, H. Bar, L. Kreplak, S. V. Strelkov, and [20] M. Riaz, M. Versaevel, D. Mohammed, K. Glinel, and U. Aebi, Nat. Rev. Mol. Cell Biol. 8, 562 (2007). S. Gabriele, Sci. Rep. 6, 34141 (2016). [10] C. Sultan, D. Stamenovi´c, and D. E. Ingber, Ann. [21] J. R. Tse and A. J. Engler, Curr. Protoc. Cell Biol. 47, Biomed. Eng. 32, 520 (2004). 10.16.1 (2010). [11] D. Tsuruta and J. C. R. Jones, J. Cell Sci. 116, 4977 (2003). [12] F. Graner and J. A. Glazier, Phys. Rev. Lett. 69, 2013 (1992). [13] T. D. Pollard and G. G. Borisy, Cell 112, 453 (2003). [22] J. Schindelin, I. Arganda-Carreras, E. Frise, V. Kaynig, M. Longair, T. Pietzsch, S. Preibisch, C. Rueden, S. Saalfeld, B. Schmid, J.-Y. Tinevez, D. J. White, V. Hartenstein, K. Eliceiri, P. Tomancak, and A. Car- dona, Nat. Methods 9, 676 (2012). [23] E. Meijering, O. Dzyubachyk, and I. Smal, Meth. Enzy- mol. 504, 183 (2012).
1907.01585	1	1907	2019-07-02T19:01:01	Strongly bent double-stranded DNA: reconciling theory and experiment	[ "physics.bio-ph", "cond-mat.soft", "q-bio.BM" ]	The strong bending of polymers is poorly understood. We propose a general quantitative framework of polymer bending that includes both the weak and strong bending regimes on the same footing, based on a single general physical principle. As the bending deformation increases beyond a certain (polymer-specific) point, the change in the convexity properties of the effective bending energy of the polymer makes the harmonic deformation energetically unfavorable: in this strong bending regime the energy of the polymer varies linearly with the average bending angle as the system follows the convex hull of the deformation energy function. For double-stranded DNA, the effective bending deformation energy becomes non-convex for bends greater than about 2 degrees per base-pair, equivalent to the curvature of a closed circular loop of about 160 base pairs. A simple equation is derived for the polymer loop energy that covers both the weak and strong bending regimes. The theory shows quantitative agreement with recent DNA cyclization experiments on short DNA fragments, while maintaining the expected agreement with experiment in the weak bending regime. Counter-intuitively, cyclization probability (j-factor) of very short DNA loops is predicted to increase with decreasing loop length; the j-factor reaches its minimum for loops of 45 base pairs. Atomistic simulations reveal that the attractive component of the short-range Lennard-Jones interaction between the backbone atoms can explain the underlying non-convexity of the DNA effective bending energy, leading to the linear bending regime. Applicability of the theory to protein-DNA complexes, including the nucleosome, is discussed.	physics.bio-ph	physics	Strongly bent double-stranded DNA: reconciling theory and experiment Aleksander V. Drozdetski,1, a) Abhishek Mukhopadhyay,1, a) and Alexey V. Onufriev1, 2, 3, b) 1)Department of Physics, Virginia Tech, Blacksburg, VA 24061, USA 2)Department of Computer Science, Virginia Tech, Blacksburg, VA 24061, USA 3)Center from Soft Matter and Biological Physics, Virginia Tech, Blacksburg, VA 24061, USA (Dated: 4 July 2019) The strong bending of polymers is poorly understood. We propose a general quantitative framework of polymer bending that includes both the weak and strong bending regimes on the same footing, based on a single general physical principle. As the bending deformation increases beyond a certain (polymer-specific) point, the change in the convexity properties of the effective bending energy of the polymer makes the harmonic deformation energetically unfavorable: in this strong bending regime the energy of the polymer varies linearly with the average bending angle as the system follows the convex hull of the deformation energy function. For double-stranded DNA, the effective bending deformation energy becomes non-convex for bends greater than ∼ 2◦ per base-pair, equivalent to the curvature of a closed circular loop of ∼ 160 base pairs. A simple equation is derived for the polymer loop energy that covers both the weak and strong bending regimes. The theory shows quantitative agreement with recent DNA cyclization experiments on short DNA fragments, while maintaining the expected agreement with experiment in the weak bending regime. Counter- intuitively, cyclization probability (j-factor) of very short DNA loops is predicted to increase with decreasing loop length; the j-factor reaches its minimum for loops of ≃ 45 base pairs. Atomistic simulations reveal that the attractive component of the short-range Lennard-Jones interaction between the backbone atoms can explain the underlying non-convexity of the DNA effective bending energy, leading to the linear bending regime. Applicability of the theory to protein-DNA complexes, including the nucleosome, is discussed. I. INTRODUCTION Deformation of polymers is ubiquitous, elastic prop- erties of these macromolecules are crucial for their dy- namics. Biopolymers are abundant in nature and play vital roles in many biological processes1 -- 4, which not only depend upon the polymer structure, but also their physical properties5 -- 7. Among biopolymers, DNA stands out as a case of its own. Understanding DNA deforma- tion is crucial for the mechanistic grasp of vital cellular functions such as packaging of DNA compactly into vi- ral capsids, the chromatin, formation of protein/DNA complexes and regulation of gene expression2,8. An all- important example of DNA deformation, relevant to a variety of biological processes that depend on its elas- tic properties, is DNA looping, which occurs in many prokaryotic9 and eukariotic10 systems. A number of reg- ulatory proteins can loop DNA into various bent confor- mations, critical for regulation of many biological pro- cesses involving DNA11. Most notably, DNA is strongly bent in the nucleosome12,13, which is the fundamental unit of genome packing: accessibility to genomic infor- mation in eukareotes is modulated by the strength of DNA-protein association14,15. Note that the majority of eukariotic genomic DNA (75-80%) is packed tightly into nucleosomes10. Nanostructures made directly of DNA16 or those that use DNA as a scaffold17, can be influenced18 by its mechanical properties on short length scales, pro- a)These two authors contributed equally b)Electronic mail: [email protected] viding yet another impetus to understand the strong bending regime of the DNA. Experimental evidence on cyclization of DNA frag- ments shorter than ∼100 base-pairs points to the fact that strongly bent DNA -- most relevant from a biological perspective -- is considerably more flexible than expected from established models (worm-like chain) that work well within the weak bending regime. Yet, despite decades of experimental and theoretical effort, the story of how this arguably most important polymer behaves under defor- mation is far from complete, with controversies and new developments abound19 -- 25. The current state and the relevant terminology are briefly reviewed below. The bending flexibility of a polymer is conventionally quantified in terms of its persistence length, Lp, a length scale below which the polymer behaves more or less like a rigid rod. Specifically, Lp is defined as length of the poly- mer segment over which the time-averaged orientation of the polymer becomes uncorrelated; for fragments smaller than Lp, the thermal fluctuation alone is not enough to induce significant (∼ 1 rad) bending26. Here we use this definition of Lp to qualitatively separate the two bending regimes: if no significant bending is observed on length scales shorter than Lp, the polymer can be deemed weakly bent; otherwise the bending is assumed to be strong. For double-stranded DNA, a variety of experimental techniques27 -- 32, revealed that Lp ≈ 150 bp or 500 A. Based on the Lp value and the above definition of strong bending, we conclude that most of the DNA in eukareotes is strongly bent. Indeed, since the nucleosome contains a stretch of double-stranded DNA of ∼ 150 bp looped almost twice, the DNA in this complex can be considered as strongly bent. Response of DNA to mechanical stress has been stud- ied extensively 2,8,27,30,32 -- 49, leading to a consensus in modeling the weak bending regime. Arguably the most widely used simplified model of DNA bending is the worm-like chain (WLC) model. In the original WLC32,50, the polymer is modeled as a continuous, isotropic elastic rod with its deformation energy being a quadratic func- tion of the deformation angle. In the discrete version of WLC model, the bending energy of the polymer consist- ing of N segments of length l is given by: Echain = N −1 Xi 1 2 kBT Lp l θ2 i (1) where θi is the angle between two consecutive segments (see inset of Fig. 1). While this simplistic model lacks some features of the real DNA, such as sequence depen- dence of its local mechanical properties, it nevertheless captures the key physics of weak polymer bending, which explains why the model is robust and is widely adopted to interpret experiment. Various theoretical models of DNA bending, including those that explicitly account for the sequence-dependence51 -- 54, were consequently developed that also assumed harmonic (quadratic) angular defor- mation energy of DNA. There is very little doubt that the Hookean, "elastic rod" models accurately describe many polymers in the weak bending regime1, including the double-stranded DNA32,44. Indeed, lowest order term of a Taylor series expansion of any well-behaved func- tion around its local minimum is quadratic, which means that for small deviations from equilibrium, the response function can be considered harmonic. However, by the same logic it should be expected that beyond a certain threshold the bending energy may no longer be approxi- mated by the quadratic term alone; investigations of pos- sible influence of non-harmonic terms on the mechanical properties of double-stranded DNA is a relatively new area. Historically, only very large fragments (hundreds to thousands of base-pairs) were investigated27,28, which are well described by the WLC regardless of what hap- pens on short length-scales38. However, within the last decade or so, the prevailing view of DNA as a Hookean polymer was challenged by experiments that were able to investigate the flexibility of DNA on scales smaller than several Lp. Counter- intuitively, small DNA fragments (≈ 100bp) were found to have much higher probability of cyclization (sponta- neous formation of loops) than that predicted by the WLC theory42. This discovery sparked considerable con- troversy, which still remains unresolved. What is partic- ularly puzzling is that strongly bent DNA appears less rigid than the DNA in the Hookean regime. Some of the follow-up experimental and theoretical work supports the validity of WLC even for tightly bent DNA20,32,55, while others still show that short, tightly bent DNA is much more flexible19,38,56 than previously thought, in a manner that can not be described by a harmonic model38. Several theoretical models have been proposed to ac- count for the unexpectedly high flexibility of strongly 2 bent double-stranded DNA. One popular model57 -- the meltable WLC or MWLC -- postulated that the extra flexibility stems from formation of small local "bubbles" of single stranded DNA, which is much softer than the double helix. However, the degree of softening provided by the mechanism alone was later found58 to be inad- equate to fully explain the very sharp bends in DNA observed experimentally; in atomistic simulations, nega- tive super-coiling was required to induce such bubbles in DNA mini-circles59. An early model60, put forward well before the unusual DNA flexibility was discovered exper- imentally, suggested that the energy of a bent double- helix could be lowered by formation of sharp, ∼ 90o kinks that maintain the Watson-Crick pairing along the helix. Sharp kinks were indeed observed in a pioneering atom- istic simulation43 some thirty years later, but subsequent improvements in the simulation methodology indicated that these were only induced at a high bend angle equiv- alent to those occurring in circles of just 45 base-pairs61, while experimental softening of the DNA is seen experi- mentally for circles as large as ∼ 106 base-pairs19. Sharp kinks in double-stranded DNA can be introduced empiri- cally into the WLC model, e.g. by adding freely-bending hinge elements to the WLC chain, leading to a kinkable WLC, or KWLC39. A non-linear empirical bending po- tential that allows for the possibility of ∼ 90o kinks in double-stranded DNA was recently proposed20, but its physical origins, the critical value of the DNA curva- ture at which the kink occurs, and the corresponding energy gain remained unknown20. At the same time, a purely linear empirical bending potential was shown38 to describe the softer DNA seen in AFM experiments, al- though the origin of the linear regime and its parameters (e.g. critical bend angle where the linear regime begins ) remained unclear. Are the kinking and the linear regime just two manifestations of a deeper underlying principle? In summary, the nature of the effective bending energy of double-stranded DNA in the strong bending regime as well as the precise connection to the observed softening of the polymer is not fully clear. The influence of mechani- cal constraints on this connection remains unexplored. It is unclear how the softening of strongly bent DNA stems from its atomic-level structure and interactions. From a more philosophical standpoint, is hard to believe that very special models are needed to describe the bending of the DNA; rather it is more like that the curious case of the DNA is just a special case of a broader underlying theory applicable to all polymers. In this work we propose, and verify against available experiment, a unified theoretical description of poly- mer bending that treats the weak and strong bending regimes on the same footing, guided by a simple physi- cal principle. The proposed framework does not rely on ad-hoc postulates; instead, it shows how the apparent softening of strongly bent DNA follows naturally from a specific mathematical property of the experimentally- derived bending energy. Simulations suggest an atomistic explanation for the specific shape of the bending energy function. II. METHODS A. DNA bending energy from experimental data A statistically significant, diverse set of several hun- dred PDB structures of protein-DNA complexes was in- vestigated previously in Ref.32. The probability distri- bution of the experimental DNA bending angles was used in Ref.32 to approximate the bending energy E(θ) (per base pair) as a fourth oder polynomial: E(θ) = 203.1θ2 − 552.7θ3 + 416.8θ4 (where θ is in radians and E(θ) is in units of kT ). Here we use this E(θ) to rep- resent the experimental effective bending energy of the double-stranded DNA, blue line in Fig. 2. B. Atomistic MD simulations of closed DNA loops To avoid "end effects", and make a close connection with DNA cyclization experiments, we employed closed DNA circles to estimate their effective bending energy E(θ) per base pair. DNA circles of various sizes (50-400 bp) were generated using NAB62 (AmberTools) for se- quence poly(dA).poly(dT), helical repeat of 10 bp, and other parameters of B-DNA as specified in NAB. We de- liberately chose this simple, uniform sequence to focus on the basic physics of DNA deformation. All of the atomistic MD simulations were performed within AMBER-12 package, using ff99bsc0 force-field. The Generalized Born (GB-HCT, AMBER option igb=1) implicit solvation model was used to treat solvation ef- fects, including 0.145M of monovalent salt. No long- range cut-off was employed. The model's performance in atomistic simulations of DNA, including studies of its deformation41, is well established63. Two critical advan- tages of the implicit solvation over the more traditional explicit solvation64 made the former the method of choice in this work. These are the superior simulation efficiency for large DNA structures65 and the straightforward man- ner in which their energies, including free energy of sol- vent re-arrangement, can be estimated63 within the im- plicit solvation framework. All DNA circles were initially minimized for 1000 steps with "P" atoms restrained their original positions with a force constant of 1.0 kcal/mol/A2 to enforce the circular shape. Each system was then heated to 300K and equi- librated for 100 ps with the same restraints as for the minimization. Shake was used to constrain the hydrogen atoms; we employed 2 fs time-step for the atomistic simu- lations. Finally, we generated 1 ns long MD trajectory for each circle at 300K, with "P" atoms also restrained with a force constant of 0.1 kcal/mol/A2, sufficient to support the near perfect circular shape of the fragment, but al- lowing for local re-arrangements. The energies and their components, including the electrostatic, VDW, bond, 3 etc. were saved every 20 fs, and averaged over the whole trajectory. The relatively short simulation time allowed us to simulate even the largest of the circles; it is justified by the use of the strong positional restraints, which per- mit only local, very fast structural re-arrangements. For smaller circles we verified that increasing the simulation time by an order of magnitude had negligible effect on the computed averages. In Fig. 4, the energy per bp was computed as the difference between per bp potential en- ergies of the given circle and the largest circle simulated, which is virtually unbent. C. Coarse-grained simulations of DNA loops ESPResSo66 was used to create and simulate coarse- grained closed loops of DNA of different sizes, from 6 to 600 bp long. A single bead of the appropriate mass rep- resents one base-pair of B-DNA; the bead-bead distance was set to 3.3 A , corresponding to the average distance between base pairs in canonical DNA. The bonds between the beads were made virtually inextensible (very large coefficient of the quadratic bond stretching energy); the bond angle potential (effective bending energy) between neighboring beads was defined to have the same form as in Fig. 2, that is correspond to the bending potential inferred from the experimental data32. No further bead- bead interactions or constraints on the loop geometry were imposed. The loops were simulated at T = 300K, and energy-minimized using steepest descent. D. Coarse-grained simulations of confined DNA fragments a. Protein-DNA complex. ESPResSo66 was used to create and simulate a 20 bead long fragment of "DNA" bound to a spherical charged "protein", Fig. 5. The beads and their interactions were set up as described above, with the following modifications. The end beads were not linked to create a loop. Each bead carried a unit charge qs =-1; The bead charges interacted only with a positive charge Q of the "protein", represented by a spherical impenetrable constraint of radius R. In addition, two impenetrable walls were placed above and below the charge Q to minimize out-of-plane bending of the "DNA". The confining charge Q was varied from 10 to 1000, effectively sampling two orders of magnitude of confinement strength (defined here as Q/qs). The constraint radius R was also varied to sample various curvature values of the "protein", and thus various total bending angles of the confined "DNA", Fig. 5. b. A nucleosome model. For the nucleosome model, the system described above was modified to mimic the confinement of DNA around realistic histone core. The DNA fragment size was increased to 147 bp, and the non- bonded interactions between monomers were turned on for an additional realism67. The fragment was confined around a cylinder of fixed diameter R =∼100A, and the walls were placed ∼50A apart (approximate dimensions of the nucleosome complex67). III. RESULTS AND DISCUSSION A. The proposed unified framework of polymer bending We begin with a useful analogy from classical ther- modynamics that connects system's stability to convex properties of its governing potential. For example, for a system to be stable against a macroscopic fluctuation in energy, the entropy of the system as a function of energy, S(E), must be concave (non-convex). Any chord con- necting two points on a graph of S(E) must lie below the curve itself in order to satisfy the second law of thermo- dynamics (maximum S). Conversely, the inverse function E(S) must be convex. If, however, E(S) is not convex over some region, the system phase-separates once this region is reached, with the properties of the two phases corresponding to the end points of the convex hull of the non-convex region. The actual, physical average energy of the system follows the convex hull, which makes the energy manifestly convex. This very general reasoning, with appropriate choice of the perturbation coordinate and potential, is applicable to phase transition of single species polymers (Flory-Huggins Theory68), as well as to stretching of polymers21 and other materials69. Here we use the analogy to develop a general framework that de- scribes polymer response to bending, weak and strong, on the same footing. Echain = PN Consider a polymer chain made of N ≫ 1 inextensi- ble, identical monomer segments with effective bending deformation energy E(θi) for each bending site, where θi is the angle between two successive segments (see in- set of Fig. 1). Here we assume that the effective E(θ) takes into account all the interactions, short- and long- range, between the monomers. For notational simplicity, in what follows we ignore the difference between N and N − 1 for large N . The total energy of the polymer is i E(θi), and without loss of generality we assume no intrinsic bends, i.e. E(0) = 0. Just like in WLC, we assume isotropic bending energy, which is a reasonable assumption for DNA fragments longer than 2 helical repeats or 20 bp20. For the moment, we further assume no torsional degrees of freedom. In order to in- duce an average non-zero bend in the chain, the polymer must be constrained, and the problem of finding the equi- librium polymer conformation is reduced to minimizing Echain, subject to the specific constraint of the problem. Here we assume that entropic effects are relatively small at length scales of interest (. Lp) -- an assumption that we explicitly confirm below by numerical experiments. We begin by considering a very special case of a uni- formly bent polymer -- constrained to have the same con- stant curvature along the entire chain. By construction, such a polymer consists of identically bent segments with each bending angle θi equal to the average deformation 4 angle, ¯θ = N −1PN i θi, and its total energy is N E(¯θ). monomers remains constant, α = PN Next, consider a more realistic situation where the polymer bending is enforced by a much less restrictive constraint: that the sum of the bend angles between the i θi = const. Note that this constraint alone does not fully define the geom- etry of the polymer. A closed planar loop, with the first and last segments linked, is a relevant example for which i θi = 2π, from elementary geometry of polygons, see also the SI. Mathematically, the problem of finding the minimum energy conforma- tion of the polymer is that of energy minimization under the specific constraint: the constraint is satisfied; PN Echain = N E(¯θ) = min { i=1 θi=N ¯θ=α PN N Xi E(θi)} (2) where we make a clear distinction between E, which is the average bending energy per bending site in the minimum energy state of the polymer, and E corresponding to the uniform bending. Using Lagrange multipliers, Eq. 2 can be reduced to min{E(θ1)+· · ·+E(θN )−λ(θ1 +· · ·+θN − α)}. Differentiating with respect to θi gives a set of equa- tions ∂θiE(θi) − λ = 0 (for all i) which leads to a set of equalities ∂θ1 E(θ1) = ∂θ2E(θ2) = · · · = ∂θN E(θN ). For a convex functional form of E(θ), ∂θE(θ) monotonically increases with θ, and therefore the equalities are satis- fied only if θ1 = θ2 = · · · = θN : the polymer is always uniformly bent, that is each segment is bent through the same angle θ = ¯θ and E(¯θ) = E(¯θ). However, for a non- convex function such as one shown in Fig. 1, there can be more than one value of θ that satisfies ∂θ1E(θ1) = ∂θ2E(θ2) = · · · = ∂θN E(θN ): ∂θi E(θa) = ∂θiE(θb) for some θa < θb. Of special importance are θa and θb that mark the beginning and the end of the convex hull of E(θ) -- the segment of a straight line tangent to the non- convex function at two points, such that for any argu- ment between these two points the value of the function at the argument is greater than that of the convex hull line at the same argument, Fig. 1. One can demonstrate, see SI, that for bend angles ¯θ in the convex hull inter- val, θa < ¯θ < θb, a uniformly bent chain is no longer the stable minimum energy conformation of the polymer. Instead, the stable minimum is achieved when the distri- bution of bend angles is bi-modal: each segment is bent through one of the two bending angles θa or θb. This gen- eral point is illustrated in SI for a model polymer chain described by a non-convex bending potential relevant to the case of the DNA. In what follows we derive an explicit expression for E(θ) for θa < ¯θ < θb. In the minimum energy con- formation, let 0 < p < 1 represent the fraction of all the bending sites that are in the state θb and 1 − p the fraction of the remaining sites in the state θa. The to- tal bending angle in terms of θa and θb is then given by N pθb + N (1 − p)θa = N ¯θ = α, and the bending en- ergy per monomer in the non-convex region is E(¯θ) = 5 ) T k ( p b / y g r e n E 10 8 6 4 2 0 WLC u ll v e x h n c o E(θ) from experiment MD (energy minimized) MD (T=300K) 0 θ a 10 30 Bending angle, θ (degrees) 20 θ b 40 FIG. 2: DNA effective bending energy E(θ) (per bp) extracted from the probability distribution32 of DNA bends that naturally occur in protein-DNA complexes (blue line), and the average energy of unrestrained DNA closed loops simulated via coarse-grained MD with the same E(θ) (crosses). Green symbols: energy minimized (simulated annealing) loops. Black symbols: loops simulated at T=300K (the corresponding angular probability distribution and example structures are given in the SI). In both cases, the average loop energy as a function of average bend angle ¯θ = θ follows the convex hull of E(θ). The small deviation of the T=300K points from the convex hull are a result of ensemble average sampling and insignificant out-of-plane bending seen in the simulation. in Fig. 2, is the value of θa, which enhances robustness of the theory to inevitable imperfections32 of the input bending energy profile. For example, a uniform re-scaling E(θ) → λE(θ) would leave the x-coordinates θa and θb of the convex hull double-tangent segment unchanged be- cause the derivatives would be re-scaled by the same λ. Further discussion of the robustness of ECH model to its parameters can be found below and in SI. B. Bending of a circular loop, weak and strong While many different types of constraints can be phys- ically realized, one of the most important ones is the closed loop constraint, which is also used in DNA cy- clization experiments30,32,42 critical39 for investigating the strong bending regime. Consider the case of a sin- gle closed loop α = PN i θi = 2π. From Eq. 4, the total bending energy of a closed loop of total length L (number of base pairs, corresponding to "N" in Eq. 2 ) is given by Eloop = E(¯θ)L. Since ¯θ = α L , the bending energy of the loop is: L = 2π if L > 2π θa 2π2kBT Lp L Eloop(L) =  kBT Lpθa(cid:16)2π − 1  < L < 2π θa (5) Note that, where defined, the new function Eloop depends on just one new parameter: θa -- lower boundary of the 2 Lθa(cid:17) if 2π θb FIG. 1: Two different forms for a bending energy profile of a homopolymer. Shown is the (effective) bending energy per site E(θ). If the profile is purely convex down (black curve), the minimal energy conformations of the polymer is uniform bending (all sites are identically bent). If the function has a non-convex region (blue curve), non-uniform bending is more energetically favorable. In this case the total energy of the system follows the convex hull of the energy curve (red line). pE(θb) + (1 − p)E(θa). Rewriting p = (¯θ − θa)/(θb − θa), we arrive at E(¯θ) = ¯θ − θa θb − θa (E(θb) − E(θa)) + E(θa) (3) Therefore, in the non-convex region, the actual polymer energy per bending site, E(¯θ) corresponding to the stable minimum energy state, is a linear function of the average deformation ¯θ. Clearly, E(¯θ) < E(¯θ) within the convex hull interval, Fig. 1. To arrive at a general theory that can account for both the weak and strong bending regimes simultaneously, we use the form of Eq. 3 for the strong bending regime, while retaining WLC for the weak bending. In the proposed Energy Convex Hull(ECH ) model, the average per seg- ment (e.g. per base-pair) bending energy is described by an everywhere differentiable piece-wise polynomial func- tion: quadratic WLC (Eq. 1) for ¯θ < θa, and a linear function -- convex hull of E(θ) -- for θa < ¯θ < θb: 1 2 kBT Lp ¯θ2 if ¯θ ≤ θa E(¯θ) =   kBT Lpθa(¯θ − 1 2 θa) (4) if θa < ¯θ < θb where Lp is the accepted persistence length, well estab- lished for the weak bending regime; here it dimension- less, expressed in terms of the number of bending sites (e.g. number of base pairs for DNA loops). In this work we are not interested in the extreme strong bend- ing regime ¯θ > θb, since for the DNA this regime would correspond to loops smaller than 10 bp. Such small loops are likely physically impossible due to steric constraints, and are much smaller than those observed in cyclization studies19,70. Thus, the only key parameter that ECH the- ory inherits from the input effective bending energy, E(θ) non-convex domain. Although we tacitly assumed the loop to be confined to a 2D plane to simplify the deriva- tions, our unconstrained coarse-grained simulations of closed loops at 300K demonstrate, Fig. 2, that the as- sumption has little effect on our key conclusions. C. Application to double-stranded DNA The preceding discussion was not restricted to the case of DNA: non-uniform, two-phase bending, and the corre- sponding linear bending regime can be a feature of any polymer. However, since DNA is arguably the most im- portant polymer, and it exhibits looping in many differ- ent biological systems, we will focus on double-stranded DNA for the rest of the study. An effective bending en- ergy (per bp) calculated from a statistical analysis of ex- perimental PDB structures of DNA-protein complexes32 is shown in Fig. 2. This effective bending energy func- tion has a non-convex region, and thus a convex hull, the end points of which are θa = 2.2◦ and θb = 35.8◦, cor- responding to fragment lengths of L ∼ 160 and ∼ 10 bp respectively for DNA closed loops. Coarse-grain molecular dynamics simulations at 300K (see Supplementary Material) demonstrate that poly- mers with this effective bending energy between monomers exhibit all of the key features discussed above. For large loop sizes the bending angles are small (weak bending) -- the system samples the convex (harmonic) re- gion of the energy function, Fig. 2, and the distribution of bend angles is uni-modal. However, as the loop size decreases, the average angle per bending site ¯θ increases, eventually crossing the θa threshold. Once this happens, the energy of the system per bending site increases lin- early with ¯θ, and the distribution of bend angles becomes bi-modal, until the system reaches the upper boundary of the convex hull at θb. D. Comparison with DNA cyclization experiments are Most results cyclization experimental in terms of ex- pressed32,42 the Jacobson-Stockmayer j-factor, which estimates the probability that a linear polymer of length L forms a closed loop by joining its cohesive ends30,71. While Monte-Carlo based numerical approaches to compute j-factor exist23, here we use a well-established72,73 analytical expression for the j-factor of an unconstrained closed loop: j(L) ≃ exp(cid:18)− Eloop kBT + L 4Lp(cid:19) (6) k L3 L (cid:19)5 p (cid:18) Lp exp(cid:16) L where 1 L3 tribution, averaged over possible looping geometries, and L (cid:17)5 p (cid:16) Lp 4Lp(cid:17) accounts for the entropic con- kB T (cid:17) is the energy penalty of bending the DNA fragment to form the loop. We note that the k depends, exp(cid:16)− Eloop ) M ( r o t c a f - j 10-8 10-12 10-16 10-20 10-24 10-28 1e-09 1e-10 1e-11 60 40 50 40 50 70 Length, L (bp) 60 80 70 90 6 Experiment ECH WLC 80 90 100 110 100 110 FIG. 3: DNA cyclization j-factors computed using the proposed model (green line) and WLC (blue line) are compared with recent experiment19 (red dots, L > 60 bp). Experimental values of persistence length, Lp = 150 bp and θa = 2.2◦ (Fig. 2 ) were used; the value of k in Eq. 6 was obtained independently for each model as best fit against two experimental data points for fragment length L = 101 and 106 bp, see Supplementary Material. The envelopes of the j-factor (brown dashed lines) for ECH and WLC are shown in the inset. Predicted envelope for ECH j-factor has a minimum near 45 bp. The experimental data points L = 50 and L = 40 bp were shared by Taekjip Ha (see ref.19) in private communication to assess model performance after the model had been constructed. in a complex manner, on the loop closing geometry and can be expected to remain invariant over a relatively short range of loop lengths L, within the same experi- ment. To make a direct connection with cyclization exper- iments for non-integer numbers of helical repeats, we modulate the torsionally independent loop energy from Eq. 6 with cos(2πL/h), where we assumed the helical repeat h = 10 bp per turn. The agreement with the cy- clization experiment is robust with respect to the precise value of the helical repeat, see SI. This form of the mod- ulating factor is adopted from Ref.72 to account for the periodic variation of the j-factor due to the torsional com- ponent of the energy72. This simple way of accounting for non-integer numbers of helical repeats is sufficient for the purpose of testing key predictions of ECH vs. WLC, and does not affect the comparison with the over-all (en- velope, average) behavior39 of experimental j-factors, see also Table S1 in SI. We use Eloop(L) defined in Eq. 5 for ECH and Eloop(L) = 2π2kBT Lp L for all L in the case of WLC. The proposed ECH model and WLC are compared with the most recent experiment19 in Fig. 3. As seen from Fig. 3, ECH leads to an excellent agree- ment with the cyclization experiment, while the j-factors predicted by conventional WLC are off by several or- ders of magnitude in the strong bending regime (WLC is known to work well in the weak bending regime where it coincides with ECH by construction). The agreement of ECH with the experiment is robust to the value of its key input parameter θa, see below and SI. of very short a. Cyclization loops. Counter- intuitively, the predicted envelope function for ECH j- factor, which is essentially Eq. 6, has a minimum near 45 bp and begins to increase for even smaller loops, whereas for WLC j-factor decreases sharply for small loops. This completely counter-intuitive behavior of the cyclization probability for very tight loops predicted by ECH is borne out by experiment, Fig. 3; its physical origin is explained below. The two experimental points at L=50 and L=40, which support the counter-intuitive prediction of the theory, were not available to us until after the ECH framework was fully developed and tested against published data19 for larger circles. The over-all variation of the j-factor as a function of the loop length for both models is governed by the in- terplay between the entropic and the mechanical bend- ing energy costs Eloop(L) of forming the loop. For small loops, the entropic penalty of forming the loop decreases with the loop size L; however, Eloop(L) → ∞ for small L within WLC, which leads to a steep decrease in the over- all cyclization probability. In contrast, ECH loop energy, Eq. 5, approaches a constant for L → 0, which explains why the corresponding j-factor reaches a minimum and then begins to increase for small enough L, Fig. 3. This very different qualitative behavior of WLC and ECH j- factors for small loops can be used as a discriminating experimental test of the models. The predicted minimum value of the j-factor can be used to further discriminate between models that exhibit the minimum: for example, both KWLC39 and a recent version23 of MWLC predict the minimum, but the loop sizes at which the minima occur are substantially different from the ∼ 45 base pairs predicted by ECH . Within the proposed ECH framework of polymer bend- ing, the central role is played by convexity proper- ties of the effective bending energy between individual monomers. For the DNA, we used the energy profile in- ferred from statistical analysis of experimental structures of protein-DNA complexes (Fig. 2) -- the energy has a clear non-convex region, responsible for the "softer", lin- ear bending mode of short DNA loops. The same general considerations will hold for any effective bending energy that has a distinct non-convex region regardless of its origin24, including a kinkable WLC (kWLC) potential20. Thus, even though ECH explains experimental results perceived to be in contradiction with WLC, there is no fundamental contradiction between the new framework and the conceptual basis of WLC. E. Origin of the non-convex bending energy of DNA. To investigate, qualitatively, the physical origin of the non-convexity of the DNA effective bending energy we employed all-atom Molecular Dynamic (MD) simulations of uniformly bent DNA circles of a wide range of sizes, from small to very large, corresponding to almost unbent 7 DNA, see "Methods". Specifically, we examined the aver- age bending energy per base pare. The total bending en- ergy profile obtained from these simulations, along with the breakdown into components of different physical ori- gin, are shown in Fig. 4; one can clearly see a prominent non-convex region, in qualitative agreement with the ex- periment, Fig. 2. The key parameter θa ≈ 1.5◦ from the MD simulations, which is not all that different from the value of 2.2◦ inferred from the experimental data, Fig. 2. Some discrepancy is likely due to sequence effects74,75, force-field issues76,77, or the fact that the experiment- based potential in Fig. 2 may itself deviate from reality to some extent, as noted in the original publication32. Importantly, the use of MD-derived θa = 1.5◦ in Eqs. 5 and 6 results, see SI, in virtually the same close agree- ment with the cyclization experiment we have seen Fig. 3, which is based on θa = 2.2◦ derived from experiment. This insensitivity of the prediction of ECH to the value of its key input parameter points again to the robustness of the framework. The following qualitative conclusions can be made from the MD-based analysis of the DNA bending, Fig. 2. For small bending angles, the total energy is reasonably well approximated by a quadratic function. However, once the bending reaches the transi- tion angle θa, the VDW energy decreases at a rate faster than the increase of the other terms combined, which re- sults in a non-convex region of the total E(θ), Fig. 4. It is this sharp decrease in the VDW contribution that gives rise to the existence of a non-convex region in the DNA bending energy. Further analysis, (inset in Fig. 4), reveals that it is the attractive component of the VDW interactions between DNA backbone atoms (backbone- backbone), rather than base stacking, that is critical to the counter-intuitive sharp decrease in the total bending energy, see Supplementary Material for further atomistic details. The key role of the backbone-backbone VDW term suggests that it is the overall structure of DNA, rather than sequence details, that is responsible for the existence of the convex hull in the polymer's effective bending profile; variations in the DNA sequence may al- ter the range of bending angles over which the convex hull exists. It is worth mentioning that local "bubbles" of bro- ken WC bonds do not occur in our atomistic MD sim- ulations of DNA circles where the uniform bending is enforced by constraints on the phosphorous atoms, see "Methods". Yet, these simulations yield a non-convex profile of the DNA bending energy, Fig. 4), which, as we have demonstrated, always leads to the existence of linear "soft" bending regime. Thus, the simulations sug- gest that local DNA melting may not be necessary to explain the high flexibility of strongly bent DNA and the stark deviation of experimental j-factors from WLC pre- dictions. We stress that we do not rule out "bubbles" of broken WC bonds in actual sharply bent DNA; instead, we predict that if WC bond breaking were suppressed experimentally, the qualitative picture of sharply bent DNA being much softer would remain, with the experi- 8 of strongly bent DNA follows the convex hull of E(θ) -- may still hold in a model of protein-DNA complex, Fig. 5 and "Methods". We vary the total positive charge Q of the cylindrical core to modulate the electrostatic attrac- tion of the negatively charged polymer (monomer charge qs) to the core surface of the "protein", and, hence, the degree of the polymer confinement. In the limit of very strong confinement (Q/qs → ∞), the polymer is forced to be confined to a circular, uniformly bent path on the surface of the cylindrical core, and has very few degrees of freedom left to explore in this regime, solid red line in Fig. 5. The average bending energy in this case fol- lows the given functional form of E(θ) (red dashed line in Fig. 5), and ECH clearly does not apply. As we decrease the confinement strength, the polymer is allowed to as- sume non-uniform bending conformations while lowering its total bending energy. The effective bending energy per monomer begins to approach the convex hull (solid green and blue lines), making ECH more applicable. In the case of the weakest confinement (solid purple line), the polymer is still loosely bound to the core, but is al- lowed to relax almost completely. This is the limiting case described by our ECH model: the resulting energy per monomer follows the convex hull fairly closely We argue that it is this low confinement regime, where ECH is relevant, that describes the real nucleosome13 -- arguably the most important DNA-protein complex. To illustrate, we model a "variable confinement nucleosome" by a coarse-grained 147-bp DNA fragment placed next to a cylinder with relative dimensions of the actual histone core67, see Methods; as the core charge Q is increased, the whole fragment starts to wrap around the cylinder once the confinement strength is Q/qs ≥ 90. At this value of the DNA confinement, the energy cost of pulling away a fragment of ∼ 20 bp in our model is ≈ 10kBT , which is comparable to ≈ 6kBT estimated from experiment2 for the fragment of the same length in the actual nucleosome. Moreover, even for higher degrees of confinement, up to Q/qs ≃ 200 (≃ 20kBT to pull away a 20 bp fragment), the corresponding blue lines in Fig. 5 still approximate the convex hull, and so ECH is still likely applicable, at least qualitatively. FIG. 4: The effective DNA bending energy, per base pair, and its physical components as a function of the bending angle θ, inferred from all-atom MD simulations of DNA circles of variable lengths (50-400 bp). The main contribution to the non-convexity of the bending energy comes from the Van der Waals (VDW) interactions. The backbone-backbone part of these interactions contribute the most to the non-convexity due to a sharp increase in the attractive energy component for 3◦ < θ < 4◦, as shown in the inset. For reference, a WLC fit for the small angle bends (≈ 1 − 3.5◦) (grey dashed line) yields the persistence length of 58.2 nm (≈ 172 bp), reasonably close to the experimental value of ≈ 50 nm (≈ 150 bp). mental j-factors still deviating from the WLC in a way qualitatively similar78 to what is currently observed in experiment, Fig. 3. An analogy can be made here with the physics behind the DNA overstretching plateau79,80, where the polymer extension occurs at constant force, and the stretching energy grows linearly with the polymer extension. This peculiar regime can be explained21 via the same main ar- gument used in the current work -- the existence of a non- convex region in the polymer deformation energy. In the case of DNA overstretching, experiments have demon- strated convincingly81 that WC bond breaking is not re- quired for the existence of the characteristic plateau on the force-extension diagram. F. Beyond closed loops: a protein-DNA "complex". IV. CONCLUSION The proposed framework is based on one main assump- tion: despite constraints, the polymer chain is still free to explore sufficient conformational space to search for minimum energy. So far, we focused on DNA loops because of direct connection to key cyclization experi- i θi = α = 2π is min- imally restrictive. However, other realistic scenarios of DNA bending, notably in protein-DNA complexes, may involve different types of constraints that can confine the polymer strongly enough to potentially violate the main assumption to various degrees. Here we investigate to which extent our main conclusion -- deformation energy ments; the single constraint PN It is now well established that slightly bent DNA be- haves like an elastic rod -- the deformation energy is a quadratic (harmonic) function of the deformation. How- ever, recent experiments demonstrated that strong bend- ing of small DNA fragments could no longer be described within this classical model. Here we have proposed a novel framework for bending of polymers, which is based on the consideration of con- vex properties of the effective bending energy between successive monomers. Within the framework, the bend- ing energy is harmonic for small bends, but once the average deformation reaches the convex hull of the ef- 9 energy, leading to the linear bending regime. We use MD simulations only for general reasoning, which is robust to details of the simulation protocol. In this work our focus is the main principle; future re- finements of the ECH theory may be able to account for details not considered here, such as sequence dependence of the DNA bending energy, the influence of torsional stress and supercoiling, etc. We have also just barely touched upon structural consequences of ECH , such as the number and distribution of "kinks" in tightly bent DNA. An analysis of these features will likely lead to more verifiable predictions of the theory. Likewise, we have derived specific mathematical expressions for bend- ing under only one type of constraint; other relevant types of constraints need to be considered in more de- tail to complete the theory. Based on our analysis, the key conceptual features of ECH will likely hold. The new theory does not contradict the conceptual ba- sis of the classical models of DNA bending such as WLC, but also agrees with recent experimental cyclization data on strongly bent small DNA circles19. A completely counter-intuitive prediction that cyclization probability reaches a minimum for very small loops has proved to be consistent with additional experimental data points, not available to us when we made the prediction. We believe that the novel general framework can be used to analyze, at least conceptually, many other scenar- ios of strong polymer bending, and should help interpret future experimental observations. V. FUNDING This work was supported in part by the National Insti- tutes of Health [R01 GM099450] and the National Science Foundation [MCB-1715207]. VI. ACKNOWLEDGMENTS We thank Taekjip Ha for sharing with us unpublished experimental data. We thank Igor Tolokh for many de- tailed and insightful comments. The authors acknowledge Advanced Research Com- puting at Virginia Tech for providing computational re- sources and technical support that have contributed to the results reported within this paper. VII. AUTHOR CONTRIBUTIONS AD, AM and AVO performed the research and wrote the manuscript. AVO designed the research. FIG. 5: Polymer bending in a "protein-DNA complex" model with variable strength of polymer confinement and curvature, see "Methods". The red circle represents the cylindrical charged core of the "protein" to which the oppositely charged "DNA" (black chain) is attracted. Under weak confinement, the system follows the convex hull of the effective E(θ), while approaching E(θ) (red dashed line) itself for strong confinement. Shown is the average energy per bead against the average bending angle θ, at different confinement strengths governed by the ratio Q/qs of the confining charge Q to the opposite charge qs of the confined polymer. The intrinsic bending of the polymer is described by (experimental) E(θ) from Fig. 2. fective bending energy function, a "phase transition" to the strongly bent regime occurs, in which the system's energy is a linear function of the average bending angle. In this regime, which persists for as long as the aver- age deformation is within the convex hull interval, the two states of bending co-exits: some segments are bent weakly, while others are bent strongly, with the propor- tion of the latter increasing with the increased average bend (e.g. shorter loops). The transition point from the harmonic to the linear bending regime occurs at the beginning of the convex hull segment -- this point plays a special role in the new theory. These general consid- erations are expected to hold for any polymer with an effective bending energy that has a distinct non-convex region, regardless of its origin20,24. For generic "sequence-averaged" double-stranded DNA considered here, we conclude that the effective bending deformation energy becomes non-convex for strong bends greater than ∼ 2◦/bp, which corresponds to circular loops shorter than ∼ 160 bp. The conclusion about the DNA bending energy being non-convex relies on an analysis of a large number of experimental protein- DNA complexes, and is consistent with the shape of the bending energy inferred from atomistic MD simulations. The simulations also yield a qualitatively similar value for the bend angle that marks the onset of the linear bend- ing regime. Further, atomistic simulations of DNA circles reveal that the attractive short-range Lennard-Jones in- teractions between the backbone atoms are key for the underlying non-convexity of the DNA effective bending SUPPLEMENTARY MATE- RIAL 10 VIII. CONVEX VS. NON-CONVEX BENDING ENERGY FUNCTIONS Below is an argument for the difference in bending be- havior of polymers characterized by a convex vs. non- convex bending energy. As in the main text, we assume that the polymer deformation energy has only the bend- ing component, which is isotropic. In all cases, bending i θi = N ¯θ = const, where ¯θ is the average bending angle of the polymer chain made of N segments. is induced by constraint PN Without the loss of generality, consider the uniformly bent conformation of a polymer with N = 2, and let's investigate its stability to perturbation. A perturba- tion ∆θ that reduces the bend angle at one site means that the other bending site must increase its bend angle by ∆θ in order to satisfy the constraint N ¯θ = const. This perturbation changes the total energy Echain by E(¯θ + ∆θ) + E(¯θ − ∆θ) − 2E(¯θ). If E(θ) is a con- the black curve in Fig. 6), the per- vex function (e.g. turbed system will have a higher energy than in the initial uniformly bent case. This is because, by defini- tion, a convex curve always lies below its chords, so that 2E(¯θ) < E(¯θ + ∆θ) + E(¯θ − ∆θ). Thus, the perturbation leads to increase in system energy which implies that the system was at a stable equilibrium. Therefore, the min- imum energy conformation of a polymer with a convex effective bending energy is always that of a uniformly bent chain. However, if the function E(θ) has a non-convex region, e.g. the blue curve in Fig. 6, then for any ¯θ in that re- gion, and ∆θ that does not take the system outside of it, 2E(¯θ) > E(¯θ + ∆θ) + E(¯θ − ∆θ) , which means it is possible to lower the energy of the polymer further by non-uniform bending. Namely, one site is now bent through ¯θ − ∆θ, and the other through ¯θ + ∆θ, with the new value of the average chain energy per site, 1 2 Echain falling on the midpoint of the line (dashed red line in Fig. 6) connecting the two new bending states on the en- ergy curve. The longer the cord connecting the two per- turbed states at ¯θ − ∆θ and ¯θ + ∆θ, the larger the energy gain 2E(¯θ) − E(¯θ + ∆θ) − E(¯θ − ∆θ) due to the non- uniform bending, for as long as the cord is completely below the E(θ) curve. The largest energy gain, and thus lowest possible Echain, is achieved for the limiting cord that is the convex hull of E(θ) -- the line segment tangent to the non-convex function at two points, such that be- tween these two points the value of the function is greater FIG. 6: Two different forms for a bending energy profile of a homopolymer. Shown is the (effective) bending energy per site E(θ). If the profile is purely convex down (black curve), the minimal energy conformations of the polymer is uniform bending (all sites are identically bent). If the function has a non-convex region (blue curve), non-uniform bending is more energetically favorable. In this case the total energy of the system follows the convex hull of the energy curve (red line). than that at any point of the line segment. For this lim- iting case, ¯θ − ∆θ = θa, ¯θ + ∆θ = θb. IX. CONCAVE VS. CONVEX CLOSED LOOPS. The exterior angle sum theorem is valid for convex polygons, directly applicable to the sum of the tangent vectors of a convex closed curve. We do not consider non- convex (concave) closed curves here because these can not minimize the deformation energy of a closed inextensible elastic loop, at least as long as the effective bending en- ergy E(θ) is a monotonic function of θ, or curvature κ. (The existence of a non-convex region in E(θ) does not imply a non-monotonic E(θ). ) The outline of an intuitive proof idea is as follows. Any concave curve can be transformed -- "banged out" -- to eliminate a local non-convex region while bringing the bending energy down in the process, see Fig. 7 for an illustration of the process for a shallow "dimple". For a deeper "dimple", the procedure may involve two steps. First, the entire convex portion of the curve is "stretched out" via a uniform re-scaling of its curvature κ(s) → λκ(s), 0 < λ < 1. The bending energy of the con- vex portion of the curve will become lower during this step. Since the curve is inextensible, and there are no breaks, points A and B will move further apart as a re- sult, thus also reducing the curvature of the non-convex "dimple" that spans the AB segment, and hence lower- ing its bending energy. After that, the step in Fig. 7 is applied, making the resulting curve convex, and lowering its bending energy further. We do not pursue a more formal proof based on variational calculus. 11 bp long. The closed loops were created and simulated as discussed with the Main text. The loops were simulated at T = 300K to create 100,000 snapshots. The corre- sponding probability distribution of bend angles for each loop size is shown in Fig. 8. Note that for a large, but finite N ≫ 1, the predictions in Fig. 8 should be interpreted as applicable to the entire conformational ensemble of bent loops; in particular, the predicted linear dependence of the fractions of strongly and weakly bent fragments apply to the ensemble aver- ages. For a given value of ¯θ, between θa and θb, one will observe a distribution of tightly bent fragments among the loops, a given loop can have none or more than one; bent conformations we observed in our MC simulations were qualitatively consistent with the above picture. In this work we did not pursue a detailed analysis of "struc- tural" consequences of ECH model, in part because of an uncertainty associated with the value of θb appropriate for double-stranded DNA. We hope to revisit this issue in the future. FIG. 7: The bending energy of the concave closed curve (top) can be reduced by reflecting the concave portion of the curve across the convex hull line (AB, red), to produce a fully convex curve (bottom). The procedure reduces the bend angles θA and θB to, respectively, θ∗ A = θA − 2α and θ∗ B = θB − 2β, while keeping the curvature unchanged everywhere else along the curve. For a monotonic E(θ) the net result is a lower bending energy of the loop. X. NON-CONVEX BENDING ENERGY FUNCTION LEADS TO BI-MODAL DISTRIBUTION OF BENDING ANGLES One clear consequence of non-convexity of the bending potential (Fig. 6), is that the corresponding distribution of polymer bend angles becomes bi-modal once the av- erage bending angle ¯θ is within the convex hull region. A weak enough bending is always uniform, for as long as the average bend angle ¯θ is below θa. As the average bend angle becomes just slightly larger than θa, most of the segments are still bent weakly through θa, but a small fraction becomes strongly bent through θb. As the constraint forces the system to bend further, the frac- tion of the strongly bent segments increases linearly with ¯θ, until, eventually all the segments are strongly bent through θb. Beyond that point the system re-enters the uniform bending regime again. Here we confirm this ex- pectation quantitatively, for a coarse-grained DNA model with one bead per bp. Specifically, we have analyzed an- gular distribution of different sized loops, from 6 to 600 FIG. 8: Angular probability distributions at 300K resulting from the non-convex bending potential (Fig. 2 of the main text) used in coarse-grained simulations of DNA closed loops of variable size. As the loop size (indicated in the top right corner) decreases, the average bending angle per base pair increases. When the average angle falls into the convex hull range, the angular distribution becomes bi-modal with peaks at θa and θb, corresponding to the weakly and strongly bent states, respectively. The bending through the larger of the two values, θb, can be interpreted as "kinking". The fractional occupancy of both of these states of bending is shown in the inset as a function of the average bend angle ¯θ. Circles: occupancy of the strongly bent state. Squares: occupancy of the weakly bent state. Out-of-plane motion likely affects angular probability distribution of the largest (600 bp) loop, which may explain the shift, compared to expectation, of the position of the corresponding distribution peak. 12 The values of ktop, kbot in the above equations are in- ferred from fitting Eq. 8 to experimental J(L) points; since we have added an oscillatory part to j(L) of Eq. 7, we need two parameters in the functional form of J(L). For the fit we have chosen two experimental data points for loop lengths L = 101 bp (the top envelope of j(L)) and 106 bp (the bottom envelope). The reason we chose J(L) that correspond to the largest L available is because the resulting fit presents the most stringent test for the model in the tight bending regime where L is small. Note that each curve in Fig. 3 of the Main Text, and Figs. 10a, 10b, has its own set of best fits values of ktop, kbot, which explains why the WLC and ECH curves do not coincide in the limit of large loops, where ECH approaches WLC. TABLE I: j-factor ratios, J(L1)/J(L2), predicted using ECH and WLC models compared with experiment19. L1(bp) 40 71 80 90 L2(bp) 50 101 101 101 Experiment ECH 1.50 0.993 2.01×10−1 1.51×10−1 3.28×10−1 2.17×10−1 3.56×10−1 5.59×10−1 WLC 1.12×10−6 3.08×10−6 6.10×10−4 3.73×10−2 The ECH envelope functions, Eqs. 9, have a minimum, which can be derived directly from the expression of j- factor (Eq. 7). The minimum occurs at L = 5 + Lpθ2 2 a 1 4Lp (11) Using the value of Lp = 150 bp and θa = 2.2◦, the mini- mum of j-factor is found at L ≃ 45 bp. No such minimum exists in the WLC case in the range of L of interest to us. Note that ratios of J-factors at integer values of helical repeats can be predicted directly from Eq. 7, which does not contain the oscillatory components. The constant k also cancels, which presents a convenient way to compare ECH theory directly to experiment, Table I. XII. ROBUSTNESS TO MODEL DETAILS The key parameter of the ECH theory is the transi- tion point, θa from the quadratic to the linear bending regime. As seen from Fig. 10, a good agreement with the experimental j-factors is achieved over a relatively broad range of θa values. Specifically, both θa = 2.2◦ inferred from experiment, and θa = 1.5◦ obtained from atomistic MD simulations (see Main Text) yield nearly the same agreement with experiment. In both cases, the counter- intuitive prediction of increased cyclization probability for very short loops hold. Predictions of ECH framework are also robust to the precise value of the helical repeat parameter h used in Eqs. 8 and 7 to account for the torsional stress created by loops with non-integer number of helical repeats. The FIG. 9: Examples of a 60- base-pair (left) and 80- base-pair (right) loop from the conformational ensemble shown in Fig. 8. J-FACTOR ENVELOPE FUNCTIONS AND FITTING XI. TO EXPERIMENT The difference between ECH and WLC models is en- coded in the j-factor via j(L) ≃ k L3 L (cid:19)5 p (cid:18) Lp exp(cid:18)− Eloop(L) kBT + L 4Lp(cid:19) (7) where Eloop is the bending energy of forming a loop within each model, see Main Text. For quantitative com- parison with experiment we follow the standard approach and augment the above formula with an oscillatory term that accounts for periodic variations in J(L) due to tor- sional rigidity. Specifically, the torsional dependence of the j-factor is represented by a cosine function, similar to that of the original work72, to oscillate between the upper and lower envelope functions, with a period of h = 10 bp, see Fig. 3 in the Main Text. J(L) = 1 2hJtop(cid:16)1 + cos(cid:16) 2πL 10 (cid:17)(cid:17) + Jbot(cid:16)1 − cos(cid:16) 2πL 10 (cid:17)(cid:17)i (8) where the "top"/"bottom" envelope curves have the functional form of the j-factor in Eq. 7, Jtop/bot(L) = ktop/bot L (cid:19)5 p (cid:18) Lp L3 e(cid:16)Lpθa(2π− 1 2 Lθa)+ L 4Lp (cid:17) (9) Here we are only interested in loops of length L < 160 bp, since for L > 160 bp ECH = WLC by construction for DNA. Thus, the curve representing ECH in Fig. 3 of the main text is Eq. 8, with the corresponding top/bottom envelope functions given by Eqs. 9. For the case of WLC we replace Eloop(L) with the ap- propriate expression, so Jtop/bot used in Eq. 8 are now given by: Jtop/bot(L) = ktop/bot L (cid:19)5 p (cid:18) Lp L3 e(cid:16)2π2 Lp L + L 4Lp (cid:17) (10) 10-8 10-12 ) M ( r o t c a f - j 10-16 10-20 10-24 10-28 40 10-8 ) M ( r o t c a f - j 10-16 Basepair per turn = 10 Experiment WLC ECH (θ ECH (θ a=2.2) a=1.5) 50 60 70 Length, L (bp) 80 90 100 110 (a) Basepair per turn = 10.5 Experiment WLC ECH (θ ECH (θ a=1.5) a=2.2) 10-24 50 Length, L (bp) 100 (b) FIG. 10: Robustness of ECH model predictions (green curves ) to its input parameters. In (a), j-factors are estimated using base-pair per helical repeat h = 10, and in (b) h = 10.5. In each panel, the ECH predictions are based on two different values of the key input parameter. Solid green lines correspond to θa = 2.2◦ inferred from experiment, and dashed green lines correspond to θa = 1.5◦ from MD simulations. The fitting procedure is the same as described above, the two experimental points used to obtain the asymptotes are indicated by dashed black circles. For (a) these points are the same as in the main text. use of h = 10.5, more appropriate for DNA in solution82, vs. h = 10 corresponding to classical B-form, has little affect on the agreement of the predicted j-factors with the experiment. 13 FIG. 11: The increase of the bending angle of the DNA duplex causes a decrease in the average distance between atom pairs that contribute significantly to the total VDW EV DW energy, modeled here as the Lennard-Jones (LJ) potential; the atoms move deeper into the LJ potential well. The shape of the well is conducive of a sharp decrease in the total VDW energy upon small changes in the atom-atom distance rij caused by the DNA bending. Once the average distance passes the LJ well minimum, the VDW energy begins to increase again. XIII. ORIGIN OF THE NON-CONVEX BENDING AT THE ATOMIC LEVEL To illustrate, at the atomic level, the origin of the non- convex behavior of the backbone-backbone VDW energy, consider pairs of oxygen (O) atoms in the backbone. We choose these atoms because they have one of the lowest VDW energy minima out of all atom pairs in the DNA, and have most atom pairs within the effective short range of the interaction. The O-O VDW energy as a function of pairwise distance rij has a potential well at rmin ∼ 3.3A, Fig. 11. Further analysis of pairwise interatomic dis- tances reveals that as the double helix bends, the geom- etry of the backbone deforms in a way that more oxygen atoms pairs fall into this well (∼ 3.1 − 3.5A): beyond θa the accumulation of the attractive contributions begins to sharply lower the total interaction energy compared to the native, unbent state where all the oxygen atoms are significantly further apart (>4.0A) than the well mini- mum. As the helix bends further, these O-O pairs even- tually pass the LJ minimum, and begin to climb onto the repulsive wall, which increases the total VDW inter- action energy, as seen in the corresponding figure of the main text. 1A. Y. Grosberg, A. R. Khokhlov, and L. W. Jelinski, American Journal of Physics 65, 1218 (1997). 2H. G. Garcia, P. Grayson, L. Han, M. Inamdar, J. Kondev, P. C. Nelson, R. Phillips, J. Widom, and P. A. Wiggins, Biopolymers 85, 115 (2007). 3C. Bustamante, Y. R. Chemla, N. R. Forde, and D. Izhaky, Annual Review of Biochemistry 73, 705 (2004). 14 4P. Nelson, Proceedings of the National Academy of Sciences 96, 14342 (1999). 5K. Van de Velde and P. Kiekens, Polymer Testing 21, 433 (2002). 6J. Gosline, M. Lillie, E. Carrington, P. Guerette, C. Ortlepp, and K. Savage, Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences 357, 121 (2002). 7S. Kasas, A. Kis, B. M. Riederer, L. Forr´o, G. Dietler, S. Catsicas, ChemPhysChem 5, 252 (2004). 8J. P. Peters and L. J. Maher, Quarterly reviews of biophysics 43, 23 (2010). 9J. D. Gralla, Cell 66, 415 (1991). 10T. J. Richmond and C. A. Davey, Nature 423, 145 (2003). 11R. Schleif, Annual review of biochemistry 61, 199 (1992). 12R. Kornberg, Science 184, 868 (1974). 13K. Luger, A. W. Mader, R. K. Richmond, D. F. Sargent, and and T. J. Richmond, Nature 389, 251 (1997). 14S. Henikoff, Nat Rev Genet 9, 15 (2008). 15A. T. Fenley, R. Anandakrishnan, Y. H. Kidane, Onufriev, Epigenetics & Chromatin 11, 11 (2018). and A. V. 42T. E. Cloutier and J. Widom, Proceedings of the National Academy of Sciences 102, 3645 (2005). 43F. Lankas, R. Lavery, and J. H. Maddocks, Structure 14, 1527 (2006). 44A. K. Mazur, Physical review letters 98, 218102 (2007). 45Y. Seol, J. Li, P. C. Nelson, T. T. Perkins, and M. Betterton, Biophysical journal 93, 4360 (2007). 46J. Strauss and L. J. Maher, Science 266, 1829 (1994). 47C. Pr´evost, M. Takahashi, and R. Lavery, ChemPhysChem 10, 1399 (2009). 48A. P. Fields, E. A. Meyer, and A. E. Cohen, Nucleic acids research 41, 9881 (2013). 49R. S. Mathew-Fenn, R. Das, and P. A. Harbury, Science (New York, N.Y.) 322, 446 (2008). 50O. Kratky and G. Porod, Recueil des Travaux Chimiques des Pays-Bas 68, 1106 (1949). 51B. D. Coleman, W. K. Olson, and D. Swigon, The Journal of Chemical Physics 118, 7127 (2003). 52F. Lankas, J. Sponer, J. Langowski, and T. E. Cheatham, Bio- A. Aksimentiev, physical Journal 85, 2872 (2003). 16C. Maffeo, and Nucleic Acids Research 44, 3013 (2016). Yoo, J. 17A. J. Mastroianni, S. A. Claridge, and A. P. Alivisatos, Journal of the American Chemical Society 131, 8455 (2009). 18S. Y. Y. Park, A. K. Lytton-Jean, B. Lee, S. Weigand, G. C. Schatz, and C. A. Mirkin, Nature 451, 553 (2008). 19R. Vafabakhsh and T. Ha, Science 337, 1097 (2012). 20A. Vologodskii and M. D. Frank-Kamenetskii, Nucleic acids re- search 41, 6785 (2013). 21A. V. Savin, I. P. Kikot, M. A. Mazo, and A. V. Onufriev, Pro- ceedings of the National Academy of Sciences 110, 2816 (2013). 22M. Zoli, The Journal of Chemical Physics 148, 214902+ (2018). 23D. Geissler, Sivak and A. P. L. 53S. B. Dixit, D. L. Beveridge, D. A. Case, T. E. Cheatham, E. Giu- dice, F. Lankas, R. Lavery, J. H. Maddocks, R. Osman, H. Skle- nar, et al., Biophysical Journal 89, 3721 (2005). 54S. Fujii, H. Kono, S. Takenaka, N. Go, and A. Sarai, Nucleic acids research 35, 6063 (2007). 55A. K. Mazur and M. Maaloum, Physical review letters 112, 068104 (2014). 56L. Czapla, D. Swigon, and W. K. Olson, Journal of Chemical Theory and Computation 2, 685 (2006), pMID: 26626674, https://doi.org/10.1021/ct060025+. 57J. Yan and J. F. Marko, Physical Review Letters 93 (2004), 10.1103/physrevlett.93.108108. 58R. A. Forties, R. Bundschuh, and M. G. Poirier, The Journal of Chemical Physics 136, 045102+ (2012). 24H. Salari, B. Eslami-Mossallam, S. Naderi, and M. Ejtehadi, The Journal of chemical physics 143, 104904 (2015). 25Y.-Y. Wu, L. Bao, X. Zhang, and Z.-J. Tan, The Journal of Chemical Physics 142, 125103+ (2015). 26A. Travers and J. Thompson, Philosophical Transactions of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences 362, 1265 (2004). 27C. G. Baumann, S. B. Smith, V. A. Bloomfield, and C. Bus- tamante, Proceedings of the National Academy of Sciences 94, 6185 (1997). 28D. M. Crothers, J. Drak, J. D. Kahn, and S. D. Levene, Methods in enzymology 212, 3 (1992). 29M. Vologodskaia and A. Vologodskii, Journal of molecular biol- ogy 317, 205 (2002). 30D. Shore, J. Langowski, and R. L. Baldwin, Proceedings of the National Academy of Sciences 78, 4833 (1981). Nucleic Acids Research 37, 4580 (2009). 59J. S. Mitchell, C. A. Laughton, Nucleic Acids Research 39, 3928 (2011). and S. A. Harris, 60F. H. C. Crick and A. Klug, Nature 255, 530 (1975). 61J. Curuksu, M. Zacharias, R. Lavery, and K. Zakrzewska, Nucleic acids research 37, 3766 (2009). 62T. Macke and D. Case, Washington DC: American Chemcial So- ciety , 379 (1998). 63A. Onufriev, in Modeling Solvent Environments, edited by M. Feig (Wiley, USA, 2010) 1st ed., pp. 127 -- 165. 64A. V. Onufriev and S. Izadi, Wiley Interdisciplinary Reviews: Computational Molecular Science 8, e1347 (2018), https://onlinelibrary.wiley.com/doi/pdf/10.1002/wcms.1347. 65R. Anandakrishnan, A. Drozdetski, R. C. Walker, and A. V. Onufriev, Biophysical journal 108, 1153 (2015), pMCID: PMC4375717. 66H.-J. Limbach, A. Arnold, B. A. Mann, and C. Holm, Computer 31P. J. Hagerman, Annual review of biophysics and biophysical Physics Communications 174, 704 (2006). chemistry 17, 265 (1988). 32Q. Du, C. Smith, N. Shiffeldrim, M. Vologodskaia, and A. Vol- ogodskii, Proceedings of the National Academy of Sciences 102, 5397 (2005). 33M. Frank-Kamenetskii, A. Lukashin, V. Anshelevich, and A. Vol- ogodskii, Journal of Biomolecular Structure and Dynamics 2, 1005 (1985). 34J. Marko and E. Siggia, Macromolecules 27, 981 (1994). 35V. A. Bloomfield, Biopolymers 44, 269 (1997). 36F. Lankas, J. Sponer, P. Hobza, and J. Langowski, Journal of molecular biology 299, 695 (2000). 37T. E. Cloutier and J. Widom, Molecular cell 14, 355 (2004). 38P. A. Wiggins, T. Van Der Heijden, F. Moreno-Herrero, and P. C. A. Spakowitz, R. Phillips, J. Widom, C. Dekker, Nelson, Nature nanotechnology 1, 137 (2006). 39P. Wiggins, R. Phillips, and P. Nelson, Physical Review E 71, 1 (2005). 40M. Biswas, T. Wocjan, J. Langowski, and J. Smith, EPL (Eu- rophysics Letters) 97, 38004 (2012). 41Y. J. Bomble and D. A. Case, Biopolymers 89, 722 (2008). 67A. Fenley, D. Adams, and A. Onufriev, Biophysical Journal 99, 1577 (2010). 68P. J. Flory, Principles of Polymer Chemistry, 1st ed. (Cornell University Press, Ithaca,USA, 1953). 69R. I. Babicheva, K. A. Bukreeva, S. V. Dmitriev, and K. Zhou, Computational Materials Science 79, 52 (2013). 70T. A. Lionberger, D. Demurtas, G. Witz, J. Dorier, T. Lillian, E. Meyhofer, and A. Stasiak, Nucleic acids research 39, 9820 (2011). 71H. Jacobson and W. H. Stockmayer, The Journal of Chemical Physics 18, 1600 (1950). 72J. Shimada and H. Yamakawa, Macromolecules 698, 689 (1984). 73J. F. Allemand, S. Cocco, N. Douarche, and G. Lia, The Euro- pean Physical Journal E 19, 293 (2006). 74R. Lavery, K. Zakrzewska, D. Beveridge, T. C. Bishop, D. A. Case, T. Cheatham, S. Dixit, B. Jayaram, F. Lankas, C. Laughton, et al., Nucleic acids research 38, 299 (2010). Irobalieva, W. Chiu, M. 75Q. Wang, R. N. F. and B. M. Nucleic Acids Research 45, 7633 (2017), L. Zechiedrich, Schmid, Pettitt, J. M. Fogg, http://oup.prod.sis.lan/nar/article-pdf/45/13/7633/22894963/gkx516.pdf. 76T. E. Cheatham and D. A. Case, Biopolymers 99, 969 (2013). 77A. Savelyev and A. D. MacKerell Jr, The journal of physical accounts for broken WC bonds if these occur in the DNA of the complexes. 79S. B. Smith, Y. Cui, and C. Bustamante, Science 271, 795 chemistry letters 6, 212 (2014). (1996). 78Quantitative details can be different if WC bond breaking is sup- pressed. Note that the effective loop bending energy of ECH the- ory in Fig. 3 comes from a statistical analysis of real protein-DNA complexes. Consequently, the ECH effective energy with param- eters used in that figure, θa = 2.2◦ and θb = 35.8◦, implicitly 80J. van Mameren, P. Gross, G. Farge, P. Hooijman, M. Modesti, M. Falkenberg, G. J. L. Wuite, and E. J. G. Peterman, Proc. Natl. Acad. Sci. USA 106, 18231 (2009). 81D. H. Paik and T. T. Perkins, Journal of the American Chemical Society, J. Am. Chem. Soc. 133, 3219 (2011). 82J. C. Wang, Proceedings of the National Academy of Sciences 76, 200 (1979). 15
1102.0604	2	1102	2012-05-11T19:44:09	A small-world of weak ties provides optimal global integration of self-similar modules in functional brain networks	[ "physics.bio-ph", "cond-mat.stat-mech", "cs.SI", "physics.soc-ph", "q-bio.NC" ]	The human brain is organized in functional modules. Such an organization presents a basic conundrum: modules ought to be sufficiently independent to guarantee functional specialization and sufficiently connected to bind multiple processors for efficient information transfer. It is commonly accepted that small-world architecture of short lengths and large local clustering may solve this problem. However, there is intrinsic tension between shortcuts generating small-worlds and the persistence of modularity; a global property unrelated to local clustering. Here, we present a possible solution to this puzzle. We first show that a modified percolation theory can define a set of hierarchically organized modules made of strong links in functional brain networks. These modules are "large-world" self-similar structures and, therefore, are far from being small-world. However, incorporating weaker ties to the network converts it into a small-world preserving an underlying backbone of well-defined modules. Remarkably, weak ties are precisely organized as predicted by theory maximizing information transfer with minimal wiring cost. This trade-off architecture is reminiscent of the "strength of weak ties" crucial concept of social networks. Such a design suggests a natural solution to the paradox of efficient information flow in the highly modular structure of the brain.	physics.bio-ph	physics	A small-world of weak ties provides optimal global integration of self-similar modules in functional brain networks Lazaros K. Gallos and Hern´an A. Makse Levich Institute and Physics Department, City College of New York, New York, NY 10031, USA Mariano Sigman Integrative Neuroscience Laboratory, Physics Department, FCEyN, Universidad de Buenos Aires, Buenos Aires, Argentina Abstract The human brain is organized in functional modules. Such an organization presents a basic conundrum: modules ought to be sufficiently independent to guarantee functional specialization and sufficiently connected to bind multiple processors for efficient information transfer. It is com- monly accepted that small-world architecture of short lengths and large local clustering may solve this problem. However, there is intrinsic tension between shortcuts generating small-worlds and the persistence of modularity; a global property unrelated to local clustering. Here, we present a possible solution to this puzzle. We first show that a modified percolation theory can define a set of hierarchically organized modules made of strong links in functional brain networks. These modules are "large-world" self-similar structures and, therefore, are far from being small-world. However, incorporating weaker ties to the network converts it into a small-world preserving an underlying backbone of well-defined modules. Remarkably, weak ties are precisely organized as predicted by theory maximizing information transfer with minimal wiring cost. This trade-off architecture is reminiscent of the "strength of weak ties" crucial concept of social networks. Such a design suggests a natural solution to the paradox of efficient information flow in the highly modular structure of the brain. 2 1 0 2 y a M 1 1 ] h p - o i b . s c i s y h p [ 2 v 4 0 6 0 . 2 0 1 1 : v i X r a 1 I. INTRODUCTION One of the main findings in Neuroscience is the modular organization of the brain which in turn implies the parallel nature of brain computations [1 -- 3]. For example, in the visual modality, more than thirty visual areas analyze simultaneously distinct features of the vi- sual scene: motion, color, orientation, space, form, luminance and contrast among others [4]. These features, as well as information from different sensory modalities, have to be integrated, as one of the main aspects of perception is its unitary nature [1, 5]. This leads to a basic conundrum of brain networks: modular processors have to be sufficiently isolated to achieve independent computations, but also globally connected to be integrated in coherent functions [1, 2, 6]. A current view is that small-world networks provide a solution to this puzzle since they combine high local clustering and short path length [7 -- 9]. This view has been fueled by the systematic finding of small-world topology in a wide range of human brain networks derived from structural [10], functional [11 -- 13], and diffusion tensor MRI [14]. Small-world topology has also been identified at the cellular- network scale in functional cortical neuronal circuits in mammals [15, 16] and even in the nervous system of the nematode Caenorhabditis elegans [8]. Moreover, small-world property seems to be relevant for brain function since it is affected by disease [17], normal aging and by pharmacological blockade of dopamine neurotransmission [13]. While brain networks show small-world properties, several experimental studies have also shown that they are hierarchical, fractal and highly modular [2, 3, 18, 19]. As there is an intrinsic tension between modular and small-world organization, the main aim of this study is to reconcile these ubiquitous and seemingly contradictory topological properties. Indeed, traditional models of small-world networks cannot fully capture the coexistence of highly modular structure with broad global integration. First, clustering is a purely local quantity which can be assessed inspecting the immediate neighborhood of a node [8]. On the contrary, modularity is a global property of the network, determined by the existence of strongly connected groups of nodes that are only loosely connected to the rest of the network [2, 3, 20, 21]. In fact, it is easy to construct modular and unclustered networks or, reciprocally, clustered networks without modules. Second, the short distances of a small-world may be incompatible with strong modularity which typically presents the properties of a "large-world" [22 -- 28] characterized by long 2 distances which effectively suppress diffusion and free flow in the system [27]. While a clustered network preserves its clustering coefficient when a small fraction of shortcuts are added (converting it into a small-world) [8], the persistence of modules is not equally robust. As we show below, shrinking the network diameter may quickly destroy the modules. Hence, the concept of small-world may not be entirely sufficient to explain the modular and integration features of functional brain networks on its own. We propose that a solution to modularity and broad integration can be achieved by a network in which strong links form large-world fractal modules, in agreement with [2, 3, 18, 19], which are short-cutted by weaker links establishing a small-word network. A modified percolation theory [29, 30] can identify a sequence of critical values of connectivity thresholds forming a hierarchy of modules which progressively merge together. This proposal is inspired by a fundamental notion of sociology termed by Granovetter as "the strength of weak ties" [31, 32]. According to this theory, strong ties (close friends) clump together forming modules. An acquaintance (weak tie) becomes a crucial bridge (a shortcut) between the two densely knit clumps (modules) of close friends [31]. Interestingly, this theme also emerges in theoretical models of large-scale cognitive archi- tecture. Several theories suggest integration mechanisms based on dynamic binding [6, 33] or on a workspace system [1, 34]. For instance, the workspace model [1, 34] proposes that a flexible routing system with dynamic and comparably weaker connections transiently con- nects modules with very strong connections carved by long-term learning mechanisms. II. RESULTS A. Experimental design and network construction We capitalize on a well-known dual-task paradigm, the psychological refractory period [35]. A total of 16 subjects responded with the right hand to a visual stimulus and with the left hand to an auditory stimulus (see SI Appendix). The temporal gap between the auditory and visual stimuli varied in four stimulus onset asynchrony, SOA= 0, 300, 900 and 1200 ms. The sequence of activated regions which unfolds during the execution of the task has been reported in a previous study [36]. The network analysis relies on the time-resolved BOLD-fMRI response based on the phase 3 signal obtained for each voxel of data [37]. We first compute the phase of the BOLD signal for each voxel with methods developed previously [37]. For each subject and each SOA task, we obtain the phase signal of the i-th voxel of activity, φi(t){t=1,..,T}, over T = 40 trials performed for a particular SOA value and subject. We use these signals to construct the network topology of brain voxels based on the equal-time cross-correlation matrix, Cij, where a network link indicates a high cross-correlation in the phase activity of the two voxels (see SI Appendix). The accuracy of the calculated Cij values was estimated through a boot strapping analysis. The 95% confidence interval becomes more narrow for higher Cij values, e.g., for Cij = 0.975 it is (0.9744, 0.9760). The corresponding standard deviation is of the order of 0.003. Thus, we typically distinguish between values that differ by 0.005 (see SI Appendix and Fig. S1). To construct the network, we link two voxels if their cross-correlation Cij is larger than a predefined threshold value p [11, 12, 38]. The resulting network for a given p is a repre- sentation of functional relations among voxels for a specific subject and SOA. We obtain 64 cross-correlation networks resulting from the four SOA values presented to the 16 subjects. B. Percolation analysis Graph analyses of brain correlations relies on a threshold [11] which is problematic since small-world like properties are sensitive to even a small proportion of variation in the con- nections. The present analysis may be seen as an attempt to solve this problem. The thresholding procedure can be naturally mapped to a percolation process (defined in the N × N space of interactions Cij); a model to describe geometrical phase transitions of connected clusters in random graphs, see Chapters 2 and 3 in [29] and [30, 39]. In general, the size of the largest component of connected nodes in a percolation process remains very small for large p. The crucial concept is that the largest connected component increases rapidly through a critical phase transition at pc, in which a single incipient cluster dominates and spans the system [29, 30, 39]. A unique connected component is expected to appear if the links in the network are occupied at random without correlations. However, when we apply the percolation analysis to the functional brain network, a more complex picture emerges revealing a hierarchy of clusters arising from the non-trivial correlations in brain activity. 4 For each participant, we calculate the size of the largest connected component as a func- tion of p. We find that the largest cluster size increases progressively with a series of sharp jumps (Fig. 1A, SOA=900 ms, all participants, other stimuli are similar). This suggests a multiplicity of percolation transitions where percolating modules subsequently merge as p decreases rather than following the typical uncorrelated percolation process with a single spanning cluster. Each of these jumps defines a percolation transition focused on groups of nodes which are highly correlated, constituting well-defined modules. Figure 1B shows the detailed behavior of the jumps in a typical individual (subject la- beled #1 in our dataset [40], SOA=900 ms). At high values of p, three large clusters are formed localized to the medial occipital cortex (red), the lateral occipital cortex (orange) and the anterior cingulate (green). At a lower p = 0.979, the orange and red clusters merge as revealed by the first jump in the percolation picture. As p continues to decrease this mechanism of cluster formation and absorption repeats, defining a hierarchical process as depicted in the top panel of Fig. 1B. This analysis further reveals the existence of "stub- born" clusters. For instance, the anterior cingulate cluster (green), known to be involved in cognitive control [41, 42] and which hence cannot commit to a specific functional module, remains detached from the other clusters down to low p values. Even at the lower values of p, when a massive region of the cortex including motor, visual and auditory regions has formed a single incipient cluster (red, p ≈ 0.94), two new clusters emerge; one involving subcortical structures including the thalamus and striatum (cyan) and the other involving the left frontal cortex (purple). This mechanism reveals the iteration of a process by which modules form at a given p value and merged by comparably weaker links. This process is recursive. The weak links of a given transition become the strong links of the next transition, in a hierarchical fashion. Here, we focus our analysis on the first jump in the size of the largest connected com- ponent, for instance, pc = 0.979 in Fig. 1B. We consider the three largest modules at pc with at least 1,000 voxels each. This analysis results in a total of 192 modules among all participants and stimuli which are pooled together for the next study. An example of an identified module in the medial occipital cortex of subject #1 and SOA=900 ms is shown in Fig. 1C in the network representation and in Fig. 1D in real space. The topography of the modules reflects coherent patterns across the subjects and stimuli as analyzed in SI Appendix (see Fig. S2). 5 FIG. 1. Percolation analysis. (A) Size of the largest connected component of voxels (as measured by the fraction to the total system size) versus p for the 16 subjects (SOA=900 ms). The other three SOA values give similar results. The inset presents a detail around p ≈ 0.95. (B) Detail for a representative individual. As we lower p the size of the largest component increases in jumps when new modules emerge, grow, and finally get absorbed by the largest component. We show the evolution of the modules by plotting connected components with more than 1,000 voxels. The hierarchical tree at the top of the plot shows how clusters evolve by merging with each other. (C) A typical module in network representation. (D) The same module as in (C) embedded in real space - this specific module projects to the medial occipital cortex, see SI Appendix for the spatial projection of all modules. C. Scaling analysis and Renormalization Group To determine the structure of the modules we investigate the scaling of the "mass" of each module (the total number of voxels in the module, Nc) as a function of three length- scales defined for each module: (i) the maximum path length, (cid:96)max, (ii) the average path length between two nodes, (cid:104)(cid:96)(cid:105), and (iii) the maximum Euclidean distance among any two 6 0.940.960.981.00p0.00.10.20.30.40.50.6Fractionofthelargestcomponent0.800.850.900.951.00p0.00.20.40.60.81.0Fractionofthelargestcomponent0.940.950.960.970.00.20.40.60.8ABCDnodes in the cluster, rmax. The path length, (cid:96), is the distance in network space defined as the number of links along the shortest path between two nodes. The maximum (cid:96)max is the largest shortest path in the network. Figure 2A indicates power-law scaling for these quantities [22, 29]. For instance: Nc(rmax) ∼ (rmax)df , (1) defines the Euclidean Hausdorff fractal dimension, df = 2.1± 0.1. The scaling with (cid:96)max and (cid:104)(cid:96)(cid:105) is consistent with Eq. (1) as seen in Fig. 2A. The exponent df quantifies how densely the volume of the brain is covered by a specific module. Next, we investigate the network properties of each module, applying Renormalization Group (RG) analysis for complex networks [22 -- 26]. This technique allows one to observe the network at different scales transforming it into successively simpler copies of itself, which can be used to detect characteristics which are difficult to identify at a specific scale of observation. We use this technique to characterize sub-modular structure within each brain module [2]. We consider each module identified at pc separately. We then tile it with the minimum number of boxes or sub-modules, NB, of a given box diameter (cid:96)B [22], i.e., every pair of nodes in a box has shortest path length smaller than (cid:96)B. Notice that the calculations are performed in network space, where path lengths are defined across the network links without the need for an embedding dimension. Covering the network with minimal NB sub-modules represents an optimization problem which is solved using standard box-covering algorithms, such as the Maximum Excluded Mass Burning algorithm, MEMB, which was introduced in [22, 23, 43] to describe the self similarity of complex networks ranging from the WWW, biological and technical networks (see SI Appendix and Fig. 2B describing MEMB; the code can be downloaded from [40]). The requirement to minimize the number of boxes is important since the resulting boxes are characterized by the proximity between all their nodes and minimization of the links connecting the boxes [27]. Thus, the box-covering algorithm detects boxes/submodules that also tend to maximize modularity. The repetitive application of box-covering at different values of (cid:96)B is a RG transformation [22] that yields a different partition of the brain modules in submodules of varying sizes (Fig. 2B). Figure 2C shows the scaling of NB versus (cid:96)B averaged over all the modules for 7 FIG. 2. Strong ties define fractal modules. (A) Number of voxels or mass of each module, Nc, versus (cid:96)max (red •), (cid:104)(cid:96)(cid:105) (green (cid:4)), and rmax (blue (cid:7)). Each point represents a bin average over the modules for all subjects and stimuli. We use all the modules appearing at pc. The straight lines are fittings according to Eq. (1). The variance is the statistical error over the different modules. The variance is similar in the other data. (B) Detection of submodules and fractal dimension inside the percolation modules. We demonstrate the box-covering algorithm for a schematic network, following the Maximum Excluded Mass Burning algorithm in [22, 43] (SI Appendix). We cover a network with boxes of size (cid:96)B which are identified as sub-modules. We define (cid:96)B as the shortest path plus one. (C) Scaling of the number of boxes NB needed to cover the network of a module versus (cid:96)B yielding dB. We average over all the identified modules for all subjects. (D) Degree distribution averaged over all the brain modules. The individual degree distributions for each module (Fig. S4 and SI Appendix) roughly follow a power law with an average exponent γ = 2.11± 0.04. (E) Dependence of the scaling factor s((cid:96)B), defined through k(cid:48) = s((cid:96)B)k for the renormalized degree k(cid:48), on (cid:96)B. The exponent dk = 1.5 characterizes how the node degree changes during the renormalization process. (F) Quantification of the modularity of the brain modules. The identified percolation modules at pc are composed of submodules with a high level of modularity as can be seen by the scaling of Q((cid:96)B) with (cid:96)B that yields a large modularity exponent dM = 1.9 ± 0.1. Deviations from linear scaling are found at large (cid:96)B due to boundary effects since the network is reduced to just a few submodules. 8 Amax<>,,r(mm)rmaxmax<>100101102103104Modulemass,NCmax100101102OriginalTiledat=2BTiledat=3B10-410-310-210-1100NB/N100B101102=1.9BCB100101102100101102103Modularity,=1.9B()100101102k10-610-510-410-310-210-1100P(k)γ=2.1100B101102=1.5s()10-210-1100BDEFall individuals and stimuli. This property is quantified in the power-law relation [22]: NB((cid:96)B) ∼ (cid:96) −dB B , (2) where dB is the box fractal dimension [22 -- 26] which characterizes the self-similarity between different scales of the module where smaller-scale boxes behave in a similar way as the original network. The resulting dB averaged over all the modules is dB = 1.9 ± 0.1. D. Morphology of the brain modules The RG analysis reveals that the module topology does not have many redundant links and it represents the quantitative statement that the brain modules are "large-worlds". However this analysis is not sufficient to precisely characterize the topology of the modules. For example, a two-dimensional complex network architecture and a simple two-dimensional lattice are compatible with the scaling analysis and the value of the exponents described in the previous section. To identify the network architecture of the modules we follow established analysis [18, 44] based on the study of the degree distribution of the modules, P (k), and the degree-degree correlation P (k1, k2). The form of P (k) distinguishes between a Euclidean lattice (delta function), an Erdos-Renyi network (Poisson) [30], or a scale-free network (power-law) [44]. We find that the brain modules have a broad degree distribution [11, 44] with an approximate power-law P (k) ∼ k−γ. The statistical analysis provides strong evidence for a power law form and rules out exponential decay (see SI Appendix). In Fig. S4 we present a number of P (k) curves for different modules, along with their best fittings. In the SI Appendix we describe the calculation method that takes into account all the clusters and finally yields an average exponent γ = 2.11 ± 0.04. An 'average' curve for the distribution is shown in Fig. 2D, where the exponent γ is not a direct fit to this curve, but instead represents the result of the accurate calculation. This result indicates that the modules have a scale-free fractal structure, different from a simple two-dimensional lattice, where P (k) should be a narrow function. The embedding of scale-free networks in a finite-dimension real space constitutes a prob- lem which has attracted recent attention [45 -- 47]. Scale-free networks may arise from a 2-dimensional lattice with added dense connectivity locally, where the weights and connec- 9 tivity are inversely proportional to the Euclidean distance on the lattice. To investigate this possibility we study the correlation function of the phases of the voxels as a function of Euclidean distance in real space: C(r) = (cid:104)cos(φ1 − φ2)(cid:105) versus r = (cid:126)r1 − (cid:126)r2. This function can be interpreted as the correlation between two spins with orientation determined by the phase φi of the voxel at location (cid:126)ri (average is over all pairs at distance r). We find (see SI Appendix and Fig. S3) that C(r) decays algebraically with distance. Thus, our results indicate that modules are scale-free networks which can be embedded in a lattice with an added long-range connectivity. How can fractal modularity emerge in light of the scale-free property, which is usually associated with small-worlds [18]? In a previous study [23], we introduced a model to account for the simultaneous emergence of scale-free, fractality and modularity in real networks by a multiplicative process in the growth of the number of links, nodes and distances in the network. The dynamic follows the inverse of the RG transformation [23] where the hubs acquire new connections by linking preferentially with less connected nodes rather than other hubs. This kind of "repulsion between hubs" [24] creates a disassortative structure -- with hubs spreading uniformly in the network and not crumpling in the core as in scale- free models [44]. Hubs are buried deep into the modules, while low degree nodes are the intermodule connectors [24]. A signature of such mechanism can be found by following hubs' degree during the renor- malization procedure. At scale (cid:96)B, the degree of a hub k changes to the degree of its box k(cid:48), through the relation k(cid:48) = s((cid:96)B)k. The dependence of the scaling factor s((cid:96)B) on (cid:96)B defines the renormalized degree exponent dk by s((cid:96)B) ∼ (cid:96) [22]. Scaling theory defines precise relations between the exponents for fractal networks [22], through γ = 1 + dB/dk. For the case of brain modules analyzed here (Fig. 2E), we find dk = 1.5 ± 0.1. Using the values of dB and dk for the brain clusters, the prediction is γ = 2.26 ± 0.11, which is close to the calculated value of γ = 2.11 ± 0.04 from Fig. 2D. −dk B The previous analysis reveals the mechanism of formation of a scale-free network, but it does not assure a fractal topology. Fractality can be determined from the study of the degree-degree correlation through the distribution, P (k1, k2) to find a link between nodes with (k1, k2) degree. This correlation characterizes the hub-hub repulsion through scaling exponents de and (see SI Appendix and Fig. S5) [23, 48]. In a fractal, they satisfy = 2 + de/dk. A direct measurement of these exponents yields de = 0.51± 0.08 and = 2.1± 0.1 10 (Fig. S5). Using the measured values of de and dk, we predict = 2.3± 0.1, which is close to the observed exponent. Taken together, these results indicate a scale-free fractal morphology of brain modules. Such structure is in agreement with previous results of the anatomical connectivity of the brain [2, 3] and functional brain networks [11]. E. Quantifying submodular structure of brain modules Standard modularity decomposition methods [20, 21] based on maximization of the mod- ularity factor Q as defined in [2, 20, 21, 27, 28] are particularly suitable to uncover the submodular structure. For example, the Girvan-Newman method [20] yields a value of Q ∼ 0.82 for the brain clusters, indicating a strong modular substructure. The box covering algorithm benefits from detecting submodules (the boxes) at different scales. Then, we can study the hierarchical character of modularity [2, 27, 28], and detect whether modularity is a feature of the network that remains scale-invariant (see SI Appendix and Fig. S6 for a comparison of the submodular structure obtained using Girvan-Newman and box covering). The minimization of NB guarantees a network partition with the largest number of in- tramodule links and the fewest intermodule links. Therefore, the box covering algorithm maximizes the following modularity factor [27, 28]: NB(cid:88) i=1 Q((cid:96)B) ≡ 1 NB Lin i Lout i , (3) which is a variation of the modularity factor, Q, defined in [20, 21]. Here, Lin i and Lout represent the intra and intermodular links in a submodule i, respectively. Large values of i → 0) correspond to high modularity [27]. We make the whole modularization Q (i.e. Lout method available at [40]. i Figure 2F shows the scaling of Q((cid:96)B) averaged over all modules at percolation revealing a monotonic increase with a lack of a characteristic value of (cid:96)B. Indeed, the data can be fitted with a power-law form [27]: Q((cid:96)B) ∼ (cid:96)dM B , (4) which is detected through the modularity exponent, dM . We study the networks for all the subjects and stimuli and find dM = 1.9 ± 0.1 (Fig. 2F). The lack of a characteristic length- scale expressed in Eq. (4) implies that submodules are organized within larger modules such 11 FIG. 3. Transition from fractal to small-world networks. (A) Left panel shows a typical percolation module in network space. The colors identify sub-modules obtained by the box-covering algorithm with (cid:96)B = 15. This fractal module contains 4097 nodes with (cid:104)(cid:96)(cid:105) = 41.7, (cid:96)max = 139, and rmax = 136 mm. When a small fraction prew of the links are randomly rewired [8], the modular structure disappears together with the shrinking path length. The rewiring method starts by selecting a random link and cutting one of its edges. This edge is then rewired to another randomly selected node, and another random link starting from this node is selected. This is again cut and rewired to a new random node, and we repeat the process until we have rewired a fraction prew of links. The final link is then attached to the initially cut node, so that the degree of each node remains unchanged. (B) Small-world cannot coexist with modularity. The large diameter and modularity factor, Eq. (4) for (cid:96)B = 15, of the fractal module in (A) (left panel) diminish rapidly upon rewiring a tiny fraction prew ≈ 0.01 of links, while the clustering coefficient still remains quite large. (C) The transition from fractal to small-world to random structure is shown when we plot the mass versus the average distance for all modules for different prew values as indicated. The crossover from power-law fractal to exponential small-world/random is shown. 12 Large-worldfractalnetworkmaxmaxSmall-worldnetworkRandomnetworkp=rew0prewp=rew0.01p=rew0.1p=rewAB10-110-210-310-4100C0.00.20.40.60.81.0C/C(0)(0)(0)//prew100101102103104Modulemass,NC<>10010110200.010.11that the inter-connections between those submodules repeat the basic modular character of the entire brain network. The value of dM reveals a considerable modularity in the system as it is visually apparent in the sample of Fig. 3A, left panel, where different colors identify the submodules of size (cid:96)B = 15 in a typical fractal module. For comparison, a randomly rewired network (Fig. 3A, right and central panels) shows no modularity and has dM ≈ 0. Scaling analysis indicates that dM is related to Lout ∼ (cid:96)dx B , which defines the outbound exponent dx characterizing the number of intermodular links for a submodule [27] (dx is related to the Rent exponent in integrated circuits [3]). From Eq. (4), we find: dM = dB − dx, which indicates that the strongest possible modular structure has dM = dB (dx = 0) [27]. Such a high modularity induces very slow diffusive processes (subdiffusion) for a random walk in the network [27]. Comparing Eq. (4) with (2), we find dx = 0, which quantifies the maximum degree of modularity in the brain modules. F. Small-world or large-world fractal modularity An important consequence of Eqs. (1) and (2) is that the network determined by the strong links above the first pc-jump lacks the logarithmic scaling characteristic of small- worlds and random networks [8]: (cid:104)(cid:96)(cid:105) ∼ log Nc, (5) A fractal network poses much larger distances than those appearing in small-worlds [22]: a distance (cid:96)max ∼ 100 observed in Fig. 2A (red curve) would require an enormous small-world network Nc ∼ 10100, rather than Nc ∼ 104, as observed for fractal networks in Fig. 2A. The structural differences between a modular fractal network and a small-world (and a random network) are starkly revealed in the panels of Fig. 3A. We rewire the fractal module on the left panel by randomly reconnecting a fraction prew of the links while keeping the degree of each node intact [8]. Figure 3B quantifies the transition from fractal (prev = 0) to small-world (prev ≈ 0.01 − 0.1) and eventually to random networks (prev = 1), illustrated in Fig. 3A: we plot (cid:96)max(prew)/(cid:96)max(0), the clustering coefficient C(prew)/C(0) and Q(prew)/Q(0) for a typ- ical (cid:96)B = 15 as we rewire prew links in the network. As we create a tiny fraction prew = 0.01 of short-cuts, the topology turns into a collapsed network with no trace of modularity left, 13 FIG. 4. Weak ties are optimally distributed. (A) Three modules identified at pc = 0.98 for the subject in Fig. 1B. The colors correspond to different submodules as identified by the box covering algorithm at (cid:96)B = 21. (B) When we lower the threshold to p = 0.975, weak ties connect the modules. The three original modules as they appear in (A) are plotted in red, orange and purple and the light blue nodes are the nodes added from (A) as we lower p. Blue lines represent the added weak links with distance longer than 10 mm. The weak links collapse the three modules into one. (C) Sketch of the different critical values of the shortcut exponent α in comparison with df . (D) Cumulative probability distribution P (rij > r). The straight line fitting yields an exponent α − 1 = 2.1 ± 0.1 indicating optimal information transfer with wiring cost minimization [50]. Certain clusters occupy two diametric parts of the brain. In practice, these are two modules that are connected through long-range links. These links increase significantly the percentage of links at large distances rij, since they are superimposed on top of the regular distribution of links within unfragmented clusters. This behavior is manifested as a bump in the curve. while C(0.01) still remains quite high (Fig. 3B). The rewired networks present the expo- nential behavior of small-worlds [8], and also random networks as prev increases, obtained 14 ABp=0.98p=0.975CD2(~4.2)+1(~3.1)(~2.1)Small-worldFractalOptimizationwithlocalpositionknowledge(Milgram,Internet)Wiringcostoptimization,withglobalroutingknowledge(brain,airportnetwork)100101102r(mm)10-410-310-210-1100P(rij>r)-1=2.1from Eq. (5): Nc ∼ exp(cid:0)(cid:104)(cid:96)(cid:105)/(cid:96)0 (cid:1), (6) where Nc is averaged over all the modules (Fig. 3C). The characteristic size is very small and progressively shrinks to (cid:96)0 = 1/7 when prew = 1. The hallmark of small-worlds and random networks, exponential scaling Eq. (6), is incompatible with the hallmark of fractal large-worlds, power-law scaling Eq. (2). Similarly, while we find a broad domain where short network distances coexist with high clustering forming a small-world behavior, modularity does not show such a robust behavior to the addition of shortcuts. G. Short-cut wiring is optimal for efficient flow Figure 3B suggests that modularity and small-world cannot coexist at the same level of connectivity strength. Next, we set out to investigate how the small-world emerges. When we extend the percolation analysis lowering further the threshold p below pc, weaker ties are incorporated to the network connecting the self-similar modules through short-cuts. A typical scenario is depicted in Fig. 4A showing the three largest percolation modules identified just before the first percolation jump in the subject #1 shown in Fig. 1B at p = 0.98. For this connectivity strength, the modules are separated and show submodular fractal structure indicated in the colored boxes obtained with box covering. When we lower the threshold to p = 0.975, Fig. 4B, the modules are now connected and a global incipient component starts to appears. A second global percolation-like transition appears in the system when the mass of the largest component occupies half of the activated area (see e.g. Fig. 1). For different individuals, global percolation occurs in the interval p = [0.945, 0.96] as indicated in the inset of Fig. 1A. Our goal is to investigate whether the weak links shortcut the network in an optimal manner. When the cumulative probability distribution to find a Euclidean distance between two connected nodes, rij, larger than r follows a power-law: P (rij > r) ∼ r−α+1, (7) statistical physics makes precise predictions about optimization schemes for global function as a function of the short-cut exponent α and df [26, 49, 50]. Specifically, there are three critical values for α as shown schematically in Fig. 4C. If α is too large then shortcuts will not 15 be sufficiently long and the network will behave as fractal, equal to the underlying structure. Below a critical value determined by α < 2df [26], shortcuts are sufficient to convert the network in a small world. Within this regime there are two significant optimization values: (i) Wiring cost minimization with full routing information. This considers a network of dimension df , over which short-cuts are added to optimize communication, with a wiring cost constraint proportional to the total shortcut length. It is also assumed that coordinates of the network are known, i.e. it is the shortest path that it is being minimized. Under these circumstances, the optimal distribution of shortcuts is α = df + 1 [50]. This precise scaling is found in the US airport network [51] where a cost limitation applies to maximize profits. (ii) Decentralized Greedy searches with only local information. This corresponds to the classic Milgram's "small-world experiment" of decentralized search in social networks [49], where a person has knowledge of local links and of the final destination but not of the intermediate routes. Under these circumstances, which also apply to routing packets in the Internet, the problem corresponds to a greedy search, rather than to optimization of the minimal path. The optimal relation for greedy routing is α = df [26, 49]. Hence, the analysis of P (rij > r) provides information both on the topology of the resulting network and on which transport procedure is optimized. This distribution reveals power-law behavior Eq. (7) with α = 3.1 ± 0.1 when averaged over the modules below pc (Fig. 4D). Given the value obtained in Eq. (1), df = 2.1, this implies that the network composed of strong and weak links is small-world (α < 2df ) [26] and optimizes wiring cost with full knowledge of routing information (α = df + 1) [50]. III. DISCUSSION The existence of modular organization which become small-world when short-cut by weaker ties is reminiscent of the structure found to bind dissimilar communities in social networks. Granovetter's work in social sciences [31, 32] proposes the existence of weak ties to cohese well-defined social groups into a large-scale social structure. The observation of such an organization in brain networks suggests that it may be a ubiquitous natural solution to the puzzle of information flow in highly modular structures. Over the last decades, wire length minimization arguments have been used successfully to explain the architectural organization of brain circuitry [52 -- 56]. Minimizing wire length 16 is in fact of paramount importance, since about 60% of the cortical volume is taken up by wire (axons and dendrites) [57]. This turns out to optimize conduction rate, posing a strict packing limitation of the amount of wire in cortical circuits [57]. Our finding of a distribution of weak links which minimizes wiring cost is hence in line with a previous literature, consistently showing that neural circuit design is under pressure to minimize wiring length. However, some important nuances of the specific optimization procedure ought to be considered. First, we specifically showed that at the mesoscopic scale, short-cut distribution optimizes wiring cost while maintaining network proximity. This is consistent with the organization of large-scale neural networks in which total wiring can in fact be decreased by about 32% (in 95 primate cortical areas) and up to 48% in the global neuronal network of the nematode Caenorhabditis elegans [58]. This extra wiring cost comes from long-range connections which achieve network benefits of shortening the distance between pairs of nodes [58]. Our results are in agreement with this observation, suggesting that simultaneous op- timization of network properties and wiring cost might be a relevant principle of brain architecture. In simple words, this topology does not minimize the total wire per-se, simply to connect all the nodes; instead it minimizes the amount of wire required to achieve the goal of shrinking the network to a small-world. A second intriguing aspect of our results, which is not usually highlighted, is that this minimization assumes that broadcasting and routing information are known to each node. How this may be achieved -- what aspects of the neural code convey its own routing information -- remains an open question in Neuroscience. BOLD fMRI is an indirect measure of brain activity which relies on multiple vascular and biophysical factors which couple the neural response to the haemodynamic signal [59]. Even if in fMRI research it is always assumed that haemodynamic signals reflect metabolic demand generated by local neuronal activity, recent studies have shown reliable haemody- namic signals that entrains to task structure independently of standard neural predictors of haemodynamics [60]. Hence, our results, as any other fMRI analysis, have to be taken cautiously and may partly reflect the underlying structure of vascular motives. Specifically, the human cortical vascular system has a large number of arterial anastomoses which show a seemingly looking fractal structure in the mm to cm range [61]. Precise measurements of fractality have been reported at the micrometer scales in volumes of the order of a few mm3 [62, 63], which corresponds to approximately a voxel volume, where branching structure of 17 microcapilarities then generates fractals. Hence, it is possible that the fractal organization of brain modules is inherited from the vascular system itself. Although we cannot readily test the influence of the vascular system at a large scale, it is still possible to address this concern at a microscopic scale, by discarding neighboring correlations. Neighboring voxels are expected to carry some shared signal due to spatial autocorrelations from the microvascular network. To assure that our results do not rely on neighbouring correlations which might be particularly spurious, we coarse-grained the original fMRI signal by doubling the lattice spacing, reducing the number of voxels by a factor of 8 and repeat the calculations. The results are consistent with the percolation picture of fractal modules, albeit with an expected lower pc. Such a renormalized pc is expected from renormalization theory to change under coarse-graining, while the main results on long-range links, such as the value of the exponents, are insensitive to this type of coarse-graining. We also investigate whether the map of fractal dimension dB reflects a meaningful orga- nization based on known facts of functional properties of the cortex and the specific task which subjects are performing. We found a topographical organization of fractality in the human brain (Fig. S7). The right portion of the anterior cingulate, SMA and the right PPC regions involved in routing of information and cognitive control [41, 42], which are ex- pected to have a more complex functional organization, are the clusters with higher fractal dimension. The left-right asymmetry is interesting since, in this specific task, the left hand response is queued for a few hundred millisencods and has to be temporally connected to working memory and inhibitory circuits. While not fully conclusive, this analysis suggests a functional role of the network architectures described here. Another similar concern is that the recovered brain modules may be a manifestation of the fractal structure of the underlying three-dimensional vortex grid or of the cortex. However, since the dimensions of the grid (d = 3) and of the cortex (d = 2.7)[64] are both sufficiently different from 1.9 and the connectivity distribution of the modules is much broader than the typical Euclidean fractal cortex (which should be narrow around k ∼ 6) or a 3d lattice (k = 6), we may safely assume that these objects have their own structure. Moreover, we also observed modules with similar fractal dimension in subcortical structures suggesting that these results do not simply reflect anatomical properties of the cortical mantle. A hierarchical modular organization of the brain composed of modules within mod- ules has been invoked in [2, 3] to describe the brain structure. The present results sup- 18 port these previous findings, while, at the same time, provide a new view by integrating the results with the (non-critical) properties of small-worlds and placing self-similarity in the framework of scaling theory, universality and Renormalization Groups [65]. In this framework, brain modules are characterized by a set of novel scaling exponents, the sep- tuplet: (df , dB, dk, de, dM , γ, α) = (2.1, 1.9, 1.5, 0.5, 1.9, 2.1, 3.1), and the scaling relations dM = dB − dx, relating fractality with modularity, α = df + 1, relating global integration with modularity, γ = 1+dB/dk, relating scale-free with fractality, and = 2+de/dk, relating degree correlations with fractality. One advantage of this formalism is that the different brain topologies can be classified into universality classes under RG [65] according to the septuplet (df , dB, dk, de, dM , γ, α). Universality applies to the critical exponents but not to quantities like (pc, C, (cid:96)0) which are sensitive to the microscopic details of the different experimental situations [65]. In this framework, (non-critical) small-worlds are obtained in the limit (df , dB, dk, de, dM , dx) → (∞,∞,∞, 0, 0,∞). A path for future research will be to test the universality of the septuplet of exponents under different activities covering other areas of the brain, e.g., the resting-state correlation structure [66]. In conclusion, we propose a formal solution to the problem of information transfer in the highly modular structure of the brain. The answer is inspired by a classic finding in sociology: the strength of weak ties [31]. The present work provides a general insight into the physical mechanisms of network information processing at large. It builds up on an example of considerable relevance to natural science, the organization of the brain, to establish a concrete solution to a broad problem in network science. The results can be readily applied to other systems -- where the coexistence of modular specialization and global integration is crucial -- ranging from metabolic, protein and genetic networks to social networks and the Internet. ACKNOWLEDGMENTS LKG and HAM thank the NSF-0827508 Emerging Frontiers Program for financial sup- port. MS is supported by a Human Frontiers Science Program Fellowship. We thank D. 19 Bansal, S. Dehaene, S. Havlin, and H.D. Rozenfeld for valuable discussions. [1] Dehaene S, Naccache L (2001) Towards a cognitive neuroscience of consciousness: basic evi- dence and a workspace framework. Cognition 79:1-37. [2] Meunier D, Lambiotte R, Bullmore ET (2010) Modular and hierarchically modular organiza- tion of brain networks. Front. Neurosci. 4:200. [3] Bassett DS, et al. (2010) Efficient physical embedding of topologically complex information processing networks in brains and computer circuits. PLoS Comput. Biol. 6:e1000748. [4] Felleman DJ, van Essen DC (1991) Distributed hierarchical processing in the primate cerebral cortex. Cereb. Cortex 1:1-47. [5] Treisman A (1996) The binding problem. Current opinion in neurobiology 6:171-178. [6] Tononi G, Sporns O, Edelman GM (1994) A measure for brain complexity: relating functional segregation and integration in the nervous system. Proc. Nat. Acad. Sci. USA 91:5033-5037. [7] Sporns O, Chialvo DR, Kaiser M, Hilgetag CC (2004) Organization, development and function of complex brain networks. Trends Cognit. Sci. 8:418-425. [8] Watts D, Strogatz S (1998) Collective dynamics of 'small-world' networks. Nature 393:440- 442. [9] Bassett DS, Bullmore ET (2006) Small-world brain networks. Neuroscientist 12:512-523. [10] He Y, Chen ZJ, Evans AC (2007) Small-world anatomical networks in the human brain re- vealed by cortical thickness from MRI. Cereb. Cortex 17:2407-2419. [11] Eguiluz VM, Chialvo DR, Cecchi GA, Baliki M, Apkarian AV (2005) Scale-free brain functional networks. Phys. Rev. Lett. 94:018102. [12] Achard S, Salvador R, Whitcher B, Suckling J , Bullmore ET (2006) A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs. J. Neurosci. 26:63-72. [13] Achard S, Bullmore ET (2007) Efficiency and cost of economical brain functional networks. PLoS Comput. Biol. 3:e17. [14] Hagmann P, et al. (2007) Mapping human whole-brain structural networks with diffusion MRI. PLoS One 2:e597. [15] Song S, Sjostrom PJ, Reigl M, Nelson S, Chklovskii DB (2005) Highly nonrandom features of 20 synaptic connectivity in local cortical circuits. PLoS Biol. 3:e68. [16] Yu S, Huang D, Singer W, Nikolic D (2008) A small world of neuronal synchrony. Cereb. Cortex 18:2891-2901. [17] Stam CJ, Jones BF, Nolte G, Breakspear M, Scheltens P (2007) Small-world networks and functional connectivity in Alzheimer's disease. Cereb. Cortex 17:92-99. [18] Ravasz E, Somera AL, Mongru DA, Oltvai ZN, Barab´asi A-L (2002) Hierarchical organization of modularity in metabolic networks. Science 297:1551-1555. [19] Palla G, Derenyi I, Farkas I, Vicsek T (2005) Uncovering the overlapping community structure of complex networks in nature and society. Nature 435:814-818. [20] Girvan M, Newman MEJ (2002) Community structure in social and biological networks. Proc. Nat. Acad. Sci. 99:7821-7826. [21] Fortunato S (2010) Community detection in graphs. Phys. Rep. 486:75-174. [22] Song C, Havlin S, Makse HA (2005) Self-similarity of complex networks. Nature 433:392-395. [23] Song C, Havlin S, Makse HA (2006) Origins of fractality in the growth of complex networks. Nature Phys. 2:275-281. [24] Goh KI, Salvi G, Kahng B, Kim D (2006) Skeleton and fractal scaling in complex networks Phys. Rev. Lett. 96:018701. [25] Radicchi F, Ramasco JJ, Barrat A, Fortunato S (2008) Complex networks renormalization: flows and fixed points. Phys. Rev. Lett. 101:148701. [26] Rozenfeld HD, Song C, Makse HA (2010) The small world-fractal transition in complex net- works: A renormalization group approach. Phys. Rev. Lett. 104:025701. [27] Gallos LK, Song C, Havlin S, Makse HA (2007) Scaling theory of transport in complex bio- logical networks. Proc. Nat. Acad. Sci. USA 104:7746-7751. [28] Galvao V, et al. (2010) Modularity map of the human cell differentiation network. Proc. Nat. Acad. Sci. USA 107:5750-5755. [29] Bunde A, Havlin S, editors (1996) Fractals and Disordered Systems 2nd edition (Springer- Verlag, Heidelberg). [30] Bollob´as B (1985) Random Graphs (Academic Press, London). [31] Granovetter MS (1973) The strength of weak ties. American Journal of Sociology 78:1360- 1380. [32] Onnela J-P, et al. (2007) Structure and tie strengths in mobile communication networks. Proc. 21 Nat. Acad. Sci. USA 104:7332-7336. [33] Tononi G, Edelman GM (1998) Consciousness and complexity. Science 282:1846-1851. [34] Baars BJ (1997) In the theater of consciousness: The workspace of the mind. (Oxford Uni- versity Press, USA). [35] Pashler H (1994) Dual-task interference in simple tasks: Data and theory. Psychological Bul- letin 116:220-220. [36] Sigman M, Dehaene S (2008) Brain mechanisms of serial and parallel processing during dual- task performance. J. Neurosci. 28:7585-7598. [37] Sigman M, Jobert A, Dehaene S (2007) Parsing a sequence of brain activations of psychological times using fMRI. Neuroimage 35:655-668. [38] Salvador R, et al. (2005) Neurophysiological architecture of functional magnetic resonance images of human brain. Cereb. Cortex 15:1332-1342. [39] Stanley HE (1971) Introduction to phase transitions and critical phenomena. (Oxford Univer- sity Press, Oxford). [40] The entire experimental dataset and modularization and fractal codes are available at http://lev.ccny.cuny.edu/∼hmakse/brain.html [41] Zylberberg A, Dehaene S, Roelfsema PR and Sigman M (2011) The human Turing machine: a neural framework for mental programs. Trends in Cognitive Sciences 15: 293-300. [42] Duncan, J (2010) The multiple-demand (MD) system of the primate brain: mental programs for intelligent behaviour. Trends in Cognitive Sciences 14:172-179. [43] Song C, Gallos LK, Havlin S, Makse HA (2007) How to calculate the fractal dimension of a complex network: the box covering algorithm. J. Stat. Mech. P03006. [44] Barab´asi A-L, Albert R (1999) Emergence of scaling in random networks. Science 286:509-512. [45] De Arcangelis L, Herrmann HJ (2010) Learning as a phenomenon occurring in a critical state. Proc. Nat. Acad. Sci. USA 107:3977-3981. [46] Li D, Li G, Kosmidis K, Stanley HE, Bunde A, Havlin S (2011) Percolation of spatially constraint networks. Euro. Phys. Lett. 93:68004. [47] Daqing L, Kosmidis K, Bunde A, Havlin S (2011) Dimension of spatially embedded networks. Nature Phys. 7: 481 -- 484. [48] Gallos LK, Song C, Makse HA (2008) Scaling of degree correlations and its Influence on diffusion in scale-free networks, Phys. Rev. Lett. 100:248701. 22 [49] Kleinberg J (2000) Navigation in a small world. Nature 406:845. [50] Li G, Reis SDS, Moreira AA, Havlin S, Stanley HE, Andrade Jr JS (2010) Towards design principles for optimal transport networks. Phys. Rev. Lett. 104:018701. [51] Bianconi G, Pin P, Marsili M (2009) Assessing the relevance of node features for network structure. Proc. Nat. Acad. Sci. USA 106:11433-11438. [52] Cowey A (1979) Cortical maps and visual perception: The Grindley Memorial Lecture. The Quarterly Journal of Experimental Psychology, 31:1-17. [53] Linsker R (1986) From basic network principles to neural architecture: emergence of spatial- opponent cells. Proceedings of the National Academy of Sciences 83:7508. [54] Mitchison G (1991) Neuronal branching patterns and the economy of cortical wiring. Proceed- ings of the Royal Society of London B 245:151. [55] Cherniak C (1995) Neural component placement. Trends in neuroscience 18:522-527. [56] Chklovskii DB (2000) Optimal sizes of dendritic and axonal arbors in a topographic projection. Journal of neurophysiology 83:2113. [57] Chklovskii DB, Schikorski, T and Stevens, CF (2002) Wiring optimization in cortical circuits Neuron 34: 341 -- 347 [58] Kaiser M, Hilgetag CC (2006) Nonoptimal component placement, but short processing paths due to long-distance projections in neural systems. PLoS Comput. Biol. 2:e95. [59] Logothetis NK, Pauls J, Augath M, Trinath T, and Oeltermann A (2001) Neurophysiological investigation of the basis of the fMRI signal. Nature 412:150-157. [60] Sirotin YB and Das A (2009) Anticipatory haemodynamic signals in sensory cortex not pre- dicted by local neuronal activity. Nature 457:475-479. [61] Duvernoy HM, Delon S and Vannson JL (1981) AnticipCortical blood vessels of the human brain. Brain Research Bulletin 7:519-579. [62] Cassot F, Lauwers F, Fouard C, Prohaska S, and Lauwers-Cances V (2006) A novel three- dimensional computer-assisted method for a quantitative study of microvascular networks of the human cerebral cortex. Microcirculation 13:1-18. [63] Risser L, Plourabou´e F, Steyer A, Cloetens P, Le Duc G, and Fonta C (2006) From homoge- neous to fractal normal and tumorous microvascular networks in the brain. Journal of Cerebral Blood Flow & Metabolism 27:293-303. [64] Kiselev VG, Hahn KR, Auer DP (2003) Is the brain cortex a fractal? NeuroImage 20:1765- 23 1774. [65] Stanley HE (1999) Scaling, universality, and renormalization: Three pillars of modern critical phenomena. Rev. Mod. Phys. 71:S358-S366. [66] Raichle ME, et al. (2001) A default mode of brain function. Proc. Natl. Acad. Sci. USA 98:676-682. 24 SUPPORTING INFORMATION I. FMRI METHODS AND NETWORK CONSTRUCTION A total of 16 participants (7 women and 9 men, mean age, 23, ranging from 20 to 28) were asked to perform two tasks with the instruction that they had to respond accurately and fast to each of them. The first task was a visual task of comparing a given number (target T1) to a fixed reference, and, second, an auditory task of judging the pitch of an auditory tone (target T2) [36]. The two stimuli are presented with a stimulus onset asynchrony (SOA), i.e., the delay in the onset of T1 and T2, varying from: SOA=0, 300, 900 and 1200 ms. In the number-comparison task, a number varying randomly among four values (28, 37, 53, 62) was flashed on a computer screen and subjects had to respond, with a key press using the right hand, whether the number was larger or smaller than 45. In the auditory task, subjects had to respond whether the tone was high or low frequency with a key press using the left hand. Full details and preliminary statistical analysis of this experiment have been reported in [36]. The study is part of a larger neuroimaging research program headed by Denis Le Bihan and approved by the Comit´e Consultatif pour la Protection des Personnes dans la Recherche Biom´edicale, Hopital de Bicetre (Le Kremlin-Bicetre, France). Subjects performed a total of 160 trials (40 for each SOA value) with a 12 s inter-trial interval [37]. The 160 trials were performed in five blocks of 384 s with a resting time of ∼ 5 min between blocks. For each trial, we recorded whole-brain fMRI images at a sampling time, TR = 1.5 s producing 8 fMRI images between two consecutive trials. From these images we computed the phase and amplitude of the hemodynamic response of each trial as explained in [37]. The experiments were performed on a 3T fMRI system (Bruker). Functional images sensitive to blood oxygenation level dependent (BOLD) contrast were obtained with a T2∗-weighted gradient echoplanar imaging sequence [repetition time (TR) = 1.5 s; echo time = 40 ms; angle = 90◦; field of view (FOV) = 192 × 256 mm; matrix = 64 × 64]. The whole brain was acquired in 24 slices with a slice thickness of 5 mm. High- resolution images (three-dimensional gradient echo inversion-recovery sequence, inversion time = 700 mm; FOV = 192 × 256 × 256 mm; matrix = 256 × 128 × 256; slice thickness = 1 mm) were also acquired. To estimate the periodicity and phase of the event-related BOLD response, the data from 25 each subject were submitted to a first-level model in which the signal from each trial (8 TRs of 1.5 s) was fitted with three regressors: a constant, a sine, and a cosine function at the above period. To facilitate intersubject averaging across possible differences in anatomical localization, the regression weights of the sines and cosines were stereotactically transformed to the standardized coordinate space of Talairach and Tournoux ([Montreal Neurological Institute] MNI 152 average brain) to spatially normalize for individual differences in brain morphology. Normalized images had a resolution of 8 mm3. Normalized phase images were transformed with the inverse tangent function to yield a phase lag expressed in radians for each voxel i and each trial t = 1, .., T over T = 40 trials: φi(t) ∈ [0, 2π] [37], indicating phase lags in the interval [0, 12]s. We calculate cross-correlations between different brain areas based on these phases [11, 12, 38]. We determine the equal-time cross-correlation matrix C with elements Cij measuring the cross-correlation between the phase activity φi(t) of the i-th and j-th voxel over T = 40 trials for each subject and SOA condition: T(cid:88) t=1 Cij = 1 T cos(φi(t) − φj(t)). (8) By construction, the elements satisfy −1 ≤ Cij ≤ 1, where Cij = 1 corresponds to perfect correlations, Cij = −1 corresponds to perfect anticorrelations, and Cij = 0 describes a pair of uncorrelated voxels. The entire experimental dataset is available in [40]. For our analysis, we create a mask where we keep voxels which were activated in more than 75% of the cases, i.e., in at least 48 instances out of the 64 total cases considered. The obtained number of activated voxels is N ≈ 60, 000, varying slightly for different indi- viduals and stimuli. The 'activated or functional map' exhibits phases consistently falling within the expected response latency for a task-induced activation [36]. As expected for an experiment involving visual and auditory stimuli and bi-manual responses, the responsive regions included bilateral visual occipito-temporal cortices, bilateral auditory cortices, mo- tor, premotor and cerebellar cortices, and a large-scale bilateral parieto-frontal structure, see SI-Section "Spatial projection of the modules" below . In the present analysis, we do not explore the differences in networks between different conditions. Rather, we consider them as independent experiments, generating a total of 64 different networks, one for each condition of temporal gap and subject. 26 FIG. S1. Boot strap analysis. The interval between the two curves corresponds to the 95% confidence interval for the calculation of the mean fraction of links (cid:104)p(cid:105) as a function of (cid:104)p(cid:105). The inset zooms in the regime around the values used in Fig. 4A. The use of fMRI neighboring voxels can be expected to carry some shared signal due to spatial autocorrelations (vascular, subject motion or scanner noise), which could give rise to spurious correlations over short distance. To test for this effect, we double the lattice spacing, reducing the voxels by a factor of 8 and repeat the calculations. The results are consistent with the percolation picture of Fig. 1, albeit with a lower pc, while the main results on long-range links are insensitive to this type of artifacts. II. BOOT STRAP ANALYSIS In order to estimate the accuracy of the correlation calculations, we performed a non- parametric boot strap analysis. We consider the set of the 40 trials per subject and SOA value. We perform the boot strap analysis for each possible pair of voxels. The correlation between two voxels for each of those trials serves as our original sample of 40 correlation values. We then draw 10000 re-samples from this sample with substitution. The arithmetic mean is calculated for each re-sample. Calculating the average value of all these means gives the boot strap estimate for the mean correlation. The 95% boot strap confidence interval is calculated by the distribution of the 10000 mean values at the 0.05 and 0.95 points of the distribution, respectively. The above process yields the confidence interval for the correlation value between two 27 0.970.980.990.970.980.990.80.850.90.951<p>0.80.850.90.95195% Confidence intervalvoxels. A different pair of voxels may have very different value of correlation, so in Fig. S1 we present the 95% bootstrap confidence interval as a function of the average value of correlation. The interval becomes smaller, i.e. the accuracy of the calculation increases, for larger p values. Considering the networks of Fig. 4A and B, for example, the intervals for p = 0.975 and p = 0.98 correspond to (see inset) [0.9744, 0.9760] and [0.9795, 0.981], respectively. III. SPATIAL PROJECTION OF THE MODULES The complex network representation reveals functional links between brain areas, but cannot directly reveal spatial correlations. Since voxels are embedded in real space, we also study the topological features of modules in three dimensions, where now voxels assume their known positions in the brain and links between them are transferred from the corresponding network, i.e., they are assigned according to the degree of correlation between any two voxels, Eq. (8), which is independent of the voxels proximity in real space. The above procedure yields a different spatial projection of the modules for each subject; an example for subject #1 and SOA=900 ms in the medial occipital cortex is shown in Fig. 1D. We study each of these percolation modules separately and find that they all carry statistically similar patterns. The topography of the identified modules reflects coherent patterns across different subjects, as shown next. Fig. S2A shows a medial sagital view of the largest four percolation modules for all the participants under stimulus SOA=0. In virtually all subjects we observe a module covering the anterior cingluate (AC) region, a module covering the medial part of the posterior parietal cortex (PPC) and a module covering the medial part of posterior occipital cortex (area V1/V2), along the calcarine fissure. We measure the likelihood that a voxel appears in the largest percolation module among all the participants in Fig. S2A by counting, for each voxel, the number of individuals for which it was included in one of the first four percolation modules. The spatial distribution of the first percolation modules averaged over all the subjects depicted in Figs. S2B and S2C shows that modules in the three main modes, V1/V2, AC and PPC, are ubiquitously present in percolation modules and, to a lesser extent, voxels in the motor cortex (along the central sulcus) are slightly more predominantly on the left hemisphere. The correlation 28 FIG. S2. The emerging modules have consistent spatial projections. (A) Spatial distri- bution of the four largest percolation modules (yellow, orange, red, brown) appearing at the first percolation jump, pc, for each subject under stimulus SOA=0. Most modules are localized in the same regions: anterior cingulate, posterior medial-occipital, posterior parietal and thalamus. (B) and (C) These panels show the number of times that the largest percolation cluster for each of the 16 subjects appears in a given voxel. White bleached regions correspond to voxels which are active in the 16 subjects, while the red regions correspond to voxels shared by half of the subjects. The anterior cingulate, a fundamental node in cognitive control, is the only region shared by all subjects. networks obtained from each subject yield modules with consistent topographic projections. IV. BOX COVERING ALGORITHM FOR FRACTAL DIMENSION IN NET- WORK SPACE For a given percolation module, the detection of submodules or boxes follows from the application of the box-covering algorithm for self-similar networks [22,43]. The algorithm can be downloaded at [40]. In box covering we assign every node to a box or submodule, by finding the minimum possible number of boxes, NB((cid:96)B), that cover the network and whose 29 ABCdiameter (defined as the maximum distance between any two nodes in this box) is smaller than (cid:96)B. We implement the Maximum Excluded Mass Burning (MEMB) algorithm from [43] for box covering. The algorithm uses the basic idea of box optimization, where we require that each box should cover the maximum possible number of nodes, and works as follows: We first locate the optimal 'central' nodes which will act as the origins for the boxes. This is done by first calculating the number of nodes (called the mass) within a distance rB from each node. We use, (cid:96)B = 2rB + 1. The node that yields the largest mass is marked as a center. Then we mark all the nodes in the box of this center node as 'tagged'. We repeat the process of calculating the mass of the boxes starting from all non-center nodes, and we identify a second center according to the largest remaining mass, while nodes in the corresponding box are 'tagged', and so on. When all nodes are either centers or 'tagged' we have identified the minimum number of centers that can cover the network at the given rB value. Starting from these centers as box origins, we then simultaneously burn the boxes from each origin until the entire network is covered, i.e. each node is assigned to one box (we call this process burning since it is similar to burning algorithms developed to investigate clustering statistics in percolation theory [29,30]). In Fig. 2A we show how box-covering works for a simple network at different (cid:96)B values. RG is then the iterative application of this covering at different (cid:96)B. V. CORRELATION FUNCTION Connections between voxels are determined according to the value of the correlation between the two voxels, as described above. This value may also depend on the physical (Euclidean) distance between the two voxels, since areas that are close to each other should interact stronger. We studied the correlation function, C(r) of the phases of the voxels: C(r) = (cid:104)cos(φ1 − φ2)(cid:105), (9) where φi denotes the phase of voxel i. The distance r is the Euclidean distance between the two voxels 1 and 2 and the average is taken over all pairs at distance r. This function can be interpreted as the correlation between two spins with orientation determined by the phases 30 FIG. S3. Spatial correlation function. This function measures the correlation C(r) between the phase of two voxels that are at a Euclidean distance r apart, as a function of r. As shown in the inset, it decays as a power law with slope 0.75 ± 0.02. φi of the voxels. We notice that this correlation function is usually studied in Ising-like spin models. We find that C(r) decays algebraically with distance, as shown in Fig. S3, and follows a power law form, C(r) ∼ r0.75. The value of the exponent 0.75± 0.02 was calculated through standard OLS regression. Notice that this function does not go to 0 asymptotically, but reaches a value of 0.1, which represents the average correlation (notice that in the definition of the correlation, the average value was not subtracted). This indicates that long-range correlations remain strong even at large distances. Further analysis is required to elaborate on this point, which is currently outside the scope of our present study. VI. EXPONENTS CALCULATION In Fig. 2D of the main text we show an aggregate average of the degree distributions for all clusters. This curve exhibits the general trends of the P (k) distribution, demonstrating for example the heavy tail, but it cannot be used for a direct determination of the exponent γ. In our work we studied the properties of 192 network clusters, as described in the main text. The calculation of the scaling exponents was done separately for each network. The resulting set of 192 values was then analyzed through non-parametric boot strap analysis, in order to get the average value of the exponent and the corresponding confidence intervals. 31 100101102r (mm)10-1100C(r)020406080100r (mm)0.00.20.40.60.81.0C(r)As an example, in Fig. S4 we show the degree distributions for 9 different clusters. In the plots, it is clear that there is always a plateau at small k values, while in many cases there is an asymptotic exponential cutoff. We fitted these distributions assuming that a power law describes the data within a given interval only. For this, we used a generalized power-law P (k; kmin, kmax) = k−γ ζ(γ, kmin) − ζ(γ, kmax) , (10) where kmin and kmax are the boundaries of the fitting interval and the Hurwitz ζ function form is given by ζ(γ, α) =(cid:80) i(i + α)−γ. (cid:33) We used the maximum likelihood method, following e.g. Clauset et al, SIAM Review, 51, 661 (2009). The fit was done in an interval where the lower boundary was kmin. For a given kmin value we were fixing the upper boundary to kmax = wkmin, where w is a parameter. We calculated the slopes in successive intervals by continuously increasing kmin and varying the value of w from 4 to 30. In this way, we sampled a large number of possible intervals. For each one of them we calculated the maximum likelihood estimator through the numerical solution of γ = argmax (cid:32) N(cid:88) −γ ln ki − N ln [ζ(γ, kmin) − ζ(γ, kmax)] (11) where ki are all the degrees that fall within the fitting interval and N is the total number of nodes with degrees in this interval. The optimum interval was determined through the i=1 Kolmogorov-Smirnov test. For the goodness-of-fit test, we used the Monte-Carlo method described in Clauset et al. For each possible fitting interval we generated 10000 synthetic random distributions following the best-fit power law. We then calculated the value of the Kolmogorov-Smirnov (KS) test for each one of them and measured the fraction pf it of realizations where the real data KS value was smaller than the synthetic SK value. We accepted the power-law hypothesis when this ratio was larger than pf it > 0.2. The average ratio over all clusters that were retained was pf it = 0.65. In this way, it is possible that we could accept more than one exponents for a given cluster at different intervals. In all these cases, the different γ values were very close to each other and we considered the final exponent to be the average of the individual exponents. Standard boot strap analysis on the resulting set of the individual cluster values yielded the exponent γ = 2.11 ± 0.04, with a 95% confidence interval [2.039, 2.178]. 32 FIG. S4. Degree distribution for network clusters. A number of degree distribution functions P (k) are shown for different clusters. The red lines correspond to the best power-law fitting, and the blue ones to an exponential fitting. (A) Degree distribution P (k) in logarithmic axes. The power-law slopes correspond to the exponent γ, and are shown on the plots. (B) The same distributions and fittings in semi-logarithmic axes. 33 10-310-210-110-310-210-110010110-410-310-210-1100101100101102γ=2.131γ=2.079γ=1.906γ=2.441γ=2.857γ=1.969γ=2.283γ=2.025γ=1.92810-310-210-110-310-210-102040608010-410-310-210-1020406080020406080100kP(k)P(k)kABThe same analysis was performed to test for a possible exponential description of the data. We scanned the same intervals as for the case of power-law and we used the maximum likelihood method to determine the optimum exponential fitting to the form: P (k; kmin, kmax) = 1 − e−λ e−λkmin − e−λkmax e−λk. (12) We again used KS statistics to determine the optimum fitting intervals and also the goodness- of-fit. In all the cases where the power-law was accepted, the exponential fitting gave an average ratio of pf it = 0.017, which rules out the possibility of an exponential distribution. VII. SCALING ANALYSIS The structure of a fractal network can be characterized by a set of scaling exponents. They define the scaling of many important system properties. Some of these properties and the corresponding exponents are as follows: a. The degree distribution: P (k) ∼ k−γ, where γ is the degree exponent [44]. b. The scaling of the mass with size: NB ∼ (cid:96) −dB B , which defines the fractal exponent dB [22]. c. The degree-degree distribution P (k1, k2) ∼ k k− 2 , where is the degree-degree exponent, and can be measured through Eb(k) ∼ k, which is the integration of P (k1, k2) over k2 [48]. −γ+1 1 d. The probability that modules are connected through their hubs, E ∼ (cid:96)−de defines the B hub-hub exponent de [23]. e. The scaling of the degree of the modules with the size of the modules: s ∼ (cid:96) −dk B , which defines the dk exponent [22]. f. The scaling of the modular factor as defined in Eq. (3): Q((cid:96)B) ∼ (cid:96)dM B , through the modularity exponent dM [27,28]. Scaling theory then defines precise relations between the exponents valid for fractal scale- free networks: g. γ = 1 + dB/dk [22], h. = 2 + de/dk [48], and i. dM = dB − dx [27,28]. We have measured directly all the exponents (see Fig. 2 and Fig. S5) for the brain modules 34 FIG. S5. Calculation of the scaling exponents. (A) Hub-hub exponent de through the scaling of E((cid:96)B). B Degree-degree exponent through the dependence of Eb(k) on the degree k [48]. and find: γ = 2.11 ± 0.04, de = 0.51 ± 0.08, dB = 1.9 ± 0.1, dk = 1.5 ± 0.1, = 2.1 ± 0.1, dM = 1.9 ± 0.1. Using these values in the known scaling relations above (g) and (h), we predict γ = 2.26 ± 0.11 and = 2.34 ± 0.06, which are reasonably close to the calculated exponents γ = 2.11 and = 2.1 from the direct measures. This set of results gives support to a scale-free fractal morphology of the brain modules. Notice that a Euclidean 2d lattice would be obtained in the limit γ → ∞, dk=0, → ∞. VIII. MODULARITY ANALYSIS In the main text of the paper we have described our modularity analysis of the brain clusters according to the MEMB technique. The modular properties of the same clusters can be also analyzed through techniques that partition a network according to maximization of modularity. We employed the Girvan-Newman method [20], which locates the point where the modularity measure, Q, is maximum. The definition of Q according to [20] is: (cid:32) NM(cid:88) i=1 − li L (cid:19)2(cid:33) (cid:18) di 2L Q = , (13) where NM is the number of modules, L is the number of links in the network, li is the number of links within the module i, and di is the sum of the degrees in this module. A value of Q = 0 corresponds to a completely random configuration or to the case of one module only. For the brain clusters we found an average modularity value of Q = 0.82. This is an indication of strong modularity within each cluster. A direct comparison between MEMB 35 110100k10-410-310-210-1Eb(k)ε=2.11101000.010.1de=0.5B()BEABFIG. S6. Modular properties of the brain clusters. Comparison between the partition provided by the MEMB method (at (cid:96)B = 15) with the corresponding partition using the Garvin- Newman method [20]. The modularity index from the Newman definition Q is around 0.82, as found by the latter method. Both methods yield similar sub-modules. and the Girvan-Newman method shows that they result in quite similar partitions. We cal- culated that 92% of the total links belong within a given module in both methods. A visual comparison is shown in Fig. S6. The maximization of modularity verifies the modular char- acter of the clusters. The use of the MEMB, though, provides us with the extra advantage of modifying the scale at which we observe the modules to determine whether the modular structure is scale-invariant, i.e., if it is composed of modules inside modules. 36 Girvan-Newmanmodules(Q=0.82)MEMBmodules(=15)BFIG. S7. Topographical map of module fractality. For each voxel, we calculate the average fractal dimension of the clusters to which it belongs, considering only voxels which form part of a cluster for at least eight subjects, to assure that mean values are not heavily determined by individual contributions. While the average over all clusters is dB = 1.9 ± 0.1, the dimension of each cluster exhibits small variations around this value which allows us to identify consistent differences among them. The clusters in the auditory cortex present the smaller fractal dimension dB, while parietal and motor clusters show intermediate values of dB. The right portion of the SMA and the right PPC were the clusters with the higher fractal dimension. 37 Auditory cortex1020304050601.91.6Right Parietal CortexAnterior Cingulate Supplementary Motor Cortex
1605.08944	3	1605	2018-07-29T05:53:26	Molecular force spectroscopy of kinetochore-microtubule attachment {\it in silico}: Mechanical signatures of an unusual catch bond and collective effects	[ "physics.bio-ph", "q-bio.SC" ]	Measurement of the life time of attachments formed by a single microtubule (MT) with a single kinetochore (kt) {\it in-vitro} under force-clamp conditions had earlier revealed a catch-bond-like behavior. In the past the physical origin of this apparently counter-intuitive phenomenon was traced to the nature of the force-dependence of the (de-)polymerization kinetics of the MTs. Here first the same model MT-kt attachment is subjected to external tension that increases linearly with time until rupture occurs. In our {\it force-ramp} experiments {\it in-silico}, the model displays the well known `mechanical signatures' of a catch-bond probed by molecular force spectroscopy. Exploiting this new evidence, we have further strengthened the analogy between MT-kt attachments and common ligand-receptor bonds in spite of the crucial differences in their underlying physical mechanisms. We then extend the formalism to model the stochastic kinetics of an attachment formed by a bundle of multiple parallel microtubules with a single kt considering the effect of rebinding under force-clamp and force-ramp conditions. From numerical studies of the model we predict the trends of variation of the mean life time and mean rupture force with the increasing number of MTs in the bundle. Both the mean life time and the mean rupture force display nontrivial nonlinear dependence on the maximum number of MTs that can attach simultaneously to the same kt.	physics.bio-ph	physics	Molecular Force Spectroscopy of kinetochore-microtubule attachment in-silico: mechanical signatures of an unusual catch-bond and collective effects Dipanwita Ghanti,1 Shubhadeep Patra,2 and Debashish Chowdhury1 1Department of Physics, Indian Institute of Technology Kanpur, 208016, India 2ISERC, Visva-Bharati, Santiniketan 731235, India Measurement of the life time of attachments formed by a single microtubule (MT) with a single kinetochore (kt) in-vitro under force-clamp conditions had earlier revealed a catch-bond-like behav- ior. In the past the physical origin of this apparently counter-intuitive phenomenon was traced to the nature of the force-dependence of the (de-)polymerization kinetics of the MTs. Here first the same model MT-kt attachment is subjected to external tension that increases linearly with time until rupture occurs. In our force-ramp experiments in-silico, the model displays the well known 'mechanical signatures' of a catch-bond probed by molecular force spectroscopy. Exploiting this new evidence, we have further strengthened the analogy between MT-kt attachments and common ligand-receptor bonds in spite of the crucial differences in their underlying physical mechanisms. We then extend the formalism to model the stochastic kinetics of an attachment formed by a bundle of multiple parallel microtubules with a single kt considering the effect of rebinding under force-clamp and force-ramp conditions. From numerical studies of the model we predict the trends of variation of the mean life time and mean rupture force with the increasing number of MTs in the bundle. Both the mean life time and the mean rupture force display nontrivial nonlinear dependence on the maximum number of MTs that can attach simultaneously to the same kt. I. INTRODUCTION A mitotic spindle [1 -- 3] is an example of a self-organized [4] multi-component molecular machine [5] that carries out mitosis [6], i.e., the process of segregation of repli- cated chromosomes, in eukaryotic cells. These machines are assembled at the right place at the right time and disassemble after serving their biological function. Even after the mitotic spindle is fully assembled, its size, posi- tion, orientation as well as the architecture keep chang- ing dynamically with time, as required for its function. Many of these changes of the spindle are driven by its own components that transduce input energy to gener- ate these forces. Understanding its "emergent mechan- ics" [7], i.e., how its mechanical properties emerge from the complex dynamics, interactions and feedback of its energy-consuming active building blocks [8], is one of the aims of research on molecular bio-mechanics at the inter- face of physics and biology. Assembling a mitotic spindle requires formation of molecular joints between specific components. One of the major components of a spindle, that also forms all the key molecular joints in it, is a stiff filament called microtubule (MT) [9] each of which has a tubular struc- ture. On the surface of each sister chromatid, that results from DNA replication, a proteinous complex called kine- tochore (kt) is located [10]. During the self-assembling of the spindle each kt attaches with one or more MTs; the actual number varies from one species to another. In this paper we study kt-MT attachment in a mitotic spindle as an example of a transient joint in a multi-component molecular machine. This molecular joint plays important roles not only in the morphogenesis [11] but also in the subsequent emergent mechanics of the mitotic spindle. Force plays all the three roles, namely, input, output and signal, for different components of the same kt-MT In order to force exert attachment [12]. Force exerted by MTs, the key force generators in mitosis, on the kt is essential for proper positioning of the chromosomes in the initial stages of mitosis. Equally important is the opposing tensions ex- erted by the MTs attached to the two sister chromatids that pull the two sister chromatids apart and away from each other in the late stages of mitosis [6]. through polymeriza- tion/depolymerization, the tip of each MT remains free to polymerize/depolymerize [13] and rapid turnover of its monomeric subunits continues even when the kt- MT attachments remains intact. A kt-MT attachment would rupture spontaneously, even in the absence of any externally applied tension, by a thermally activated hopping over a barrier that separates the bound state from the unbound state in the energy landscape [14]. Moreover, the kt-MT attachment experiences quite high levels of tension at various stages of mitosis. How the structural integrity of a kt-MT joint is maintained during the entire lifetime of the spindle, in spite of these potentially disrupting tendencies, is one of the wonders of spindle operation. In this paper we study theoretical models of molecular joints formed by the attachment of N (N ≥ 1) parallel MTs to a single kt by treating it as an unusual "ligand- receptor" bond. In the words of Martin Karplus [15], "the ligand can be as small as an electron, an atom or diatomic molecule and as large as a protein". In prin- ciple, this definition of a ligand can be extended even further to include a MT [9], whose tubular filamentous structure consists of a hierarchical organization of many proteins. Analogously, the kt, a macromolecular com- plex hub, is a receptor for a MT. The protocols of 'force- clamp' and 'force-ramp' that we implement in the com- puter simulations of our models may be regarded as in- silico analogs of the corresponding in-vitro experiments 8 1 0 2 l u J 9 2 ] h p - o i b . s c i s y h p [ 3 v 4 4 9 8 0 . 5 0 6 1 : v i X r a carried out with common chemical ligand-receptor bonds [16]. In a force-clamp experiment the magnitude of the ex- ternally applied tension F on a pre-formed kt-MT at- tachment is kept fixed ('clamped') and the duration for which the attachment survives before getting fully rup- tured is defined as its lifetime. Similarly, in a "force- ramp" protocol [17] the magnitude of the tension is in- creased ('ramped up') with time in a well-defined man- ner till the bond ruptures at a value of the tension that is identified as the rupture force. Because of the intrin- sically stochastic nature of the process of rupture, both the lifetime and rupture force of a kt-MT joint are ran- dom variables that fluctuate from one kt-MT joint to another identical joint. By computing the probability distributions of the lifetime and rupture force and, then, analyzing the data in the light of the analogy with ligand- receptor bonds, we address some fundamental questions on the biomolecular mechanics of the kt-MT joint in a mitotic spindle. Our in-silico studies have been motivated by the in- vitro biophysical experiments [18 -- 20] carried out over the last few years using reconstituted kinetochores of bud- ding yeast which happens to be the simplest because each kt can attach with only a single MT [18]. Under force- clamp conditions, created in-vitro using optical trap [21], the mean lifetime of the reconstituted kt-MT attachment of budding yeast was found to increase with increasing tension up to a moderate level beyond which the mean lifetime decreased with further increase of tension. Such nonmonotonic variation of the average lifetime with in- creasing strength of the pulling force is reminiscent of catch-bonds formed by wide varieties of ligands with their respective receptors [22 -- 26]. Akiyoshi et al.[21] could account for the catch-bond- like behaviour of the reconstituted kt-MT attachment, as displayed by experimental data, with a 2-state model [27]. However, this simple model reveals neither the structural nor the kinetic origins of this behavior. Sub- sequently, a minimal theoretical model was developed by Sharma, Shtylla and Chowdhury (from now onwards referred to as SSC model) [28] that explicitly describes the polymerization and depolymerization of the MT. The SSC model reproduced the universally accepted 'mechan- ical signatures of catch bonds' in force-clamp experiments and elucidated the crucial role of MT kinetics (particu- larly its force-dependence) that makes this catch-bond unique and unusual. In the first part of this paper, we present further evidence in favour of this catch-bond-like behavior by demonstrating that the SSC model also reproduces the well known 'mechanical signatures of catch-bonds' in force-ramp experiments. In the second part of this pa- per we push the analogy with ligand-receptor bonds even further to situations where a bundle of parallel MTs (i.e., multiple "ligands") form non-covalent bonds with a sin- gle kt (i.e., the "receptor"). Studies of the case N > 1 are important for several reasons. First, it is a natural 2 curiosity because such systems are very common in bi- ological systems. Except unicellular eukaryote budding yeast, cells of most of the organisms, including mammals, have multiple MTs attached with single kt. For exam- ple, about 20-40 parallel MTs are bound to each kt in the metaphase spindles of mammalian cells. Analyzing this extended version of the model with N > 1 under both force-clamp and force-ramp conditions we make new the- oretical predictions. Second, from the perspective of physics, this system provides a unique opportunity to explore collective effects in force generation. Collective force generation by a bun- dle of polymerizing biofilaments like MTs have been stud- ied both experimentally as well theoretically (see ref.[29] for a recent overview). Similar fundamental questions on the collective effects of MT bundles in the MT-kt at- tachments are addressed in this paper. What makes the problem very interesting is that the kinetics of the indi- vidual MTs get influenced by others bound to the same kt in spite of the fact that there is no direct interactions between them; the interactions between the MTs are like feedbacks mediated by the kt to which all these MTs are attached. At least two physical phenomena add to the complex- ity of the process of rupture if N > 1. For example, up on detachment of a MT from the kt, the load it experi- enced before detachment must now be shared by the n (provided n > 0) MTs that are still attached to the same kt according to some load-sharing formula. Since, as seen in the special case N = 1, increasing load on a single MT does not necessarily destabilize it, the extra load is likely to have a nontrivial effect on the overall stability of the attachment. Moreover, as long as n ≥ 1, a detached MT can reattach thereby, probably, prolonging the lifetime of the attachment. Do the mean values of the lifetime and rupture force increase simply monotonically, perhaps lin- early, with increasing N or is the variation with N more nontrivial? This question is addressed in the second half of this paper In this paper we also mention the experimental meth- ods that, at least in principle, can test the validity of our theoretical predictions. It is worth mentioning here that the focus of this work is not on throwing new light on catch-bonds in the usual ligand-receptor sys- tems that have been studied for decades. Instead, our focus is on the tension-induced rupture of the kt-MT attachment, which is an essential transient molecular joint in a functionally important multi-component intra- cellular molecular machine. However, we discuss this phenomenon from a broader perspective of molecular force spectroscopy of noncovalent ligand-receptor bonds to highlight the crucial differences in spite of the super- ficial similarities. 3 nm. However, there is a small offset of about 0.92 nm be- tween the dimers of the neighboring protofilaments. SSC model adopts the single protofilament model [38] where each MT is viewed as a single protofilament that grows helically with an effective dimer size 8/13 nm. Thus, following the SSC model, throughout this paper, we rep- resent a MT as a strictly one-dimensional lattice with the lattice size 8/13 nm. In this model the instantaneous overlap between the outer surface of the MT and the inner surface of the coaxial cylindrical sleeve is represented by a continuous variable y(t) which is a function of time t. The total length of the coupler is L so that 0 ≤ y(t) ≤ L. Two main postulates of this model are as follows [28]: Postulate (a): increasing overlap y lowers the energy of the system and that this lowering of energy is propor- tional to y so that the kt-MT interaction potential Vb(y) is assumed to have the form (see Fig.1(b)) Vb(y) = −By, (1) where B is the constant of proportionality. Accordingly, the magnitude of the depth of the potential at y = L is BL. Postulate (b): the external force F suppresses the rate of depolymerization β of the MT and that β decreases exponentially with increasing F following β(F ) = β0exp(−F/F(cid:63)), (2) where β0 is depolymerization rate in the absence of any external force and the parameter F(cid:63) is a characteristic force that determines the sharpness of the decrease of β(F ) with F . The postulate (a) is essentially a limiting case of the Hill model in the sense that the "roughness" of the interface between the outer surface of the MT and inner surface of the sleeve is neglected in the minimal version of the SSC model. The postulate (b) is qualitatively supported by the in-vitro experiments of Franck et al. [39]. The decrease of the rate β with the external force F need not be exponential; all the conclusions drawn from the SSC model in ref.[28] remain valid as long as the de- crease of β with increasing F is sufficiently sharp. The external tension (using the correct notation for its direc- tion) corresponds to an effective potential Vf = F y (see Fig.1(b)). Note that the overlap y(t) can be viewed as the po- sition of a hypothetical Brownian particle in a one- dimensional space and subjected to an external poten- tial V (y) = −By + F y. Accordingly, the kinetics of this model kt-MT attachment can be formulated in terms of a Fokker-Planck (FP) equation [40] ∂P (y, t) ∂t = D ∂2P (y, t) ∂y2 − v(F ) ∂P (y, t) ∂y (3) for the probability density P (y, t), where the net drift FIG. 1. (a) A schematic depiction of the kt-MT attachment in the presence of external force. (b) Hypothesized effective potentials Vb(y) and Vf (y, t) are plotted against the instan- taneous length of overlap y(t). (c) Net drift velocity v(F ) is plotted against F for three different values of F(cid:63). Parameters used are listed in the table-I. (d) Linearly increasing force (F = at); different straight lines correspond to different rates of loading. The kt-MT attachment survives the increasing tension up to a certain time and then gets ruptured. II. SSC MODEL: A BRIEF REVIEW In this section we present a brief summary of the SSC model and review its main results obtained earlier un- der force-clamp condition. This summary will help in motivating the adaptations that are appropriate for the- oretical analysis of the force-ramp scenario presented in section IV. The SSC model [28] is a minimal model in the sense that it does not make any assumption about the molec- ular constituents or structure of the kt-MT attachment. It merely assumes a cylindrical, effectively "sleeve-like", coupler (in the spirit of the Hill sleeve model [30]) that is coaxial with the MT and has a diameter slightly larger than that of the MT (see Fig.1(a)). The sleeve may be an abstract representation of the Dam1 ring [31] while the "rigid rod", that connects the sleeve with the kineto- chore, captures the effects of Ndc80 proteins [32 -- 34]. No further structural or kinetic details of the Hill model and its later extensions [35] (see refs.[36, 37] for reviews) have been incorporated in the minimal models of the kt-MT attachments studied here. Each microtubule is a cylindrical hollow tube with a di- ameter of approximately 25 nm. Globular proteins called α and β tubulins form hetero-dimers that assemble se- quentially to form a protofilament. Normally 13 such protofilaments, arranged parallel to each other, form a microtubule. The length of each α-β dimer is about 8 velocity B − F B − F Γ Γ v(F ) = +(α−β(F ))(cid:96) = +(α−β0e−F/F(cid:63) )(cid:96) (4) of the hypothetical Brownian particle involves a phe- nomenological coefficient Γ, that characterizes the vis- cous drag on it, and (cid:96) is the increase of the length of MT caused by the addition of each of its sub- units. The diffusion constant D in (3) gets contributions from two different physical processes. First, on length scales much longer than (cid:96), the stochastic polymerization- depolymerization of a MT can be described in terms of the drift velocity v and an effective diffusion constant [41] DM T = (cid:96)2(α + β)/2. (5) even when the MT tip is not attached, or tethered, to any surface. The second contribution that exists even in the absence of polymerization-depolymerization of the MT is the diffusive motion of the kinetochore plate itself [42]. Considering (cid:96) = 8/13 nm, α=30 s−1 and β << α, the effective diffusion constant DM T is approximately 5 nm2/s. Even if one includes the maximum possible value of β(F ), i.e., β0 = 350s−1 [30, 42 -- 44] in (5), the effec- tive diffusion constant DM T increases to about 70 nm2/s which is still about an order of magnitude smaller than the contribution coming from the diffusional movement of the kinetochore plate which is typically 700 nm2/s [30, 42, 43]. Therefore, throughout this paper, we as- sume the diffusion constant D to be independent of the external tension F . The FP equation (3) can also be re-cast as an equation of continuity ∂P (y, t) ∂t = − ∂J(y, t) ∂y (6) for the probability density P (y, t) with the probability current density where U(cid:48)(y) = d U (y)/dy and effective potential U (y) is given by (β(F ) − α) D + (cid:96) y. (8) Note that the terms involving F and B in eq.(8) are of energetic origin whereas the remaining two terms involv- ing α and β are of kinetic origin. The attachment survives as long as y remains non-zero; the rupture of the attachment is identified with the at- tainment of the value y = 0 for the first time. For the calculation of the lifetime of the attachment a unique ini- tial condition is required. In ref.[28] the authors assumed (cid:20) ∂P (y, t) (cid:20) ∂P (y, t) ∂y ∂y J(y, t) = −D = −D (cid:20) F − B kBT U (y) kBT = P (y, t) − v(F ) D U(cid:48)(y) kBT + P (y, t) . (7) (cid:21) (cid:21) (cid:21) 4 that initially (i.e., at time t = 0) the MT is fully inserted into the sleeve, i.e., y(t = 0) = L (initial condition). (9) Since the MT is not allowed to penetrate the kinetochore plate, the overlap y cannot exceed L. This physical con- dition is captured mathematically by imposing the re- flecting boundary condition J(y, t)y=L = 0. An absorbing boundary condition P (y, t)y=0 = 0 (10) (11) is imposed at y = 0 for the calculation of the life times. In terms of the hypothetical Brownian particle, the FP equation for y(t) can be viewed as that for the position of a hypothetical Brownian particle, subjected to an ex- ternal potential V (y) = −By + F y, in a one dimensional space with a reflecting boundary at y = L, an absorbing boundary at y = 0 and the initial condition y(t = 0) = L. Starting from the initial condition y = L, the time taken by the kt-MT attachment to attain vanishing over- lap (y = 0) for the first time was identified as the life time of the attachment. Thus, the calculation of the lifetime is essentially that of a first passage time for a hypothetical Brownian particle: the time it takes to reach y = 0 for the first time starting from y = L at t = 0. This life- time fluctuates from one kt-MT attachment to another; the distribution of the lifetime contains all the statistical information. In ref.[28] the authors calculated the exact distribu- tion of the lifetimes analytically in the Laplace space and hence the mean lifetime < t > to be ev(F )L/D − 1 < t >= (cid:20) (cid:21) (12) D − L v(F ) v2(F ) FIG. 2. Using WPE method, continuous 1-D space of length L is discretized. In this discretized space MT tip is mov- ing either in the forward or in the backward direction using transition rates wf (j) and wb(j). For the convenience of numerical computation of the distribution of the lifetimes by computer simulation, the SSC model was discretized in ref.[28] following pre- scriptions proposed earlier by Wang, Peskin and Elston (WPE) [45, 46]. WPE method is a numerical algorithm 5 samples each with a randomly chosen initial condition. The results of these computations are plotted in Fig.3. The "catch-bond" behavior is still observed. In the second type, the lifetimes are first calculated for a fixed initial condition y(t = 0) = L0 (1 ≤ L0 ≤ L) and then these lifetimes are averaged over a large number of samples, all for the same initial overlap L0, getting the mean life time < τ > (L0) corresponding to the fixed L0. The process is then repeated for several different values of L0 to get < τ > as a function of L0 (0 ≤ L0 ≤ L). The results of these computations are plotted in the inset of Fig.3. As expected on physical grounds, the mean lifetime of the attachment increases with increasing initial overlap L0, attaining its largest value (≈ 758s) for L0 = L i.e., where the MT is initially fully inserted into the coupler. Note that the mean lifetime corresponding to L0 = L is about seven times that of the attachments where the initial overlap is selected at random. Such lower values of < τ > for random initial overlaps is expected on the physical grounds that in several initial configurations the MT begins with an initial overlap y(t = 0) < L and, hence, expected to rupture sooner that those with initial overlap y(t = 0) = L. in which a FP equation is discretized into a discrete Markovian jump process by finite differencing of the FP equation. Following WPE, space was discretized into M cells, each of length h = L/M and the continuous effec- tive potential U (y) was replaced by its discrete counter- part (cid:18) β0e−F/F∗ − α (cid:19) (cid:21) (cid:20) (F − B) kBT Uj kBT = + (cid:96) D yj (13) where yj denotes the position of the center of the j-th cell. In the discrete formulation, instead of a FP equation, a master equation describes the kinetics of the system in terms of discrete jumps of the hypothetical Brownian par- ticle from the center of a cell to that of its adjacent cells, either in the forward or in the backward direction. The rates of forward and backward jumps ωf (j) and ωb(j) on the discretized lattice were given by [45] ωf (j) = D h2 exp (cid:18) kB T − δ Uj − δ Uj kB T (cid:19) = 1 h − 1 B − F Γ (cid:19) (cid:18) + (cid:96)(α − β) − δ Uj − 1 kBT exp (14) ωb(j) = D h2 (cid:18) exp (cid:19) δ Uj kB T δ Uj kB T = 1 h − 1 F − B (cid:19) (cid:18) δ Uj + (cid:96)(β − α) − 1 kBT Γ exp where δ Uj = Uj+1 − Uj (15) (16) Excellent agreement between the results derived from the analytical theory and computer simulations was reported in ref.[28]. III. FORCE-CLAMP: DEPENDENCE OF LIFE TIME FOR N=1 ON INITIAL CONDITIONS In ref.[28], where the SSC model was presented, the authors reported the results for the model only under force-clamp conditions. However, as summarized in the preceding section, the lifetimes of the attachments were calculated beginning always with the unique initial con- dition y(t = 0) = L. In order to test whether the con- clusions drawn in ref.[28] are sensitive to the choice of the initial condition, we have now carried out a detailed investigation of the distribution of the lifetimes with two different types of initial conditions. In one of these, any integer lying between 1 and L is chosen, with equal probability, to be the initial value of the overlap y. The mean lifetime is obtained by av- eraging the lifetimes over a sufficiently large number of FIG. 3. Mean lifetime, under force-clamp condition, plotted against external force F for randomly chosen initial position of the coupler. In the inset mean lifetime is plotted by varying the initial position of the coupler L0 for a fixed force F = 0.5pN. The numerical values of all the other parameters used in the simulation are listed in the table I except, N = 1, F∗ = 1pN and B = 2pN. • Physical origin of the catch-bond-like behavior The external pulling force F has two opposite effects on the MT. From the expression (4) for the net drift veloc- ity v(F ) we see that, on the one hand, the MT is bodily pulled out of the coupler by F . On the other hand, be- cause of the suppression of the depolymerization by the external pull F , if the depolymerization rate β falls below that of polymerization the tip of the MT exhibits a net growth. Moreover, if the suppression of depolymerization is so strong that the net rate of tip growth into the cou- ●●●●●●●●●●●●●●●●●●●●▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲020400200400600L0(nm)<t>(s)00.511.52050100F(pN)<t>(s)pler (increase in y) can more than compensate the rate of bodily exit of the MT from the coupler (decrease of y) the growing MT tip moves deeper inside the coupler (re- sulting in the net increase of y) when subjected to exter- nal tension. Such an increase of y (indicated by increase of v(F ) in Fig.1(c)), instead of the naively expected de- crease, upon application of F would be interpreted as an effective increase of the stability of the kt-MT attachment with increasing strength of the applied force F . However, as the strength of F increases, β(F ) gradually saturates. Since β practically stops decreasing further with the fur- ther increase of F the bodily movement of the MT out of the coupler at higher values of F can no longer be compensated by the tip growth into the coupler; the net decrease of y (indicated by decrease of v(F ) in Fig.1(c)) with further increase of F in this regime manifests as decrease in the stability of the kt-MT attachment. How- ever, monotonic decrease of v(F ) with increasing F seen in Fig.1(c)) results for larger values of F(cid:63) because of weak suppression of depolymerization by the external tension. The physical scenario that emerges from the above interpretation of the dependence of v(F ) on F(cid:63) is also consistent with the expression (12) for the mean lifetime < t > where v(F )L acts like an effective barrier height. For small enough F(cid:63), the nonmonotonic variation of v(F ) with F manifests as a nonmonotonic variation of the bar- rier height v(F )L, resulting in a nonmonotonic variation of the mean lifetime < t > with F which has been in- terpreted as a catch-bond-like behavior. In contrast, for sufficiently large F(cid:63) the monotonic decrease of v(F ) with F results in a monotonic decrease of the effective bar- rier height v(F )L which, in turn, causes the monotonic decrease of the mean lifetime < t > with F that has been interpreted as a slip-bond-like behavior. Thus, to summarize, whether the attachment behaves like a catch bond or a slip bond depends crucially on the magnitude of F(cid:63), which determines the extent of suppression of de- polymerization for a given F , i.e., how sharply the de- polymerization rate β(F ) falls with increasing tension F . IV. RUPTURE OF KT-MT ATTACHMENT UNDER RAMP FORCE FOR N=1 In ref.[28] the external tension F was assumed to be independent of time t; this condition corresponds to a "force-clamp" situation in the experiments. In this section the time-dependent external tension F (t) is as- sumed to increase according to a well defined protocol; this corresponds to a "force-ramp" in experiment (see Fig.1(d)). We adopt the postulates (a) and (b) of the SSC model. For the sake of simplicity, we assume a lin- ear ramp force, namely, F (t) = a t where a is the load- ing rate. The instantaneous external tension F (t) can be derived from the corresponding instantaneous poten- tial landscape, Vf (y, t) = F (t)y . The effective potentials Vb(y) and Vf (y, t) at an arbitrary instant of time are plot- ted in Fig.1(b). Net instantaneous potential V (y, t) felt 6 by the kinetochore is V (y, t) = Vb(y) + Vf (y, t). For the theoretical treatment of the kt-MT attach- ment subjected to a ramp force F (t), we adapt the corre- sponding theory for ligand-receptor bond rupture, devel- oped originally by Bell [47] and subsequently extended by Evans and Ritchie [48] and by Evans and Williams [49] (see also the reviews in refs.[50 -- 52]). In the pres- ence of a given tension F , let kof f (F ) be the rate (i.e., probability per unit time) of unbinding of a MT from the kt. Because of the specific choice of the initial condition y(t = 0) = L and the absorbing boundary condition at y = 0, no rebinding of the MT is possible and, there- fore, rebinding rate remains kon(F ) = 0 throughout this section. Denoting the probability that y (cid:54)= 0 (i.e., MT is at- tached to the kt) at time t by the symbol Pon(t), the equation governing the time evolution of Pon(t) is dPon(t) dt = −kof f (F )Pon(t). (17) Hence, in terms of kof f (F ), the survival probability S(t) of the attachment (i.e., the probability that the hypo- thetical Brownian particle has not reached y = 0 before time t) can be expressed as [51] S(t) = exp (cid:48) kof f (F (t ))dt (18) (cid:48)(cid:21) (cid:20) − (cid:90) t 0 (cid:18) (cid:90) ∞ (cid:20) Moreover, in terms of kof f (F ) the probability density ρf p(F ) of the rupture forces is expressed as [51] kof f (F (cid:48))dF (cid:48)(cid:19)(cid:21) (cid:90) F 0 (19) ρf p(F ) = kof f (F ) a exp − 1 a Mean Rupture force is given by [51] < F >= F ρf p(F ) dF (20) 0 Thus, for the calculation of S(t) and ρf p(F ) the analyt- ical expression for kof f (F ) is required. For kof f (F ) we use the expression for the inverse of the average lifetime of a single kt-MT attachment in the SSC model, reported in ref.[28], namely, kof f (F ) = 1 < t > = v2(F ) D(ev(F )L/D − 1) − Lv(F ) (21) where the expression v(F ) is given by Eq.(4). The expres- sion (21) was derived under force-clamp condition and, therefore, strictly valid when the force does not vary at all with time. Use of this expression for kof f (F ) in the cal- culation of S(t) and ρf p(F ) is based on the assumption that the expression (21) is a good approximation even when the tension varies with time. Obviously, the de- viations of kof f (F ) from this expression in force-clamp conditions is expected to be insignificant provided the rate of increase of F is sufficiently small. Substituting Eq.(21) into the Eqs.(18) and (19) we get, respectively, Parameter Inter-space between MT binding site l [30, 42, 43] Total length of coupler L [53 -- 56] Polymerization rate α [30, 42 -- 44] Maximum Depolymerization rate β0 [30, 42 -- 44] Characteristic force of Depolymerization F(cid:63) [28] Attractive force between kt-MT B [28] Diffusion constant D [30, 42, 43] Viscous drag coefficient Γ [30, 42, 43, 57] Values 8/13 nm 50 nm 30 s−1 350 s−1 0.8 pN 1.9 pN 700 nm2s−1 6pN s µm−1 TABLE I. Values of the parameters for kt-MT system the survival probability S(t) and the rupture force den- sity ρf p(F ) by numerically evaluating the respective in- tegrals. 7 Thus, the theoretical results for the case N = 1 have been derived from numerical integrations of eqns.(18)- (19) which have been plotted throughout this section by lines. For computer simulation of the model, we dis- cretize the FP equation of the SSC model following WPE prescription [45, 46] as explained above [28]. Instead of a constant force, a time-dependent external force F = at is imposed. Carrying out computer simulations of this dis- cretized version of the model we directly compute the sur- vival probability S(t) and the distribution ρf p(F ) of the rupture forces. Throughout this section, discrete symbols have been used to plot the data obtained from computer simulations of the discretized model. Parameter values that we used for numerical calculations are listed in table I. FIG. 4. Probability density of rupture force of the kt-MT attachment with N = 1, under force-ramp condition, for four different loading rates, namely, (a) a = 3 × 10−4pN s−1 (violet rhombus), (b) a = 1 × 10−3pN s−1 (green square), (c) a = 3 × 10−3pN s−1(blue triangle) and (d) a = 3 × 10−2pN s−1 (red circle) are plotted. The continuous curves have been plotted by numerical integration of the Eq.(19) whereas the discrete data points have been obtained from computer simulations of the discretized version of the same model. Numerical values of all the other parameters are listed in table-I. In the Fig.4 the rupture force distribution obtained from numerical integration of the eqns.(18)-(19) of the continuum theory and those obtained from computer simulation of the discretized model are plotted for four different loading rates. At loading rates as low as a = 3× 10−4pN s−1 (violet), the most probable rupture force 8 rates the probability of survival remains high, and prac- tically unaffected by the applied force, upto quite high values of the force and, accordingly, the most probable rupture force is also expected to be high. In contrast, sharp drop in the survival probability with increasing force is also reflected in the vanishingly small most prob- able rupture force at very low loading rates. In the Fig.5(b) we have plotted mean rupture force as a function of loading rate. Mean rupture force increases with increasing loading rate. The increase of mean and most probable rupture force with increasing loading rate is also observed in case of common ligand-receptor at- tachments [51]; it follows from the mathematical form of the equation dPon(F ) dF = − 1 a kof f (F )Pon(F ) (22) which is nothing but the equation (17) expressed in terms of force F rather than time t. Eqn.(22) implies that the rate of decay of the bound state of the bond is inversely proportional to the loading rate a. Consequently, the kt-MT attachment persists up to higher values of force when subjected to faster loading rates. The continuous black line in Fig.5(b) has been ob- tained using Eq.(20). As the loading rate exceeds about 100 pN/s, the black line begins to deviate from the cor- responding data points obtained from simulations. This increasing deviation indicates increasing failure of the ap- proximation made by substituting the force-clamp values of kof f (F ) for evaluating the integrals in Eq.19. However, surprisingly, even at ten times faster loading rates the er- ror made by this approximation is within about 20%. Irrespective of the actual loading rate, a slip bond ex- hibits a single peak at F = Fmp in the rupture force distribution ρf p(F ) at a given rate of loading. In this case, the most probable rupture force Fmp → 0 corre- sponding to a → 0 and Fmp increases with loading rate a. The trend of variation of ρf p(F ) with the loading rate a is qualitatively different in case of catch bonds. For the latter, at sufficiently low values of a, the distribu- tion ρf p(F ) exhibits a high peak at F = 0 and a much lower peak at a larger nonzero value of F while ρf p(F ) remains very small over a wide range of F in between these two peaks. With the increase of the loading rate a the second peak at the non-zero F increases in height while a concomitant lowering of the peak at F = 0 oc- curs. The occurrence of two peaks in ρf p(F ) for a given a is regarded as the 'mechanical signature' of catch bond in force-ramp experiments [22, 23, 58]. The shape of ρmp(F ) plotted in Fig.4 for four different values of loading rate a is, thus, an unambiguous evidence in favour of the catch-bond-like behavior exhibited by the kt-MT attachment (for N = 1) also in our force- ramp experiment in-silico. Several different molecular mechanisms proposed so far can account for the observed signatures of catch bond in conventional ligand-receptor systems [22 -- 26]. However, the distinct mechanism that we have summarized above in the context of force-clamp FIG. 5. (a)Survival probability for different loading rates; the continuous curves in (a) have been plotted by numerical integration of the Eq.(18). (b) Mean rupture force is plotted against the logarithm of the loading rate; the continuous black line in (b) has been plotted by numerical integration of the Eq.(20). The same symbols in (a) and Fig.4 correspond to the same set of values of the model parameters. Numerical values of all the other parameters are listed in table-I. is vanishingly small. At such slow loading rates the rup- ture of the attachment is mostly spontaneous dissocia- tion caused by thermal fluctuation and is very rarely driven by the applied tension. However, as the load- ing rate increases a second peak at a non-zero value of the force begins to emerge. At moderate loading rates like a = 1 × 10−3pN s−1 (blue line and triangle ) and a = 3×10−3pN s−1 (green line and square)), a large frac- tion of the kt-MT attachments survive upto a high force before getting ruptured while another significant fraction of the attachments still dissociate at a vanishingly small force. But, at sufficiently high rates of loading, for ex- ample at a = 3 × 10−2pN s−1 (red), an overwhelmingly large fraction survives up to a high force while very few attachment get ruptured by very weak forces. In the Fig.5(a) the survival probabilities are plotted at the same loading rates for which the rupture force distri- butions have been plotted in Fig.4. At very high loading 9 studies of kt-MT attachment is responsible also for the catch-bond-like behavior displayed in the Figs.4 and 5. In principle, our theoretical predictions for N = 1 can be tested using the reconstituted kinetochore of budding yeast in-vitro [19] applying standard techniques of dy- namic force spectroscopy [16]; a typical set up would use an optical trap with controlled ramp protocol [17]. In the force-clamp set up with optical trap, the bead-trap separation is maintained at a fixed value with a computer controlled feedback while the change in the length of the MT is recorded by monitoring the movement of the spec- imen stage [17]. A force-ramp set up, where the bead- trap separation is gradually increased with time, has also been designed by modifying the force-clamp software [17]. This force-ramp can be used to test the corresponding theoretical predictions made in this paper. However, the slow loading rate required to observe the theoretically predicted behavior may still pose technical challenges. V. EXTENDED SSC MODEL OF MT- SINGLE KT ATTACHMENT FOR N > 1 FIG. 6. Three microtubules (green cylinders) are attached to a single kinetochore (violet wall) in the presence of external tension on kinetochore (inspired by Fig.1(b) of ref.[65]). In this section we extend the SSC model to capture some key features of the energetics and kinetics of a dy- namic attachment formed between a single kt and a bun- dle of N parallel MTs. As mentioned in the introduction, this extension is motivated by the fact that, in almost all organisms, except for budding yeast, each kt can nor- mally attach to multiple MTs simultaneously. However, in none of the organisms, other than budding yeast, the Dam1 ring, or any analogous complete ring-like struc- ture, have been detected so far. Therefore, kt-MT cou- pling based on a real complete sleeve or ring seem highly unlikely in these systems [59]. Based on the ultrastructure of vertebrate kinetochores, [60 -- 62] and in-vitro molecular force spectroscopy [63], it is widely believed that flexible filamentous MT-binding proteins [43, 65], that are components of a kinetochore, can form load-bearing attachments with MTs. The 'binders' appear as one of the core concepts in several recent models that include also the "lawn" model [67], "sliding foot" model [68] , etc. These binders can engage a MT from all angles (see Fig.6). Moreover, unlike the synchronous attachments and detachments of the pos- tulated MT-binding sites on the inner surface of Hill's sleeve [30], the attachment and detachment of these flex- ible filamentous binders are, in general, not synchronous. Furthermore, these filaments do not link among them- selves permanently to form any rigid ring-like or sleeve- like structure. Nevertheless, based on the observations in their in- vitro experiments and Monte Carlo simulations, Powers et al. [63] argue that an effectively biased diffusion mech- anism, similar to that postulated by Hill [30], can still emerge from the fibrous kt-MT linkers even if no rigid sleeve-like structure exist at the surface of a kinetochore. Therefore, the effective potential landscape has also been speculated [64] to be qualitatively similar to that in the Hill sleeve model. Because of the possibility that the binders engage the MT surface practically uniformly and because of the finite maximum stretchable length of the binders, we assume that an effective sleeve-like region may be created (see Fig.6). It is worth pointing out that the effective potential in the Hill sleeve model is corrugated because movement of the sleeve along the MT requires breaking and subse- quent re-establishment of the bonds between MT-binding sites on the inner surface of the sleeve and their specific binding sites on the outer surface of the MT. In the sim- plest version of the SSC model used earlier in this paper, only the tilt of the corrugated potential was retained by assuming a linear potential energy landscape; the corru- gation, which manifests as 'molecular friction', was ig- nored. Even this simplified potential energy landscape was found to be adequate to get a deep insight into the physical mechanism of the catch-bond-like behaviour of the kt-MT attachment. In the same spirit, the effective potential energy land- scape for every individual kt-MT attachment is assumed here also to be linear. Even during a period when y re- mains fixed individual binders can attach to- or detach from the MT surface. Consequently, unlike the original Hill-sleeve model, a major component of the force pulling the MT towards the kt surface could be of entropic ori- gin [66, 67]. A kt-ward pull exerted by a binder bound to curled protofilament at the tip of a depolymerizing MT can suppress the curling, and hence the rate of depoly- merization of the MT just as the Dam1 ring does in case of budding yeast. Thus, both the two postulates (a) and (b), encapsulated by the eqns.(1) and (2), respectively, are assumed to remain valid for each individual MT, pro- vided Vb is interpreted as a potential of mean force. We study the collective strength and stability of this attachment formed by a bundle of parallel MTs by com- puter simulation of molecular force spectroscopy under both force-clamp and force-ramp conditions. To our knowledge, no experimental data are available at present to make direct comparison with the predictions of the general model (N > 1) analyzed in this section. How- ever, very recent experimental breakthroughs [69] sug- gest that both force-clamp and force-ramp experiments with reconstituted mammalian kinetochores in-vitro may become possible in near future. In this extended SSC model at any arbitrary instant of time t, a single kt is attached to n(t) (1 ≤ n(t) ≤ N ) parallel MTs, each through its respective coupler, where N is the maximum number of MTs that can attach to the kt simultaneously. For simplicity, all the couplers are assumed to have identical length L. The MTs are not directly coupled by any lateral bond (transverse to their axis). Instead, all the collective effects arise from their indirect coupling via the kinetochore to which n(t) MTs are attached. The physically motivated assumption of the model, which couple their kinetics is that at any instant of time t, the externally applied load tension F is shared equally among the n(t) MTs that are attached to the kt at that instant through their respective couplers, i.e., F/n(t). We consider two possible scenarios for the rupture of a joint formed by a kt initially with multiple MTs. In the first, once a MT detaches, its re-attachment to the same kt is not allowed. Number of MTs attached with kt, starting from the initial maximum value N , varies irreversibly as N → N − 1 → N − 2 → N − 3 → ...............2 → 1 → 0 (23) In the second scenario, once a MT detaches it can reat- tach again to the same kt and can grow inside the coupler because of its polymerization. So, in this case, the num- ber of MTs attached to the kt varies reversibly as N (cid:42)(cid:41) N − 1 (cid:42)(cid:41) N − 2 (cid:42)(cid:41) N − 3 (cid:42)(cid:41) ...............2 (cid:42)(cid:41) 1 → 0 (24) where the last step is irreversible because of the absorbing boundary condition imposed at n = 0. Extending the WPE prescription [45, 46] used earlier for the single MT-kt attachment, space is now discretized into M cells, each of length h = L/M . Then the time- dependent discrete effective potential is given by (cid:20)(cid:18) (cid:19) Unj kBT = F n(t) − B kBT β0e + (cid:96) − F /n(t) F(cid:63) − α D (cid:21) yj (25) where n(t) is the number of MTs attached to the kt at the instant of time t. Accordingly, the corresponding forward (wf n(j)) and backward (wbn(j)) transition rates can be written by substituting Unj in the place of Uj in 10 FIG. 7. Survival probability is plotted as a function of time t, under force clamp condition (a) in the absence of rebind- ing, for three different values of the tension, F = 0.01 pN, (red circle), F = 0.5 pN (blue square) and F = 1 pN (green triangle), and (b) in the presence of rebinding, for three val- ues of the tension F = 0.01 pN (red circle), F = 0.6 pN (blue square) and F = 1.5 pN (green triangle). In the insets of both the figures the corresponding distributions of the lifetimes are shown. The numerical values of all the other parameters used in the simulation are listed in the table I except, N = 40, F∗ = 1 pN and B = 1 pN. the Eqns.(14),(15). In our simulation of both the scenar- ios mentioned above, initially, all the N MTs are fully inserted into the kt coupler. In the first scenario, using the transition rates given by wf n(j) and wbn(j), the position of a MT tip inside its coupler is updated. But, once an attachment rup- tures, its reattachment to the kt is not allowed; therefore, detached MT is no longer monitored in our simulation. However, the simulation is continued till the last surviv- ing MT-kt attachment ruptures. This first passage time is identified as the life time of the molecular joint consist- ing of N MTs with a single kt. The process is repeated 106 times, starting from the same initial condition, to obtain the distribution of the lifetimes. In the same sce- nario, under the force-ramp condition (F = at) we collect the data similarly to obtain the distribution of rupture forces (i.e., the force at which the tip of the last surviving MT exits from its coupler). In the alternative scenario, the transition rates wf n(j) and wbn(j) govern the kinetics of the tip of each MT as 11 FIG. 8. The mean life time < t > is plotted against (a) ten- sion F for N = 40 and (b) number N for F = 10 pN, B = 0.5 pN, each for both the scenarios, namely with rebinding (blue circle) and without rebinding (red triangle). (b) We found best fit of our simulation data with the curve < t >∝ N 0.53, represented by black continuous line. The numerical values of all the other parameters used in the simulation are listed in the table I except, N = 40, F∗ = 1 pN and B = 1pN. Error bars represent standard deviation of the simulation data. FIG. 9. Survival probabilities, under force-ramp condi- tion, for three different loading rates a = 18pN/s (red circle), 20pN/s (blue square) and 22pN/s (green triangle) are plot- ted (a) in the absence of rebinding and (b) in the presence of rebinding . In the inset the corresponding distributions of the rupture forces are shown. The numerical values of all the other parameters are listed in the table I. long as it moves inside the corresponding coupler. How- ever, once the attachment between a MT and the kt, through the coupler, ruptures it must get an opportunity to reattach through its natural kinetics of polymerization and depolymerization outside the coupler. Therefore, in this scenario, the continuing forward and backward movement of the tip of a detached MT outside its cou- pler is monitored in our simulation. During this period the force-free kinetics of the MT tip outside its coupler is implemented in our simulation by replacing the potential (25) by the simpler potential (cid:20) β0 − α (cid:21) D Vj kBT = (cid:96) yj (26) and simultaneously replacing the transition rates wf n(j) and wbn(j) by wf 1(j) = D h2 − δV j exp(− δVj kB T − 1) kB T (27) and wb1(j) = D h2 δVj kB T exp( δVj kB T − 1) , (28) respectively, where δVj = Vj+1 − Vj. If, through this kinetics outside the coupler, a MT succeeds in re-entering its coupler its kinetics reverts back to that governed by the transition rates wf n(j) and wbn(j). Thus, starting from the initial state the time evolution of all the MTs are monitored till the instant when, for the first time, none of the MTs is attached to the kt; this first-passage time is identified as the lifetime of the attachment. Repeating this process we have obtained the distributions of the lifetimes in the second scenario. Similarly for the ramp force we have obtained the distribution of the rupture force which is defined as the force at which, for the first time, none of the MTs is attached to the kt. A. Results on life time distribution under clamp force for N > 1 In Fig.7 (a) and (b) survival probabilities of an attach- ment, consisting initially of 40 MTs and a single kt, have been plotted as a function of time for the two cases where rebinding is (a) forbidden and (b) allowed, respectively. The attachment survives for longer duration in interme- diate range of the clamp force (F = 0.5pN, blue square) than at the high and low strength of the tension. In the inset the corresponding distributions of the lifetimes of the attachments are also shown. The trends of variation of the survival probability with the clamp force indicates a catch-bond-like behavior. In- deed, this catch-bond-like behavior can be seen directly in Fig.8(a) where the mean life time < t >, plotted against the clamp force F , displays a maximum at a non- zero finite value of F irrespective of whether rebinding of the MTs is allowed or forbidden. The physical cause of the catch-bond-like behavior is the same as that pointed out in the special case N = 1. Moreover, as expected on physical grounds, for any given F , the mean life time 12 < t > is higher if rebinding is allowed as compared to the mean life time in the absence of rebinding. In the Fig.8(b) mean lifetime is found to increase with the number of microtubule (N). This is consistent with one's intuitive expectation. Besides, for any given value of N , allowing rebinding of the MTs results in a higher life time. However, the interesting point is that the mean life time does not exhibit trivial linear increase with N . Instead, it increases nonlinearly (more precisely, sub- linearly) with N in both the cases. Though the parallel MTs do not interact with one another laterally but only by equal sharing of the instantaneous load, the nonlinear behavior is a collective emergent property of the inter- acting system. Recent reconstitution of mammalian kt in-vitro [69] have raised the hope of indicates promising new routes for testing our results for N > 1. B. Results on rupture force distribution under force ramp for N > 1 FIG. 10. Survival probability under force-ramp condition and probability density of rupture force (in the inset) of the kt- MT attachment with N = 40 for both in the presence and absence of rebinding for four different loading rates, namely, (a) a = 1 × 10−2pN s−1, (b) a = 2 × 10−2pN s−1, (c) a = 3 × 10−2pN s−1 and (d) a = 0.1pN s−1 are plotted. The numerical values of all the other parameters used in the simulation are listed in the table I except, N = 40, F∗ = 0.5pN, B = 1.5pN and α = 50s−1. 13 reports on different ligand-receptor bonds [22, 23, 58, 70]. The contrast of the qualitative trends of variation of the rupture force distributions in the Figs.9 and 10 also em- phasizes the role of the importance of the energetics and kinetics of the MTs in the catch-bond-like behavior. In the Fig.11(a) and (b) the average rupture force is plotted, respectively, against the loading rate a (for a given N ) and against N (for a given loading rate a). The log-scale along the X-axis in Fig.11(a) is used to cover a wide range of loading rates in the most suitable man- ner. The higher survival probability caused by reattach- ment of MTs is more pronounced at slower loading than at faster loading. This trend of variation follows from the fact that at faster loading detached MTs get smaller chances of reattaching before the complete rupture of the attachment. What is interesting from the quantitative point of view is that the average rupture force increases nonlinearly with increasing loading rate. For high loading rate average rupture force < F > follows a linear trend [71], but here nonlinear behavior arises because, faster loading rates allow less time for the dissociation and de- polymerization processes, ultimately leading to rupture of MT-kt bonds. Finally, the increase of the mean rup- ture force with increasing N also seems to be nonlinear. VI. DISCUSSION, SUMMARY AND CONCLUSION In this paper we have developed theoretical models of molecular joints formed by N (> 1) parallel MT filaments with a single kt by extending the SSC model [28] that was developed for the special case N = 1. By carrying out extensive kinetic Monte Carlo simulations of our theoret- ical models of kt-MT attachments, we have computed the probability distributions, and hence the mean values, of the lifetimes and rupture forces which are the two main characteristic statistical properties of such transient at- tachments. The SSC model with N = 1 [28] not only reproduced the catch-bond-like behaviour of the kt-MT attachments observed in force-clamp experiments in-vitro for bud- ding yeast [21], but also elucidated a plausible under- lying mechanism that gives rise to this counter-intuitive phenomenon. However, in ref.[28], the lifetimes of the at- tachments were calculated for only a single unique initial condition. In the first half of this paper we have presented new results for some other initial conditions to convinc- ingly establish that the qualitative conclusions drawn in ref.[28] are valid for all possible initial conditions. More- over, we have presented further evidence in favour of the catch-bond-like behavior, for the same N = 1 case, by reporting 'mechanical signatures' of typical catch-bond observed in in our in-silico force-ramp experiments. In the second half of this paper we have extended the SSC model to the more general case N > 1. In this case, the possibility of re-attachment of a detached MT to the same kt, before the last surviving MT gets detached, can FIG. 11. Mean rupture force is plotted against (a) the loading rate a for a fixed N = 40, and (b) N for a fixed loading rate a = 10 pN/s, F∗ = 1 pN and B = 1 pN. Logarithmic scale is used along the X-axis in (a) to cover a very broad range of a. In the inset of (a) the mean rupture force is plotted for relatively lower loading rates where the difference in the data for the two cases, namely with and without rebinding, are more pronounced and clearly visible. (b) The best fit to our simulation data is obtained with the curve < F >∝ N 0.39, represented by black continuous line. Error bars (small green dot) represent standard deviation. The numerical values of all the other parameters are listed in the table I. In the Fig.9(a) and (b) survival probabilities (and the corresponding rupture force distribution in the insets) are plotted, respectively, in the absence and presence of rebinding for three different loading rates a = 18pN/s, 20pN/s and 22pN/s. Survival probability remains high upto a certain force beyond which it drops quite sharply. The rupture force distribution in this figure does not dis- play the bimodal form seen earlier in Fig. 4 and Fig. 5 for N=1. In contrast, in the Fig.10, where the survival probabilities and the corresponding rupture force distri- bution (in the insets) are plotted for a slightly different set of values of the key parameters F∗ and B, a bimodal form is found. Moreover, the trend of variation of rup- ture force distribution and survival probability is similar to those observed in Fig. 4 and Fig. 5(a) for N=1 sce- nario. There is a minor difference between the bimodal forms of the rupture force distributions in Fig.4 for N=1 and Fig.10 for N > 1; the first peak in the former appears at F = 0 whereas that in the latter corresponds to a non- zero value of F . However, both are consistent with earlier prolong the lifetime. We present simulation data to es- tablish that, in spite of this additional complexity that did not exist in the special case N = 1, the kt-MT attach- ment still exhibits a catch-bond-like behavior in a part of the parameter space of this model. As a byproduct of this investigation we also find that both the mean life- time and mean rupture force scale nonlinearly with N . This result is important from the perspective of collective phenomena. Although in our models there is no direct lateral interaction among the MTs, the indirect interac- tions among the MTs is mediated by the kt to which all the MTs are attached. This indirect interactions give rise to the non-trivial nonlinear scaling of the mean lifetime and mean rupture force with N . Similar trends of vari- ation of non-covalent bond rupture characteristics with increasing number of ligands have been observed in the past [72]. The SSC model [28], and its extensions reported in this paper, are minimal models based on two key assumptions encapsulated by the Eqs.(1) and (2). The first postu- late (1) incorporates the main feature of the energetics of MT-coupler interactions that implicitly depends also on the structure of the kt-MT coupler. The second postu- late (2) captures the most essential aspect of the kinetics of depolymerization of microtubules under load tension. These minimal models draw heavily on biased diffusion of Hill's sleeve [30] and conformational wave based on curl- ing of depolymerizing tip of MT [73]. Both these models, however, were proposed long before the composition and structure of kt could be explored at the molecular level [10]. We have argued that our minimal models can also be interpreted so as to make these consistent with the recent structural models of mammalian kinetochores be- cause our minimal models do not explicitly assume any specific structure of the kt-MT coupler. A mechanical model, in terms of beads connected by springs, was devel- oped by Bertalan et al.[74]; in this model, the attachment is assumed to be formed by the insertion of the curling protofilament hook into the loops formed by the kineto- chore fibrils. It has not been possible to identify measures that would differentiate between our kinetic models and the more explicit structural model developed by Bertalan et al. [74]. One of the unique features of the polymerization kinet- ics of a MT is its dynamic instability [75]. A polymer- izing MT keeps growing in length till it suffers a "catas- trophe" whereby it abruptly begins to depolymerize. A depolymerizing MT would, eventually, disappear unless its rapid shrinkage is stopped by a process called 'rescue" following which it resumes polymerization. The theory for this phenomenon of dynamic instability, that began with Hill's pioneering work [76], has been re-formulated and improved over the subsequent years [77 -- 84] (see ref.[38, 85, 86] for reviews). The 2-state model that Akiyoshi et al.[21] used to account for their experimental data explicitly describe switching of the MT between the growing and shrinking stages because of catastrophe and rescue. This model 14 was extended by Zhang [87] assigning additional distinct mechano-chemical states that enable capturing the de- pendence of the MT catastrophe rate on the GTP-tubulin concentration. However, neither of these two versions of the 2-state model throw light on the physical origin of the phenomenon in terms of the structure and dynam- ics of the kt-MT attachment. Any explicit description of the kinetics of the growing and shrinking MTs separately would require equations that govern the time evolutions of probability densities P±(x, t) of the polymerizing (+) and depolymerizing (-) MTs. In contrast, the SSC model, as well as the extended versions studied in this paper, de- scribe MT kinetics in terms of a single probability den- sity P (x, t) = P+(x, t) + P−(x, t). The assumption that P (x, t) alone provides an alternative, but adequate, de- scription of the generic features of the molecular force spectroscopy of kt-MT attachment is an assumption that is well justified by the results. In recent years, strain-dependent detachment of molec- ular motors like dynein and myosin have revealed catch- bond like stabilization of the track-bound state of the motor by externally applied tension [88 -- 91]. Such catch- bonds have important biological functions in cell adhe- sion, mechanosensation, mechano-transduction, immune response, bacterial mechanics, etc. [92 -- 95]. Here we have modelled and analyzed the kt-MT attachment by drawing analogy with common ligand-receptor bonds. Elsewhere we have invoked similar analogies [96, 97] for studying transient attachments formed by MTs in the mitotic spindle, namely at the cell cortex [98] and at the spindle pole [99]. Conceptually, this a leap forward be- cause the MTs, the analogs of ligands, are self-organized supra-molecular structures made of building blocks each of which itself is a macromolecule while the kt, the coun- terpart of a receptor, is also a complex structure made of made macromolecules. In case of common ligands at least three different ge- ometries can be distinguished; (a) N ligands in parallel where each one is subjected to a load F/N if the load F is shared equally by all, (b) N ligands in series where all the ligands are subjected to the same load F , and (c) N ligands in 'zipper' configuration where only the bond at the leading edge bears the entire load F while no load is experienced by the others. Moreover, in case of parallel geometry, the flexibility of the long ligands can have significant effect on the manner in which the load is shared. In contrast, each MT is quite stiff. Our model with N > 1 corresponds to the 'parallel' geometry where, at any instant, the load is shared equally by those MTs that are still attached to the kt at that instant of time. We also stress that, in spite of these superficial simi- larities, there are several crucial differences in the under- lying physical mechanisms because of which none of the mechanism responsible for the catch-bonds in common ligand-receptor systems [26, 27, 100 -- 102] is directly ap- plicable to the kt-MT attachment. The main sources of these differences arise from the fact that (i) each MT tip can grow or shrink because of ongoing polymerization or depolymerization of the MT and (ii) the rate of depoly- merization is strongly suppressed by externally applied tension. It is precisely for this reason that we regard the kt-MT attachments as "unusual" in spite of the fact they display the usual signatures of catch-bonds. manuscript. DC also thanks Raymond Friddle, Gaurav Arya, D. Thirumalai and Shaon Chakrabarti for useful correspondences. This work has been supported by a J.C. Bose National Fellowship (DC) and "Prof. S. Sampath Chair" Professorship (DC). 15 ACKNOWLEDGEMENT One of the authors (DC) thanks Charles Asbury for valuable comments on a shorter preliminary draft of this REFERENCES [1] T. Wittmann, A. Hyman and A. Desai, Nat. Cell Biol. 3, E28 (2001). [2] E. Karsenti and I. Vernos, Science 294, 543 (2001). [3] K.J. Helmke, R. Heald and J.D. Wilbur, Int. Rev. Cell Mol. Biol. 306, 83 (2013). [27] V. Bargeson, D. Thirumalai, PNAS 102, 1835 (2005). [28] A. K. Sharma, B. Shtylla and D. Chowdhury, Phys. Biol.11, 1478 (2014). [29] T. Bameta, D. Das, D. Das, R. Padinhateeri and M. Inamdar, Phys. Rev. E 95, 022406 (2017). [4] E. Karsenti, Nat. Rev. Mol. Cell Biol. 9, 255 (2008). [5] J. Frank (ed.), Molecular Machines: Workshops of the [30] T. Hill, Proc. Natl. Acad. Sci. U.S.A. 82, 4404 (1985). [31] G.J. Buttrick and J.B.A. Millar, Chromosome Res. 19, cell (Cambridge University Press, 2011). [6] J.R. McIntosh, M.I. Molodotson and F.I. [32] S. Westermann, D.G. Drubin and G. Barnes, Annu. 393 (2011). Ataullakhanov, Quart. Rev. Biophys. (2012). Rev. Biochem. 76, 563 (2007). [7] S. Dumont and M. Prakash, Mol. Biol. of the Cell 25, [33] T.N. Davis and L. Wordeman, Trends in Cell Biol. 17, 3461 (2014). 377 (2007). [8] S. Reber and A.A. Hyman, Cold Spring Harb. Perspct. [34] E.A. Foley and T.M. Kapoor, Nat. Rev. Mo. Cell Biol. Biol. 7, a015784 (2015). 14, 25 (2013). [9] J. L.D. Lawson and R.E. C. Salas, Biochem. Soc. Trans. [35] A. Efremov, E.L. Grishchuk, J.R. McIntosh and F.I. 41, 1736 (2013). Ataullakhanov, PNAS 104, 19017 (2007). [10] I.M. Cheeseman, Cold Spring Harb. Perspect. Biol. 6, [36] C.L. Asbury, J.F. Tien and T.N. Davis, Trends in Cell a015826 (2014). Biol. 21, 38 (2011). [11] S. Petry, Annu. Rev. Biochem. 85, 659 (2016). [12] E.C. Yusko and C.L. Asbury, Mol. Biol. of the Cell 25, 3717 (2014). [13] Margolis and Wilson, Nature (2981). [14] E. Evans, Faraday Discuss., 111, 1 (1998). [15] M. Karplus, J. Mol. Recognit. 23, 102 (2010). [16] A.R. Bizzarri and S. Cannistraro (eds.) Dynamic Force Spectroscopy and Biomolecular Recognition, (CRC Press, 2012). [37] E.L. Grishchuk, in: Centromeres and Kinetochores, ed. B.E. Black (Springer, 2017). [38] H. Bowne-Anderson, M. Zanic, M. Kauer and J. Howard, Bioessays 35, 452 (2013). [39] A. D. Franck, A.F. Powers, D.R. Gestaut, T. Gonen, T. N. Davis and C.L. Asbury, Nat. Cell Biol. 9, 832, (2007). [40] H. Risken, The Fokker-Planck Equation (Springer, 1996). [17] A.D. Franck, A.F. Powers, D.R. Gestaut, T.N. Davis [41] L.A. Mirny and D.J. Needleman, Meth. Cell Biol. 95, and C.L. Asbury, Methods 51, 242 (2010). 583 (2010). [18] S. Biggins, genetics 194, 817 (2013). [19] B. Akiyoshi and S. Biggins, Chromosoma 121, 235 [42] A. P. Joglekar and A.J. Hunt, Biophys. J. 83, 42 (2002). [43] B. Shtylla and J. P. Keener, SIAM J. Appl. Math. 71, (2012). 1821 (2011). [20] K.K. Sarangapani and C.L. Asbury, Trends in Genet. [44] J. C. Waters, T.J. Mitchison, C.L. Rieder and E. D. 30, 150 (2014). [21] B. Akiyoshi, K. K. Sarangapani, A. F. Powers, C. R. Nelson, S. L. Reichow, H. Arellano-Santoyo, T. Gonen, J. A. Ranish, C. L. Asbury and S. Biggins, Nature 468, 576 (2010). [22] W. E. Thomas, Annu. Rev. Biomed. Eng. 10, 39 (2008). [23] W. E. Thomas, V. Vogel and E. Sokurenko, Annu. Rev. Biophys. 37, 399 (2008). [24] E.V. Sokurenko, V. Vogel and W.E. Thomas, Cell Host & Microbe 4, 314 (2008) Salmon, Mol. Biol. Cell. 7, 1547 (1996). [45] H. Wang, C. Peskin and T. Elston, J. Theo. Biol. 221, 491 (2003). [46] H. Wang, T. C. Elston, J. Stat. Phys. 128, 35 (2007). [47] G.I. Bell. Science, 200, 618, (1978). [48] E. Evans and K. Ritchie, Biophy J.,72, 1541, (1997). [49] E Evan and P. Williams, in: Phys. of Biomolecules and Cells, eds. H. Flyvbjerg, F. Julicher, F. Ormos and F. David (Springer and EDP Sciences, 2002). [50] E. Evans, Annu. Rev. Biophys. Biomol. Struct. 30, 105 [25] O.V. Prezhdo and Y.V. Pereverzev, Acc. Chem. Res. (2001). 42, 693 (2009). [26] S. Chakrabarti, M. Hinczewski and D. Thirumalai, J. Struct. Biol. 197, 50 (2017). [51] R. W. Friddle, in: Dynamic Force Spectroscopy and Biomolecular Recognition, ed. A.R. Bizzarri and S. Can- nistraro (CRC Press, 2012). 16 [52] G. Arya, Molec. Simul. 42, 1102 (2016). [53] S. Gonen, B. Akiyoshi, M.G. Iadanza, D. Shi, N. Dug- gan, S. Biggins, T. Gonen, Nat. Struct. Mol. Biol. 19, 925 (2012). [54] A. P. Joglekar, D. Bouck, K. Finley, X. Liu, Y. Wan, J. Berman, X. He, E.D. Salmon and K.S. Bloom, J. Cell Biol. 181, 587 (2008). [55] K. Johnston, A. Joglekar, T. Hori, A. Suzuki, T. Fuka- gawa, and E.D. Salmon, J. Cell Biol. 189, 937 (2010). [56] A. P. Joglekar, D. C. Bouck, J. N. Molk, K. S. Bloom and E. D. Salmon, Nat. Cell Biol. 8, 581 (2006). [57] W. F. Marshall, J. F. Marko, D. A. Agard and J. W. Sedat, Curr. Biol. 11, 569 (2001). [75] A. Desai and T.J. Mitchison, Annu. Rev. Cell Dev. Biol. 13, 83 (1997). [76] T.L. Hill, Proc. Natl. Acad. Sci. USA 81, 6728 (1984). [77] M. Dogterom and S. Leibler, Phys Rev Lett. 70, 1347 (1993). [78] D. J. Bicout, Phys. Rev. E. 56, 6656-6667, (1997). [79] T. Antal, P.L. Krapivsky, S. Redner, M. Mailman and B. Chakraborty, Phys. Rev. E 76, 041907 (2007). [80] Sumedha, M.F. Hagan and B. Chakraborty, Phys. Rev. E 83, 051904 (2011). [81] D.J. Needleman, A. Gronen, R. Ohi, T. Maresca, L. Mirny and T. Mitchison, Mol. Biol. Cell 21, 323 (2010). [82] J. Brugues, V. Nuzzo, E. Mazur and D.J. Needleman, [58] E. Evans, A. Leung, V. Heinrich and C. Zhu, PNAS Cell 149, 554 (2012). 101, 11281 (2004). [83] K. Ishihara, K.S. Korolev and T.J. Mitchison, eLife 5, [59] J.R. McIntosh, E. O'Toole, K. Zhudenkov, M. Mor- phew, C. Schwartz, F.I. Ataullakhanov and E.L. Gr- ishchuk, J. Cell Biol. 200, 459 (2013). e19145 (2016). [84] F. Decker, D. Oriola, B. Dalton and J. Brugues, preprint (2018). [60] Y. Dong, K.J. VandenBeldt, X. Meng, A. Khodjakov [85] F.I. Ataullakhanov, K.S. Melnik and A.A. Butylin, Bio- and B.F. McEwen, Nat. Cell Biol. 9, 516 (2007). phys. (Moscow), 58, 120 (2013). [61] B.F. McEwen and Y. Dong, Cell. Mol. Life Sci. 67, 2163 [86] H. Bowne-Anderson, A. Hibbel and J. Howard, Trends (2010). [62] J.R. McIntosh, E.L. Grishchuk, M.K. Morphew, A. K. Efremov, K. Zhudenkov, V.A. Volkov, I.M. Cheeseman, A. Desai, D.N. Mastronarde and F.I. Ataullahkhanov, Cell, 135, 322 (2008). [63] A.F. Powers, A.D. Franck, D.R. Gestaut, J. Cooper, B. Gracyzk, R.R. Wei, L. Wordeman, T.N. Davis and C.L. Asbury, Cell 136, 865 (2009). Cell Biol. 25, 769 (2015). [87] Y. Zhang, J. Biol. Chem. 286, 39439 (2011). [88] T.. Erdmann and U.S. Schwarz, Phys. Rev. Lett. 108, 188101 (2012). [89] T. Erdmann , K. Bartelheimer and U.S. Schwarz, Phys. Rev. E 94, 052403 (2016). [90] Y. Inoue and T. Adachi, Phys. Rev. E 93, 042403 (2016). [64] S. Santaguida and A. Musacchio, EMBO J. 28, 2511 [91] A. Nair, S. Chandel, M. Mitra, S. Muhuri and A. Chaud- (2009). [65] J. P. Keener and B. Shtylla, Biophys. J. 106, 998 (2014). [66] A.V. Zaytsev, F.I. Ataullakhanov and E.L. Grishchuk, huri, Phys. Rev. E 94, 032403 (2016). [92] F.J. Vernerey and U. Akalp, Phys. Rev. E 94, 012403 (2016). Cell. Mol. Bioengg. 6, 393 (2013). [93] D.E. Leckband and J. de Rooij, Annu Rev. Cell Dev. [67] A.V. Zaytsev, L.J.R. Sundin, K.F. DeLuca, E.L. Gr- ishchuk and J.G. DeLuca, J. Cell Biol. 206, 45 (2014). [68] P.L. Janczyk, K.A. Skorupka, J.G. Tooley, D.R. Mat- son, C.A. Kestner, T. West, O. Pornillos and P.T. Stuck- elberg, Dev. Cell 41, 438 (2017). [69] J.R. Weir et al. Nature 537, 249 (2016). [70] J. Kim, C.Z. Zhang, X. Zhang and T.A. Springer, Na- ture 466, 992 (2010). [71] P. Williams, Anal. Chim. Acta.,479, 107, (2003). [72] T.A. Sulchek, R.W. Friddle, K. Langry, E.Y. Lau, H. Albrecht, T.V. Rattp, S.J. DeNardo, M.E. Colvin and Al. Noy, PNAS 102, 16638 (2005). [73] D.E. Koshland, T.J. Mitchison and M.W. Kirschner, Biol. 30 291 (2014). [94] M. Huse, Nat. Rev. Immunol. 17, 679 (2017). [95] A. Persat et al. Cell 161, 988 (2015). [96] D. Ghanti, R.W. Friddle and D. Chowdhury, preprint (2017). [97] D. Chowdhury et al. (2018). [98] H.Y. Wu, E. Nazockdast, M.J. Shelley and D.J. Needle- man, Bioessays 39, 1600212 (2016). [99] K. K. Fong, K.K. Sarangapani, E.C. Yusko, M. Riffle, A. Llauro, T.N. Davis and C.L. Asbury, Mol. Biol. Cell (2017). [100] S. Chakrabarti, M. Hinczewski and D. Thirumalai, PNAS 111, 9048 (2014). Nature 331, 499 (1988). [101] S. Rakshit and S. Sivasankar, Phys. Chem. Chem. Phys. [74] Z. Bertalan, C.A.M. La Porta, H. Maiato and S. Zap- 16, 2211 (2014). peri, Biophys. J. 107, 289 (2014). [102] B. Liu, W. Chen and C. Zhu, Annu. Rev. Phys. Chem. 66, 427 (2015).
1510.04722	1	1510	2015-10-15T21:28:46	Disordered nano-wrinkle substrates for inducing crystallization over a wide range of concentration of protein and precipitant	[ "physics.bio-ph", "cond-mat.soft" ]	There are large number of proteins, the existence of which are known but not their crystal structure, because of difficulty in finding the exact condition for their crystallization. Heterogeneous nucleation on disordered porous substrates with small yet large distribution of pores is considered a panacea for this problem, but a universal nucleant, suitable for crystallizing large variety of proteins does not really exist. To this end, we report here a nano-wrinkled substrate which displays remarkable ability and control over protein crystallization. Experiments with different proteins show that on these substrates, crystals nucleate even at very low protein concentration in buffer. Small number of very large crystals appear for precipitant concentrations varied over orders of magnitude ~0.003-0.3M; for some proteins, crystals appear even without addition of any precipitant, not seen with any other heterogeneous substrates. In essence, these substrates significantly diminish the influence of the above two parameters, thought to be key factor for crystallization, signifying that this advantage can be exploited for finding out crystallization condition for other yet-to-be-crystallized proteins.	physics.bio-ph	physics	This document is the unedited author's version of a submitted work that was subsequently published by Langmuir, after peer review. To access the final edited and published work, go to Langmuir, 2013, 29 (13), 4373–4380, DOI 10.1021/la305135y Disordered nano-wrinkle substrates for inducing crystallization over a wide range of concentration of protein and precipitant By Anindita Sengupta Ghatak1 and Animangsu Ghatak1,2,* [1] Department of Chemical Engineering and DST Unit on Soft Nanofabrication, Indian Institute of Technology, Kanpur, 208016 (India) [2] Max Plank Institute for polymer research, Ackermannweg 10, D-55128, Mainz, Germany [] Prof. A. Ghatak Corresponding-Author, Author-Two, E-mail: [email protected] ABSTRACT There are large number of proteins, the existence of which are known but not their crystal structure, because of difficulty in finding the exact condition for their crystallization. Heterogeneous nucleation on disordered porous substrates with small yet large distribution of pores is considered a panacea for this problem, but a universal nucleant, suitable for crystallizing large variety of proteins does not really exist. To this end, we report here a nano-wrinkled substrate which displays remarkable ability and control over protein crystallization. Experiments with different proteins show that on these substrates, crystals nucleate even at very low protein concentration in buffer. Small number of very large crystals appear for precipitant concentrations varied over orders of magnitude ~0.0030.3M; for some proteins, crystals appear even without addition of any precipitant, not seen with any other heterogeneous substrates. In essence, these substrates significantly diminish the influence of the above two parameters, thought to be key factor for crystallization, signifying that this advantage can be exploited for finding out crystallization condition for other yet-to-be-crystallized proteins. 1. INTRODUCTION Crystallization of protein molecules from supersaturated solution is a complex phenomenon, dictated by several variables: rate of evaporation of water, concentration of proteins, chemical nature of precipitants and additives, temperature of the surrounding environment, pH of the buffer solution and even physical/chemical heterogeneity of the substrate. As a result, determining the exact recipe and range of values of different parameters for generating diffraction quality crystals of protein poses a significant scientific challenge. Therefore, despite many attempts, still there are several proteins which have not yet been crystallized with size suitable for X-ray diffraction. Naturally, the search for a suitable process/material, which can decrease the potential energy barrier for formation of the critical nuclei, has been a subject of vigorous experimental and theoretical research13, leading to the realization that a chemically and/or physically modified heterogeneous substrate48 can possibly serve as a universal nucleant9. An example of such substrate is porous material with pore sizes approximately same as that of a critical nucleus; for example, porous silicon and glass surfaces1013, porous polymeric membrane have pores of sizes 40-200 nm14, bioactive gel-glass5, nano-porous materials15,16 with pore sizes in the range of 210 nm are indeed found to be effective nucleating agent. A direct correlation exists between the pore size and the radius of gyration gR of the protein pd molecules: while pores with d  p R g do not contribute to nucleation, those with d  p R g do not diminish the energy barrier sufficiently enough for nucleation to occur. Furthermore, uniformity in pore size too is detrimental as it does not facilitate breaking of symmetry and induce nucleation. Uniformity in diameter of nucleant also allows only protein molecules of specific size to be crystallized15. Therefore, it is hypothesized that disordered porous materials are most suitable for promoting nucleation5,10 thus underlining the need of a substrate with very small pores but with wide distribution in pore size. It is in this context that we present here a novel nano-wrinkled substrate, the local radius of curvature of which varies over a wide range from - 10nm to 10nm and over a length scale of 125nm. These substrates are prepared in a two stage process. In the 1st stage, a thin strip of an elastomer, e.g. poly(dimethylsiloxane) (PDMS) is stretched to a desired extension ratio,  ~ 1.2 – 1.4 and is plasma oxidized to generate a thin silica crust and is then released instantaneously leading to longitudinal surface folds, because of the difference in deformability of the stiff skin and the soft under-layer17-19. The atomic force microscopy (AFM) images in figure 1a and b depict these 1st generation wrinkles (hereon referred to as “S1” surface) which appear far from sinusoidal in contrast to similar experiments but with slower release of slightly plasma-oxidized films17. The wrinkles remain asymmetric with cylindrical folds merging along a deep furrow forming a sulcus like structure; these are also the locations where the curvature of these surfaces becomes maximum. The maximum curvature max however does not exceed the inverse material length scale   1 , because, surface tension does not allow larger curvature to be achieved20. For PDMS with surface tension,  = 22 mN/m and shear modulus  = 0.6 MPa, max is found to be ~0.02 nm-1, thus defining a limiting lengthscale for most soft-lithography methods. Increasing the modulus of the elastomer does not help increasing the max , because then the advantage of using a soft, deformable material for patterning is compromised. We overcome this difficulty by choosing a material which is of low modulus at the bulk, but stiff at the vicinity of its surface. In particular, wrinkled substrates prepared in the 1st stage (Figure 1ab) is used for this purpose, because, the presence of the thin crust increases its modulus close to the surface without altering it at the bulk. When we subject such a substrate to a second cycle of above process, finer 2nd generation wrinkles as represented by Figure 1cd (hereon referred to as “S2” surface) with  as large as 0.1 nm-1 appear at the location of merger of folds. In addition, the standard deviation of the distribution of curvature too remains very large. Importantly, curvature varies continuously from positive to the negative maxima, both of which are useful for crystal nucleation4, an advantage not available with conventional porous substrates, for which, the surface remains essentially flat except at the location of pores, where it remains large but nearly equal13 for all pores. 2. EXPERIMENTAL SECTION 2.1. Materials. Ferritin, a protein extracted from horse spleen and thaumatin extracted from bacteria thaoumatococcus daniellii were purchased from Sigma chemicals. Glucose isomerase obtained from bacteria streptomyces rubiginosus and xylanase from trichoderma reesei were purchased from Hampton research. ADA buffer and Na/K tartrate were purchased from Hampton research. CdSO4.8H2O, Sodium azide, CaCl2.2H2O, NaCl, Tris HCl, PEG 8000, (NH4)2SO4 were all purchased from Marc. Izit dye was purchased from Hampton Reseach. The proteins were used for the experiments without further purification. Deionized water obtained from Mili Q system was used throughout the study. Sylgard 184 elastomer was obtained from Dow Corning and octadecyltrichlorosilane (OTS) molecules were obtained from Sigma chemicals. 2.2. Method of preparation of protein solutions. 2.2.1. Ferritin. The protein solution of 50mg/ml was obtained from the supplier, which was diluted to 20mg/ml with deionized water. In addition, 10% 3CdSO4. 8H2O, 3 mM NaN3 and (NH4)2SO4 in the concentration range of 0.0  2.0 M were used for the crystallization experiments. Each drop of the pre-crystallization solution was prepared by mixing 0.8 µl of each of the above CdSO4. 8H2O, NaN3 and (NH4)2SO4 solution and 2.4 µl of protein solution. Thus the protein concentration in the resultant pre-crystallization solution was 10 mg/ml whereas the concentration of (NH4)2SO4 lie in the range of 0.0  0.3 M. 2.2.2. Thaumatin. The protein was obtained from Sigma chemicals as dry flakes, which was dissolved in 0.1M ADA buffer (pH = 6.5) to attain a concentration of 30 mg/ml. Na/K tartrate in concentration range: of 0.01.5 M was used as the precipitant for crystallizing the protein. 2 µl of each of the protein solution and that of Na/K tartrate was mixed in the pre-crystallization solution. Thus the resultant solution had a protein concentration of 15 mg/ml and the precipitant concentration in the range of 0.0  0.75 M. 2.2.3. Glucose Isomearse. The protein solution obtained at 33mg/ml was diluted to 10mg/ml concentration with deionized water. 0.1M HEPES (pH = 7.5) and (NH4)2SO4 in the concentration range of 0.2  2.0M were used as precipitants for crystallization. Each drop of the pre- crystallization solution contained 1.0 µl of each of the HEPES, (NH4)2SO4 solution and 2 µl of the protein solution. The resultant solution thus had 5 mg/ml of protein and 0.05  0.5 M of (NH4)2SO4. 2.2.4. Xylanase. The protein solution obtained at 36mg/ml was diluted to 10 mg/ml with deionized water. For the crystallization, we used 0.1M NaCl, 0.05M Tris HCl (pH = 8.5), 15% (w/v) PEG 8000, and CaCl2 in the concentration range of (0.21.2) M. Each drop contained 0.7 µl of each of NaCl, Tris HCl, PEG 8000, CaCl2. 2H2O solution and 2.8 µl of the protein solution. Thus the protein and precipitant concentration in the resultant pre-crystallization solution was maintained at 5 mg/ml and 0.0025  0.15 M respectively. 2.3. Method of preparation of nano-wrinkled surfaces. Thin strips of PDMS film (length 25 mm, width 10 mm, thickness ~ 350µm) were uniaxially stretched to a desired extension ratio using a home-made gadget, following which, the films were exposed to radio frequency oxygen plasma (Harrick Plasma, model PDC-32G, pressure ~0.05 Torr, power ~6.8W, time ~ 4 min). The stretched films were then released instantaneously without using any motion control device. This process led to the appearance of surface wrinkles because of difference in stiffness of the thin silicate crust (elastic modulus ~76 GPa) and the elastic substrate (modulus ~3 MPa)19. Wrinkles of different periodicity and amplitude were generated by varying the extension ratio  ~ 1.2 – 1.4. Here-on we refer to these surfaces with 1st generation wrinkles as the “S1” surface. The PDMS films patterned with the 1st generation wrinkles were subjected to a repeated sequence of the above three steps (time for plasma oxidation ~ 1 min), which yielded the 2nd generation wrinkles. We refer to these surfaces as “S2 ” surface. During preparation of both the 1st and 2nd generation samples, the PDMS strip was stretched uniaxially along identical direction and to identical extension ratio. In addition to these two different types of substrates, we used also flat wrinkle free surface of PDMS which we turned hydrophilic by small degree of plasma oxidization (pressure ~0.05 Torr, power ~6.8W, time ~ 1 min). We refer to these surfaces as “S” which we used as the control surface. 2.4. Atomic Force Microscopy (AFM) measurement. The wrinkled substrates were imaged in tapping mode by using Pico Plus Atomic force microsope, (Molecular Imaging) at room temperature. The scan size was 2m x 2m and 5m x 5m. 2.5. Method for obtaining curvature from the AFM images of nano-wrinkled substrates. In order to calculate the curvature, such a profile was first divided into several segments, each of which was fitted to a polynomial equation: y   0  nxa 0 n n . After several trials, a 10th order polynomial was found to perfectly capture the surface profiles, including the junction of the folds where curvature was maximum. The curvature was then calculated as yd 2 2 dx and plotted in figure 2(b). Because of uni-axial extension, the curvature along z is expected to be negligible relative to that along x , except at the vicinity of the defects, therefore was not considered for calculation. 2.6. Method of carrying out x-ray diffraction analysis. X-ray diffraction data were collected from the crystals by using Rigaku MicroMax007HF X-ray source with a copper rotating-anode generator equipped with Varimax optics, a MAR345dtb image-plate detector and an Oxford Cryosystem 700 series cryostream. 3. RESULTS AND DISCUSSION 3.1. Energy barrier for nucleation on nano-wrinkled substrates. An estimate of rate of nucleation on these substrates can be obtained by considering that the free energy barrier F for nucleation of crystals is a function of curvature  at any location, so that the rate of nucleation can be expressed as5,21: R   exp   * F   B Tk . Here,  is the rate (time-1) at which material gets transported to the nucleation site and Bk and T are respectively the Boltzmann constant and the absolute temperature. At a location infinitesimally away from where it is maximum, the curvature can be expressed by Taylor series expansion as:   max    x   x  xx  max   , where, max ' xx  max ' xx max is the gradient of curvature and x is a characteristic length-scale which can be represented by the wavelength  of the wrinkles. These two parameters intrinsically accounts for the local variation in surface topography, in contrast to an equivalent porous substrate, for which the asphericity in the shape of pores are required to be accounted via a shape factor, an empirical quantity5. The energy barrier is expected to be minimum at the vicinity of maximum curvature max but increase away from it according to the following expansion about the minimum energy barrier: * F     F  *   max   D    max  22 , where D is the stiffness of the energy landscape. Combining the above two expressions, we obtain a relation for the energy barrier in terms of the gradient in curvature: * F     F  *   max    xxD '   max  2 2  2 . If  p is the probability density function of occurrence of curvature,  then the rate of nucleation of a crystal can be estimated as     exp         F * Tk B     p   d . An expression for the rate of nucleation is then obtained by considering normal distribution for  p with  as the mean curvature and 0 as the standard deviation: R      exp    F *     Tk  B   d  p     exp F  max  *  Tk B D         '  xx  Tk 2 B max   2 2   max  2 2  0  2      The above relation suggests that the rate of heterogeneous nucleation is a function of several topographical parameters of the substrate. For example, the mean curvature  and the distribution in curvature 0 are important, as increase in both these quantities, enhances the nucleation rate R ; in addition, gradient in curvature ' too is important as R is expected to increase with decrease in ', an effect not observable with porous substrates. Furthermore, since the curvature of wrinkled substrates varies from positive and negative, the above relation suggests that the crystals are expected to nucleate faster on a concave surface for which the curvature turns negative, a conclusion also arrived from the route of molecular dynamic simulations4 of crystallization of colloidal crystals. In particular, spatial variation in curvature over a large range allows protein molecules to self-select a curvature during aggregation to form a cluster. 3.2. Curvature of wrinkles. We then obtain the numerical estimates of the distribution of curvature and the spread of these distributions for various substrates used in our experiments by analyzing their corresponding images using the image processing software, ImageJ. We first obtain from the AFM image, the height profile of a wrinkled surface along a straight horizontal line drawn along x at any random location along z . The plot in Figure 1e shows a typical profile obtained for a 2nd generation wrinkle prepared with 4.1 (referred by S240%). The profile of curvature (Figure 1f) appears oscillatory with several positive and negative peaks. The number of peaks appears more than the number of folds, because of roughness of the surface over length-scales much smaller than that of the wavelength of the wrinkles and the noise associated with AFM imaging and the subsequent measurement. In order to have realistic estimation of the distribution of curvature, we filter out the peak curvatures having absolute value smaller than a threshold limit, only ones exceeding this limit are considered for further calculation. For any substrate, the threshold is set to be 20% of max attained, e.g., for the S240% surface, this limit is found to be 0.025 nm-1. Such a set of peak curvatures compiled after analyzing 150 randomly chosen line-profiles show that curvature for this surface varies from 0.027 nm-1 to 0.21 nm-1. Figure 1g shows probability distribution of curvature of several such wrinkled surfaces. For all cases, with increase in curvature from the threshold value, the probability of its occurrence first increases, reaches a maxima and then decreases to a small value for very large curvature. The curvature at which this maxima occurs is considered as the mean curvature, 0 and the standard deviation of the distribution 0 is then calculated by considering it to be a normal distribution. The maximum and mean curvatures and the standard deviation, all differ for different surfaces and vary distinctly for 1st and 2nd generation wrinkles (Figure 1h). For example, for the S240% surface, the probability of  0.1nm-1 is found to be @ 0.4%, whereas for a S230% surface, the maximum curvature, max is found to be 0.096 nm-1. For S130% and S140% surfaces, max is found to be 0.042 nm-1 and 0.05nm-1 respectively. Apart from max being large, the other important parameter is 0 which for the S240% and S230% surfaces are found to be 0.017 and 0.0144 nm-1 respectively. These values are significantly larger than the corresponding S1 wrinkles, nm-1. For the S220% wrinkles, all these parameters were found to be 0062 .0 0  and .0 0055 smaller by about an order of magnitude than the other cases. (a) η=270nm (b) η=225nm nm 30 nm 30 10 0 0 1000 2000 nm (c) η=125nm 10 0 0 1000 2000 nm (d) η=125nm nm 20 10 0 nm 20 0 500 1000 nm 10 0 0 500 1000 nm (e) 20 y nm  10 0 0 (f) 0.10 0.00   1-nm -0.10 0 (g) 0.3 0.2  f 0.1 0.0 0.001 1 500 1000 1500 2000 500 0 0.01  nm -1 0.1 1000 nmx  0.05 (h)  0  nm 1- 0.04 0.03 0.02 0.01 0.00 1500 2000 0.020 0.015 0 0  0.010 0  nm 0.005 1- 0.000 1 2 3 4 5 Figure 1: (a  d) Atomic force microscopy (AFM) images of wrinkle patterns generated on PDMS films. Images (a) and (c) represent respectively the 1st and 2nd generation wrinkles obtained by stretching the PDMS film to extension ratio: = 1.3. Similarly images (b) and (d) represent the same for = 1.4. (e) The AFM images a-d were used for obtaining typical profiles of the surface of a wrinkled film along a horizontal line drawn perpendicular to the longitudinal axes of the wrinkles. The specific surface corresponding to the data in plot (e) was prepared via double extension of 40%-40% (represented by S240%). The inset depicts that small segment of the profile was fitted to a polynomial curve, which was then used for obtaining the curvature. (f) Several such small segments were used to obtain the curvature  as a function of spatial distance x . Several line-profiles were chosen randomly from the AFM image for obtaining the corresponding curvatures which were found to be oscillatory consisting of several positive and negative peaks. The positive peaks larger than a threshold value was considered and the probability of occurrence of such a peak curvature was plotted in figure (g). The semi-log plot in Figure (g) shows several such distribution functions of positive-peak curvatures for different wrinkled surfaces. Symbols ,  and  represent wrinkled surfaces S240%, S230% and S220% respectively; symbols  and  represent respectively S140% and S130%. These distribution functions were assumed to be normal and the corresponding mean  and standard deviation 0 were obtained. The bar chart (h) depicts these values for different surfaces considered in plot (g). 3.3. Schematic of experimental set-up. In order to determine, how wrinkle patterns of different 0 and 0 affect nucleation of protein crystals, these substrates were used in the crystallization set up8 of Figure 2 consisting of two rigid plates placed parallel to each other with a small gap ~120 µm maintained by two spacers. Here the top plate was silanized by coating it with self-assembled monolayer (SAM) of octadecyltrichlorosilane (OTS) molecules, which renders this surface hydrophobic, while the nano-wrinkled poly-(dimethylsiloxane) (PDMS) film attached to a rigid glass slide was used as the bottom surface, which was hydrophilic. A small volume of 4l of the protein-buffer- precipitant solution was placed between these two substrates to form a liquid disk of diameter ~2.0 mm. The whole set up was placed inside the controlled environment of a constant temperature bath maintained at 20ºC. In contrast to conventional hanging drop method, this set- up has the advantage that it allows controlled evaporation of water via fine adjustment of the gap between the film and the plate. The resultant crystals are amenable for easy harvesting for diffraction analysis or further downstream processes. Since, several liquid disks can be formed side by side, the experiment allows also for easy scale up of screening of the crystallization condition. Protein solution hydrophobic silanized glass plate hydrophilic nano-wrinkled PDMS film Rigid substrate spacer Figure 2. Schematic of the crystallization set up with parallel plate geometry 3.4. Crystallization of Ferritin. Depending upon the topography of the patterned PDMS film, crystallization was observed to proceed differently for the protein molecules used in our experiments. Figure 3al show the results for iron storing protein ferritin (molecular weight = 474kDa) crystallized on flat, wrinkle free substrate S and on surfaces with 1st and 2nd generation wrinkles, i.e. S140% and S240% respectively. In conventional hanging drop method22, ferritin has usually been crystallized using cadmium sulfate (80mM 3CdSO4. 8H2O), sodium azide (3mM) and ammonium sulphate (1.0M) as the precipitants. In our experiment, we varied concentration of (NH4)2SO4 from 0.0  0.3 M, keeping that of other species in the buffer constant (10mg/ml ferritin, 1.6% 3CdSO4. 8H2O and 0.5mM NaN3). Negligible nucleation was observed on smooth, featureless surface S when the concentration of ammonium sulfate, Cammonium_sulfphate was smaller than 0.003 M (Figure 3a); beyond this limit, small crystals of ferritin appeared within twelve hours, which grew to their maximum size within 2 days (Figure 3b). Large number of crystals appeared and grew with longest dimension, l finally reaching 30.2 ± 3.2 µm for Cammonium_sulphate ~0.03M (Figure 3c). The crystals were even larger with l ~ 63.2 ± 20 µm and 84.8 ± 6.8 µm for (NH4)2SO4 concentration 0.16 M and 0.3 M respectively (Figure 3d). On S140% surface with wrinkles of intermediate curvature, crystals were larger, with longest dimension attaining l ~ 150.4 ± 3.4 µm for Cammonium_sulfate = 0.3 M (Figure 3h). The crystal size decreased with decrease in Cammonium_sulfate, finally getting negligibly small for 0.0M. In contrast to these surfaces, crystal sizes were significantly larger on the S240% surface (Figure 3il). For example, crystals of size l ~ 55.3 ± 12.3 µm appeared even when no (NH4)2SO4 was used15; large number of tiny crystals of dimension ~ 20.5 ± 4.2 m appeared with 3CdSO.8H2O as the only precipitant present. Over a large range of Cammonium_sulphate ~ 0.02  2.0M, the crystals were observed to appear within ~24 hours and the extent of nucleation and the final size of crystals remained almost unaltered with crystal size attaining l ~ 210 ± 5µm. These results signified vanishing effect of precipitant concentration on crystal nucleation on the S240% wrinkled surfaces. In another set of experiments, in which protein concentration was systematically varied, the crystal was observed to appear within 60 hours with protein concentration as low as 2 mg/ml on the S240% surface without use of any (NH4)2SO4 but with 1.6% by weight of 3CdSO4.8H2O and 0.5mM NaN3. No such crystal appeared on the S surface (Figure S2) at similar conditions. Importantly the crystal appeared not at the edge of the drop but away from it, towards the centre, implying that crystallization did not occur via evaporation induced super-saturation of the protein solution, but surface pattern induced decrease in the energy barrier for nucleation. Notice that the crystals appeared somewhat different when crystallized on different wrinkled surfaces at different concentrations of the precipitant. For example, the crystals on S240% surfaces, especially the one that appear in optical micrograph of figure 3(l) appeared dendritic. In order to confirm if these crystals indeed represented Ferritin, we carried out x-ray diffraction. The diffraction pattern, presented in figure S3, showed that the crystal was of protein and not of salt and they could be diffracted up to 6Å resolution. In fact, similar dendrite shape of ferritin crystal has been observed and analyzed by others as well23, who have attributed the dendritic growth of ferritin to oligomers of different chain length present in the protein solution. In our experiment, the crystal was constrained to grow within a confined space between the two parallel plates, which too appear to have influence on the crystal shape. 0.0 M S (a) (e) S140% (i) S240% 0.0 M (m) S (q) S240% Ferritin: Cammonium_sulfate 0.03 M 0.003 M (b) (c) (f) (j) (g) (k) Thaumatin: CNa/K_tartrate 0.5 M 0.1 M (n) (o) 1000 (u) 0.3 M (d) (h) (l) l 150 µm 0.75 M (p) 100 l  μm 10 1 0.0 (v) 100 l μm  10 Ferritin S S240% 0.1 Cammonium_sulfate (mol/lit) 0.3 0.2 0.4 no crystal Thaumatin S S240% (s) (t) (r) l 1 0.0 150 µm 0.2 0.6 CNa/K_tartrate (mol/lit) 0.4 0.8 Figure 3. (a-l) The optical micrographs show ferritin crystals grown on three different substrates: S, S1 and S2 and at different concentration of ammonium sulfate, Cammonium_sulfate = 0 to 0.3M as precipitant. Similarly, optical micrographs (m – t) represent tetragonal bipyramid shaped thaumatin crystals grown on two different substrates: S (mp), S240% (qt) by varying the precipitant concentration, CNa/K_tartrate in the range 0  0.75 M. In figure (l), l represents the longest dimension of crystal. Plot of l as a function of precipitant concentration is plotted in figures (u) and (v) for ferritin and thaumatin respectively. The symbols  and □ represent substrates S and S240%, respectively. 3.5. Crystallization of Thaumatin. The above observations were made also with protein thaumatin (molecular weight = 22kDa), which was crystallized from a buffer solution in which sodium/potassium tartrate (Na/K tartrate) was used as the precipitant24. Figure 3mt shows typical optical micrographs of crystals on S and S240% surfaces with precipitant concentration, CNa/K_tartrate varied over the range 0.0  0.75 M. On the control surface S, no crystal was observed to appear for CNa/K_tartrate < 0.1M (Figure 3mn); here the protein solution dried out completely over twelve hours. Beyond this threshold, the crystals appeared with their sizes increasing with the precipitant concentration. The crystal size maximized at CNa/K_tartrate ~ 0.5 M (Figure 3o), beyond which, it decreased slightly; very small crystals appeared for CNa/K_tartrate ~ 0.75 M as shown by micrograph in Figure 3p. The results were, however, remarkably different on the S240% surface (Figure 3qt), for which the crystals of thaumatin appeared even without use of any precipitant (Figure 3q) unlike any example in the literature. In fact, the size of these crystals was already larger than those achieved on the S surface for various precipitant concentrations (Figure 3uv). Thus, in effect, the S240% surface eliminated the need of a precipitant for this protein to crystallize. However, in order to produce even larger crystals, the CNa/K_tartrate was optimized by varying it similarly, as on the S surface. Final result of these experiments was that crystals as large as 200.2 ± 19.6 µm, 155.6 ± 12.7 µm and 123.1 ± 9.3 µm appeared respectively for CNa/K_tartrate ~ 0.1M, 0.5M and 0.75M (Figure 3rt). These sizes far exceeded what could be achieved over the S surface, nevertheless were similar over three orders of magnitude in variation of CNa/K_tartrate. Similar to ferritin, for thaumatin too crystals appeared, albeit small, on the S240% surface even when the protein concentration was decreased by an order of magnitude to 1.5mg/ml with precipitant concentration as small as CNa/K_tartrate = 0.2M. Here the crystals appeared within 48 hours. But the same did not happen on the featureless S surface (Figure S2), signifying once again the ability of the nano-wrinkled surface in crystallizing from a very dilute solution of protein and precipitant. It is worth noting that these concentrations of protein and the precipitant were significantly smaller than any reported value in the literature ~2 mg/ml)15. 3.6. Crystallization of Glucose Isomerase and Xylanase II. Glucose isomerase (molecular weight = 173 kDa) was crystallized from a solution containing 5mg/ml protein and 25 mM HEPES. In addition, (NH4)2SO4 was used as the precipitant, the concentration of which was varied over Cammonium_sulphate ~ 0.05  0.5 M. On the control surface S, very small trigonal crystals appeared within 10 hours for Cammonium_sulphate > 0.24 M as depicted by the optical micrographs in Figure 4(ac). With varying concentration of (NH4)2SO4, the final size of the crystals was found to vary from 18.1 ± 4.3 µm to 33.0 ± 4.8 µm. However, on the S240% surface, crystals of size ~50 m appeared within 24 hours at precipitant concentration as low as 0.05M; the size of the crystal grew to be as large as l = 100.3 ± 20.3 µm for 0.5M (NH4)2SO4 as shown in Figure 4(df), significantly larger than that achieved on the substrates S and on several other heterogeneous surfaces25. A typical x-ray diffraction pattern for a single glucose isomerase crystal grown on these nano-wrinkled substrates has been presented in Figure 4g for which maximum resolution of 2.5Å could be achieved. It is worth noting that similar crystal morphology could be observed earlier with significantly larger concentration of protein ~22 mg/ml on other commonly used heterogeneous substrates e.g. human hair26. It is to be noted that in all experiments described above, we used freshly prepared nano-wrinkled substrates. In order to see if these nano-wrinkled surfaces can be used also over a longer period, a typical S240% substrate was stored inside desiccators in dry and dust-free condition for three weeks, following which it was used for the crystallization experiment. The optical micrographs presented in figure 4(h) and (i) show the crystals which appear on these surfaces suggesting that ability of the S240% substrates to crystallize remain unaltered over this period, although nucleation gets somewhat uncontrolled. Glucose Isomerase : Cammonium_sulphate (a) 0.05M No crystal 0.25M (b) 0.5M (c) S (d) S240% (g) 150µm (f) (i) (e) (h) 150µm S240% (three weeks later) 150µm Figure 4. Optical images (af) depict crystals of Glucose Isomerase grown by varying the precipitant concentration, Cammonium_sulphate = 0.05 to 0.5 M on substrate S and S240%. No crystal appeared of S surface for Cammonium_sulphate < 0.25 M. (g) X-ray diffraction pattern from single glucose isomerase crystal. While in all the above cases, the PDMS surfaces were freshly prepared, optical images (hi) show that crystals appear also on a S240% surface which was stored in a dust and moisture free environment for about three weeks. The concentration of Glucose Isomerase in the protein solution was maintained at 5 mg/ml for all these cases Similarly, protein xylanase II (molecular weight = 21kDa) too crystallized from the same initial concentration of 5 mg/ml. Plate like crystals were obtained from a solution containing 12.5 mM NaCl, 6.25 mM Tris HCl, 1.8% (w/v) PEG 8000 in which, concentration of CaCl2 was varied from 2.5  150 mM. Figure S4(bc) depicts the optical micrograph of xylanase II crystals on the control surface S, in which large number of tiny crystals appeared corroborating the observation with the other protein molecules. With increasing concentration of precipitant, the crystal size increased to a certain extent on substrate S140% (optical micrographs in Figure S4(ef)). However, the effect of precipitant became irrelevant on S240% on which, for all concentrations of the precipitant, only few crystals nucleated which grew to a significantly larger dimension ~ 208.4 ± 6.0 µm as presented in the optical micrographs in Figure S4(gi) and Figure S5(b). The plate shaped crystals of Xylanase II were very thin with insignificant third dimension. In fact, they were so fragile that it was not possible to pick up the crystals from the mother liquor and to mount on the diffractometer for X-ray diffraction analysis. However, we could carry out dye test with Izit dye on these crystals, the optical micrographs presented in Figure S6 confirmed that these crystals indeed represented Xylanase II and not any salt. Furthermore, similar plate like morphology of these crystals has been observed to form earlier in experiments in which polysryrene nano-spheres are used as heterogeneous nucleant27. 4. SUMMARY The work presented here has several novel aspects which we will now summarize. (a) First of all, it is known that lengthscale of sub-microscopic features in soft materials is limited by their surface tension because surface tension does not allow very large curvature to be achieved in soft solids. Here we have overcome this limitation by introducing a simple two step process of preparing a substrate on PDMS. In fact, we have generated surface features which have curvature as large as 0.2 nm-1. Furthermore, this method has allowed us to generate wrinkles, the mean curvature and the spread in distribution of curvature both can be varied over a large range. (b) Our theoretical analysis shows that both large curvature and large spread in curvature of a nucleant are useful for diminishing the energy barrier for nucleation of crystals. A substrate with such features is expected to trap protein molecules of wide range of sizes to crystallize and thus be a universal nucleant. We have shown that our nano-wrinkled surfaces comprising of folds with very large curvature, both positive and negative curvature and large distribution of curvature is useful for crystallizing protein molecules with sizes varying over a large range from 20 kDa to 480 kDa. (c) Experiments with protein molecules of wide range of molecular weight on identically prepared wrinkled surface, e.g. the S240% substrate, show that these molecules can indeed self- select a suitable lengthscale for heterogeneous nucleation which leads to crystallization over a wide range of protein and precipitant concentrations, not achieved with any other known substrate. (d) It is known that homogeneous nucleation occur only when concentration of the solute exceeds supersaturation, at which large number of crystals appear at a time, which do not grow to a very large size. However, for heterogeneous nucleation, crystal can appear at a much smaller extent of supersaturation14,28. In fact, supersaturation may not even be required in the bulk protein solution, as the nano-wrinkles having curvature similar to the radius of gyration of the protein molecule can lure these molecules resulting in protein concentration at the vicinity of the substrate significantly larger than that in the bulk. This effect is similar to the classical Kelvin effect observed in the context of capillary condensation, in which water is observed to condense inside a hydrophilic capillary even though the humidity in the atmosphere remains at sub- saturation level. Because of this effect, very few crystals appear which once nucleated, grow to a very large size thus consuming most of the solute molecules. As a result, further nucleation of crystal is prevented leading to highly controlled nucleation. (e) For all different protein molecules, crystals could be nucleated from protein solutions having wide range of precipitant concentration, importantly with very low concentration of precipitant. At least for one protein molecule, crystals appeared even without use of any precipitant. These results suggest that use of these nano-wrinkled surfaces is expected to simplify the screening process for crystallization of many yet-to-be-crystallized proteins, considered a bottleneck in molecular biology and its applications in therapeutics, disease diagnosis and its prevention. (f) By carrying out experiment on three week old S240% substrate, we have shown that the nano-wrinkled surfaces can be used for crystallization over a long period of time. More systematic experiments, however, needs to be carried out in order to understand the ageing effect of the substrates on their ability to crystallize protein molecules. (g) Beside protein crystallization, these surfaces are expected to have significant impact also in the general area of bio-mineralization, and variety of other operations involving nucleation, e.g. boiling, droplet condensation, protein separation and patterning. ACKNOWLEDGEMENTS Authors thank Miss Saloni Sharma, the staff of DST Unit on Soft Nanofabrication, IIT Kanpur for her help in carrying out the AFM scans and Dr. Krishnacharya Khare for useful discussions. ASG thank Mr. Vaibhav Singh Bais for his help in carrying out the X-ray diffraction analyses. ASG acknowledges Department of Science and Technology (DST), Government of India for the grant DST/CHE/20080204. AG acknowledges support from the Humboldt Foundation in the form of a fellowship for this work. ASSOCIATED CONTENT Images of different types of surface defects on nano-wrinkled substrates, images showing effect of nano-wrinkles on crystallizing thaumatin and ferritin from very low protein concentration in respective protein solutions, x-ray diffraction patterns of ferritin and glucose isomerase crystals, images depicting crystallization of Xylanase II on different surfaces using calcium chloride as the precipitant and in different precipitant concentrations, plot showing size of glucose isomerase and xylanase II crystals as a function of precipitant concentration on different surfaces, image showing plate shaped xylanase II crystal after dye test. This information is available free of charge via the Internet at http://pubs.acs.org/. REFERENCES 1) McPherson, A. (1999) Crystallization of Biological Macromolecules (Cold Spring Harbor Lab. Press, Woodbury, NY). 2) McPherson, A.; Shlichta, P. Heterogeneous and epitaxial nucleation of protein crystals on mineral surfaces. Science 1988, 239, 385–387. 3) Saridakis, E.; Chayen, N. E. Towards a 'universal' nucleant for protein crystallization. Trends Biotechnol.Trends Biotechnol. 2009, 27, 99−106. 4) Cacciuto, A.; Auer, S.; Frenkel, D. Onset of heterogeneous crystal nucleation in colloidal suspensions. Nature 2004, 428, 404−406. 5) Chayen NE.; Saridakis E.; Sear RP. Experiment and theory for heterogeneous nucleation of protein crystals in a porous medium. Proc Natl Acad Sci U S A. 2006, 103, 597–601. 6) Falini, G.; Fermani, S.; Conforti, G.; Ripamonti, A. Protein crystallization on chemically modified mica surfaces. Acta Crystallogr.2002, D58, 1649–1652. 7) Liu, Y.-X.; Wang, X.-J.; Lu, J.; Ching, C.-B. Influence of the Roughness, Topography, and Physicochemical Properties of Chemically Modified Surfaces on the Heterogeneous Nucleation of Protein Crystals. J. Phys. Chem. B 2007, 111, 13971–13978. 8) Sengupta Ghatak A.; Ghatak, A. Controlled Crystallization of Macromolecules using Patterned Substrates in a Sandwiched Plate Geometry. Ind Eng Chem Res. 2011, 50, 1298412989. 9) García-Ruiz, J. M. Nucleation of protein crystals. J Struct. Biol. 2003, 142, 22–31. 10) Chayen NE.; Saridakis E.; El-Bahar R.; Nemirovsky Y. Porous silicon: an effective nucleation-inducing material for protein crystallization. J Mol Biol. 2001, 312, 591–595 (2001). 11) Stolyarova S.; Saridakis E.; Chayen NE; Nemirovsky Y. A Model for Enhanced Nucleation of Protein Crystals on a Fractal Porous Substrate. Biophys J. 2006, 91, 3857– 3863. 12) Rong L.; Komatsu H.; Yoshizaki I.; Kadowaki A.; Yoda S. Protein crystallization by using porous glass substrate. J Synchrotron Radiat. 2004, 11, 27–29. 13) Page, A. L; Sear, R. P. Heterogeneous Nucleation in and out of Pores. Phys. Rev. Lett. 2006, 97, 065701. 14) Curcio, E.; Fontananova, E.; Di Profio, G.; Drioli, E. Energetics of Protein Nucleation on Rough Polymeric Surfaces. J. Phys. Chem. B, 2006, 110, 12438-12445. 15) Shah, U.V.; Williams, D. R.; Heng, J.Y.Y. Selective Crystallization of Proteins Using Engineered Nanonucleants. Cryst. Growth Des. 2012, 12, 13621369. 16) Shah, U. V.; Allenby, M. C.; Williams, D. R., Heng, J.Y.Y. Crystallization of Proteins at Ultralow Supersaturations Using Novel Three-Dimensional Nanotemplates. Cryst. Growth Des. 2012, 12, 17721777. 17) Yang, S.; Khare, K.; Lin, P.-C. Harnessing Surface Wrinkle Patterns in Soft Matter. Adv. Funct. Mater. 2010, 20, 2550–2564. 18) Chung, J. Y.; Nolte, A. J. M. & Stafford, C. M. A Versatile Platform for Measuring Thin- Film Properties. Adv. Mater. 2011, 23, 349–368. 19) Efimenco, K.; Rackaitis, M.; Manias, E.; Vaziri, A.; Mahadevan, L.; Genzer, J. Nested self-similar wrinkling patterns in skin. Nat. Mater. 2005, 4, 293–272. 20) Jagota, A.; Paretkar, D.; Ghatak, A. Surface-tension-induced flattening of a nearly plane elastic solid. Phys. Rev. E 2012, 85, 051602. 21) Debenedetti, P. G. 1996, Metastable Liquids (Princeton University Press, Princeton). 22) Granier, T.; Gallois, B.; Dautant, A.; D'estaintot, BL.; Précigoux, G. Comparison of the structures of the cubic and tetragonal forms of horse spleen apoferritin. Acta Crystallogr. 1997, D53, 580587. 23) Gutiérrez-Quezada, A. E.; Arreguín-Espinosa, R.; Moreno, A. 2010, Handbook of crystal growth (Springer). 24) Ko, T.P.; Day, J.; Greenwood, A.; McPherson, A. Greenwood, A. & McPherson, A. Structures of three crystal forms of the sweet protein thaumatin. Acta Crystallogr. 1994, D50, 813825. 25) Nederlof, I.; Hosseini, R.; Georgieva, D.; Luo, J.; Li, D.; Abrahams, J. P. A Straightforward and Robust Method for Introducing Human Hair as a Nucleant into High Throughput Crystallization Trials. Cryst. Growth Des. 2011, 11, 1170  1176. 26) Chayen, N. E.; Lloyd, L. F.; Collyer, C. A.; Blow, D. M. Trigonal crystals of glucose isomerase require thymol for their growth and stability. J. cryst. Growth 1989, 97, 367374. 27) Kallio, J. M.; Hakulinen, N.; Kallio, J. P.; Niemi, M. H.; Kärkkäinen, S.; Rouvinen, J. The Contribution of Polystyrene Nanospheres towards the Crystallization of Proteins. PLoS ONE 2009, 4, 17. 28) Mullin, J. W. Crystallization; Elsevier Butterworth-Heinemann: Oxford 2001. 0 0 0.02  0  1- nm 0.01 0.04  0  1- nm 0.02 0.00 1 2 3 4 5 0.00 nm 500 1000 For Table of Contents Use Only S140% S240% nm 30 0 150 µm 0 1000 2000 nm 20 10 0 0 nm Table of Content Figure
1209.1757	1	1209	2012-09-02T19:32:32	Solitons and Collapse in the lambda-repressor protein	[ "physics.bio-ph", "cond-mat.soft", "q-bio.BM" ]	The enterobacteria lambda phage is a paradigm temperate bacteriophage. Its lysogenic and lytic life cycles echo competition between the DNA binding $\lambda$-repressor (CI) and CRO proteins. Here we scrutinize the structure, stability and folding pathways of the $\lambda$-repressor protein, that controls the transition from the lysogenic to the lytic state. We first investigate the super-secondary helix-loop-helix composition of its backbone. We use a discrete Frenet framing to resolve the backbone spectrum in terms of bond and torsion angles. Instead of four, there appears to be seven individual loops. We model the putative loops using an explicit soliton Ansatz. It is based on the standard soliton profile of the continuum nonlinear Schr\"odinger equation. The accuracy of the Ansatz far exceeds the B-factor fluctuation distance accuracy of the experimentally determined protein configuration. We then investigate the folding pathways and dynamics of the $\lambda$-repressor protein. We introduce a coarse-grained energy function to model the backbone in terms of the C$_\alpha$ atoms and the side-chains in terms of the relative orientation of the C$_\beta$ atoms. We describe the folding dynamics in terms of relaxation dynamics, and find that the folded configuration can be reached from a very generic initial configuration. We conclude that folding is dominated by the temporal ordering of soliton formation. In particular, the third soliton should appear before the first and second. Otherwise, the DNA binding turn does not acquire its correct structure. We confirm the stability of the folded configuration by repeated heating and cooling simulations.	physics.bio-ph	physics	Solitons and Collapse in the λ-repressor protein Andrey Krokhotin,1, ∗ Martin Lundgren,1, † and Antti J. Niemi1, 2, ‡ 1Department of Physics and Astronomy, Uppsala University, P.O. Box 803, S-75108, Uppsala, Sweden 2 Laboratoire de Mathematiques et Physique Theorique CNRS UMR 6083, F´ed´eration Denis Poisson, Universit´e de Tours, Parc de Grandmont, F37200, Tours, France The enterobacteria lambda phage is a paradigm temperate bacteriophage. Its lysogenic and lytic life cycles echo competition between the DNA binding λ-repressor (CI) and CRO proteins. Here we scrutinize the structure, stability and folding pathways of the λ-repressor protein, that controls the transition from the lysogenic to the lytic state. We first investigate the super-secondary helix-loop- helix composition of its backbone. We use a discrete Frenet framing to resolve the backbone spectrum in terms of bond and torsion angles. Instead of four, there appears to be seven individual loops. We model the putative loops using an explicit soliton Ansatz. It is based on the standard soliton profile of the continuum nonlinear Schrodinger equation. The accuracy of the Ansatz far exceeds the B-factor fluctuation distance accuracy of the experimentally determined protein configuration. We then investigate the folding pathways and dynamics of the λ-repressor protein. We introduce a coarse-grained energy function to model the backbone in terms of the Cα atoms and the side-chains in terms of the relative orientation of the Cβ atoms. We describe the folding dynamics in terms of relaxation dynamics, and find that the folded configuration can be reached from a very generic initial configuration. We conclude that folding is dominated by the temporal ordering of soliton formation. In particular, the third soliton should appear before the first and second. Otherwise, the DNA binding turn does not acquire its correct structure. We confirm the stability of the folded configuration by repeated heating and cooling simulations. PACS numbers: 05.45.Yv 87.15.Cc 36.20.Ey I: INTRODUCTION residues. The transition between the lysogenic and the lytic state in bacteriophage λ infected E. coli cell is the paradigm genetic switch mechanism. It is described in numerous molecular biology textbooks and review arti- cles [1]-[7]. The interplay between the lysogeny main- taining λ-repressor (CI) protein and the CRO regulator protein that controls the transition to the lytic state is a simple model for more complex regulatory networks, including those that can lead to cancer in humans. In the present article we describe the physical prop- erties of the λ-repressor protein, that controls the lysogenic-to-lytic transition. We investigate in detail the stability of its native conformation, the dynamics of the folding process, and the landscape of folding pathways. We find that the folded configuration displays a struc- ture which is unique among all known protein structures. We also conclude that the folding pathways are entirely dominated by the loop regions. In particular, the tempo- ral ordering of loop formation appears to be the decisive factor for the protein's ability to reach its native fold. If solitons form in a wrong order the protein may misfold. Full crystallographic information of the experimental λ-repressor structure that we use in our investigation is available in Protein Data Bank (PDB) [8] under the code 1LMB. This structure is a homo-dimer with 92 residues in each of the two monomers. It maintains the lysogenic state by binding to DNA with a helix-turn-helix motif that is located between the residue sites 33-51. Through- out this article we shall use the PDB indexing of the For the statistical analyses that are presented here, we utilize a subset of PDB data that consists of the canonical set of structures with better than 2.0 A resolution. We have compared the results with the subset that contains only those proteins with better than 2.0 A resolution and with less than 30% sequence similarity. Our conclu- sions are independent of the data set, and for illustrative purposes we here use the canonical 2.0 A set. This article is composed as follows: We first explain how to describe the geometry of a generic folded protein in terms of its backbone central Cα carbons. We propose that a coarse-grained energy function, that models the backbone geometry, can be constructed with only the Cα coordinates as dynamical variables. We argue that a variant of the discrete non-linear Schrodinger (DNLS) equation is a suitable Master Equation to describe folded proteins, in terms of its dark soliton solution. We then proceed to utilize this general framework to study the structure of the λ-repressor protein. We show that the λ-repressor backbone is composed from seven individual soliton solutions of the DNLS equation, within the ac- curacy of crystallographic structure measurements. In the same vein we propose, that protein folding can be described in terms of a coarse grained model, based on relaxation dynamics. We utilize this to investigate the folding dynamics of the λ-repressor. We conclude that the temporal ordering of soliton formation is important for reaching the correct native state. A wrong ordering in soliton formation can be a cause for misfolding. We observe that the second soliton has a peculiar structure 2 1 0 2 p e S 2 ] h p - o i b . s c i s y h p [ 1 v 7 5 7 1 . 9 0 2 1 : v i X r a that sets it apart from any other known structure in all proteins. II: METHODS: A: Backbone geometry Our analysis of the λ-repressor protein will be based on an effective, coarse grained energy function approach that has been recently developed in [9]-[13]. In this ap- proach the protein geometry is described in terms of the backbone Cα atoms. The ensuing bond and torsion an- gles assume the role of the dynamical variables. These angles are constructed as follows: Let ri be the coordi- nate sites of the Cα carbons, where the index i = 1, ..., N runs over all amino acids. For a given protein these co- ordinates can be read from the PDB. For each site i, we introduce the unit tangent vector ri+1 − ri ri+1 − ri ti = (1) the unit binormal vector bi = ti−1 − ti ti−1 − ti and the unit normal vector ni = bi × ti (2) (3) The orthogonal triplet (ni, bi, ti) determines a discrete version of the Frenet frame at each position ri of the backbone. The backbone bond angles are κi ≡ κi+1,i = arccos (ti+1 · ti) (4) and the backbone torsion angles are τi ≡ τi+1,i = sign{bi−1×bi·ti}·arccos (bi+1 · bi) (5) Conversely, if these angles are all known, we can use  = ni+1 bi+1 ti+1 cos κ cos τ cos κ sin τ − sin κ  − sin τ sin κ cos τ sin κ sin τ cos κ cos τ 0  ni bi ti i+1,i (6) to iteratively construct the frame at position i + i from the frame at position i. Once we have all the frames, we obtain the entire backbone from k−1(cid:88) rk = ri+1 − ri · ti (7) i=0 and standard β-strand is Without any loss of generality we may set r0 = 0, and choose t0 to point into the direction of the positive z-axis. β − strand : (13) (14) We note that the relation (7) does not involve the vec- tors ni and bi. Consequently we may rotate the (ni, bi) frame vectors, without affecting the backbone, by select- ing an arbitrary linear combination of these two vectors independently at each site i. For this we introduce a lo- cal SO(2) transformation that rotates the (ni, bi) by an angle ∆i so that the ti remain intact,  cos ∆i = n  b t i sin ∆i 0 − sin ∆i cos ∆i 0 1 0 0 n  b t n  b t i → e∆iT 3 2 i (8) where the SO(3) generators are (T i)jk = ijk, [T i, T j] = ijkT k We combine n and b into the complex vector n + ib and rewrite (8) as ni + ibi → ei∆i (ni + ibi) ≡ e1 i + ie2 i (9) The frame rotation (8) corresponds to the following transformation in the bond and torsion angles, κi T 2 → e∆iT 3 (κiT 2) e−∆iT 3 τi → τi + ∆i−1 − ∆i (10) (11) Since the transformation (10), (11) leaves ti intact it has no effect on the backbone. A priori, the fundamental range of the bond angle κi is κi ∈ [0, π]. For the torsion angle the range is τi ∈ [−π, π). Consequently we may identify (κi, τi) with the canonical latitude and longitude angles of a two-sphere S2. However, in the sequel we find it useful to extend the range of κi into [−π, π] mod(2π), but with no change in the range of τi. We compensate for this two-fold covering of S2 by introducing the following discrete Z2 symmetry κk → − κk τi → τi − π for all k ≥ i (12) We note that this is a special case of (10), (11), with ∆k = π ∆k = 0 for k ≥ i + 1 for k < i + 1 The regular protein secondary structures correspond to definite values of (κi, τi). For example standard α-helix is α − helix : 2 τ ≈ 1 (cid:26)κ ≈ π (cid:26)κ ≈ 1 τ ≈ π 3 FIG. 1: The distribution of the Cα-Cα bond length in PDB. The fluctuations around the average value d ≈ 3.8 A are small, the secondary peak around 2.9 A is due to cis-proline. Note logarithmic scale. Similarly we can describe all the other regular secondary structures such as 3/10 helices, left-handed helices etc. with definite constant values of κi and τi. Loops are con- figurations that interpolate between these regular struc- tures, so that along a loop the values of (κi, τi) are vari- able. Finally, we compute the average value of the bond length in (7) using PDB. The result shown in Figure 1 is constructed using the canonical set of protein structures which are measured with better than 2.0 A resolution. The average value is concentrated around ri+1 − ri = d ≈ 3.8 A (15) FIG. 2: (Color online) In the Frenet frames the fluctuations in the direction of the vectors ui around their average value, given by (18), (20) and denoted by the (red) lateral vector, are relatively small. The left-hand side of the horse-shoe corre- sponds to β-strands, the right-hand side to α-helices and the connecting region at the bottom corresponds to loops. The minuscule isolated island corresponds to the "left-handed α" region. can compute the values of θi from the PDB. As shown in Figure 1 the range of variations in θi are quite small, it fluctuates around < θ > ≈ 1.98 (rad) (18) In our theoretical analysis we use the fixed bond length value (15). We also impose the forbidden volume (steric) constraint ri − rk ≥ 3.8 A for i − k ≥ 2 (16) The longitude ϕi in (17) measures distance from the direction of the Frenet frame normal vector ni. It is the azimuthal angle between ni and the projection of ui on the normal plane spanned by (ni, bi). Under the frame rotation (8) we have between the backbone Cα atoms. This condition is well respected by folded protein structures in PDB. B: Side-chain geometry Following [13] we characterize the side-chain directions in terms of directional vectors ui that point from the Cα towards the corresponding Cβ carbon. At each Cα sin θi · cos ϕi sin θi · sin ϕi  ui = cos θi (17) The latitude angle θi counts deviation from the direc- tion of the corresponding Frenet frame tangent vector ti. When θi = 0 the ti and ui are parallel. Note that the angle θi remains invariant under the rotation (8). We ϕi → ϕi + ∆i (19) and consequently the values of ϕi depend on the framing. From PDB we find that in the Frenet frames the values of ϕi are subject to relatively small fluctuations around the average value < ϕ > ≈ −2.43 (rad) (20) As shown in Figure 2, it is remarkable that the direc- tion of ui nutates in a very regular horse-shoe shaped manner, that reflects the underlying secondary structure environment. This proposes that there is a strong local coupling between the two angular variables θi and ϕi, that depends on the secondary structure. The notable expection from (18), (20) is the left- It is visible in Figure 2 as a mi- handed loop region. ) ÅBond Length (22.533.544.5N entries 110210310410510nuscule isolated region, with < θ > ≈ 2.25 (rad) < ϕ > ≈ −1.90 (rad) (21) But this is also quite close to the values in (18), (20). C: Backbone energy and solitons Proteins display a hierarchy which is determined by the spatial length scale. As the length scale increases, shorter distance dynamical variables become gradually disengaged. Following the general concept of universal- ity [14]-[17], we utilize this hierarchy of scales to system- atically coarse grain the energy function. At each level of hierarchy we retain explicitly only those variables that remain relevant. The variables that are less and less so as the distance scale increases are accounted for effectively, through the functional form and the values of parame- ters in the various individual energy contributions that involve the remaining relevant variables only. In [9]-[13], see also [18], it has been argued that a Lan- dau free energy function that aims to compute the over- all fold geometry can be based solely on those variables that determine the positions of the central Cα atoms. Since the fluctuations in the bond lengths are minimal, see Figure 1, the leading order contribution to the en- ergy then involves only the bond and torsion angles for the Cα backbone. The functional form of the energy can be uniquely determined by symmetry considerations. For this we note that any backbone energy function that involves only the bond and torsion angles must remain invariant under the SO(2) transformation (10), (11). In- deed, previously it has already been shown [9]-[13] how this SO(2) gauge invariance allows us to uniquely deduct the functional form of the energy function, in the long distance limit where any higher order non-local contri- bution can be ignored. In this limit the energy function can only involve the following terms [9]-[11], E = − N−1(cid:88) (cid:26) N(cid:88) 4 sum together with the three first terms in the second sum comprise exactly the energy of the standard DNLS equa- tion when expressed in terms of the Hasimoto variable of fluid mechanics [11], [20]. The fourth (bτ ) is a conserved quantity in the DNLS hierarchy, the "momentum", and the fifth term (aτ ) is the conserved "helicity". The last (cτ ) term is the (non-conserved) Proca mass term that we include for completeness. The energy function (22) does not purport to explain the fine details of the atomary level mechanisms that give rise to protein folding. Rather, it examines the proper- ties of a folded protein backbone in terms of universal physical arguments along the lines of [14]-[17]. Indeed, the functional form (22) is deeply anchored in the el- egant mathematical structure of integrable hierarchies [20]. Within this framework no term beyond those in the integrable hierarchy can be added without violating the underlying general and elegant mathematical principles. In this sense, (22) is the universal long distance limit that would emerge from any microscopic level Schrodinger op- erator, whenever we truncate the Landau free energy to explicitly include only the backbone bond and torsion angles. Remarkably, we have found that despite the very gen- eral nature of argumentation that leads us to adopt the energy function (22), it is fully capable of describ- ing folded protein backbones with sub-atomic precision of around 0.5 A, and even better [21]. This is due to the observation [10], [11], that (22) describes proteins in terms of solitons that are the paradigm structural self- organizers. Indeed, solitons are tremendously robust in their ability to preserve the form under both quantum mechanical and thermal fluctuations. We derive the relevant soliton profile as follow: We first introduce the τ -equation of motion = dτ κ2 i τi − bτ κ2 i − aτ + cτ τi = 0 ∂E ∂τi from which we solve 2 κi+1κi + i + q · (κ2 2κ2 i=1 i=1 + dτ 2 i − bτ κ2 i τ 2 κ2 i τi − aτ τi + cτ 2 τ 2 i i − m2)2 (cid:27) τi[κ] = aτ + bτ κ2 i cτ + dτ κ2 i (23) (22) Even though there are four parameters in (23) one of them, the overall scale, cancels out. In the sequel we shall choose aτ = −1.0 so that for an α-helix (13) we have The detailed derivation of (22) is presented in [9]. It in- volves the introduction of frame (gauge) invariant com- binations of the Frenet frame bond and torsion angles [18], [19]. For the present purposes it suffices to observe, that (22) coincides with the one dimensional discretized Abelian Higgs Model Hamiltonian in the unitary gauge, in terms of the Frenet frame bond and torsion variables. We can also recognize (22) as a variant of the discrete nonlinear Schrodinger (DNLS) equation [20]: The first τi[α] = 1 + bτ κ2 i cτ + dτ κ2 i ≈ 1 mod (2π) and for a β-strand (14) τi[β] = 1 + bτ κ2 i cτ + dτ κ2 i ≈ π mod (2π) When we use (23) to eliminate the torsion angles we where get for the bond angles the energy N(cid:88) (cid:19) i=1 · E[κ] = − N−1(cid:88) V [κ] = − 2 κi+1κi + i=1 (cid:18) bτ cτ − aτ dτ (cid:18) b2 (cid:19) τ + 8qm2 dτ 2bτ − [12] is given by the naive discretization of the continuum dark NLSE soliton [20], 5 (µ1 + 2πN1) · eσ1(i−s) − (µ2 + 2πN2) · e−σ2(i−s) eσ1(i−s) + e−σ2(i−s) (29) Here s is a parameter that determines the backbone site location of the soliton center. This is the center of the fundamental loop that we describe by the soliton. The µ1,2 ∈ [0, π] are parameters. In the case of proteins the values of µ1,2 are entirely determined by the adjacent helices and strands. Far away from the soliton center we have (cid:26) µ1 κi → mod (2π) −µ2 mod (2π) i > s i < s For α-helices and β-strands the µ1,2 values are deter- mined by (13), (14). Negative values of κi are related to the positive values by (12). The N1 and N2 constitute the integer parts of µ1,2 and for simplicity we shall take N1 = N2 ≡ N . This in- teger is like a covering number, it determines how many times κi covers the fundamental domain [0, π] when we traverse the soliton once. We introduce this integer for the following reason: The Ansatz (29) is monotonic but in general the values of κi ∈ [0, π] that we obtain from PDB are not. Whenever we encounter a site i where κi in the PDB data fails to be monotonic, we either add or subtract 2π to its value, to regain a monotonic data pro- file. This shift does not have any effect on the backbone geometry. In this manner we utilize the multi-valuedness of κi to convert any sequence {κi} into a monotonic one, that we can then approximate by the Ansatz (29). Note that for µ1 = µ2 and σ1 = σ2 we recover the hy- perbolic tangent. In this case the two regular secondary structures before and after the loop are the same. More- over, only the (positive) σ1 and σ2 are intrinsically loop specific parameters. They specify the length of the loop and as in the case of the µ1,2 they are combinations of the parameters in (22). Similarly, in the case of the torsion angle there is only one loop specific parameter in (23). The overall, common scale of the four parameters is again irrelevant and two of the remaining three parameters are fixed by the regular secondary structures that are adjacent to the loop. Long protein loops, and entire super-secondary protein structures such as a helix-loop-helix can be constructed by combining together solitons (28), (23). A typical short super-secondary structure that is described by a single soliton involves at least 15-20 different amino acids, of- ten even more. As a consequence our DNLS equations and the explicit soliton Ansatz compute the 45-60 spa- tial coordinates of the ensuing Cα carbons in terms of of the five essential universal parameters in (22). This im- plies that the DNLS equations comprises a highly under- determined system of equations, and the key physical principles of our approach are experimentally testable. 2κ2 i + V [κi] (24) κi = 1 cτ + dτ κ2 · κ2 + q · κ4 (25) The functional form of (25) is familiar from various stud- ies in mathematical physics: The first term relates to the potential energy in a Calogero-Moser system. The second and third terms have the conventional form of a symmetry breaking double-well potential. Depending on the parameter values we may be either in the broken sym- metry phase where κ and τ both acquire a non-vanishing and constant ground state value, or in the symmetric phase where κ vanishes. In applications to proteins, regular structures such as helices (13) and strands (14) correspond to different bro- ken symmetry ground states of the energy. Furthermore, the numerical value of the first term in (25) is often small in comparison to the second and third, and so is b2 τ in comparison to 8qm2 so that for an α-helix (13) and for a β-strand m ≈ π 2 m ≈ 1.0 (26) (27) In [10], [11] it has been shown that loops, which are re- gions where (κi, τi) are variable, can be constructed in terms of the dark soliton solution of the generalized dis- crete nonlinear Schrodinger equation that derives from the energy (24), κi+1 = 2κi − κi−1 + dV [κ] dκ2 i κi (i = 1, ..., N ) (28) where we set κ0 = κN +1 = 0. This is the Master equation that we use to compute the shape of a folded protein Cα backbone. D: Soliton Ansatz We do not know the explicit form of the dark soli- ton solution to the present discrete nonlinear Schrodinger equation. But a numerical approximation can be easily constructed using the procedure described in [11]. Fur- thermore, since it turns out that in the case of proteins the first term in (25) is small, an excellent approximation i − bθκ2 i θ2 κ2 i θi − aθθi + cθ 2 θ2 i + . . . (30) energy In [21] it has been shown using the Ansatz (29) that over 92% of PDB configurations can be constructed in terms of 200 explicit soliton profiles. This makes a strong case that the solitons of the DNLS equation are the mod- ular building blocks of folded proteins [21]. E: Side-chain energy function In order to account for the Cβ contribution to the pro- tein free energy, we augment (22) by terms that involve the variables (θi, ϕi) in (17). We shall assume that side- chain directions are locally slaved to the backbone. By an explicit analysis of PDB structures using Frenet frames this can be confirmed to be the case [22], as shown in Figure 2. The Cβ latitude angles θi are gauge i.e. frame invari- ant: The θi are entirely determined by the tangent vec- tors ti, and consequently can not depend on the choice of framing. To the leading order we may then assume that each θi interacts locally, with the corresponding κi only. The leading order contribution is obtained by Tay- lor expanding a general interaction potential around the (κi dependent) equilibrium values of the θi, (cid:27) N(cid:88) (cid:26) dθ 2 i=1 Eθ = where the additional terms are of higher order in powers of κi and θi. From this we solve for θi, θi = aθ + bθκ2 i cθ + dθκ2 i (31) Again, as in the case of (23), the overall scale cancels which leaves us with only three independent parameters. As visible from Figure 2, the fluctuations in θi around the average value (18) are small. From this Figure we also learn [22] that these fluctuations are slaved to the backbone geometry, which is dictated by the κi. This confirms that the present approximation (30) is reason- able. Unlike θi, the Cβ longitude angle ϕi does not remain intact under the frame rotation (8) but transforms ac- cording to (19). Consequently it can form a SO(2) frame (i.e. gauge) invariant combination with the backbone torsion angle (5). Two examples of gauge invariant com- binations are and τi − ϕi−1 + ϕi i(cid:88) k=1 τk ϕi + (32) We prefer to proceed with the second one, it is local in ϕi. (The first is a difference of the second.) As in (22) we specify the unitary gauge, which amounts to selecting the Frenet framing along the backbone. 6 As visible in Figure 2, the fluctuations in ϕi are about as small as those in θi. Moreover, in the combinations (32) the torsion angles τi are determined locally by the bond angles according to (23). Consequently we may also Taylor expand the ϕi contribution to the energy, and following (30) we conclude that the leading order contribution is of the form N(cid:88) (cid:26) dϕ 2 i=1 Eϕ = i − bϕκ2 i ϕi − aϕϕi + κ2 i ϕ2 cϕ 2 ϕ2 i (cid:27) + . . . (33) (34) This slaves ϕi to the backbone κi according to ϕi = aϕ + bϕκ2 i cϕ + dϕκ2 i Again only three of the four parameters are independent, the overall scale drops out. F: Total energy We combine (22), (30) and (33) to arrive at the total E = Eκ + Eτ + Eθ + Eϕ 2 κi+1κi + i + q · (κ2 2κ2 (cid:26) N(cid:88) i=1 i − bτ κ2 i τ 2 κ2 i τi − aτ τi + i − bθκ2 i θ2 κ2 i θi − aθθi + + i=1 = − N−1(cid:88) (cid:26) dτ N(cid:88) (cid:26) dθ N(cid:88) (cid:26) dϕ N(cid:88) i=1 i=1 + 2 2 (cid:27) (35) (36) (37) (38) i − m2)2 (cid:27) (cid:27) cτ 2 τ 2 i cθ 2 θ2 i (cid:27) + . . . (39) + 2 i=1 i − bϕκ2 i ϕi − aϕϕi + κ2 i ϕ2 cϕ 2 ϕ2 i Since the variations in (θi, ϕi) are much smaller than those in τi, the ensuing contributions Eθ and Eϕ are also much smaller than Eτ . Furthermore, according to (39) the backbone torsion angles τi and the side-chain angles (θi, ϕi) are all slaved to the backbone bond angles κi. As a consequence the DNLS equation (28) is the Mas- ter Equation that entirely determines the geometry of a folded protein: The Cα−Cβ backbone is constructed by first solving for κi. The remaining two angles (θi, ϕi) are then constructed in terms of the κi using (23), (31) and (34). G: Parameters The energy function (35)-(39) involves a number of pa- rameters. Eventually, we would like to compute their numerical values directly from the amino acid sequence. But at the moment this has not yet been achieved. A priori it appears that the number of parameters needed in (35)-(39) to describe an entire protein back- bone, could be quite large. However, due to the presence of the dark soliton the number of parameters is actu- ally remarkably small: For each super-secondary struc- ture such as a helix-loop-helix, whenever the loop can be described in terms of a single soliton solution to (28), the potential (25) has only four independent parameter com- binations. In addition of q and m that characterize the second and third term, there are only two essential pa- rameters in the first term. Three of these four parameters can be given the following interpretations. Two of them determine the values of κi in the ground states such as (13), (14) that are located along the backbone before and after the soliton i.e. the type of the helix that precedes and follows the soliton. The third parameter determines the length of the loop. The fourth parameter can then be included as one of the three independent parameters in the torsion profile (23). It can be attributed to the length of the soliton, in terms of torsion. In addition, in the soli- ton profile of κi there is the parameter that specifies the position of the soliton along the backbone. This is an additional parameter that emerges from the periodicity of the lattice structure (lattice translation invariance). Since the overall scale in (23) cancels out, the two re- maining parameter combinations in addition of the loop length, become determined by the values of τi in the ground states surrounding the soliton i.e. the type of the helix as in (13), (14). In this way we arrive at the conclusion that for the backbone, the only loop specific parameters are those that determine the lengths of the solitons. All additional parameters in the energy function determine the regular secondary structures such as (13) and (14). The pro- files of all loops are completely fixed by the unique dark solution solution to (28). Similarly, we conclude from (31), (34) that in the equa- tions that determine the Frenet frame orientations of the Cβ carbons, there is only one loop specific parameter in both θi and ϕi. In each equation, the overall scale fac- tor cancels out and the values of the additional two in- dependent parameters are fully specified by the regular secondary structures adjacent to the loop. Since (35)-(39) aims to predict the 45-60 space coordi- nate of Cα, and the corresponding 30-45 directional co- ordinates of the Cβ in terms of 11 essential parameters, we have a highly underdetermined system of equations. This implies that the physical principles of our approach are experimentally testable. 7 In [21] we have found that most crystallographic pro- tein structures in PDB can be described in a modu- lar fashion and with experimental B-factor precision, by combining together no more than 200 explicit soliton pro- files. We propose that by learning how to compute the parameter values directly from the sequence, the geo- metric shape of most folded proteins can be constructed simply by solving the Master equation (28). H: Fluctuations around solitons As such, the equations (28), (23), (31), (34) describe the critical points of the energy function (35). This en- ergy function should be duly interpreted as describing the effective long distance limit of the full microscopic second-quantized Schrodinger operator. As such, (35) then relates to proteins in the sense of a semi-classical approach. This kind of semi-classical description is com- mon in quantum field theories. There, it is often boldly used to describe phenomena at length scales that are sev- eral orders of magnitude smaller than anything which may have any direct relevance to proteins. We now wish to estimate the short distance scale, at which we expect the semiclassical approach to proteins based on (35), to break down due to quantum mechanical zero-point fluc- tuations. The backbone profile (28), (23) describes the Cα lat- tice in the limit where thermal fluctuations vanish. But even near zero temperature the protein remains subject to residual zero-point fluctuations that can not always be ignored. It is difficult to estimate and even harder to accurately calculate the amplitude of these zero-point fluctuations. For a realistic order of magnitude estimate we inspect the distribution of the B-factors that char- acterize experimental uncertainties in PDB data. We use all those PDB structures where the crystallographic measurements have been made at temperatures less than 50K. The result is displayed in Figure 3. It shows that for the Cα carbons the zero point fluctuations have a lower bound which is in the vicinity of 0.15 A. Consequently we estimate that the precision of our semi-classical ap- proach can at best be of the same order of magnitude. We account for these zero point fluctuations by dressing our (semi)classical backbone profiles with a tubular dominion that has a radius of 0.15 A. In particular, this tubular do- minion accounts for the bond length fluctuations, around the average distance (15) between the neighboring Cα carbons. As can be seen from Figure 1, the fluctuations in the Cα− Cα distances are normally within range of 0.15 A. 8 FIG. 3: The number of entries in PDB with temperature below 50K vs. Debye-Waller fluctuation distance. Note the logarithmic scale. III: λ-REPRESSOR PROTEIN AS A MULTISOLITON CONFIGURATION A: Loop spectrum of 1LMB We start our soliton-based investigation of the λ- repressor protein by analyzing its (κi, τi) spectrum. This will identify the putative soliton content. We use the first chain of the homo-dimer with the PDB code 1LMB. The structure is conventionally interpreted as a four loop configuration, and the second loop is the DNA binding one. From the PDB file we read the Cα coordinates. We compute the tangent vectors from (1) and the binormal vectors from (2), and the bond and torsion angles from (4) and (5). We construct these angles using the standard convention that κi ∈ [0, π]. We locate the regions where the torsion angles τi are subject to large fluctuations. In these regions we judiciously implement the transforma- tion (12). This identifies the soliton structures in the loops. Both the motivation and the technical details of the soliton identification procedure are described in [10] and [13]. In the left hand side of Figure 4 we show the (κi, τi) spectra that we obtain from the PDB data using the stan- dard differential geometric convention that curvature is positive κi > 0. Each of the four figures describes the spectra over one of the four loops of 1LMB, as they are identified in PDB. We observe that in each Figure 4 on the left hand side, there is a region where the torsion angle τi fluctuates rapidly between positive values and negative values. On general grounds [10], [13] a region where the torsion angle is solely positive and only subject to small variations is a putative regular secondary struc- ture. On the other hand, a region where the torsion angle fluctuates between positive and negative values is an in- dicative of an inflection point [13], this kind of fluctuation FIG. 4: (Color online) On the left column, we show the (κi, τi) spectrum for the four PDB loop structures of 1LMB, that we compute with the prevailing differential geometric convention that curvature is a nonnegative quantity. Black is the bond angle κi, grey (red) is the torsion angle τi. On the right we display the corresponding spectra after we have identified the soliton profiles in κi, using (12). All PDB loop structures except for the second, display the profile of a soliton pair. Only the second PDB loop can be identified as a single soliton state. suggests the presence of solitons. By judiciously apply- ing the gauge transformation (12) in the regions where τi fluctuates between positive and negative values, we find that the κi profiles in each of the left hand side Figure 4 indeed describe solitons: Comparing with (13), (14) we conclude that the first loop structure in the PDB data is a pair of solitons separated by a short α-helix. The second PDB loop is a single soliton that interpolates between two α helices. The third loop structure in the PDB data is a pair of solitons separated by a short β-strand. Fi- nally, the fourth PDB loop is a pair of solitons, separated by a short α-helix. Notice that our spectral analysis based structure iden- tification is a refinement of the conventional one, which utilizes techniques such as the presence or absence of hy- drogen bonds to conclude whether a site is part of a reg- ular secondary structure or part of a loop. Consequently ) ÅFluctuation Amlitude (00.10.20.30.40.50.60.70.80.91N entries 110210Index 202224262830323436Angles (radians) -3-2-10123qIndex 202224262830323436Angles (radians) -3-2-10123qIndex 363840424446Angles (radians) -3-2-10123Index 363840424446Angles (radians) -3-2-10123Index 4550556065Angles (radians) -3-2-10123Index 4550556065Angles (radians) -3-2-10123Index 6668707274767880Angles (radians) -3-2-10123Index 6668707274767880Angles (radians) -3-2-10123very short helical structures that become clearly visible in our (κi, τi) spectral analysis, can be interpreted differ- ently in more conventional approaches. B: Soliton Ansatz We proceed to evaluate the parameters in (29), (23) that give our best fit to those seven soliton profiles, iden- tified by analyzing the (κi, τi) spectrum. In Table 1 we TABLE I: Our best parameter values for each of the seven solitons in Figure 2. Note that m1, m2 are determined mod (2π). Soliton (8,30) (26,39) (36,47) (46,59) (56,65) (62,74) (74,87) c1 2.00441 2.94889 2.89729 2.97927 2.96486 2.94948 2.89725 c2 1.99595 2.95201 2.90755 3.00015 2.97087 2.94547 2.89945 m1 26.65124 70.67882 39.27387 1.07948 26.69087 20.43071 89.50870 m2 26.68412 70.60369 39.22546 1.52942 26.25280 20.38220 89.55252 9 soliton profiles could also be given asymmetric integer parts. The numerical values of the bond angles κi that we compute from (29) using the parameter values in Ta- ble I, are to be reduced onto the fundamental domain [0, π] using the mod(2π) multivaluedness. We recall from Section II D, that the introduction of non-vanishing values of N1 and N2 enables us to account for the non-monotonic profile that the bond angles of the PDB configuration display when restricted into the fundamental domain: At each point where the profile of the PDB data becomes non-monotonic, we simply add (or subtract) 2π until we obtain a monotonic profile. In this manner, by allowing the bond angle to take values over a larger domain, the κi profile of the PDB data over each soliton can be made monotonic which improves the accuracy of the fitted soliton profile (29). See Figure 5 where we display the second PDB loop together with its approximation by the Ansatz (29) in the extended range, as an example. s a/b d/b Soliton 24.50259 −9.9921 · 10−2 4.2191 · 10−5 (8,30) (26,39) 30.49642 −1.5114 · 10−7 1.0662 · 10−11 (36,47) 41.39325 −5.3794 · 10−7 7.4566 · 10−11 5.1477 · 10−7 −5.1529 · 10−7 (46,59) 53.67225 (56,65) 57.85123 −9.62942 · 10−8 1.45097 · 10−12 (62,74) 70.22069 −9.27151 · 10−7 3.05202 · 10−10 1.8457 · 10−11 (74,87) 75.56315 −7.13705 · 10−7 present our best parameter values. The parameters can be computed by various different techniques. Here we have chosen a Monte Carlo search that is fast and gives very good accuracy. In Table 1 we also identify the corre- sponding super-secondary structures by their PDB back- bone indices. Note that since our approach is based on the spectral analysis of bond and torsion angles and since the definition of a bond angle involves three sites while that of the torsion angle involves four, three residue in- dices at both ends of the 1LMB chain are absent in the (κi, τi) list. Notice also that we have introduced some overlap between the different super-secondary structures. This ensures that we can eventually combine them to- gether into the full 1LMB backbone. Moreover, in the case of all except the fourth soliton, we have utilized the multi-valuedness in the definition of the bond angle to extend the range of its values outside of the fundamen- tal domain κi ∈ [0, π]. This corresponds to selecting non-vanishing integers N1, N2 in the Ansatz (29). For simplicity we have limited our Monte Carlo search to the symmetric case N1 = N2, but for better accuracy the FIG. 5: The data points correspond to the second loop of 1LMB, after we have properly extended the range of the bond angles, by using the inherent multivaluedness of these angles. The interpolating function describes the Ansatz (29) with the parameter values given in Table I. We compute the torsion angles τi from (23), before implementing the 2π reduction of the bond angles. We then reduce the ensuing values of the τi into the fun- damental domain τi ∈ (−π, π] using the 2π periodicity. The underlying multivaluedness entirely accounts for the fluctuations in the τi profile. In Figure 6 we compare the explicit backbone profiles that we have computed from (29), (23), (6), (7) using the parameter values in Table I, with the 1LMB data in PDB. We find that for each soliton, our backbone pro- files describe the structural motifs of 1LMB with a pre- cision that is substantially better than the experimental precision which is determined by B-factors. This persist even when we account for the 0.15 A estimate of the zero point fluctuations around our solitons. With these highly precise soliton profiles we can unambiguously con- Index 36384042444648 (radians) ik-40-30-20-1001020304010 TABLE II: The (minimal) soliton sites that we have used in searching for matching structures in PDB, together with the number of matches. The search is limited to those x- ray structures that have a resolution better than 2.0 A and a match is a configuration that deviates less than 0.5 A in total root mean square distance (RMSD) from the soliton. Soliton 1 2 3 4 Sites (20,28) (27,36) (36,46) (50,58) Matches 9601 4 810 159 Soliton Sites Matches 5 6 7 (55,63) 1552 (66,75) (74,82) 1342 406 the DNA recognition helix, is unique to the λ-repressor protein. The only matching structures are located in the different PDB entries of the λ-repressor protein itself. We also note that for this soliton the B-factors are slightly higher than for any of the other six solitons along the 1LMB backbone. Finally, by following [12] we have combined the seven solitons together to describe the sites 8-90 (PDB index- ing) of the entire 1LMB backbone. The result is shown in Figure 7. The overall RMSD accuracy that we get in FIG. 7: (Color online) The PDB backbone configuration of 1LMB (dark red) and its multi-soliton (light green) approxi- mation, for sites 8-90. The total RMSD distance is 0.45 A. this manner is 0.45 A. This could be further improved by adjusting the parameters while combining the solitons. But as such, the accuracy displayed by the configuration in Figure 7 is below the experimental B-factors. Con- sequently any improvement is senseless, in light of the quality of the experimental data. FIG. 6: (Color online) The distance between the PDB back- bone of the first 1LMB chain and its approximation by the seven solitons (29), (23) as a function of the residue number. The black line denotes the distance between the soliton and the corresponding PDB configuration, the grey area around the black line describes the estimated 0.15 A zero point fluc- tuation distance around solitons, obtained from Figure 3. The grey (red) line denotes the Debye-Waller distance that is com- puted form the B-factors in PDB. clude that 1LMB has a total of seven solitons. Two of the α-helices and the sole β-strand are so short that the ensuing soliton pairs are interpreted as single loops in the conventional, hydrogen bond based analysis of the PDB data. The only motif where our construction identifies a single PDB loop as a single isolated soliton, is the DNA binding one. All of the remaining three loops consist of a pair of solitons with profile (29), (23) that are separated from each other either by a very short α-helix in case of the residues (23, 33) and (69, 90), or by a very short β-strand in case of residues (51, 61). We have performed a statistical analysis on the oc- currence of our seven solitons in all PDB proteins. In Table 2 we list the number of matches that each of these solitons has when we search PDB for configurations that deviate from the given soliton by an overall root mean square distance (RMSD) which is less than 0.5 A. We have chosen this cut-off value since it is representative of the Debye-Waller fluctuation distance in the experi- mental 1LMB data; see Figure 6. The interesting obser- vation is that the second soliton of 1LMB that precedes Site Index1015202530Angstroms00.20.40.60.81yqSite Index26283032343638Angstroms00.20.40.60.81yqSite Index363840424446Angstroms00.20.40.60.81yqSite Index46485052545658Angstroms00.20.40.60.81yqSite Index56575859606162636465Angstroms00.20.40.60.81yqSite Index62646668707274Angstroms00.20.40.60.81yqSite Index74767880828486Angstroms00.20.40.60.81yq11 FIG. 8: (Color online) The multi-soliton solution to (40) (line) and the PDB values of 1LMB (dots) along the Cα backbone, for the backbone bond angles C: Soliton solution and 1LMB We proceed to construct the parameters corresponding to the 1LMB backbone, for both the Cα-atoms and the side-chain Cβ-atoms, using the energy function (35). We first solve (28) to obtain the bond angle i.e. κi-profile. We then construct the backbone torsion angles τi and the side-chain (θi, ϕi) angles from (23), (31), (34). We con- struct a single solution of (28), that describes the entire chain. For simplicity of construction, we restrict the values of κi into the fundamental domain κi ∈ [0, π]. As in the case of the Ansatz (29), a higher precision could be obtained by allowing κi to take values beyond the fundamental domain. But with (35), it turns out that we can reach the B-factor accuracy by constructing the solution of (28), using the fundamental domain only. This is because the solution of (28) is even better suited for modeling proteins than the Ansatz (29). To construct the parameters that describe the back- bone κi profile of 1LMB, we use the iterative learning algorithm of [11]. It determines the parameters by locat- ing the fixed point of (cid:111) ] − (κ(n) i+1 − 2κ(n) i + κ(n) i−1) (cid:110) κ(n+1) i = κ(n) i − i V (cid:48)[κ(n) κ(n) i (40) that minimize the RMSD distance to 1LMB. Here i }i∈N denotes the nth iteration of an initial configu- {κ(n) i }i∈N and is some sufficiently small but oth- ration {κ(0) erwise arbitrary numerical constant. We select = 0.01. A fixed point of (40) clearly satisfies the DNLS equation (28). Following [11] we utilize step-functions to construct an initial configuration for κi. The ensuing initial pro- file of κ(0) is chosen to have the same overall profile as the properly gauge transformed 1LMB that we display in Figure 4 right hand side column. A Monte Carlo routine is set up to determine the parameters. For this we have developed a package that we call Propro [24]. It imple- ments our parameter learning algorithm for a given pro- tein structure, largely automatizing the entire process. i In Figure 8 we show a Propro screen capture of the κi profile that describes the final multi-soliton solution that yields the shortest overall RMSD distance between the solution to (40) and the 1LMB structure, for the back- bone Cα carbons. In Figure 9 we show the corresponding τi profile, computed from (23). The backbone Cα RMSD distance between out multi-soliton solution and 1LMB is 0.52 A. This is slightly larger that what we obtained with the Ansatz (29). But this time we have not extended the values of κi outside of the range κi ∈ [−π, π]. In Figure 10 we display the distance between the 1LMB and the soliton solution to (28), (23) for the individual Cα atoms. For the most part the distance between our multi-soliton solution and 1LMB is below the Debye-Waller fluctua- tion distance. The only real exception is located at the FIG. 9: (Color online) The profile of torsion angle computed from (23) (line) and the corresponding PDB values of 1LMB (red dots) along the Cα backbone. site 31, where the distance between 1LMB Cα carbon and the soliton solution is close to 1.6 A. This site is lo- FIG. 10: (Color online) The distance between the 1LMB back- bone and our multi-soliton configuration, constructed by solv- ing (28) (black line). The grey shaded area around the black line describes the estimated 0.15 A zero point fluctuation dis- tance around the multi-soliton solution. The grey (red) line describes the experimentally measured Debye-Waller fluctua- tion distance Site Index 1020304050607080Angstroms00.20.40.60.811.21.41.61.82yqcated at the second soliton. We recall that this soliton is unique for 1LMB (see Table II) and that it was also singled out by the Ansatz (29). Our analysis indicates that something takes place at this soliton that warrants a more careful experimental analysis. The properties of this soliton might have a role in the transition from lyso- genic to lytic state. We proceed to extend our multi-soliton to describe both the positions of the Cα carbons, and the directions of the Cβ carbons. For this we use (31) and (34). The fi- nal configuration has a combined Cα - Cβ distance of 0.6 A to 1LMB. In Figure 11 we compare the corresponding structures, the 1LMB backbone is translucent. FIG. 11: (Color online) Comparison between the Cα - Cβ multi-soliton solution (opaque, with colors online) and the 1LMB structure (translucent, grey online). The RMSD dis- tance is 0.6 A. In Figure 12 we have a close-up of the region around PDB site 31, where the difference between the multi- soliton solution and the 1LMB configuration is largest. Finally, in Table III we list the parameters in (35)-(39), for all the seven solitons. IV: COLLAPSE STUDIES OF 1LMB In the backbone energy (36), (37) we have retained only those variables that are relevant to our description of the Cα geometry. The derivation of (36), (37) is based on the general concept of universality [14]-[17], in combi- 12 FIG. 12: (Color online) Close-up of Figure 11 around site 31, the location of the second soliton. Multi-soliton is opaque, the PDB structure of 1LMB is translucent. nation with the requirement that the energy must be in- dependent of the coordinate frame where it is computed. Consequently, by construction, our energy function cor- rectly describes the leading order long distance contri- bution to any energy function that is grounded on more detailed atomic level considerations. All the variables and interactions that are less relevant for the description of the Cα geometry, are accounted for effectively through the functional form and the parameter values of the in- dividual contributions to (36), (37). We shall try and approach protein dynamics in the same universal manner. We average over all very short time scale oscillations, vibrations and other tiny fluctu- ations in the positions individual atoms that are basi- cally irrelevant to the way how the folding progresses over those time scales that are biologically relevant. The general concept of universality [14]-[17] proposes us to adopt a Markovian Monte Carlo time evolution with the following universal, coarse grained heat bath probability distribution [25], [26], [27] P = x 1 + x with x = exp{− ∆E kT } (41) Here ∆E is the energy difference between consecutive MC time steps, that we compute from (36), (37). We select the numerical value of the temperature factor kT so that the model describes the appropriate phase. In [28] it has been shown that (36), (37) is capable of describing the three phases of polymers. At low values of kT we are in the phase of collapsed proteins. As the value of kT increases and reaches the Θ-point value, there is a transition to random coil phase. When the temperature reaches even higher values, there is a cross-over to self- avoiding random walk. It turns out that in the collapsed phase, below the Θ- point temperature, the universal aspects of folding dy- namics are quite independent of the numerical value of kT . For concreteness, we perform our simulations in the collapsed using the value kT = 10−15 Note that the overall normalization of kT can always be changed by an overall normalization of the parameters in Table III. We shall assume that during the folding process there are no re-arrangements in the backbone covalent bond structure, such as chain crossings. For this we introduce a self-avoidance condition that keeps the distance between any two backbone Cα atoms at least as large as the length of a typical van der Waals radius which is around ∼1.3 Angstrom. Note that we do not propose that (41) is capable of describing the atomic level dynamics of the folding pro- cess. Such minuscule details are highly sensitive to the initial atomic configuration. A detailed knowledge of the time evolution of a particular atom during the collapse can hardly have any practical relevance for the under- lying physical principles and phenomena. Thus, for the purpose of conceptually understanding the temporal evo- lution of a protein towards its native conformation, the dynamics described by (41) is sufficient. We argue that the combination of (35), (41) correctly captures the uni- versal statistical aspects of protein collapse over the bio- logically relevant temporal and spatial scales. A: Antiferromagnetism and folding nuclei In our approach, proteins have a modular structure. A folded protein is built by combining together solitons of the discrete non-linear Schrodinger equation (28), one after another. From the point of view of the energy func- tion (36), (37), a uniform helical configuration is one with κi ≈ m and with the value of τi computed from (23) this is a ground state of the energy. In particular, a straight linear rod is a special case, it is a ground state when m = 0. As usual, the physical principles that give rise to pro- tein folding are best analyzed in the absence of other processes and interfering agents. For this reason, in the present sub-section, we study the soliton formation along a helical backbone, by inspecting how the folded config- uration builds from the ground state of the energy. Con- sequently we use a straight helical structure as our initial configuration. In the case of the λ-repressor protein we are particularly interested in the formation of the third, DNA binding soliton. The parameter values in Table III are uniform over each putative super-secondary structure. In particular, there is no information on the loop locations. In a pro- tein, the placement of a loop often correlates with the po- sition of certain amino acids, such as proline and glycine, 13 that act as folding nuclei. In order to model a folding nucleus we introduce a transient parameter σ that sends the first term in (36) into − N−1(cid:88) 2 κi+1κi → N−1(cid:88) (2σ − 1) · 2 κi+1κi (42) i=1 i=1 Initially, we set σ = 0 for all i except at the links (i, i + 1) along which the putative soliton centers are located. At these links we start with σ = +1. This corresponds to an anti-ferromagnetic i.e. repulsive nearest neighbor interaction between the ensuing two sites. During the early stages of the simulation, we decrease the values of σ so that after some number of steps we reach the uniform final value σ = 0 for all links along the entire chain. For each super-secondary structure the value of m in Table III determines the regular secondary structure which is located either before or after the corresponding loop, and the value of the parameter q in (36) determines the propensity of this structure to form. The stability of α helices and β strands is due to hydrogen bonds that form during the collapse, and consequently the value of q can be interpreted as a measure of the strength of hy- drogen bond interactions. In our simulations we wish to start from an initial configuration with no initial hy- drogen bonds. We conform to this by setting all q = 0 initially. We then switch on the hydrogen bond inter- actions by increasing the values of the parameters q to those given in Table III. We find that this stabilizes the regular secondary structures. We wish to investigate the effects that the temporal ordering of loop formation has on folding and on misfold- ing. For this we compare different orderings in removing σ and in switching on the values of the q. In the case of the lysogeny maintaining λ-repressor protein we have considered various scenarios to conclude that there is the following general pattern: The seven solitons that we display in Figure 4 (right column) tend to form as pairs, with (2,3), (4,5) and (6,7) each a soliton- soliton pair. The first soliton is also made with a pair. But after the formation, the pair of this soliton moves away and disappears through the N -terminal of the back- bone. The alternative, where the first and second soliton form as a pair seems to give rise to a misfolded state that furthermore appears to be unstructured. We now de- scribe in detail two generic examples that illustrate this general pattern: First example: In our first generic example we start from a uniform, straight helical structure. In the Fig- ure 13a we show the initial κi profile. We note that there is substantial latitude in choosing the initial val- ues of κi and τi. To begin with, we also set all q = 0 so that there are no hydrogen bonds to stabilize the he- lical structure. All the remaining parameters have the values that are listed in Table III. We introduce an an- tiferromagnetic coupling σ = +1 at links (i, i + 1) with i = 16, 23, 33, 46, 49, 63, 67. This models the folding nuclei at the putative positions of the centers of the soli- tons. During the first part of the simulation, say during the first 2, 000, 000 Monte Carlo steps, we adiabatically remove the folding nuclei by linearly decreasing the val- ues of σ until we reach the final ferromagnetic values σ = 0 at all sites. At the same time we turn on the hy- drogen bonds, by increasing the values of the couplings q from zero to the values given in Table III. After the first 2, 000, 000 steps all parameters along the backbone then have the values shown in Table III. We remind that according to our observations there is a lot of latitude in details of the procedure. We find that the last four solitons form as the pairs (4,5) and (6,7). At the putative location of the third soliton, we also observe the formation of a soliton-soliton pair. But the soliton which is located closer to the N - terminal moves towards left as shown in Figure 13, and becomes anchored at the site 23 where it forms the sec- ond soliton of 1LMB. A new soliton pair appears at the putative location of the first soliton, one of these solitons disappears through the N -terminal and the entire back- bone stabilizes rapidly into the correct native state. See Figure 13. Second example: In this example we simulate a sce- nario where the first pair is formed before the third soli- ton. The initial configuration is a folded structure, with the fully formed soliton pairs (1,2), (4,5) and (6,7) i.e. these solitons have the same (κi, τi) values as the 1LMB. But the DNA binding third soliton is absent and instead there is a helix extending from the second to the fourth soliton, see Figure 14 and 15. The simulation starts with an antiferromagnetic σ = 1 at link i = 34, and with no hydrogen bond interactions i.e. q = 0 between second and fourth solitons. There is then an initial production of a soliton-soliton pair as seen in Figure 15b and we find two possibilities: If the hydrogen bonds form slowly i.e. q grows to its final value slowly in comparison to the removal of σ, we arrive at a final fold where there is an additional soli- ton (loop) around site 39. We display the bond angle profile in Figure 14c. The protein is now misfolded. On the other hand, if the hydrogen bonds are formed more rapidly, we arrive at a configuration where the third and fourth soliton annihilate each other. No soliton is formed, but due to steric restraints there is a slightly irregular helical region between sites 31 and 45. The final config- uration is shown in Figure 15. We conclude that if the (1, 2) pair forms before the third soliton, the protein becomes misfolded into a state with more than one conformational substrate. We pro- pose that this could be confirmed experimentally. It might relate to the transition from the lysogenic to the lytic state in λ-phage. 14 FIG. 13: (Color online) Time series of soliton (loop) forma- tion along the 1LMB backbone in our simulation. The black line shows the time evolution of the solitons, the grey (red) line is the PDB profile. Time increases from top down. In the first Figure from the top we have the initial configuration, a straight helical structure. The solitons 4-7 form as the two soliton pairs (4,5) and (6,7). At the position of soliton 3, there is also a pair formation. But the pair of soliton 3 propagates along the chain towards the N -terminal, until it becomes an- chored at site i = 23 where it forms the second soliton. A new soliton pair forms at site 16 and one of these two soli- tons moves and disappears through the N -terminal, leaving us with the first soliton and the correctly folded backbone Index1020304050607080 (radians)ik-3-2-10123Index1020304050607080 (radians)ik-3-2-10123Index1020304050607080 (radians)ik-3-2-10123Index1020304050607080 (radians)ik-3-2-10123Index1020304050607080 (radians)ik-3-2-10123Index1020304050607080 (radians)ik-3-2-10123Index1020304050607080 (radians)ik-3-2-1012315 files in Figure 16 proposes the following mechanism for the lysogenic-lytic transition: Under lysogenic conditions where the λ-repressor protein prevails, the soliton pairs of the λ-repressor protein that are located immediately prior and after the DNA binding domain are relatively stable. But when there is a change in the environmen- tal conditions that excites phonon fluctuations along the protein chain such as raise in temperature or UV radia- tion, or maybe an enzymatic action that remains to be identified, either of these soliton pairs can discharge by a saddle-node bifurcation. This bifurcation disturbs the structure of the immediately adjacent DNA binding mo- tif to the extent that the protein looses its capability to maintain the lysogenic phase. Since each of the cor- responding motifs in the CRO protein are topologically more stable single soliton configurations, they are much more insensitive to effects such as local phonon excita- tions due to UV radiation and thermal effects, and con- sequently the lytic phase can take over. we display the FIG. 14: (Color online) If hydrogen bonds form slowly, there is an extraneous soliton that disturbs binding between 1LMB and DNA, and probably no binding is possible. Grey line is the 1LMB profile, and black with red dots is the simulated profile. FIG. 15: (Color online) If the hydrogen bonds are formed early, the backbone enters in an unstructured state where the 3 soliton become mis-formed. Consequently there is no soliton that would dock the 1LMB with DNA. Grey line is the 1LMB profile, and black with red dots is the simulated profile. FIG. 16: (Color online) The resolved (κi, τi) spectrum for the first PDB helix-loop-helix of 1LMB (left) and the corre- sponding structure of in 2OVG (right). The bond angle κ is black, torsion angle τ is grey (red). The bond angle spectra reveal that in 1LMB the loop is a bound state of two close-by solitons while in 2OVG there is only one soliton profile. B: Comparison of λ-repressor and CRO soliton structures In Figure 16 we compare the first two solitons in 1LMB to the first loop in the CRO regulator protein that con- trols the transition to the lytic state. For the latter, we use the PDB structure with code 2OVG. In the CRO protein, the first loop is topologically more stable, in the sense that it consists of a single soliton. As such the loop in 2OVG is more stable than in 1LMB. For a plane curve, a single soliton can be made or deleted only by transport- ing it through one of the end points of the curve. On the other hand, a pair of solitons such as the one in the left hand side of Figure 16 is not topologically stable but can be more easily created or removed locally, by a saddle- node bifurcation that brings the two solitons together. This removes the corresponding loop by converting it (in this case) into a single long α-helix. A comparison between the λ-repressor and CRO pro- first PDB helix-loop-helix motif and for comparison we display the corresponding structure in the CRO protein with PDB code 2OVG. In the case of λ-repressor the mo- tif is clearly a bound state of two solitons while in the case of CRO we have a single isolated soliton. C: Heating and cooling in 1LMB We apply (41) to theoretically investigate what takes place in 1LMB when we heat it into the random coil phase, and then re-cool it back to the collapsed phase. According to Anfinsen [29] the protein should return to its original conformation. We start from the multi-soliton configuration that de- scribes the PDB structure 1LMB with parameter values given in Table III. Unlike in the previous subsection, dur- ing the entire heating and cooling cycle we now keep all Index202224262830323436Angles (radians) -3-2-10123yqIndex810121416182022Angles (radians) -3-2-10123yqthe parameter values intact. In particular, neither dur- ing the heating nor during the cooling do we introduce any transient antiferromagnetic parameter σ as in the previous subsection. Nor do we change the parameter values q during the present process. As a consequence the position of any soliton and the size of any helical structure becomes determined dynamically, without any explicit folding nuclei. Moreover, since the parameter values q do not change, the strength of the underlying hydrogen bond interactions remains constant during the simulations. Any deformation or adjustment in the heli- cal structures will be entirely due to thermal fluctuations during the heating and cooling cycle. We introduce the heat bath dynamics (41). The tem- perature kT is assumed to be globally determined, and the heating and cooling should proceed slowly over the biologically relevant time scales. This ensures that the entire protein structure is kept at an equal temperature value so that we can ignore any effects due to local tem- perature variations. During heating and cooling cycle we follow the back- bone evolution by computing the root-mean-square dis- tance (RMSD) between the Cα coordinates r0i of the 1LMB backbone and those of the multi-soliton configu- ration ri, as a function of the temperature (cid:115) 1 (cid:88) N i 16 FIG. 17: At temperature values log kT < −4 the protein re- turns to its original conformation after the heating and cool- ing cycle. Between −4 < log kT < −2 a small fraction starts to misfiled. At log kT ≈ −2 there is a rapid transition, so that proteins that have been heated to higher temperatures become misfolded. there is a small fraction of cycles at the end of which the protein becomes misfolded. Finally at around the temperature value kTC2 there is a rapid cross-over and if the heating temperature exceeds RM SD(T ) def = (ri0 − ri)2 (43) T > TC2 We start from a low temperature value that corresponds to the collapsed phase. We again choose the numerical value kT = 10−15 (44) for the initial configuration. We adiabatically increase the temperature kT to some high value during 500,000 MC steps. We then keep the system at this high tem- perature during 1,000,000 MC steps, and finally cool it back to the original temperature value (44) during an- other 500.000 steps. The relatively large number of MC steps in our cycle at the high temperature ensures that the protein becomes fully thermalized to that tempera- ture value. We have made simulations with several hundreds of re- peated heating and cooling cycles, always starting with (44) and heating to different high temperature values that are well above the Θ-point, where the protein enters the random walk phase. From Figure 17 we learn that as long as the high temperature value remains below kTC1 ≈ 10−4 (45) the protein always returns back to its original shape upon cooling. But between values in the range 10−4 ≈ kTC1 < kT < kTC2 ≈ 10−2 the final conformation is always misfolded. The misfold- ing is caused by deformation of one or more of the loops, often due to the wrong ordering in loop formation. In Figure 18 we show as an example, how (43) evolves in average during a typical heating cycle that reaches very near the critical temperature value (45). We find that during the heating there is a rapid increase in the RMSD distance. This is due to the transition to the random coil phase. When the system is kept at the high temperature, there are substantial thermal fluctuations. The shaded region (blue online) denotes the one standard deviation extent of these fluctuations. During the cooling we have a collapse transition, and at the end the value of (43) becomes small with practically no fluctuations, indicat- ing that the protein has returned to the original 1LMB like conformation. In Figure 19 we display a generic ran- dom coil structure that we observe during the heating phase. It has the look of a typical random coil with no regular, helical structure remaining. The structure is not static, but subject to very strong thermal fluctuations in its shape. SUMMARY In summary, we have investigated the structure and folding pathways of the lysogeny maintaining λ-repressor protein. Our approach is based on an effective energy MaxkT10log-5-4-3-2-101% RMSD<102040608010017 tinuum non-linear Schrodinger equation. We are able to re-construct the entire backbone of the λ-repressor pro- tein with a precision that compares and even exceeds the precision of the experimental crystallographic structure with PDB code 1LMB, when we determine the precision in terms of the Debye-Waller B-factor fluctuation dis- tance. The high precision of our theoretical description enables us to conclude that the second soliton solution that appears in our description of the 1LMB is unique to this protein, there are no other similar solitons in the en- tire Protein Data Bank. The remaining solitons including the DNA binding one are all ubiquitous in PDB. We have also investigated the corresponding soliton structure in the CRO protein that is responsible for the lytic phase, and found that this soliton appears more stable and is also commonly found in the PDB data. These observa- tions suggest that the transition between the lysogenic and lytic life-cycles could somehow relate to the very ex- ceptional structure of the second soliton in 1LMB. We have extended our energy function to describe the collapse dynamics of 1LMB. In line with the construc- tion of the energy function, we rely on the concept of universality to propose that at biologically relevant time scales the folding dynamics can be described in terms of Glauberian relaxation dynamics, with Markovian time evolution. In this way we have found that all solitons in 1LMB appear to form as soliton-soliton pair. In par- ticular, the second soliton with its exceptional structure forms as a pair with the DNA binding third soliton; The pair of the first soliton flows away and disappears through the N -terminal of the protein. Moreover, if for some rea- son the formation of the second and third solitons is dis- rupted, for example if the second soliton forms by itself before the third soliton, we find that the third soliton can not form properly. It is either formed in combina- tion of an extra soliton, or then the protein enters in an unstructured conformation. The choice between these two alternatives is made by the strength of the hydrogen bond formation. A.N. thanks H. Frauenfelder and G. Petsko for com- munications and J. Aqvist for discussions. ∗ Electronic address: [email protected] † Electronic address: [email protected] ‡ Electronic address: [email protected] [1] M. Ptashne, A Genetic Switch, Third Edition, Phage Lambda Revisited. Cold Spring Harbor Laboratory Press, Cold Spring Harbor (2004) [2] M.E. Gottesman and R.A. Weisberg, Microbiol. Mol. Biol. Rev. 68 796 (2004) [3] A. Amir, O. Kobiler, A. Rokney, A.B. Oppenheimer and J. Stavans, Mol. Syst. Biol. 3 1744 (2007) [4] B. Snijder and L. Pelkmans, Nature Reviews, Mol. Cell. FIG. 18: (Color online) The evolution of the RMSD value (43) between the 1LMB and the heated structure during one cycle. At the high temperature value which is here kT = 10−4, the RMSD is large and subject to large thermal fluctuations; the shaded area denotes the one standard deviation fluctuation regime in the values of (43). At the end of the cycle, the protein folds back to the conformation of 1LMB. FIG. 19: A generic configuration during the high temperature regime in the cycle of Figure 18. The structure is an apparent random coil with no regular structure, and its detailed form is subject to strong thermal fluctuations. function, that we have argued describes the small fluc- tuation limit of any atomic level energy function. Our justification of the energy function is entirely based on two very general physical principles: The concept of uni- versality originally introduced in the context of phase transitions and critical phenomena, and the demand that any physical quantity must be independent of the coordi- nate system that is used for its description. Using these two universal physical principles, we have shown that the long wavelength fluctuation limit of the energy function for both the backbone Cα and side-chain Cβ atoms is fully determined in an essentially unique fashion. The energy function we have derived, computes the Cα and Cβ conformation in terms of a soliton solution to a variant of the discrete non-linear Schrodinger equation as the modular component. The explicit form of the rele- vant dark soliton solution is not known to us, but we have found that an excellent approximation can be obtained by naively discretizing the known dark soliton of the con- 18 Biol. 12 119 (2011) (1966) [5] F. St-Pierre and D. Endy, Proc. Nat. Acad. Sci 105 20705 (2008) [6] T. Wigle and S. Singleton, Bioorg. Med. Chem. Lett. 17 3249 (2007) [16] K.G. Wilson, Phys. Rev. B4 3174 (1971) [17] M.E. Fisher, Rev. Mod. Phys. 46 597 (1974) [18] A.J. Niemi, Phys. Rev. D67 106004 (2003) [19] M. Chernodub, L.D. Faddeev, A.J. Niemi, JHEP [7] M. Avlund, I. Dodd, S. Semsey K. Sneppen and S. Kr- 0812:014 (2008) ishna, Journ. Virol. 83 11416 (2009) [20] L.D. Faddeev, L.A. Takhtajan Hamiltonian methods in [8] H.M. Berman, K. Henrick, H. Nakamura, J.L. Markley, the theory of solitons (Springer Verlag, Berlin, 1987) Nucl. Acids Res. 35 (Database issue) D301 (2007) [21] A. Krokhotin, A.J. Niemi, X. Peng Phys. Rev. E85 [9] U.H. Danielsson, M. Lundgren and A.J. Niemi, Phys. (2012) 031906 Rev. E82 021910 (2010) [10] M. Chernodub, S. Hu and A.J. Niemi, Phys. Rev. E82 011916 (2010) [11] N. Molkenthin, S. Hu and A.J. Niemi, Phys. Rev. Lett. 106 078102 (2011) [22] M. Lundgren, A.J. Niemi, S. Fan, Phys. Rev. E (in print) [23] M. Lundgren, A.J. Niemi, to appear [24] Available from http://folding-protein.org [25] R.J. Glauber, Journ. Math. Phys. 4 294-207 (1963) [26] A.B. Bortz, M.H. Kalos and J.L. Lebowitz, Journ. Com- [12] S. Hu, A. Krokhotin, A.J. Niemi and X.Peng, Phys. Rev. put. Phys. 17 10-18 (1975) E83 041907 (2011) [13] S. Hu, M. Lundgren and A.J. Niemi, arXiv:1102.5658 [27] A. Krokhotin, M. Lundgren, A.J. Niemi, (to appear) [28] M. Chernodub, M. Lundgren, A.J. Niemi, Phys. Rev. Biomolecules (q-bio.BM); Phys. Rev. E (to appear) E83 011126 (2011) [14] B. Widom, J. Chem. Phys. 43 3892 (1965) [15] L.P. Kadanoff, Physics (Long Island City, NY) 2 263 [29] C.B. Anfinsen, Science 181 223 (1973) 19 TABLE III: Our best parameter values for the multi-soliton solution that models 1LMB. Notice that in line with (26), (27), the values of m1, m2 imply that all the regular structures are α-helices except for the one separating solitons 4 and 5 which is a short β-strand. We also note that the overall scale of the parameters is fixed by the normalization of the first term in (36) which we have chosen for convenience. Soliton q1 q2 1 2 3 4 5 6 7 1.12091 0.357906 0.260909 0.684119 5.15882 0.314503 0.947322 1.87372 9.69166 6.14144 4.75578 6.77828 0.38534 0.624884 m1 1.55737 1.65666 1.68182 1.47243 1.07066 1.66164 1.58568 m2 1.50013 1.54119 1.54676 1.09234 1.53464 1.6235 1.51678 Soliton a b c/2 d/2 1 2 3 4 5 6 7 -1.0 -1.0 -1.0 -1.0 -1.0 -1.0 -1.0 -32357500 -1.86689 -9.26604 -7.51371 -25.0125 -23.9299 7.9809 47.5340 438750 0.0413976 0.00101952 0.0271569 0.121919 0.0148377 0.013751 0.251803 0.597751 0.0410994 0.0181312 0.000215793 0.037091 Soliton aθ 1.24035 bθ · 10−12 -475.728 2.74724 -198463 1.32293 -5.021439·108 1.34998 -81.1641 1.38447 -6629519 1.38293 -6.28532 1.23869 -177.296 1 2 3 4 5 6 7 cθ 1.0 1.0 1.0 1.0 1.0 1.0 1.0 dθ 5.44473·10−10 0.42667 3.74966·10−6 6.59854·10−9 5.20625·10−7 1.34992·10−5 7.6271·10−7 cϕ 1.0 1.0 1.0 1.0 1.0 1.0 1.0 dϕ · 10−10 7211.16 1481.99 -51.6611 5478 1039.56 1.2685 2.09412 Soliton aϕ 1 2 3 4 5 6 7 0.971374 0.813254 0.771272 0.616865 0.89315 0.545154 0.988183 bϕ -6.7206·10−10 2.3582·10−7 2.74 ·10−6 1.2158·10−11 3.87·10−9 0.184472 3.12459·10−9
1306.6487	1	1306	2013-06-27T13:10:56	Frequency Doubling Nanocrystals for Cancer Theranostics	[ "physics.bio-ph", "physics.med-ph", "physics.optics" ]	A novel bio-photonics approach based on the nonlinear optical process of second harmonic generation by non-centrosymmetric nanoparticles is presented and demonstrated on malignant human cell lines. The proposed method allows to directly interact with DNA in absence of photosensitizing molecules, to enable independent imaging and therapeutic modalities switching between the two modes of operation by simply tuning the excitation laser wavelength, and to avoid any risk of spontaneous activation by any natural or artificial light source.	physics.bio-ph	physics	Frequency Doubling Nanocrystals for Cancer Theranostics Davide Staedler,†,(cid:107) Thibaud Magouroux,‡,(cid:107) Solène Passemard,† Sebastian Schwung,¶ Marc Dubled,§ Guillaume Stéphane Schneiter,† Daniel Rytz,¶ Sandrine Gerber-Lemaire,† Luigi Bonacina,∗,‡ and Jean-Pierre Wolf‡ Institute of Chemical Sciences and Engineering, EPFL, CH-1015, Lausanne, Switzerland, GAP-Biophotonics, Université de Genève, 22 chemin de Pinchat, CH-1211 Genève 4, Switzerland, FEE Gmbh, Struthstrasse 2, 55743 Idar-Oberstein, Germany, and SYMME, Université de Savoie, BP 80439, 74944, Annecy Le Vieux Cedex, France E-mail: [email protected] Abstract A novel bio-photonics approach based on the nonlinear optical process of second harmonic generation by non-centrosymmetric nanoparticles is presented and demonstrated on malignant human cell lines. The proposed method allows to directly interact with DNA in absence of photosensitizing molecules, to enable independent imaging and therapeutic modalities switch- ing between the two modes of operation by simply tuning the excitation laser wavelength, and to avoid any risk of spontaneous activation by any natural or artificial light source. ∗To whom correspondence should be addressed †Institute of Chemical Sciences and Engineering, EPFL, CH-1015, Lausanne, Switzerland ‡GAP-Biophotonics, Université de Genève, 22 chemin de Pinchat, CH-1211 Genève 4, Switzerland ¶FEE Gmbh, Struthstrasse 2, 55743 Idar-Oberstein, Germany §SYMME, Université de Savoie, BP 80439, 74944, Annecy Le Vieux Cedex, France (cid:107)Contributed equally to this work. 1 We demonstrate here a novel diagnostic and therapeutic (theranostic) protocol based on the nonlinear optical process of non phase-matched second harmonic (SH) generation by non-centrosymmetric nanoparticles, referred to in the following as harmonic nanoparticles (HNPs).1,2 To date, the ca- pability of these recently introduced nanometric probes of doubling any incoming frequency has not been employed for therapeutic use, although it presents several straightforward advantages, including i) the possibility to directly interact with DNA of malignant cells in absence of photo- sensitizing molecules, ii) fully independent access to imaging and therapeutic modalities, and iii) complete absence of risk of spontaneous activation by natural or artificial light sources other than pulsed femtosecond lasers. Given the unconstrained tunability of the HNPs nonlinear conversion process, this approach can be extended to selectively photo-activate molecules at the surface or in the vicinity of HNPs to further diversify the prospective therapeutic action.3 Here we show that by tuning the frequency of ultrashort laser pulses from infrared (IR) to visible (both harmless), SH generation leads respectively to diagnostics (imaging) and therapy (phototoxicity). Specifically, we report in situ generation of deep ultraviolet (DUV) radiation (270 nm) in human-derived lung cancer cells treated with bismuth ferrite (BiFeO3, BFO) HNPs upon pulsed laser irradiation in the visible spectrum, at 540 nm. We observe and quantify the appearance of double-strand breaks (DSBs) in the DNA and cell apoptosis, in the area of the laser beam. We show that DNA dam- ages are dependent on irradiation-time, laser intensity, and NP concentration. We observe that apoptosis and genotoxic effects are only observed when visible light excitation is employed, being completely absent when IR excitation is used for imaging. HNPs, a family of NPs specifically conceived for multi-photon imaging, were introduced in 2005 for complementing fluorescence imaging labels.1,4,5 Although comparatively less bright than quantum dots, HNPs possess a series of advantageous optical properties, including complete ab- sence of bleaching and blinking,1,6 spectrally narrow emission bands, fully coherent response,7 -- 9 ,and UV to IR excitation wavelength tunability.10,11 These unique characteristics have been re- cently exploited in demanding bio-imaging applications12 including regenerative research.13 The possibility of working with long wavelengths presents clear advantages in terms of tissue pene- 2 Figure 1: Multiphoton imaging of a HNPs treated sample. Lung-derived A549 cancer cells exposed for 5 h to 50 µg/mL BFO HNPs. Yellow: two photon excited fluorescence from cell membrane dye FM1-43FX. Blue: SH signal from HNPs. Scale bar:10 µm. Figure 2: Cytotoxic and oxidative effects of BFO HNPs. The nanomaterial shows good bio- compatibility with high survival rate after long-term incubation. A: cell survival by MTT for different exposure times of lung-derived A549 (black bars) and HTB-182 (white bars) cancer cell lines to 100 µg/mL BFO. B: effect of BFO at 25 and 100 µg/mL on DTT content in an abiotic environment after 1 h of incubation. C1 and C2: ROS production after 24 h exposure to 25 (white bars) or 100 (black bars) µg/mL HNPs by DHE (C1) and DCFH-DA (C2) assays. Results statistically compared to untreated cells. 3 0246810121416% of consumed DTT11.051.11.151.21.251.3A549HTB-18211.051.11.151.21.251.31.35A549HTB-182Fold-increase0204060801005h24h72hCell survival (% of control)ABC1C2********************************tration, as in the IR spectral region, imaging depth is strongly increased by reduced absorption (provided that water absorption is avoided) and weak scattering (preventing degradation of spatial and temporal laser profiles).14 As an example of HNPs based imaging, Fig. 1 displays lung-derived A549 cancer cells stained with FM1-43FX cell membrane dye exposed for 5 h to BFO HNPs at 50 µg/mL. The image was acquired upon near IR excitation at 790 nm, the two-photon excited fluorescence from the dye are shown in yellow, while the intense blue spots correspond to SH radiation emitted by HNPs. The latter have the tendency to remain attached to cell membranes without being internalized due to their relatively large size. As for other nanobiotechnological approaches, selective binding of NPs to specific cell membrane receptors would rely on the presence of targeting molecules at their surface,12 a strategy that was not implemented in this exploratory study. Given the novelty of the nanomaterial employed in this study, prior to the assessment of photo- therapeutic modality, BFO HNPs were characterized and screened for biocompatibility in terms of cytotoxicity and oxidative effect. The cytotoxic effect of 100 µg/mL BFO HNPs was assessed after 5, 24 and 72 h exposure (Fig. 2A) on two lung-derived cell lines (A549, HTB-182). BFO cytotoxicity was found acceptable in both samples as HNPs did not cause any detectable effect on cell survival after 5 h exposure, and after 24 h and 72 h cell viability remains remarkably high (>75%), comparable to that observed with HNPs composed of other nanomaterials previously screened.11 To quantify the oxidative stress induced by BFO HNPs, the catalytic activity of the NPs was first measured in a cell free environment by dithiothreitol (DTT) assay.15 BFO HNPs show a dose-dependent consumption of DTT after 1 h incubation (Fig. 2B), suggesting that they can exert catalytic production of superoxide. The production of reactive oxygen species (ROS) by BFO HNPs in cell cultures was assessed using two fluorescence assays: dihydroethidium (DHE) and carboxydichlorodihydrofluorescein diacetate (DCFH-DA).16,17 We could detect a dose-dependent increase of ROS, more pronounced in A549 cells than in HTB-182 (Fig. 2C1 and C2), which remains however low compared to that induced by other metal-based NPs.11,16 -- 18 Overall, the result of this thorough screening indicates a good biocompatibility of this nanomaterial, tested for 4 the very first time for biological applications, and sets the ground for the light-triggered HNPs-cells interaction described in the following. Immunohistochemistry images demonstrating the effect of visible irradiation on malignant Figure 3: cell lines for apoptosis (cPARP) and DNA repair (γH2AX). Human lung-derived A549 (A) and HTB-182 (B) cancer cell lines untreated (1) or exposed to 100 µg/mL BFO HNPs (2 and 3), incubated 24 h, and irradiated for 120 s. Expression of cPARP (2) or γH2AX (3) observed by IHC after further 24 h or 30 min of incubation, respectively. Positive cells are in brown, nuclei in blue, and HNPs aggregates appear as small brown spots. Scale bar: 50 µm DNA absorption is particularly efficient in the deep UV (DUV, <300 nm), as all DNA bases possess bands peaking around 260 nm with negligible intensity from 310 nm on. Irradiation of cell cultures at this wavelength results into DSBs and evokes a complex network of molecular responses, eventually resulting in DNA repair and/or cell apoptosis.19 -- 21 Histone variant H2AX is a key component of the early stage response to DNA damages, as upon UV exposure it is phosphorylated at its carboxyl terminus to form γH2AX at the DSBs sites.19,20,22,23 After the first appearance of UV-induced DNA-damages, cells first activate DNA-repair mechanisms and then apoptosis occurs to eliminate potentially hazardous cells. UV-dependent apoptosis is caused by the activation of caspase-3 and subsequently cleavage of the poly (ADP-ribose) polymerase (PARP), resulting in a cleaved-form (cPARP) with a mass of 89 kDA.23 -- 25 For the irradiation experiment, cells were plated in 35 mm Petri dishes with glass bottom for 48 h, then medium was replaced and cells were incubated for 24 h with BFO HNPs (25 or 100 µg/mL, 2 mg/mL stock solution in water) or a negative control containing the vehicle (distilled water). The sample was exposed for 30, 60, or 120 s to ultrashort (30 fs) pulses of visible light generated by a noncolliner OPA (15 mW average power, 1 KHz repetition rate) with a laser spot size of 170 µm diameter. During 5 irradiation, cells were kept in a microscope incubator. After light treatment, cells were incubated for 30 min (γH2AX assay) or 24 h (cPARP detection)22,26,27 and then fixed with 3% formaldehyde in PBS. Frequency doubling of femtosecond pulses of visible light (540 nm) by HNPs attached to cell membranes generates DUV photons in the close vicinity of cell nuclei, optimally placed for direct photo-interaction (see Fig. 1). The biological effects of such laser irradiation is reported in the immunohistochemistry (IHC) images of Fig. 3. The two image rows are associated to the two human malignant cell lines already tested for cytotoxicity, A549 and HTB-182. The control samples (A1, B1) show no expression for both reporters, confirming that they are not present under physiological conditions, while the clear effect (positive cells in brown) visible in panels 2 and 3 for treated cells indicates that a strong interaction upon irradiation takes places, showing the DNA- repairing enzyme γH2AX expression well localized within the nuclei and the cPARP reporter of cell apoptosis in the cytoplasm of the damaged cells. Figure 4: Spatial localization of the expression of cPARP and γH2AX reporters. Analysis of one rep- resentative IHC image (480 x 650 µm) for the treatments in Fig. 3. For each rectangular (80 x 130 µm) sub-region the % ratio of cells positive for the expression of cPARP (column 1) and γH2AX (column 2) is expressed according to the color-scale. Black empty square: laser spot (150 x 150 µm). The spatial localization of the IHC expression of the two reporters for DNA repair and cell apoptosis is given in Fig. 4. The laser focal spot (empty black square) is superimposed to a spatially resolved pattern of 80 × 130µm rectangles indicating in false colors the % of positive 6 cells to cPARP (A1, A2) and γH2AX (B1, B2) for the two cell lines. One can appreciate how the biological effect of visible irradiation perfectly co-localizes with the laser spot (>80 % positive cells) and rapidly decreases outside the focal region to negligible values. Figure 5: Apoptosis and DNA damage upon visible light exposure. Human lung-derived A549 (A) and HTB-182 (B) cancer cell lines exposed to vehicle (grey bars) or 100 µg/mL BFO HNPs (black bars), incubated 24 h and irradiated for 30, 60, or 120s. Expression of cPARP (upper row) or γH2AX (lower row) was observed by IHC after further 24 h or 30 min of incubation, respectively. Results are expressed as % ratio of positive cells. All comparisons to control or between cells exposed to laser with and without BFO HNPs are significant (p<0.001) if not otherwise specified. The quantitative assessment of the effects of in situ DUV generation is reported in Fig. 5. In the histograms, the number of IHC-positive cells is expressed as % ratio of total cells in the area of the laser-spot. Firstly, one can observe that BFO HNPs at 100 µg/mL without laser irradiation do not cause any increase of cPARP and γH2AX expression, ensuring that oxidative effects of BFO alone do not interfere with irradiation assays. Upon laser-exposure, the ratio of cells positive for cPARP and γH2AX clearly increases in an exposure time-dependent manner. In A549 the expression of the two proteins is comparable, whereas HTB-182 express systematically more γH2AX. Such a stronger enzymatic activity seems correlated with higher cell viability: the maximal expression of cPARP (apoptosis) is around 60% while in A549 it reaches almost 100%. The major decrease in cells viability observed upon irradiation of HNPs-treated samples, to- gether with the high spatial localization of the biological effects, makes HNPs-based approaches 7 020406080100120no laser30s60s120s020406080100120no laser30s60s120s020406080100120no laser30s60s120s% of positive cells for cPARP expressionExposure time to laser020406080100120no laser30s60s120s% of positive cells for γH2AX expressionNSNSNSNSNSNSA1B1A2B2amenable for developing therapeutic (photo-dynamic) protocols. With this goal in mind, we per- formed an additional verification, essential to ensure the possibility of independently addressing imaging and cell irradiation modalities, ensuring that the definition of the zone to be treated (which might rely on a specific NPs surface functionalization) can be preliminary safely performed with- out any risk of unwanted activation. Cells exposed to HNPs were irradiated for 5 min with laser set at 790 nm (SHG at 395 nm, outside the DNA bases absorption band) with intensity parameters equal to those of the protocol described above for visible irradiation. In this case, we did not remark any interference on cells metabolism. As reported in previous works, imaging is not limited to near infrared wavelengths but it can be performed even above 1.5 µm, with clear advantages in terms of imaging depth (thanks to decreased scattering and absorption) and long-term photo-stability.1,10 Figure 6: Effect of laser intensity. Human lung-derived A549 (A) and HTB-182 (B) cancer cell lines exposed to vehicle (gray bars) or 100 µg/mL BFO HNPs (black bars), incubated 24 h and irradiated for 30, 60 or 120 s. Expression of cPARP (graph at the top) or γH2AX (graph at the bottom) observed by IHC after further 24 h or 30 min of incubation, respectively. Results are expressed as % ratio of positive cells. All comparisons between cells exposed to laser with and without BFO are significant (p<0.001) if not otherwise specified. To complete the characterization of the HNPs+visible irradiation sinergistic effects on cells viability, we further investigated laser intensity and HNP concentration dependence. A549 and HTB-182 cell lines were exposed to 100 µg/mL HNPs or to vehicle and irradiated with the same laser average power focused onto a larger surface (400 µm diameter) corresponding to a neat 8 02040608010012030s60s120s02040608010012030s60s120s02040608010012030s60s120s% of positive cells for cPARP expressionExposure time to laser02040608010012030s60s120s% of positive cells for γH2AX expressionA1B1A2B2******NSFigure 7: Effect of HNPs concentration. Human lung-derived A549 (A) and HTB-182 (B) cancer cell lines exposed to 25 µg/mL HNPs, incubated 24 h and the irradiated for 30, 60, or 120s. Expression of cPARP (graph at the top) or γH2AX (graph at the bottom) observed by IHC after further 24 h or 30 min of incubation, respectively. Results are expressed as % ratio of positive cells. All comparisons to control or between cells exposed to laser with and without BFO are significant (p<0.001) if not otherwise specified. fivefold intensity decrease with respect to the previous protocol. As reported in Fig. 6, a substantial decrease in the expression of both cPARP and γH2AX was detected in the two lung-derived cancer cell lines. However, the difference between cells treated with BFO and with vehicle remained always significant, except for the expression of cPARP on HTB-182 cells after 30 s exposure, suggesting that BFO generate DUV at lower irradiation intensity as well. The greatest difference between cells exposed to BFO and to vehicle was always observed for the longest exposure and the decrease on cPARP expression was greater compared to the expression of γH2AX. For assessing the concentration dependence, A549 and HTB-182 cells were treated with 25 µg/mL of HNPs and irradiated according to the high intensity protocol. Also in this case, as summarized in Fig. 7, cells showed a reduced expression of cPARP, which remained anyway always significantly greater compared to that measured with vehicle. This reduction is more pronounced on HTB-182 cells. The number of positive cells for γH2AX expression decreased in the two cell lines as well. The results and controls presented altogether confirm that UV generation and related cytotoxic and genotoxic effects can be unambiguously ascribed to visible-light excited BFO HNPs. Major 9 020406080100120no laser30s60s120s020406080100120no laser30s60s120s020406080100120no laser30s60s120s% of positive cells for cPARP expressionExposure time to laserA1B1A2B2020406080100120no laser30s60s120s% of positive cells for γH2AX expression***********NSNSNSNSeffects on cell viability (up to 100% apoptosis) are observed in the irradiated areas. The differ- ence on cPARP expression between cells irradiated with lower laser intensity or exposed to lower particle concentration suggests that the synergistic effect between laser and HNPs is dominated by direct DNA photo-damages, but also implies other subsidiary mechanisms, such as a thermal effects and cell membranes disruption.28 -- 30 The apoptotic cell fraction in samples exposed to 540 nm laser but not treated with HNPs can be ascribed to direct two-photon absorption by DNA, as previously observed.31,32 Such a non HNP-specific interaction can be easily counteracted thanks to the fact that the cellular effects exerted by BFO HNPs are limited to the area of the laser spot, as highlighted in Fig. 4. If IR imaging is preliminary performed to precisely define the zone needing irradiation, treatment conserves its high specificity. It should be noted that treatment localization is expected to be greater in the proposed HNP-based approach, which is primarily based on direct UV absorption by DNA, than in photo-dynamic treatments involving upconverting and plasmonic NPs. In these two latter cases, NPs-cell interaction is mediated by ROS, which are known to diffuse through tissues.33,34 In conclusion, we have presented and demonstrated on human-derived cancer cell lines an orig- inal nano-theranostics approach based on the nonlinear optical properties of HNPs. The method proposed enables wavelength-selected imaging and direct DUV photo-interaction with nuclear DNA. The biocompatibility of BFO, a nanomaterial firstly applied here for biological applications, screened for cytotoxicity and generation of oxidative stress, was found comparable to those of other HNPs or metal-based nanoparticles currently used in biomedical studies.11,16 -- 18 It should be noted that all HNPs, possessing high nonlinear efficiency,11 can exert the effects described here, which are therefore not unique to BFO. DSBs DNA damages and induction of apoptosis are typical targets of photodynamic therapies, which normally involved the use of direct UV radiation (with poor tissue penetration and lack of specificity) or chemical photosensitizers.35,36 To date, NPs-based strategies imply using of organic sensitizers (with some notable exceptions.37) and are mediated by ROS generation.18,34,38,39 As for classical phototherapy, these approaches can generate major side effects due to the presence of 10 toxic compounds and ROS which can diffuse to nearby tissues generating oxidative stress.40,41 The approach proposed, based on nonlinear optical response by HNPs and direct photo-interaction with nuclear DNA, might avoid side effects due to organic ligands and diffusion of toxic compounds, increasing selectivity and treatment localization. Moreover, the activation of the process is intrin- sically limited to femtosecond-pulse excitation and cannot be obtained with any other artificial or natural light source, differently also from the approaches based on sequential up-conversion of light frequency. This purely physical constrain greatly decreases the risk of unspecific treatment activation, in particular for surface lesions. Finally, and very notably, the proposed strategy allows to totally decouple diagnostic modality (IR imaging) from the therapeutic photo-dynamic action (visible irradiation), by simply tuning the excitation laser wavelength. Methods Multi-photon Imaging The imaging set-up is based on a Nikon A1R-MP inverted microscope coupled with a Spectra-Physics Mai-Tai DeepSee tunable Ti:Sapphire oscillator. A Plan APO 40× WI N.A. 1.25 objective was used to focus the excitation laser and to epi-collect the nonlinearly excited signal (SH and membrane dye fluorescence). Four independent non-descanned detectors acquire in parallel the signal spectrally filtered by four tailored pairs of dichroic mirrors and in- terference filters (SH filter; 395±5.5, Chroma). Optimal pulse compression at the focal plane was adjusted by maximizing the SH signal of individual HNPs dispersed on a coverslip. Visible laser irradiation For visible irradiation, we employed a two-stage non-collinear optical parametric amplifier (TOPAS White, Light Conversion) set at 540 nm. The output pulse charac- teristics are: 30 fs pulse duration, 15 mW average power, 1 KHz repetition rate. The sample was exposed for 30, 60 and 120 s using a laser spot size (measured by a high resolution beam pro- filer) of 170 µm or 400 µm diameter. During irradiation, cells were kept at controlled temperature (37oC), CO2 concentration (5%) and humidity in a microscope incubator (Okolab UNO). 11 HNPs dispersion and characterization BFO NPs were provided by the German company FEE at high concentration in ethanol. NPs were diluted 1:50 in 500 mL ethanol and decanted for 10 days. The supernatant was then taken, ethanol evaporated, and NPs re-suspended in distilled water. Successively, NPs were dispersed by ultra-sonic bath for 24 h and quantified by Prussian blue assay. For this assay, 50 µL of BFO solution were diluted in 50 µL HCl 6 M and 100 µL of 5% potassium hexacyanoferrate (Sigma-Aldrich) in PBS were added for 15 min. After incubation, the solution absorbance was measured at 690 nm in a multiwell-plate reader (Synergy HT, BioTek) and compared with the absorbance of a calibration curve with known BFO concentration. BFO NPs were finally diluted at 2 mg/mL in water. DLS and zeta-potential measurements were carried out with a Malvern NanoZ, yielding: zeta potential -52.7 ± 3.5 mV, mean hydrodynamic diameter 165.3 ± 24 nm. Acknowledgement This research has been conducted in the framework of European FP7 Research Project NAMDI- ATREAM (NMP4-LA-2010-246479, http://www.namdiatream.eu) and partially supported by the NCCR Molecular Ultrafast Science and Technology (NCCR MUST), a research instrument of the Swiss National Science Foundation (SNSF). The authors thank MER Dr Christine Wandrey for granting access to analytical equipment and S. Afonina and J. Gâteau for assistance with fem- tosecond laser equipment. Supporting Information Available Additional experimental details are provided as supplementary information. This material is available free of charge via the Internet at http://pubs.acs.org/. 12 References (1) Bonacina, L.; Mugnier, Y.; Courvoisier, F.; Le Dantec, R.; Extermann, J.; Lambert, Y.; Boutou, V.; Galez, C.; Wolf, J. P. Applied Physics B-Lasers and Optics 2007, 87, 399 -- 403. (2) Dempsey, W. P.; Fraser, S. E.; Pantazis, P. Bioessays 2012, 34, 351 -- 60. (3) Zhao, H.; Sterner, E. S.; Coughlin, E. B.; Theato, P. Macromolecules 2012, 45, 1723 -- 1736. (4) Nakayama, Y.; Pauzauskie, P. J.; Radenovic, A.; Onorato, R. M.; Saykally, R. J.; Liphardt, J.; Yang, P. Nature 2007, 447, 1098 -- 101. (5) Pantazis, P.; Maloney, J.; Wu, D.; Fraser, S. E. Proc Natl Acad Sci U S A 2010, 107, 14535 -- 40. (6) Le Xuan, L.; Zhou, C.; Slablab, A.; Chauvat, D.; Tard, C.; Perruchas, S.; Gacoin, T.; Ville- val, P.; Roch, J. F. Small 2008, 4, 1332 -- 1336. (7) Baumner, R.; Bonacina, L.; Enderlein, J.; Extermann, J.; Fricke-Begemann, T.; Marowsky, G.; Wolf, J. P. Optics Express 2010, 18, 23218 -- 23225. (8) Hsieh, C. L.; Grange, R.; Pu, Y.; Psaltis, D. Optics Express 2010, 18, 3456 -- 3457. (9) Hsieh, C. L.; Pu, Y.; Grange, R.; Laporte, G.; Psaltis, D. Optics Express 2010, 18, 20723 -- 20731. (10) Extermann, J.; Bonacina, L.; Cuna, E.; Kasparian, C.; Mugnier, Y.; Feurer, T.; Wolf, J. P. Optics Express 2009, 17, 15342 -- 15349. (11) Staedler, D. et al. Acs Nano 2012, 6, 2542 -- 2549. (12) Culic-Viskota, J.; Dempsey, W. P.; Fraser, S. E.; Pantazis, P. Nature Protocols 2012, 7, 1618 -- 1633. 13 (13) Magouroux, T.; Extermann, J.; Hoffmann, P.; Mugnier, Y.; Le Dantec, R.; Jaconi, M. E.; Kasparian, C.; Ciepielewski, D.; Bonacina, L.; Wolf, J. P. Small 2012, 8, 2752 -- 2756. (14) Zipfel, W. R.; Williams, R. M.; Webb, W. W. Nature Biotechnology 2003, 21, 1368 -- 1376. (15) Cho, A. K.; Sioutas, C.; Miguel, A. H.; Kumagai, Y.; Schmitz, D. A.; Singh, M.; Eiguren- Fernandez, A.; Froines, J. R. Environ Res 2005, 99, 40 -- 7. (16) Frick, R.; Muller-Edenborn, B.; Schlicker, A.; Rothen-Rutishauser, B.; Raemy, D. O.; Gun- ther, D.; Hattendorf, B.; Stark, W.; Beck-Schimmer, B. Toxicol Lett 2011, 205, 163 -- 72. (17) AshaRani, P. V.; Low Kah Mun, G.; Hande, M. P.; Valiyaveettil, S. ACS Nano 2009, 3, 279 -- 90. (18) Barzilai, A.; Yamamoto, K. DNA Repair (Amst) 2004, 3, 1109 -- 15. (19) Hanasoge, S.; Ljungman, M. Carcinogenesis 2007, 28, 2298 -- 304. (20) Yuan, J.; Adamski, R.; Chen, J. FEBS Lett 2010, 584, 3717 -- 24. (21) Stergiou, L.; Eberhard, R.; Doukoumetzidis, K.; Hengartner, M. O. Cell Death Differ 2011, 18, 897 -- 906. (22) Lu, C.; Zhu, F.; Cho, Y. Y.; Tang, F.; Zykova, T.; Ma, W. Y.; Bode, A. M.; Dong, Z. Mol Cell 2006, 23, 121 -- 32. (23) Shih, M. F.; Cherng, J. Y. Molecules 2012, 17, 9116 -- 9128. (24) Takasawa, R.; Nakamura, H.; Mori, T.; Tanuma, S. Apoptosis 2005, 10, 1121 -- 30. (25) Tuchinda, C.; Lim, H. W.; Strickland, F. M.; Guzman, E. A.; Wong, H. K. Photodermatol Photoimmunol Photomed 2007, 23, 2 -- 9. (26) Lee, E. R.; Kim, J. H.; Kang, Y. J.; Cho, S. G. Biol Pharm Bull 2007, 30, 32 -- 7. 14 (27) Narayanapillai, S.; Agarwal, C.; Tilley, C.; Agarwal, R. Photochem Photobiol 2012, 88, 1135 -- 40. (28) Aragane, Y.; Kulms, D.; Metze, D.; Wilkes, G.; Poppelmann, B.; Luger, T. A.; Schwarz, T. J Cell Biol 1998, 140, 171 -- 82. (29) Huang, X.; Jain, P. K.; El-Sayed, I. H.; El-Sayed, M. A. Lasers Med Sci 2008, 23, 217 -- 28. (30) Li, J.; Guo, D.; Wang, X.; Wang, H.; Jiang, H.; Chen, B. Nanoscale Res Lett 2010, 5, 1063 -- 71. (31) Konig, K.; So, P. T. C.; Mantulin, W. W.; Gratton, E. Optics Letters 1997, 22, 135 -- 136. (32) Tirlapur, U. K.; Konig, K.; Peuckert, C.; Krieg, R.; Halbhuber, K. J. Experimental Cell Re- search 2001, 263, 88 -- 97. (33) Idris, N. M.; Gnanasammandhan, M. K.; Zhang, J.; Ho, P. C.; Mahendran, R.; Zhang, Y. Nature Medicine 2012, 18, 1580 -- U190. (34) Zhou, A.; Wei, Y.; Wu, B.; Chen, Q.; Xing, D. Mol Pharm 2012, 9, 1580 -- 9. (35) Offer, T.; Ames, B. N.; Bailey, S. W.; Sabens, E. A.; Nozawa, M.; Ayling, J. E. Faseb J 2007, 21, 2101 -- 7. (36) Soares, A. R.; Neves, M. G.; Tome, A. C.; Iglesias-de la Cruz, M. C.; Zamarron, A.; Car- rasco, E.; Gonzalez, S.; Cavaleiro, J. A.; Torres, T.; Guldi, D. M.; Juarranz, A. Chem Res Toxicol 2012, 25, 940 -- 51. (37) Minai, L.; Yeheskely-Hayon, D.; Golan, L.; Bisker, G.; Dann, E. J.; Yelin, D. Small 2012, 8, 1732 -- 1739. (38) Ungun, B.; Prud'homme, R. K.; Budijon, S. J.; Shan, J.; Lim, S. F.; Ju, Y.; Austin, R. Opt Express 2009, 17, 80 -- 6. (39) Wang, C.; Tao, H.; Cheng, L.; Liu, Z. Biomaterials 2011, 32, 6145 -- 54. 15 (40) Jomova, K.; Baros, S.; Valko, M. Transition Metal Chemistry 2012, 37, 127 -- 134. (41) Scola, N.; Terras, S.; Georgas, D.; Othlinghaus, N.; Matip, R.; Pantelaki, I.; Mollenhoff, K.; Stucker, M.; Altmeyer, P.; Kreuter, A.; Gambichler, T. Br J Dermatol 2012, 167, 1366 -- 73. 16
1607.00209	1	1607	2016-07-01T11:42:03	Nonlinear Dynamic Force Spectroscopy	[ "physics.bio-ph" ]	Dynamic force spectroscopy (DFS) is an experimental technique that is commonly used to assess information of the strength, energy landscape, and lifetime of noncovalent bio-molecular interactions. DFS traditionally requires an applied force that increases linearly with time so that the bio-complex under investigation is exposed to a constant loading rate. However, tethers or polymers can modulate the applied force in a nonlinear regime. For example, bacterial adhesion pili and polymers with worm-like chain properties are examples of structures that show nonlinear force responses. In these situations, the theory for traditional DFS cannot be readily applied. In this work we expand the theory for DFS to also include nonlinear external forces while still maintaining compatibility with the linear DFS theory. To validate the theory we modeled a bio-complex expressed on a stiff, an elastic and a worm-like chain polymer, using Monte Carlo methods, and assessed the corresponding rupture force spectra. It was found that the nonlinear DFS (NLDFS) theory correctly predicted the numerical results. We also present a protocol suggesting an experimental approach and analysis method of the data to estimate the bond length and the thermal off-rate.	physics.bio-ph	physics	Friday, July 01, 2016 Nonlinear Dynamic Force Spectroscopy Oscar Björnham1 and Magnus Andersson2,3 1Swedish Defence Research Agency (FOI), SE-906 21 Umeå, Sweden, 2Department of Physics, 3Umeå Center for Molecular Research, Umeå University, SE-901 87 Umeå, Sweden Key words: AFM, Optical tweezers, receptor, ligand Abstract Dynamic force spectroscopy (DFS) is an experimental technique that is commonly used to assess information of the strength, energy landscape, and lifetime of noncovalent bio-molecular interactions. DFS traditionally requires an applied force that increases linearly with time so that the bio-complex under investigation is exposed to a constant loading rate. However, tethers or polymers can modulate the applied force in a nonlinear regime. For example, bacterial adhesion pili and polymers with worm- like chain properties are examples of structures that show nonlinear force responses. In these situations, the theory for traditional DFS cannot be readily applied. In this work we expand the theory for DFS to also include nonlinear external forces while still maintaining compatibility with the linear DFS theory. To validate the theory we modeled a bio-complex expressed on a stiff, an elastic and a worm-like chain polymer, using Monte Carlo methods, and assessed the corresponding rupture force spectra. It was found that the nonlinear DFS (NLDFS) theory correctly predicted the numerical results. We also present a protocol suggesting an experimental approach and analysis method of the data to estimate the bond length and the thermal off-rate. n o i s r e v t p i r c s u n a M 1 Friday, July 01, 2016 Introduction In the late nineties, Evans et al. formulated the theoretical basis for dynamic force spectroscopy (DFS) (Evans and Ritchie 1997; Merkel et al. 1999) that could allow for a time varying force in the expression for bond dissociation originally given by Kramers (Kramers 1940), and later refined by Bell (Bell 1978). Since then, DFS has successfully been employed in a wide range of applications in the field of biophysics to assess information of the strength, energy landscape, and lifetime of bio-molecular interactions (receptor-ligand bonds) at the molecular scale. DFS is commonly performed using force transducers such as Optical Tweezers (OT) or Atomic Force Microscopy (AFM) instruments, which can apply loading rates from a few pN/s to several nN/s. Using these techniques and the theory for DFS, atomic information of, for example, the Streptavidin-Biotin complex (Lee et al. 1994; Yuan et al. 2000), the digoxigenin-antibody complex (Neuert et al. 2006), the Mucin 1 – antibody bond (Sulchek et al. 2005), and the unbinding force of complementary DNA (Strunz et al. 1999) have been revealed. Although the theory is very useful in its current state, a restriction of the DFS theory is the requirement of a force that increases linearly with time. An experimental system can often be configured to provide an external force that meets this constraint for a limited force interval. However, for some complexes under study there can be situations when a more general time dependency of the external force in the DFS theory is required. For example, when the rupture force surpasses the linear span of the probe in a force spectroscopy apparatus or when measurements on bonds connected to biological tissues or organelles that have an intrinsic nonlinear response to external forces are investigated. Many bacterial adhesion organelles, commonly called pili, expressed by uropathogenic, enterotoxigenic, and respiratory tract associated bacteria exhibit nonlinear force responses (Forero et al. 2006; Miller et al. 2006; Andersson et al. 2007; Chen et al. 2011; Castelain et al. 2011; Mortezaei et al. 2015a; Mortezaei et al. 2015b). In particular extensions of pili expressed by uropathogenic and enterotoxigenic bacteria, show distinct nonlinear force responses, i.e., they respond to an external force initially by unwinding the structure at a constant force and thereafter by a pseudo-elastic force response. A nonlinear force response, similar to that of a worm-like chain (WLC) polymer, is also seen when extending T4 adhesion pili expressed by Streptococcus Pneumonia (Castelain et al. 2009). The response from these adhesion organelles are thus nonlinear with extension by nature. To assess information of the adhesin using DFS, the theory needs to be refined. In this work, we extend the DFS theory to include nonlinear external forces. The theory, referred to as nonlinear dynamic force spectroscopy (NLDFS), covers positive loading rates in a nonlinear regime where the force increases continuously. NLDFS is compatible with the linear DFS theory, which is shown to be a special case. We validated NLDFS by modeling an adhesin expressed on polymers that exhibit different nonlinear force responses using Monte Carlo simulations, and by assessing the rupture force spectra for various extension velocities. The results, for all tested cases, show that the nonlinear DFS theory correctly predict the most probable rupture force obtained from the stochastic modeled data and that the methodology suggested in this work is applicable to a variety of experimental investigations. 2 Friday, July 01, 2016 Theory Linear dynamic force spectroscopy We start by briefly introduce the traditional DFS theory. For detailed information see (Evans 2001). In 1940 Kramers (Kramers 1940) used the Smoluchowski equation to conclude that the transition rate, i.e., the thermal off-rate , which gives information of how frequently a bond transits between two states separated by an energy barrier , can be described by an Arrhenius factor, (1) where is the attempt rate originating from the molecular vibrations in an overdamped condensed system, is the Boltzmann´s constant, and is the absolute temperature. Almost 40 years later Bell (Bell 1978) inserted the effect of an external force, F, to the theory of Kramers to obtain an expression for the force-dependent dissociation rate , given by , (2) where is the bond length that describes the spatial distance between the energy minima and the transition energy barrier. Evans et al. extended Bells work by including a linearly increasing force described by a constant loading rate defined as the time derivative of the force, r (1-3). They showed that with increasing force the probability for the bond to rupture increases. This implies that there will be a maximum likelihood for bond rupture for a specific force. This force, which often is referred to as the most probable rupture force, , depends on the bond length, the thermal off-rate, and the loading rate and is found as the peak force in a rupture force spectrum. We will refer to this entity as the peak force throughout the rest of this work. Evans et al. showed that the peak force can be explicitly expressed as . (3) Nonlinear dynamic force spectroscopy To reduce the complexity when analyzing experimental data obtained from nonlinear loading rates we restrict the range of the nonlinear theory to continuous forces with positive time dependent loading rates. This implies that the force is always increasing and that every force value is only present once. This restriction is not effectively limiting the usability of NLDFS since in a force spectroscopy experiment the system under study is not likely to be exposed to external forces with alternating negative and positive loading rates. To derive the necessary equations for NLDFS we start by consider the probability P of an intact bond. The probability rate of bond rupture equals the negative change of P over time and can be expressed as 3 thoffkTE,TBEkTthoffakeaBkToffkbBFxkTthoffoffkkebxFlnbBthboffBrxkTFxkkTFriday, July 01, 2016 , where r is the loading rate defined as . (4) (5) The derivative of the probability rate, which is indicative of the position of the peak force, can thereby be expressed in two ways, wiz. as (6) . (7) and as The term , is zero at the peak force, which implies that it is possible to combine Eqs. (6) and (7) to obtain the following relationship at the peak of the force rupture spectrum, . (8) Thus by inserting the expression for dP/dF from Eq. (4) into Eq. (8), we obtain which, in turn, can be written as , . (9) (10) Making use of the expression for the dissociation rate, Eq. (2), gives the resulting relation between the loading rate and the peak force as , (11) which can be reformulated as 4 offdPdPrkPdtdFdFrdt222()offoffoffoffoffdkPdkdkdPdPPkPkdtdtdtdtdt222dPddPddPdPdrddPdPdrrrrdtdtdFdtdFdFdtdFdFdFdt2ddPrdFdF2offoffdkdPdrPkdFdtdt11offoffoffdkdrkrdtkdtlnlnoffoffdkdrkdtdtlnbBFxkTthboffBrxdrkedtkTFriday, July 01, 2016 . (12) Wormlike chain model We used the WLC model as a case study to evaluate the NLDFS theory. The WLC model is commonly used to describe the nonlinear entropic driven force response of biopolymers exposed to external forces (Strick et al. 2002; Kiss et al. 2006; Bianco et al. 2007; Björnham et al. 2008; Björnham and Schedin 2009). In this model the force can be expressed as a function of the distance between the two ends of the polymer, which, for the inelastic case, is given by , (13) where lp is the persistence length, L is the Euclidian distance between the two ends of the molecule, and Lc is the contour length of the polymer. The contour length is the structural length of the polymer and equals L if the polymer is fully stretched. If the polymer is extended at a constant velocity, v, the parameter L can be expressed as . This implies that the effective loading rate can be given as . (14) (15) The NLDFS theory then predicts that the peak force should be given by Eq. (12) with r(t) being given by Eq. (15) for WLC. 5 1lnlnbBthboffBrxkTdrFxkkTdt211144BpcckTLLFlLLLvt31112BpccdFtkTdFdLvvtrtdtdLdtlLLFriday, July 01, 2016 Results and Discussion Validation of the NLDFS theory To investigate the validity of the NLDFS theory for different time-dependent external forces, it was compared to the analytic solution for the rupture probability, i.e., Eq. (4) together with Eq. (2). Further on, numerical simulations by means of Monte Carlo (MC) methods where conducted and used as validation. We set the bond length, , to 0.70 nm and the thermal off-rate, , to , which are values in the typical range for noncovalent adhesion bonds (Sulchek et al. 2005; Björnham et al. 2009). The bio-complex was exposed to three different force responses all with an extension velocity of 10.0 µm/s. To verify the simulation and the analysis procedure, three sets of data with one million measurements each, were compiled using a narrow Gaussian kernel with a width of 0.50 pN, see Figure 1. The Gaussian kernel function will push the peak of the distribution slightly towards lower forces since the analytic distribution is skewed, which in turn, will result in a net flow of probability density towards lower forces at the peak as the kernel is applied. Although the effect is negligible here, it is recommended, in both DFS and NLDFS, to carefully choose the width of the Gaussian kernel to minimize this effect and at the same time obtain a smooth curve to identify the peak force. In the first case, the force was increased linearly with time, thus resulting in a constant loading rate. This implies that Eq. (12) is reduced to Eq. (3), which is the commonly used expression for the peak force in linear DFS. For the case with a loading rate of 100 pN/s, Fig 1A shows the resulting rupture force probability spectra. The inset shows, qualitatively, the time evolution of the force. As can be seen, the analytical solution (black dashed line), which according to Eq. (12) is 70.7 pN, coincides perfectly with that of the Monte Carlo simulations (red line). Moreover, the predicted peak force (green vertical dashed line), using Eq. (12), matches the force for which the distribution has a maximum of both curves. To model nonlinear increasing forces, e.g., to mimic cases when a receptor-ligand pair is attached to a membrane or polymer, we applied both a quadraticly increasing force (elastic reversible polymer) and a force that follows that of a WLC model, i.e., Eq. (13). Figure 1B and 1C, respectively, display the rupture probability densities from the simulations using these two nonlinear forces. The two panels show that the peak forces predicted by the theory agree with that of the simulations for both the case with a quadraticly increasing force, 52.3 pN, and for the case with the WLC, 41.6 pN. Since the peak forces predicted by the theory given above are in good agreement with those of the numerical solutions for all force curves, we conclude that the NLDFS theory can accurately predict the peak force of a receptor- ligand pair connected with polymers showing linear as well as nonlinear force responses. Note that in Figure 1C, when bonds are linked via WLC polymers, the rupture probability curve shows two peaks. These two peaks can be explained by the initial slowly increasing force and the final rapidly increasing force experienced by bonds linked to WLC polymers that are extended. Thus, two effective loading rates are possible resulting in a small fraction of bonds breaking at the lower loading rate. The ones that persisted therefore break at the higher loading rate. 6 bxthoffk4110sFriday, July 01, 2016 Figure 1. An example of the rupture probability distribution using one million samples for a velocity of 10.0 µms- 1. The black dashed line is the analytical solution while the red line is the density estimate from the Monte Carlo simulations, using a Gaussian kernel with standard deviation of 0.50 pN. The vertical green line is the peak force, F, predicted by Eq. (12). The agreement is excellent except for a small deviation at the smallest forces in the WLC-case due to inherent properties of the kernel density estimation method at the boundary of the interval. The inset figure depicts the relation of the applied force with respect to time. For the linear and quadratic cases the force was given by and with the constants and set to respectively. For the WLC case the force was given by Eqs. (13) and (14) with and and , 7 1Fax22Fax1a2a10 pNμm2310pNμm10.0 µmcL3.00 nm.plFriday, July 01, 2016 Applying NLDFS theory to experimental data In practice, however, it is of limited use to calculate the peak force solely for the cases when the bond length and the thermal off-rate are known a priori. Instead, the theory must be able to serve as a tool for experimentalists to estimate these two parameters from measurement data. The question is then: how do you design an experiment protocol to extract the desired parameter values? DFS DFS provides a technique to assess the bond length and the thermal off-rate following a straightforward scheme. First, the bio-complex under study is exposed to an external force that increases linearly with time, i.e., with a constant loading rate. Eventually the bond will break, giving rise to a rupture force. However, a single rupture force is solely one sample from the probability distribution that represents the specific loading rate, the bond length, and the thermal off-rate. To obtain sufficient statistics to quantify the distribution, the rupture force must be sampled many times for a given loading rate. Second, the rupture force spectrum is constructed from the set of rupture forces obtained, whereby the peak force is identified by localizing the peak of the distribution. Third, the loading rate is changed and a new set of measurements are conducted resulting in a new value of . Thus, for every loading rate, a corresponding value of the peak force is given. The bond length and the thermal off-rate can thereafter be found by fitting Eq. (3) to this set of data. This is a well working method that has been widely used. However, as was alluded to above, a limitation of DFS is the constraint that Eq. (3) is valid only under the assumption of constant loading rates. NLDFS As described above, DFS utilizes constant values of the loading rates in the experiments. These values of loading rates are used in pair with their corresponding values of the most probable rupture forces to find estimates of the bond length and the thermal off-rate. In NLDFS, the loading rate is not constant during a measurement, which therefore requires a slightly different approach. Instead of keeping the loading rate constant, the pulling velocity, v, of the force transducer is held constant during a measurement. Hence, the rupture forces are recorded as for DFS but the peak force is paired with the corresponding velocity. To find the peak force value that corresponds to this pulling velocity, the velocity is then changed and a new set of measurements is performed. Thus, for each velocity there is a corresponding value of the peak force. These data are used together with Eq. (12) to obtain the estimated parameter values. However, to do this, an expression for the loading rate as a function of the velocity needs to be derived. For the case when the force only depends on the position, i.e., F = F(L), this can be done in the following way. First, it should be noted that the loading rate can be written as (16) where L now is a measure of the position of the force transducer, in general given by vt where both v and t are known entities. The derivative is in general a function of L that needs to be known or 8 FdFdFdLdFrvdtdLdtdLdFdLFriday, July 01, 2016 assessed, which can be found either through theoretic consideration of the system or by using measurement data. When this relation is established, Eq. (12) provides a full prediction of the expected value of the peak force given the velocity, the bond length and the thermal off-rate. Even though Eq. (12) might turn out to be an implicit function, the solution for the peak force can readily be found. This means that for every combination of the parameter values there will be one theoretical and one measured value of the peak force. Standard algorithms may then by utilized to find which parameter values that minimize the mean square error of these forces for all velocities. Protocol for NLDFS A general description of how to obtain the bond length and the thermal off-rate using measurements was shown above. We will here give a more explicit protocol how this could be done in practice. The procedure is based on Eq. (12). The loading rate needs to be formulated as a function of the velocity whereafter a fitting algorithm can be applied. For this we suggest the following approach: 1) Measure for different velocities v. It is possible to use only two different velocities but highly recommended that at least four different velocities are used to obtain better accuracy. 2) Find a relation between and v. Note that r can be expressed as function of v in an experiment, i.e., a. Relate the loading rate r to the velocity using Eq. 16. b. Use Eq. (12) to define a, possible implicit, relation between and . Using a and b, there is a relation between and that depends only on and . 3) The parameter values and can now be assessed using a standard fitting procedure with the coupled values of and . Numeric example using a WLC As a well-controlled example we numerically simulated a force spectroscopy experiment of a receptor expressed on a tip of a polymer with WLC properties that was bound to an immobilized ligand. This simulation thus mimicked an experiment using AFM or OT instrumentation. The parameter values of the WLC model were set as; bond length, xb, 0.70 nm, thermal off-rate, , 1.00·10-4 Hz, persistence length, lp, 3.00 nm, contour length, Lc, 10.0 µm, and thermal energy, kbT, 4.11 pNnm. Since the elastic stiffness of the force probe, i.e., the AFM cantilever or the bead in the optical trap, is significantly higher than the elastic properties of the modeled WLC polymer, we modeled these as infinitely stiff. We thereafter analyzed all data closely following the approach described above: 1) The peak force was identified for four different velocities . 2) We calculated t, to be used in Eq. (15), for every with the corresponding velocity v by using Eqs (13) and (14). This relation between time and the force can also be readily measured during the experiments. With the time corresponding to the peak force and the loading rate function given by Eq. (15) we had everything we needed to use Eq. (12) as a relation between and v. This means that was expressed as a function of the velocity. 9 FFvFvvFbxthoffkbxthoffkvFthoffkFvFFFFriday, July 01, 2016 3) The acquired pair values for and v were now used. A standard algorithm that finds the parameter values of the bond length and the thermal off-rate that minimizes the mean square error of the theoretical and measured values of was utilized. For each of the four different extension velocities: 10, 100, 1 000, and 10 000 µm/s; 50 MC force spectroscopy simulations were performed. The rupture forces were saved and four continuous rupture probability density distributions, using a Gaussian kernel density estimator ( ), were generated. The peak force was identified for each of the four distribution. Figure 2A shows the rupture force spectrum for the highest velocity. To estimate the bond length and thermal off-rate we thereafter numerically fitted Eq. (12) to the data using a Nelder-Mead simplex algorithm to find the parameter values that minimized the mean square error of the peak forces. The data from the simulation are shown with the fitted values in Figure 2B and Table 1. Figure 2. Panel A, a spectrum of rupture forces obtained for v = 10 000 µms-1 from 50 measurements. The peak force 88.47 pN. Panel B the best estimates of the bond length and thermal off-rate obtained from a fitting algorithm based on data from four velocities and the corresponding peak forces. Table 1. The numerical results for the simulated example with the resulting values for the thermal off-rate and the bond length. N µm/s µm/s µm/s µm/s 50 39.57 pN 55.33 pN 74.58 pN 88.47 pN 0.685 nm 1.54·10-4Hz The assessed parameter values for this simulation are close estimates of the true values, where the bond length is underestimated with only ~2.1 %, whereas the thermal off-rate is overestimated by ~54 %. The discrepancy is expected due to the stochastic nature of the receptor-bond complex and should therefore depend on the sample size. Since we conducted the analysis based on only 50 simulations, we expect this error to decrease significantly with increased sample size. To quantify the error in the parameters we conducted a statistical analysis of the data (described below). 10 F*F3pN10v100v1 000v10 000vbxthoffkFriday, July 01, 2016 Error dependencies of the sample size In the example above, measurement sets consisting of 50 rupture forces were used to identify the peak forces for each velocity. Due to the stochastic nature of the experiment, the accuracy is expected to be improved with larger data sets, i.e., the more rupture forces that are sampled the less error there will be in the parameter values. To acquire acceptable accuracy of the parameter values it is in general recommended in the literature to conduct at least ~50-100 rupture measurements (Evans 1999; Merkel et al. 1999; Björnham and Schedin 2009). Therefore, we performed simulations with 50, 70, 100 and 300 rupture force, which allowed us to quantify the expected error in the parameters as a function of the sample sizes. In addition, a control set with ten million rupture forces for each velocity were performed. The resulting mean errors from these simulations for the bond length and the thermal off-rate are presented in Figure 3 and Table 2. Since experiments normally are conducted with 50-300 samples the mean relative error of the bond length is found to be less than 4 % while the mean relative error in the thermal off-rate is ~50%. This difference in errors can be explained by the fact that the rupture force is significantly more sensitive to the bond length in comparison to the thermal off-rate, therefore this difference is less remarkable. Figure 3. Mean relative error of the parameters as function of number of measurements. The error bars show the quartiles of the stochastic distribution of retrieved parameter values. Table 2. Statistical measures of the relative errors in the resulting parameter values in comparison to the analytic values. The data was obtained by calculating the most probable rupture forces from N measurements at four different velocities and the bond length and thermal off-rate were calculated using the method described in the theory section. This procedure were then repeated 10 000 times to quantify the expected errors in the parameter values. Method Samples Iterations Mean relative error for the peak force [µm/s] Mean relative N 50 70 100 300 107 10 000 10 000 10 000 10 000 1 5.42% 4.79% 4.17% 3.06% 0.1% 3.13% 2.77% 2.38% 1.73% 0.1% 2.26% 1.99% 1.76% 1.28% 0.3% 1.81% 1.59% 1.41% 1.01% 0.1% error 3.88% 3.37% 2.81% 2.15% 0.32% 52.7% 45.5% 37.3% 26.7% 1.89% 11 Monte Carlo Monte Carlo Monte Carlo Monte Carlo Monte Carlo 110v210v310v410vbxthoffkFriday, July 01, 2016 Analytical, most probable rupture forces [pN] 41.55 58.61 74.49 89.81 - - Conclusion We have presented an extension of the standard DFS theory that can accommodate also nonlinear forces denoted NLDFS. The NLDFS theory enables investigation of a wide range of biomechanical systems that show nonlinear force responses without compromising with the well-established and frequently used linear DFS. Examples of receptor-ligand systems that can be analyzed using this theory are adhesins expressed on bacterial adhesion pili. The data analysis using NLDFS requires a slightly more advanced fitting procedure than the conventional DFS theory to acquire the parameter values of the bond length and the thermal off-rate. The reason for this is that the loading rate becomes dynamic given by the introduction of a new term, in Eq. (12). This extra term, however, disappears for constant loading rates which shows that the NLDFS theory reduces into the regular linear DFS for the case with a constant loading rate. In DFS, a set of different loading rates with the corresponding values of the peak forces are used. Finally, just as assumed in DFS experiment, we neglect the dynamic effects of the viscous drag force on the probe since the pulling velocities are slow. To conduct the equivalent procedure in NLDFS the protocol has to be modified. Instead of keeping the loading rate constant during experiments, the pulling velocity is kept constant. If the force increases linearly with distance, the loading rate in Eq. (16) is constant and the NLDFS analysis falls into the linear DFS-regime. This implies that the velocity can be used as the entity kept constant in measurements using both DFS and NLDFS theory. Evans et al. refined the concept of DFS by introducing soft polymers linking the receptor-ligand bond (Evans and Ritchie 1999). By defining a compliance function, they compared how the peak force changed with and without a soft linker. Their approach allows for analysis of bond strengths in the presence of nonlinear external forces by defining the polymer force response using a relation between the probe stiffness and the characteristic stiffness of the polymers; and utilizing an apparent loading rate, which equals a constant probe stiffness multiplied with the pulling velocity. A theoretical correlation can thereafter be established, which is used in a curve fitting procedure. The main concepts of that method and NLDFS presented in this work are similar. However, the NLDFS theory utilizes a more direct approach and introduces only a minimal modification of the linear DFS. Explicit information of the stiffness of the probe and the soft linker in the system can be readily bypassed by direct investigation of the force vs. distance curve, which also implies that nonlinear responses in the probe is treated equally as nonlinear responses in the bio-complex linker. In other words, in the approach presented here, only the force experienced by the bond under investigation is considered, disregarding the origin of the force response since it has no impact in the analysis. Another approach to deal with nonlinear loading rates is proposed in reference (Friedsam et al. 2003). Instead of using the peak force values they used the probability density function for bond rupture. This expression is, however, rather complicated and depends on; the force, the loading rate, the bond 12 lndrdtFriday, July 01, 2016 length and the thermal off-rate, where the force and the loading rate distributions are found by investigating the experimental data. Estimates of the bond length and thermal off-rate may thereby be found by fitting the function to the rupture probability data. This method uses all data points, and not only the ones close to the peak force, which is of advantage since it make use of a larger data set. On the other hand, the method is sensitive to outliers and measurement artifacts. Besides the theoretical framework presented here, a protocol for how to conduct a practical measurement and data evaluation is described by a numerical example using Monte Carlo simulations. Since the rupture forces are stochastic, the parameter values will inherently have uncertainties coupled to them. We quantified the expected uncertainties by a large number of iterative simulations that provided the magnitude of errors that one would expect in a real experiment. It was found that already at 300 experimental data points the mean relative error for the bond length is only ~2%. However, in a real experiment using AFM or OT, additional measurement error and noise will be added on top of the inherent stochastic nature of the bond under investigation. Thus, the values of the expected errors presented here are therefore to be interpreted as a best case outcome. Acknowledgements This work was supported by the Swedish Research Council (2013-5379) and from the Kempe foundation to M.A. 13 Friday, July 01, 2016 References Andersson M, Uhlin BE, Fällman E (2007) The biomechanical properties of E. coli pili for urinary tract attachment reflect the host environment. Biophys J 93:3008–14. doi: 10.1529/biophysj.107.110643 Bell G (1978) Models for the Specific adhesion of cells to cells. Science (80- ) 200:618–627. doi: 10.1126/science.347575 Bianco P, Nagy A, Kengyel A, et al (2007) Interaction forces between F-actin and titin PEVK domain measured with optical tweezers. Biophys J 93:2102–2109. doi: 10.1529/biophysj.107.106153 Björnham O, Axner O, Andersson M (2008) Modeling of the elongation and retraction of Escherichia coli P pili under strain by Monte Carlo simulations. Eur Biophys J 37:381–91. doi: 10.1007/s00249- 007-0223-6 Björnham O, Nilsson H, Andersson M, Schedin S (2009) Physical properties of the specific PapG- galabiose binding in E. coli P pili-mediated adhesion. Eur Biophys J 38:245–54. doi: 10.1007/s00249- 008-0376-y Björnham O, Schedin S (2009) Methods and estimations of uncertainties in single-molecule dynamic force spectroscopy. Eur Biophys J 38:911–922. doi: 10.1007/s00249-009-0471-8 Castelain M, Ehlers S, Klinth JE, et al (2011) Fast uncoiling kinetics of F1C pili expressed by uropathogenic Escherichia coli are revealed on a single pilus level using force-measuring optical tweezers. Eur Biophys J 40:305–316. doi: 10.1007/s00249-010-0648-1 Castelain M, Koutris E, Andersson M, et al (2009) Characterization of the biomechanical properties of T4 pili expressed by Streptococcus pneumoniae--a comparison between helix-like and open coil-like pili. ChemPhysChem 10:1533–1540. doi: 10.1002/cphc.200900195 Chen F-J, Chan C-H, Huang Y-J, et al (2011) Structural and Mechanical Properties of Klebsiella pneumoniae Type 3 Fimbriae. J Bacteriol 193:1718–1725. doi: 10.1128/JB.01395-10 Evans E (2001) Probing the relation between Force - Lifetime - and Chemistry in single molecular bonds. Annu Rev Biophys Biomol Struct 30:105–128. Evans E (1999) Looking inside molecular bonds at biological interfaces with dynamic force spectroscopy. Biophys Chem 82:83–97. doi: 10.1016/S0301-4622(99)00108-8 Evans E, Ritchie K (1997) Dynamic strength of molecular adhesion bonds. Biophys J 72:1541–1555. doi: 10.1016/S0006-3495(97)78802-7 Evans E, Ritchie K (1999) Strength of a weak bond connecting flexible polymer chains. Biophys J 76:2439–2447. doi: 10.1016/S0006-3495(99)77399-6 Forero M, Yakovenko O, Sokurenko E V, et al (2006) Uncoiling mechanics of Escherichia coli type I fimbriae are optimized for catch bonds. PLoS Biol 4:1509–1516. doi: 10.1371/journal.pbio.0040298 Friedsam C, Wehle AK, K hner F, Gaub HE (2003) Dynamic single-molecule force spectroscopy: bond rupture analysis with variable spacer length. J Phys Condens Matter 15:S1709–S1723. doi: 10.1088/0953-8984/15/18/305 Kiss B, Karsai Á, Kellermayer MSZ (2006) Nanomechanical properties of desmin intermediate filaments. J Struct Biol 155:327–339. doi: 10.1016/j.jsb.2006.03.020 Kramers H a (1940) Brownian motion in a field of force and the diffusion model of chemical reactions. Physica 7:284–304. doi: 10.1016/S0031-8914(40)90098-2 Lee GU, Kidwell DA, Colton RJ (1994) Sensing Discrete Streptavidin Biotin Interactions With Atomic-Force Microscopy. Langmuir 10:354–357. doi: 10.1021/la00014a003 14 Friday, July 01, 2016 Merkel R, Nassoy P, Leung A, et al (1999) Energy landscapes of receptor-ligand bonds explored with dynamic force spectroscopy. Nature 397:50–53. doi: 10.1038/16219 Miller E, Garcia T, Hultgren SJ, Oberhauser AF (2006) The mechanical properties of E. coli type 1 pili measured by atomic force microscopy techniques. Biophys J 91:3848–56. doi: 10.1529/biophysj.106.088989 Mortezaei N, Epler CR, P. SP, et al (2015a) Structure and function of Enterotoxigenic Escherichia coli fimbriae from differing assembly pathways. Mol Microbiol 95:116–126. doi: 10.1111/mmi.12847 Mortezaei N, Singh B, Zakrisson J, et al (2015b) Biomechanical and Structural features of CS2 fimbriae of Enterotoxigenic Escherichia coli. Biophys J 109:49–56. doi: 10.1016/j.bpj.2015.05.022 Neuert G, Albrecht C, Pamir E, Gaub HE (2006) Dynamic force spectroscopy of the digoxigenin- antibody complex. FEBS Lett 580:505–9. doi: 10.1016/j.febslet.2005.12.052 Strick TR, Dessinges M-N, Charvin G, et al (2002) Stretching of macromolecules and proteins. Reports Prog Phys 66:1–45. doi: 10.1088/0034-4885/66/1/201 Strunz T, Oroszlan K, Schäfer R, Güntherodt HJ (1999) Dynamic force spectroscopy of single DNA molecules. Proc Natl Acad Sci U S A 96:11277–11282. doi: 10.1073/pnas.96.20.11277 Sulchek T a, Friddle RW, Langry K, et al (2005) Dynamic force spectroscopy of parallel individual Mucin1-antibody bonds. Proc Natl Acad Sci U S A 102:16638–16643. doi: 10.1073/pnas.0505208102 Yuan C, Chen a., Kolb P, Moy VT (2000) Energy landscape of streptavidin-biotin complexes measured by atomic force microscopy. Biochemistry 39:10219–10223. doi: 10.1021/bi992715o 15
1006.1503	1	1006	2010-06-08T09:42:14	The role of the cytoskeleton in volume regulation and beading transitions in PC12 neurites	[ "physics.bio-ph", "q-bio.SC" ]	We present investigations on volume regulation and beading shape transitions in PC12 neurites conducted using a flow-chamber technique. By disrupting the cell cytoskeleton with specific drugs we investigate the role of its individual components in the volume regulation response. We find that microtubule disruption increases both swelling rate and maximum volume attained, but does not affect the ability of the neurite to recover its initial volume. In addition, investigation of axonal beading --also known as pearling instability-- provides additional clues on the mechanical state of the neurite. We conclude that the initial swelling phase is mechanically slowed down by microtubules, while the volume recovery is driven by passive diffusion of osmolites. Our experiments provide a framework to investigate the role of cytoskeletal mechanics in volume homeostasis.	physics.bio-ph	physics	The role of the cytoskeleton in volume regulation and beading transitions in PC12 neurites Pablo Fern´andez∗ and Pramod A. Pullarkat† ∗ E27 Lehrstuhl fur Zellbiophysik, Technische Universitat Munchen James Franck Strasse, D-85748 Garching, Germany [email protected] † Raman Research Institute C. V. Raman Avenue, Sadashivanagar, 560080 Bangalore, India [email protected] September 20, 2018 Abstract We present investigations on volume regulation and beading shape transitions in PC12 neurites conducted using a flow-chamber technique. By disrupting the cell cytoskeleton with specific drugs we investigate the role of its individual components in the volume regulation response. We find that microtubule disruption increases both swelling rate and maximum volume attained, but does not affect the ability of the neurite to recover its initial volume. investigation of axonal beading -- also known as pearling instability -- provides additional clues on the mechanical state of the neurite. We conclude that the initial swelling phase is mechan- ically slowed down by microtubules, while the volume recovery is driven by passive diffusion of osmolites. Our experiments provide a framework to investigate the role of cytoskeletal mechanics in volume homeostasis. In addition, Keywords: Volume regulation, cell mechanics, axon beading, cytoskeleton. PACS: 87.16.ln, 87.16.dp, 87.16.ad. INTRODUCTION The ability of living cells to regulate their volume is a ubiquitous homeostatic feature in biology (1, 2). Since water readily permeates through the cellular membrane, alterations in extracellular osmolarity can change the con- centration of all cytoskeletal components with severe con- sequences for the metabolism. Not surprisingly, one finds several mechanisms involved in volume regulation. In particular, many eukaryotic cell types display a short-term volume regulation response to sudden alterations in exter- nal osmolarity, the so-called regulatory volume decrease (RVD) and regulatory volume increase (RVI) (2, 3). It is generally accepted that these require the passive dif- fusion of ions. In the case of RVD, cell swelling upon hypoosmotic shock increases the membrane permeability for sodium, which diffuses out of the cell. In turn, the decrease of intracellular osmolarity drives water out and lowers cell volume (2, 3). In this mechanism water flow is driven by a difference in osmotic pressure. It is of- ten assumed that hydrostatic pressures are negligible (4), with the argument that they would make the membrane burst (1). However, though the maximal pressures sus- tained by lipid bilayers are indeed very low, the situation cannot be simply extrapolated to the living cell. The cell membrane is connected to the cytoskeleton, the biopoly- mer gel spanning across the cell interior (5 -- 7). Since the cytoskeleton is viscoelastic and contractile (8 -- 11) (also in PC12 neurites (12)), it may provide a mechanical memory of the initial state as well as a driving force for volume relaxation (13 -- 20). Though it has been argued that the cytoskeleton is too weak (typical moduli are up to 10 kPa (8, 10)) to sustain osmotic pressures (up to ∼MPa) (7), neurons under hypoosmotic shock sustain strong pres- sures for several hours (21) and swelling of erythrocytes increases and approach perfect osmometer behaviour af- Thecytoskeletoninvolumeandshaperegulation 2 ter disruption of the spectrin-actin cytoskeleton (22). This is indeed compelling evidence for a mechanical role, but in many systems the cytoskeleton seems to additionally play a signalling one. In particular, biochemical interac- tions between actin filaments and ion channels may cou- ple strain of the actin cortex to changes in channel activity (2, 23). The fact that the actin cortex is disrupted when hypoosmotic swelling begins (24, 25) seems to be due to an influx of Ca2+ (26) through mechanosensitive ion channels activated by membrane stretching (27, 28). To clarify the role of the cytoskeleton one must discern be- tween pure mechanical and mechanosensing responses, a difficult task requiring direct measurements of membrane tension. In this work, we study volume regulation in neurites, axon-like cylindrical protrusions extended by PC12 cells (29), structurally very similar to the axons produced by neurons in culture (6, 30). Neurites furnish a hitherto un- explored, yet attractive model system to investigate the role of mechanical tension in volume regulation. Their simple cylindrical geometry and low amount of invagi- nated membrane allows better volume and area calcula- tions from images. They also have a well-defined, highly organized cytoskeletal structure similar to that of axons: a central array of longitudinally arranged microtubules interconnected by microtubule binding proteins and sur- rounded by an actin cortex. Moreover, an exceptional feature of axons and neurites is their ability to undergo sudden shape transformations in response to an applied mechanical tension (20, 31, 32), an instability known as axonal beading to biologists and pearling instability to physicists. We show that axonal swelling increases mem- brane tension and that microtubules slow down the vol- ume response. We conclude that frictional forces in the cytoskeleton play an important role in axonal volume reg- ulation. EXPERIMENTAL SETUP AND METH- ODS Flow chamber The experiments have been carried out using a flow- chamber technique. A schematic of the set-up is shown in Fig. 1. A stainless-steel block and two coverslips are used to form a 10×5×1 mm3 chamber. Cells adhere on the bot- tom cover-slip. One duct of the chamber is connected to a peristaltic pump by means of long, soft silicone-rubber tubing that minimizes pressure fluctuations arising from the pump. The other duct is connected to a 3-way-valve To pump Pt-100 To water bath Normal medium Diluted medium 3-way valve Flow chamber Cells Aluminium block Microscope objective Coverslips Stainless-steel block Figure 1: A schematic diagram of the flow-chamber set-up. to select between two different media that can be pumped into the chamber. Stainless-steel tubes that are 1 mm in diameter connect the 3-way-valve to the two reservoirs where the media are stored. The chamber is intentionally made small to ensure a quick switching from one medium to the other at low flow rates, in the range of 2−4µl/s. The switching of concentrations inside the chamber was stud- ied by adding an absorbing dye to one of the solutions and monitoring the variation in the transmitted intensity with time. The concentration reaches 90% of its final value in about 10 s. The chamber, the 3-way-valve, and the stainless-steel tubes are placed inside an aluminium block with good thermal contact between each other and a wa- ter bath maintains the temperature of the block at the de- sired value. The continuous flow of pre-warmed medium keeps the chamber at constant temperature despite some heat loss through the cover-slips. Cell culture PC12 cells are from the German Collection of Microor- ganisms and Cell Lines (DSMZ) (33). They are plated on collagen coated slides and cultured in RPMI-1640 medium (Gibco) with 10% fetal bovine serum and 5% horse serum in presence of nerve growth factor (NGF) (Sigma-Aldrich Chemie, Munich, Germany) for 4 -- 5 days (34). Such young neurites are known not to have in- termediate filaments (30). Slides are coated with 3- aminopropyl triethoxysilane (Sigma). The silanised slides are covered with about 100 µl of 10% rat-tail collagen (Sigma-Aldrich) dissolved in a 70% ethanol -- 30% water solution and let dry overnight. Experimental procedure Prior to an experiment, the slide with the adherent cells is transferred to the flow-chamber. Cells are allowed to stabilize for about 5 min by circulating the experiment Thecytoskeletoninvolumeandshaperegulation 3 medium (normal medium with addition of 25 mM HEPES buffer (Invitrogen, Darmstadt, Germany)). Experiments are performed by switching from the normal experiment medium to experiment medium diluted with deionized water. The response of the cells to the sudden switch in the external concentration is observed with an Axiovert 135 microscope (Zeiss, Oberkochen, Germany) config- ured for phase-contrast imaging. Image analysis The volume and area of the neurites are analyzed from the recordings using a home-made edge detection program. Edge detection using a threshold for intensity is unreli- able due to the "halo effect" present in phase-contrast im- ages and also due to the dependence on the illumination intensity. To avoid such complications, edge pixels are recognized along the neurite by detecting the local max- ima in the gradient of intensity across the neurite. Af- ter edge detection the neurite volume and surface area are computed assuming axial symmetry for the neurite shape. Axial symmetry is verified by comparing the two detected boundaries of several neurites and is found to be a good approximation for straight neurites which are at- tached only at the two extremities. Only such neurites were selected for the experiments. Drug-induced cytoskeletal perturbation Experiments were performed in presence of cytoskeleton disrupting drugs in order to study the role of its individ- ual components. A complication in these experiments arises due to the neurites becoming fragile or losing their cylindrical geometry on cytoskeletal perturbation. This precluded experiments with the actin disrupting drug La- trunculin, which induces detachment of the growth cone. In the case of the microtubule disrupting drug Nocoda- zole (NOC) (35), which induces shape irregularities after approximately 10 min exposure, we let the drug act for 5 min before performing the hypoosmotic shock. NOC concentration was 10 µg/ml throughout. In contrast, the myosin-blocking drug Blebbistatin (36) did not signifi- cantly alter neurite shape. Thus, to ensure its effect we in- cubated neurites at 37◦ for 1 hour at a high concentration (50 µM) before transferring them to the experiment cham- ber and performing the experiment at a lower concentra- tion of 20 µM. For all experiments, drugs were present both in the normal and in the diluted medium. Since all drugs are dissolved in dimethylsulfoxide (DMSO), which itself has effects on water and ion channels (37), we per- formed control experiments in presence of DMSO 0.5%, equal to the highest DMSO concentration in any of the drug experiments. EXPERIMENTAL RESULTS Volume dynamics We begin all experiments with a hypoosmotic shock imposed by switching the cell culture medium flowing through the chamber from normal medium with a total solute concentration C0 ≃ 300 mM to a lower external value Ce (the continuous flow ensures constant external concentration at all times). In the following, for an intu- itive measure of the shock strength we will normalize the external osmolarity by its initial value and denote it by C = Ce/C0. Figure 2 shows typical responses for three different values of C at 36◦C. For weak shocks (C = 0.8) the neurite volume increases from its initial volume V0 un- til it reaches a maximum steady value Vmax. No recovery is observed for tens of minutes. For intermediate shocks, C = 0.7, the volume increases at a roughly constant rate V0 initially until it reaches a maximum value Vmax. Sub- sequently, the volume recovers almost back to V0 with a typical regulatory volume decrease response (RVD). The volume recovery is roughly exponential with a character- istic time τ . The recovery time τ is strongly temperature- dependent (shown later), but does not follow a clear trend with neurite diameter. For strong shocks, C = 0.5, the recovery is faster and there is often a remarkable "un- dershoot", where the volume goes below its initial value. Moreover, during the swelling phase the neurite under- goes a sudden and transient shape transformation from its normal cylindrical geometry to a periodically modulated one. As discussed later, this Rayleigh-like "pearling" in- stability or beading indicates an increase in membrane tension upon neurite swelling. This shape transformation, which will be discussed later, does not affect the volume response curve in any measurable way. This fact allow us to discuss the volume dynamics and shape dynamics in that order. Once the volume stabilizes to the lower external osmo- larity Ce (within about ten minutes), we perform a hy- perosmotic shock by switching back the original medium with concentration C0. The neurite shrinks, reaches a minimum volume Vmin and then comes back to its ini- tial volume in an RVI response as shown in Fig. 2. No beading shape transformation is observed in this case. In the following, we separately address the volume reg- ulation response and the peristaltic modulation. We study them as a function of osmotic shock strength, tempera- Thecytoskeletoninvolumeandshaperegulation 4 ture, and in the presence of cytoskeleton-disrupting drugs. Nonlinear swelling response. Figure 3 A shows the initial rate of change of volume divided by the initial area, V0/A0, as a function of the initial osmotic pressure dif- ference ∆Π0 = RT C0(1 − C), for temperatures 33 -- 36◦C. Assuming zero hydrostatic pressure and neglecting changes in internal osmolarity gives the following expres- sion for the initial swelling rate: V0 = A0 Lp ∆Π0 , (1) where Lp is the hydraulic permeability. Surprisingly, from the average V0 value at C = 0.7 we obtain an os- motic permeability Pf = RT δW Lp ≃ 1.4 µm/s (where δW is the molar density of water), which is about two orders of magnitude lower than the literature values for lipidic membranes and most biological cells (1, 38 -- 40) even after blockage of water channels (37). Moreover, as can be seen from Fig. 3 A, the expected linear dependence is contradicted by the strong nonlinear response observed for both hypo- and hyperosmotic shocks. We now turn to the maximum (minimum) volume at- tained in a hypoosmotic (hyperosmotic) shock. The max- imum (minimum) volume Vmax(Vmin) is to a good ap- proximation proportional to the initial volume V0 and does not depend significantly on the temperature (data not shown (41)). Thus we look at the relative maximum volume Vmax/V0 in order to minimize the effect of neu- rite diameter. As shown in Fig. 3 B, the maximum vol- ume increases nonlinearly with the initial osmotic pres- sure difference ∆Π0. The data is contrasted to the per- fect osmometer equation, corresponding to a constant to- tal amount of internal osmolites and zero hydrostatic pres- sure: V0 − Vdead Vmax − Vdead = C , (2) where the dead volume Vdead represents non-aqueous in- ternal volume. Mammalian cells have on the average a cytosolic protein concentration of ∼ 20% (5). Accord- ing to electron microscopy studies (30) the non-cytosolic volume of PC12 neurite is comprised mostly of micro- tubules and organelles and amounts to Vdead ≃ 25% V0. The black region in Fig. 3 B corresponds to V > V0/C for hypoosmotic shocks and to V < V0/C for hyperos- motic shocks; penetrating this region would require work against the osmotic gradient. As expected for a passive response, the data lies outside them. For mild shocks, C ≥ 0.7 and (hyperosmotic) C ≤ 1.4, neurites swell as much as perfect osmometers a dead volume of 25%. Therefore, any ion leakage or hydrostatic pressure are negligible during the swelling phase. However, at strong Vmax C=0.8 C=0.7 C=0.5 Vend Vend Vend 0 50 100 time (s) 150 200 Ce → C0 A 0 V / V 1.4 1.3 1.2 1.1 1 0.9 0.8 B 1.4 1.3 1.2 0 V / V 1.1 1 0.9 C0 → Ce 200 0 400 600 time (s) 800 1000 Figure 2: A: Swelling and recovery after a hypoosmotic shock performed at time t = 0. Normalised volume V /V0 as a func- tion of time for different dilutions: C = 0.5, (circles), C = 0.7 (solid line), C = 0.8 (dashed line). Temperature is in all cases 36◦C. Each curve corresponds to a different neurite. The max- imum volume attained is Vmax, and the minimum is Vend. No- tice the strong undershoot of the volume at C = 0.5, and the absence of recovery at C = 0.8. These are all general trends of the volume response. B: The full observed response when the neurite is subjected to a hypoosmotic shock and a subsequent hyperosmotic shock. At first the hypoosmotic shock C0 → Ce is applied and the neurite is allowed to swell and relax (RVD). After the relaxation phase the hyperosmotic shock Ce → C0 is applied and the neurite shrinks and recovers the initial vol- ume (RVI). Such C0 → Ce → C0 cycles can be repeated up to five times in a given neurite before the ends detach (data not shown). Thecytoskeletoninvolumeandshaperegulation 5 10 A ) s / m n ( 0 A / t d / V d 5 0 -5 B 1.5 1 0 V / M V -400 -200 ∆Π 0 0 (kPa) 200 400 C=0.5 C=0.7 C=0.8 0.5 1 1.5 2 1/C Figure 3: A: Initial swelling speed V0/A0 as a function of the initial osmotic pressure difference ∆Π0 for temperatures 33 − 36◦C. The shaded region is a guide to the eye, a spline going through the averages within a standard deviation. No- tice the nonlinear response to strong hypo- as well as hyperos- motic shocks. B: Maximum change in relative volume VM /V0 as a function of the inverse external concentration 1/C, for all temperatures (here VM stands for Vmax or Vmin as the case may be). The grey region is a guide to the eye. The dashed line corresponds to a perfect osmometer with dead vol- ume Vdead = 0.25V0, and the black regions correspond to neg- ative dead volumes. shocks (hypoosmotic C = 0.5 or hyperosmotic C = 2), neurites swell significantly less than a perfect osmome- ter would. Thus, upon strong osmotic perturbations either ions leak or a sustained hydrostatic pressure develops. Volume regulation under cytoskeleton disruption. The axonal cytoskeleton may be expected to contribute to the volume response in several ways. As discussed in the introduction, a mechanical as well as a signalling role is conceivable. In order to assess the role of individual components of the cytoskeleton, we treat neurites with the myosin blocking drug Blebbistatin (BLE) (36) and the microtubule disrupting drug Nocodazole (NOC) (35). Since these are all diluted in dimethylsulfoxide (DMSO), a compound known to alter ion channels, we also perform control experiments in presence of DMSO. Fig. 4 A shows typical responses. In presence of Noco- dazole the initial swelling rate for strong shocks C = 0.5, increases markedly, but it barely changes for C = 0.7 (see Fig. 4 B). With Blebbistatin we observe a weaker but still significant effect. For both drugs, the relationship be- tween swelling rate and initial osmotic pressure difference approaches the naively expected linear dependence given by Eq. 1. Nocodazole induced disruption of microtubules also has a strong effect on the maximum volume Vmax attained after a strong shock of C = 0.5, as shown in Fig. 4 C. Neither BLE-treatment nor DMSO alone have a signifi- cant effect on the maximum volume. For mild dilutions, C = 0.7, nocodazole has no effect, consistent with the fact that neurites swell like a perfect osmometer with 25% dead volume. Importantly, the cytoskeleton disrupting drugs do not affect the ability of the neurite to perform RVD. As can be seen in Fig. 4 A, the volume fully relaxes back to its initial value. Further evidence can be found in the Sup- plementary Material. The effect of temperature: Arrhenius behaviour. We now address the influence of temperature in the dy- namics of volume regulation. For simplicity, we describe the recovery phase by fitting single exponentials. As shown in Fig. 5, lowering the temperature from 35 to 15◦C slows down the volume dynamics by an order of magnitude. The dashed line is an Arrhenius-like equation RT , where k depends on temperature as the τ ∝ 1/k ∝ e rate constant of a thermally-activated process. This yields an activation energy ∆G ∼ 30 kT, a typical order of mag- nitude for biological processes (42). The well-defined Ar- rhenius trend is consistent with the idea that ion channels are responsible for RVD. ∆G Thecytoskeletoninvolumeandshaperegulation 6 A 0 V / V 2 1.8 1.6 1.4 1.2 1 0.8 B 0 50 100 time (s) 10 5 ) s / m n ( 0 A / t d / V d 0 0 C 1.8 1.6 1.4 0 V / M V 1.2 1 BN D 200 0 (kPa) ∆Π DBN DMSO BLE NOC 150 N B D 400 N B D 1 1.2 1.4 1.6 1.8 2 1/C Figure 4: A: Comparison of different drug treatments. Each curve is a different neurite. All experiments were performed at temperature T = 33◦C and dilution C = 0.5. Grey shaded line: DMSO control. Dashed line: BLE. Black solid line: NOC. B: Swelling speed V0/A0 as a function of the initial os- motic pressure difference ∆Π0, for temperatures 33−36◦C. D: DMSO control (n = 5 for C = 0.5, n = 14 for C = 0.7). B: blebbistatin (n = 7, n = 13). N: nocodazol (n = 8, n = 14). The grey region is a guide to the eye, corresponding to the ex- periments without drugs shown in Fig. 3 A. Data for all drugs is shown as arithmetic mean ±S. E. C: Maximum relative volume VM /V0 as a function of the osmotic pressure difference ∆Π in presence of cytoskeleton-disrupting drugs. The grey region corresponds to the data without drugs in Fig. 3 B. ) s ( e m i t y r e v o c e R 512 256 128 64 32 16 3.25 3.3 3.35 1 / T (10-3 / K) 3.4 3.45 Figure 5: Recovery time as a function of inverse temperature 1/T. The dashed line is a least-squares fit to an Arrhenius form τ ∝ e∆G / RT . Beading When neurites are subjected to strong hypoosmotic shocks they undergo a shape transformation by develop- ing a periodic array of swellings akin to beading of axons (43). We observe this in chick dorsal root ganglia (DRG) neurons as well as PC12 neurites Fig. 6. This peristaltic deformation, resembles that formed in nerves subjected to induced stretch injuries (31, 32, 44) and that observed after traumatic injuries to the brain (43). The dynamics of bead formation and the mechanism has been investi- gated recently using the osmotic shock technique (20) and shown to be similar to the pearling instability observed in synthetic membrane tubes under tension (45, 46). Here, we provide direct evidence correlating neurite tension and bead formation and describe the role of cytoskeletal com- ponents in this process. We begin by listing the main features of osmotic shock induced beading. For a given radius of the neurite, there is a critical hypoosmotic shock below which the shape remains cylindrical during the entire volume evolution, and above which a transient peristaltic modulation is ob- served. This is about C = 0.5 at 37◦C and about C = 0.7 at 25◦C for an initial neurite radius of 0.7 µm. Beading and recovery cycles (for mild shocks) can be repeated up to five times in the same neurite, after which neurites tend to detach from the substrate. Also transport of organelles can be observed during and after beading and recovery. This suggests that mild osmotic shocks leading to bead- ing and recovery causes no permanent damage to the neu- rites. On the contrary, strong shocks (C ≥ 0.5) often lead to strongly modulated beaded shapes with no recovery for several tens of minutes. Extreme shocks cause neurites to burst with organelles spewing out, corroborating a build up of membrane tension. Finally, no beading is observed Thecytoskeletoninvolumeandshaperegulation 0 A / A , 0 V / V 1.6 1.4 1.2 1 0.8 0 V / V0 A / A0 √V / A 1.1 1.05 P 1 50 100 150 time (s) 200 250 Figure 6: Top: Swelling-induced pearling instability. a: Chick-embryo neurons. b: PC12 neurites at a higher magni- fication. Bottom: Pearling instability. Relative volume V /V0, area A/A0 and shape parameter pV /V0/(A/A0) as a func- tion of time. Imposing a fast hypoosmotic shock at dilution C = 0.5 and temperature 33◦C increases membrane tension and leads to a shape instability. The instability vanishes before the volume recovers. flow 7 0 s 300 s 380 s 610 s 1300 s 1800 s Figure 7: Neurite deformations induced by the combined ef- fect of a constant laminar flow perpendicular to the neurite and different osmotic conditions. The temperature is 25◦C and Ce = 0.7C0. The sequence of events is as follows: laminar flow is started at t = 20 s and the neurite allowed to undergo viscoelastic relaxation and attain a steady state, a hypoosmotic shock is applied at t = 320 s by introducing dilute medium, the neurite is allowed to undergo volume and shape relaxation, normal medium is reintroduced after the volume has relaxed (hyperosmotic shock) at t = 1200 s, the neurite is again al- lowed to undergo volume relaxation. The flow rate is constant throughout the experiment. The dashed lines are for compari- son of curvature. The larger the average curvature of the cate- nary, the smaller the tension in the neurite (47). during hyperosmotic shocks, irrespective of the magni- tude. Axonal tension causes beading. The flow chamber technique has been used recently to quantitatively study the evolution of neurite tension in response to drag forces and to demonstrate active contractile responses in neurites (47). This is performed by imposing a constant, laminar flow perpendicular to the neurite generating a drag force. It has been shown that PC12 neurites as well as axons are viscoelastic (12) and respond to a stretching force via a relaxation process with a characteristic timescale. Un- der the influence of a flow induced drag force the neu- rite takes the form of a catenary and elongates until it reaches a final equilibrium shape (47) (see Fig. 7 t=300s). The resulting strain gives a measure of the tension in the axon. Fig. 7 shows snapshots of the neurite taken at var- ious stages during an osmotic shock experiment. Under the influence of the flow the neurite attains an equilibrium catenary shape within about a minute. When the flow is switched from the normal medium to a diluted one, keep- ing the flow rate the same, the neurite shortens until it becomes almost straight. The beading sets in during this straightening phase and the maximum beading amplitude occurs when the neurite is straight (t=380s in Fig. 7). Sub- sequently there is a relaxation process in which the bead- ing amplitude begins to decay and the flow-induced cur- vature increases. When a hyperosmotic shock is applied after the first volume relaxation process (t=1200s in Fig. 7) the neurite curvature begins to increase as the neurite shrinks and subsequently decreases to almost its normal value with a timescale very much comparable to the vol- ume recovery time. These observations clearly demon- strates that the neurite tension and the volume changes induced by osmotic treatments are correlated. Beading mechanism. The physical mechanism for bead formation is as follows (20). After applying an os- motic shock the neurite volume increases as discussed above. This results in a corresponding expansion of the measured area A(t) as shown in Fig. 6. This stretch- ing of the membrane causes an increase in the tension of the outer membrane and hence an increase in the sur- face energy, Fs = σ(A(t) − A0)2/A0. The increase in Thecytoskeletoninvolumeandshaperegulation 8 volume also costs bulk elastic energy due to the deforma- tions caused in the cytoskeletal network, which we will consider below later. It can be shown (46) that at any instant t, for a volume V (t), a peristaltically modulated shape with r(t, z) = r0 + ǫ(t) sin(qz) has a lower average surface area compared to a cylinder with the same instantaneous volume V (t), provided the wavelength λ > 2πr0. For small amplitudes, the relative area gain can be obtained as simple minded argument for this is as follows. Any per- turbation with a wavelength λ generates as pressure dif- ference given by the Laplace law, the pressure difference p balanced by the membrane tension σ. In the case of the axisymmetric modes there are two principal curvatures: that with a curvature radius equal to the neurite radius r(z), and that corresponding to the peristaltic deformation along the z-axis, with a curvature radius 1/∂2 z r(z). For large wavelengths, λ ≫ ǫ, the latter can be ignored. The pressure difference in between the crest and the trough of a peristaltic mode can be written in the small amplitude limit (ǫ ≪ r) as δS/S = ǫ(q2 − 1)/(4r2) , ∆p = σ(1/rcrest − 1/rtrough) ≃ −σǫ/r2 . where S is the surface area and q = 2πr/λ. This can be seen in Fig. 6. The ratio pV (t)/A(t) (note: the av- erage volume and area are computed for unit length) in- creases when the peristaltic mode grows. This quantity is a constant for all cylinders irrespective of radius. An in- crease in pV (t)/A(t) from this constant value indicates a decrease in area for the peristaltic shape as compared to a cylinder with identical volume. In other words, the neurite is able to reduce its interfacial energy by adopt- ing a peristaltic shape instead of a cylindrical one. As mentioned earlier any deformation costs bulk elastic en- ergy due to the cytoskeletal elasticity. Unlike the surface free energy, the bulk elastic free energy is always positive for peristaltic modes compared to a cylinder of same vol- ume (it costs bulk energy to expand as well as compress regions along the neurite). Therefore, there is a compe- tition between the surface energy and bulk energy which determines the preferred shape, giving rise to a critical tension. A rough expression for the critical tension can be obtained by comparing the average bulk energy per unit c , where E is the elastic modulus length of a cylinder Er2 of the cytoskeleton, and the corresponding surface energy σrc. This gives a critical tension σc = Erc, above which the surface contribution dominates. As seen in experi- ments, the critical tension needed to destabilise the cylin- der increases with neurite radius. Analogous shape trans- formations observed in membrane tubes, triggered by ap- plication of laser tweezers (45, 46), and in cell protrusions after treatment with latrunculins (48) have been analysed in a similar fashion. The above mentioned expression for relative gain in area shows that reduction in area increases with wave- length of the perturbation, giving a maximum area gain for λ = L/2, L the length of the neurite. Clearly, the observed value of λ is much shorter than and independent of the neurite length, but increases linearly with radius. A As indicated by the negative sign, this is an unstable flow: the pressure difference will drive water from the troughs into the crests, thereby increasing the perturbation ampli- tude ǫ. The flow rate and hence the growth of a given mode is in general proportional to the driving force ∆p/λ ∼ σǫ / λr2 . Shorter wavelength modes are thus faster to grow than longer wavelength ones. However, for very short wave- lengths the curvature along the neurite axis becomes im- portant. The respective pressure difference is given by In contrast to the previous ∆p = σ∂2 case, here the cylindrical state is stable: water flows from the crests into the troughs (which have a "negative" pres- sure) making the perturbation vanish. Moreover, the de- pendence of the flow rate on the wavelength is more pro- nounced, z r(z) ≃ σǫ/λ2. ∆p/λ ∼ σǫ/λ3 . Therefore for very small wavelengths the stabilizing flow driven by the curvature along z always dominates, and there is no instability anymore. We conclude that there is a fastest mode at an intermediate, non-zero wavelength. This can be shown to be λ ≃ 9.2 r (49). In the very dawn of linear stability analysis, this argument was applied to the case of water jets by Lord Rayleigh (50, 51). THEORY We model volume regulation in neurites taking into ac- count both mechanical and osmotic driving forces. The neurite is characterised by its volume, V (t), and the in- ternal amounts of ionic species, ni(t). For simplicity we consider only potassium and chloride, by large the most Thecytoskeletoninvolumeandshaperegulation 9 critical threshold volume; otherwise its form is not known and different expressions have been used in the literature (3, 17). In our experiments the critical volume is about 1.2 V0, as can be seen in Fig. 2. We obtain good results with the simple expression g = θ(V/1.2) V , where θ() is the step function. Finally, assuming external ion concentrations to jump i (0) the instantaneously at time t = 0 from cex full system can be written as i (0) to Ccex V = 1 V − ∆p τV (cid:18)1 + 2φ(N − 1) Π0 − C(cid:19) log(cid:18) α N 2 + (1 − α) N βC 2V 2 (7) (cid:19) (8) N = − g(V) τCl with the parameters φ = cin Cl / C0 α = cin Cl / cin K (9) (10) (11) important osmolites inside the cell. Since due to elec- troneutrality the flows of Cl− and K+ are coupled, nCl = nK , one of the two concentrations can be eliminated. The two variables of our model are the adimensional quantities V = V (t) / V0 N = nCl(t) / nCl(0) . (3) (4) The flow of water through the membrane (1) is given by V = ALp(∆Π − ∆p) , (5) where A is the neurite area, Lp the hydraulic permeabil- ity, ∆p the hydrostatic pressure, and the osmotic pressure difference is given by ∆Π ≃ RT Xi (ni/V − cex i ) in terms of external cex i and internal osmolite concentra- tions ni/V . The typical relaxation time for the volume is τV = V0 ALpΠ0 , β = cex Cl cex K / cin Cl cin K . where Π0 = RT C0 ≃ 700 kPa the osmotic pressure of normal medium. Taking 10−14 m Pa−1 s−1 for the per- meability (from Fig. 3 A) we obtain τV = 50 s. We model ion movement with a passive K+-Cl− co- transport following Ref. (3), neglecting the effect of ion pumps since these are not relevant for the short-term RVD response. Chloride flux is given by the difference in chemical potentials, nCl = −ART G log(cid:18) nClnK V 2 1 C 2cClcK(cid:19) . (6) We model the RVD response following the standard as- sumption of a volume-dependent permeability (2). The permeability G must be zero at the initial volume V0 and non-zero at a significant departure. We decompose G(V) = G g(V) , into a typical order of magnitude G and an adimensional function g(V). For the chloride relaxation time τCl = V0 cin Cl ART G , we expect a value of 10 -- 100 s taking typical values for ion conductivities (1). The volume-dependent permeabil- ity function g(V) jumps from zero to a finite value at a Taking the following physiological values for the ion con- centrations (1) cin Cl = 80 mM cin K = 150 mM cin Na = 15 mM cex Cl = 80 mM cex K = 6 mM cex Na = 110 mM , the adimensional constants become φ = 0.25, β ≃ 0.05, and α ≃ 0.5. Zero pressure model. The standard assumption in modelling volume regulation is to set the hydrostatic pres- sure to zero (3), ∆p = 0 . With this ansatz we can solve the equations for the volume response V(t). As shown in Fig. 8, the model reproduces the timescales for swelling and recovery as well as the maximum volume. Interestingly it reproduces the "un- dershoot" at C = 0.5 (see Fig. 2). However, the depen- dence of the swelling rate on osmotic pressure is clearly off. The model gives a linear response -- a feature intrinsic to any volume-dependent permeability -- whereas experi- ments show a nonlinear one (Fig. 3 A). Thecytoskeletoninvolumeandshaperegulation 10 VV0 A 1.3 1.2 1.1 B VV0 1.30 1.25 1.20 1.15 1.10 1.05 50 100 150 200 250 time s 50 100 150 200 250 time s Figure 8: A: zero pressure model for dilutions 80% (dotted line), 70% (dashed line) and 50% (solid line). B: adding non- linear friction as given by Eq. 12 reproduces the trend during the swelling phase. Compare to Figs. 2,6 Viscoelastic model. As shown by our experiments combining hypoosmotic shocks and drag forces (Fig. 6), neurites develop tension while swelling. This suggests that the nonlinear response observed in experiments may come from a nonlinear mechanical response, a ubiquitous feature of cells with structured cytoskeleton (52). We therefore assume that the pressure difference ∆p is not zero, but depends on the swelling rate as expected for a viscoelastic element. To reproduce Fig. 3 A, we assume that neurites swell at zero pressure below a critical rate but encounter internal friction for rates larger than a criti- cal value Ω. This can be written as for for V < Ω V > Ω (12) ∆p = (cid:26) 0 η sign( V)(cid:0) V/Ω − 1(cid:1) where Ω is the maximal rate of change of volume at which the neurite can swell without tension, corresponding to about 5 10−3 s−1 in our experiments, and η is the friction scale, about 2 MPa. As shown in Fig. 8, with this ansatz the model re- produces the essentially constant swelling rate observed in experiments. The nonlinear friction slows down both swelling and recovery. Interestingly, the peculiar triangu- lar shape of of the curves at C = 0.5 is much closer to that of experiments (compare to Fig. 2). DISCUSSION Microtubules mechanically slow down swelling. The initial response of neurites to hypo- as well as hyperos- motic shocks is a strongly nonlinear function of the ex- ternal osmotic pressure. Remarkably, this response be- comes much simpler after microtubule disruption: neu- rites behave as perfect-osmometers, and the swelling rate increases linearly with osmotic shock strength. The sim- plest explanation is that microtubules slow down the ini- tial volume change by mechanically opposing the hydro- static pressure difference. This is not exactly an elastic response but rather a (nonlinear) viscous one which de- pends crucially on the swelling rate. This mechanical role of microtubules may come as a surprise, as they are in general irrelevant for the (passive) mechanical response of cells. In neurites, however, their structure is differ- ent: they are arranged in bundles interconnected by mi- crotubule binding proteins (MBPs). Our results indicate that microtubules are firmly connected to the membrane, either directly by MBPs or through the actin cortex. In- terestingly, studies of volume regulation on round PC12 cells without neurites give a different picture. Whereas disruption of the actin cytoskeleton by Cytochalasin B has been reported to increase KCl efflux and diminish extent of swelling (25, 53), disruption of microtubules has no effect on RVD (25). This is consistent with our conclu- sion of a mechanical role for microtubuli, since these are organised very differently in round cells and neurites: in the latter their bundle structure provides a rigid scaffold which can oppose swelling. The mechanical response of the neurite, as described by Eq. 12, seems to be similar to that of adhering fibrob- lasts (52, 54). At slow strain rates, forces are low; above a critical strain rate, friction increases and the force be- comes much stronger. From our results, writing the crit- V0/(r0A0) ∼ 5 10−3s−1. In ical rate as a strain we get single fibroblasts, this is indeed the order of magnitude of the critical strain rate where frictional forces increase (52), which in turn agrees with the timescale of active pro- cesses (9, 11). This suggests that changes in cell shape -- including swelling -- take place at the rate allowed by spontaneous unbinding of cytoskeletal crosslinks, while faster changes are slowed down by friction between cy- toskeletal elements. RVD is osmotically driven. Our results indicate that mechanical tension does not provide the driving force for volume recovery. The pearling modulation vanishes Thecytoskeletoninvolumeandshaperegulation 11 well before the volume recovers, indicating zero mem- brane tension during RVD. Moreover, in presence of cytoskeleton-perturbing drugs the volume recovery time τ does not increase, but becomes slightly shorter. This indi- cates that volume recovery takes place via extrusion of os- molites (3). Indeed, our simple model based on two ionic species suffices to explain the essential features of the vol- ume evolution curves. It captures the broad response and even the tendency to "undershoot" at large dilutions. RVD follows an Arrhenius trend as a function of temperature. The timescales involved in the volume re- sponses are temperature dependent. The relaxation time is about an order of magnitude more temperature depen- dent than the swelling/shrinking rate, showing an expo- nential reduction as the temperature is increased. Thus the ability of the cell to respond to and negate perturbations to its volume improves drastically as the temperature ap- proaches the normal physiological values. Due to this rea- son the neurites are able to withstand strong hypoosmotic- shocks at higher temperatures when the same shock would have made them rupture at a slightly lower tem- perature. A dynamic picture of stretch injury. Similar shape transformations have been observed in nerves under the name of "beading" as a response to stretch injury (31, 32). Interestingly, electron microscopy observations of the ul- trastructure of stretch-beaded nerves show that micro- tubules are splayed out in the beads (44), consistent with our picture of a mechanical connection between micro- tubules and the membrane. The shapes of beaded nerves have been interpreted as equilibrium shapes with a con- stant curvature (31); however, our results offer an alter- native explanation, namely that the shape is given by the fastest growing mode at the time of the increase in ten- sion. If so, the shape of beaded nerves would not simply follow from structural properties (as expected for an equi- librium shape) but would also be defined by the precise way stretch-injury takes place, i.e., the rate and extent of loading. Future studies may address this question in de- tail. Conclusions Neurites respond to sudden osmotic pressure changes with a fast volume regulation response. The initial phase is characterised by a nonlinear dependence of swelling rate on the initial osmotic pressure. Cytoskeletal per- turbation, especially microtubule disruption, accelerates swelling and increases the maximum volume reached, but does not affect the relaxation phase. Taking our results to- gether, we propose that mechanical forces due to the non- linear viscosity of the cytoskeleton slow down the initial phase of change of volume. This may provide instanta- neous integrity to the neurite while osmotic mechanisms "warm up". Acknowledgements Experiments were performed at the Universitat Bayreuth, Germany, with the generous financial support of Albrecht Ott. We thank Osvaldo Chara, Pablo Schwarzbaum, Ka- rina Alleva and Wolfram Hartung for helpful discussions. References 1. Weiss, T., 1996. Cellular Biophysics. MIT Press. 2. Lang, F., G. Busch, M. Ritter, H. Volkl, S. Waldeg- ger, E. Gulbins, and D. Haussinger, 1998. Functional significance of cell volume regulatory mechanisms. Physiol. Reviews 78:247. 3. Hernandez, J., and E. Cristina, 1998. Modelling cell volume regulation in nonexcitable cells: the roles of the Na+ pump and of cotransport systems. Am. Phys- iol. Soc. C1067. 4. MacKnight, A. D. C., 1987. Volume Maintenance in Isosmotic Conditions. In R. Gilles, A. Kleinzeller, and L. Bolis, editors, Cell Volume Control: Funda- mental and comparative aspects in animal cells. Aca- demic Press, San Diego, 3 -- 43. 5. Bray, D., 2001. Cell Movements: from molecules to motility. Garland Publishing, Inc., New York, 2nd edition. 6. Alberts, B., D. Bray, J. Lewis, M. Raff, K. Roberts, and J. D. Watson, 1994. Molecular Biology of the Cell. Garland Publishing, Inc., New York, 3rd edi- tion. 7. Janmey, P. A., 1998. The cytoskeleton and cell sig- naling: component localization and mechanical cou- pling. Physiol. Rev. 78:763 -- 781. 8. Fabry, B., G. N. Maksym, J. P. Butler, M. Glo- gauer, D. Navajas, and J. J. Fredberg, 2001. Scal- ing the microrheology of living cells. Phys. Rev. Lett. 87:148102. Thecytoskeletoninvolumeandshaperegulation 12 9. Thoumine, O., and A. Ott, 1997. Time scale depen- dent viscoelastic and contractile regimes in fibrob- lasts probed by microplate manipulation. J. Cell. Sci. 110:2109 -- 2116. 10. Fern´andez, P., P. A. Pullarkat, and A. Ott, 2006. A master relation defines the nonlinear viscoelasticity of single fibroblasts. Biophys. J. 90:3796 -- 3805. 11. Pullarkat, P. A., P. A. Fern´andez, and A. Ott, 2007. Rheological properties of the Eukaryotic cell cy- toskeleton. Phys. Rep. 449:29 -- 53. 12. Bernal, R., P. A. Pullarkat, and F. Melo, 2007. Mechanical properties of axons. Phys. Rev. Lett. 99:018301. 13. Henson, J., 1999. Relationships between the Actin Cytoskeleton and Cell Volume Regulation. Mi- croscopy research and technique 47:155 -- 162. 14. Kleinzeller, A., 1965. The volume regulation in some animal cells. Arch. Biol. 76:217 -- 232. 15. Cantielo, H., 1997. Role of actin filament organiza- tion in cell volume and ion channel regulation. J. Exp. Zoology 279:425. 16. Mills, J., 1987. The Cell Cytoskeleton: Possible Role in Volume Control. In R. Gilles, A. Kleinzeller, and L. Bolis, editors, Cell volume control: fundamental and comparative aspects in animal cells. Academic Press,Inc., San Diego, 75 -- 101. 17. Strieter, J., J. L. Stephenson, L. G. Palmer, and A. M. Weinstein, 1990. Volume-activated chloride perme- ability can mediate cell volume regulation in a math- ematical model of a tight epithelium. J. Gen. Physiol. 96:319 -- 344. 18. Strange, K., 2004. Cellular volume homeostasis. Adv.Physiol.Educ. 28:155 -- 159. 19. Downey, G., S. Grinstein, A. Sue-A-Quan, B. Cza- ban, and C. Chan, 1995. Volume Regulation in leuko- cytes: requirement for an intact cytoskeleton. J. Cell Physiol. 163:96 -- 104. 20. Pullarkat, P., P. Dommersnes, P. Fern´andez, J.-F. Joanny, and A. Ott, 2006. Osmotically induced shape transformations in axons. Phys. Rev. Lett. 96:048104. 21. Wan, X., J. Harris, and C. Morris, 1995. Responses of neurons to extreme osmomechanical stress. J. Memb. Biol. 145:21 -- 31. 22. Heubusch, P., C. Jung, and F. Green, 1985. The os- motic response of human erythrocytes and the mem- brane cytoskeleton. J.Cell.Physiol 122:266 -- 272. 23. Suchyna, T., S. Besch, and F. Sachs, 2004. Dy- namic regulation of mechanosensitive channels: ca- pacitance used to monitor patch tension in real time. Phys Biol 1:1. 24. D'Alessandro, M., D. Russell, S. M. Morley, A. M. Davies, and E. Birgitte Lane, 2002. Keratin muta- tions of epidermolysis bullosa simplex alter the ki- netics of stress response to osmotic shock. J. Cell Sci. 115:4341 -- 4351. 25. Cornet, M., E. Delpire, and R. Gilles, 1988. Relations between cell volume control, microfilaments and mi- crotubules networks in T2 and PC12 cultured cells. J. Physiol. Paris 83:43 -- 49. 26. Light, D., A. Attwood, C. Siegel, and N. Baumann, 2003. Cell swelling increases intracellular calcium in Necturus erythrocytes. J.Cell Sci. 116:101 -- 109. 27. Arora, P., K. Bibby, and C. McCulloch, 1994. Slow oscillations of free intracellular calcium ion concen- tration in human fibroblasts responding to mechani- cal stretch. J.Cell.Physiol. 161:187 -- 200. 28. Guilak, F., R. Zell, G. Erickson, D. Grande, C. Rubin, K. McLeod, and H. Donahue, 1999. Mechanically induced calcium waves in articular chondrocytes are inhibited by gadolinium and amiloride. J.Orthop.Res. 17:421 -- 429. 29. Greene, L., and A. Tischler, 1976. Establishment of a noradrenergic clonal line of rat adrenal pheochromo- cytoma cells which respond to nerve growth factor. Proc.Natl.Acad.Sci.USA 73:2424 -- 2428. 30. Roger Jacobs, J., and J. Stevens, 1986. Changes in the organization of the neuritic cytoskeleton during nerve growth factor-activated differentiation of PC12 cells: a serial electron microscopic study of the de- velopment and control of neurite shape. J.Cell Biol. 103:895 -- 906. 31. Markin, V., D. Tanelian, R. Jersild, Jr., and S. Ochs, 1999. Biomechanics of stretch-induced beading. Bio- phys.J. 76:2852 -- 2860. 32. Ochs, S., R. Pourmand, and R. Jersild, Jr., 1996. Origin of beading constrictions at the axolemma: Thecytoskeletoninvolumeandshaperegulation 13 presence in unmyelinated axons and after β,β′- iminodipropionitrile degradation of the cytoskeleton. Neuroscience 70:1081 -- 1096. 33. Drexler, H. G., W. Dirks, W. R. A. F. MacLeod, H. Quentmeier, K. G. Steube, and C. C. Uphoff, 2001. DSMZ Catalogue of Human and Animal Cell Lines. Braunschweig. 34. Greene, L., M. Sobeih, and K. Teng, 1991. Method- ologiesfor the culture and experimental use of the PC12 rat pheochromocytoma cell line. In G. Banker, and K. Goslin, editors, Culturing Nerve Cells. MIT Press, Cambridge, 207 -- 226. 35. De Brabander, M., R. Van de Veire, R. Aerts, M. Borgers, and P. Janssen, 1976. The effects of methyl[5-(2-thienylcarbonyl)-1H-benzimidazol-2- yl]carbamate (R 17 934, NSC 238159), a new syn- thetic antitumoral drug interfering with microtubules, on mammalian cells cultured in vitro. Cancer Res. 36:905 -- 916. 36. Straight, A. F., A. Cheung, J. Limouze, I. Chen, N. J. Westwood, J. R. Sellers, and T. J. Mitchison, 2003. Dissecting temporal and spatial control of cytokinesis with a Myosin II inhibitor. Science 299:1743 -- 1747. 37. van Hoek, A. N., M. D. de Jong, and C. H. van Os, 1990. Effects of dimethylsulfoxide and mer- curial sulfhydryl reagents on water and solute per- meability of rat kidney brush border membranes. Biochim.Biophys.Acta 1030:203 -- 210. 38. Maric, K., B. Wiesner, D. Lorenz, E. Klussmann, T. Betz, and W. Rosenthal, 2001. Cell volume ki- netics of adherent epithelial cells measured by laser scanning reflection microscopy: determination of water permeability changes of renal principal cells. Biophys.J. 80:1783 -- 1790. 39. Huster, D., A. Jin, K. Arnold, and K. Gawrisch, 1997. Water permeability of polyunsaturated lipid mem- branes measured by 17O NMR. Biophys.J. 73:855 -- 864. 40. Lucio, A., R. Santos, and O. Mesquita, 2003. Mea- surements and modeling of water transport and os- moregulation in a single kidney cell using opti- cal tweezers and videomicroscopy. Phys. Rev. E 68:041906. 41. Fern´andez, P., 2006. Mechanics of living cells: nonlinear viscoelasticity of single fibroblasts and shape instabilities in axons. Ph.D. thesis, Universitat Bayreuth. 42. Blumenfeld, L. A., and A. N. Tikhonov, 1994. Bio- physical thermodynamics of intracellular processes: molecular machines of the living cell. Springer, Berlin. 43. D. Kilinc, G. G., and K. A. Barbee, 2008. Mechanicall-induced membrane portaion causes ax- onal beading and localized cytoskeletal damage. Exp. Neurol. 212:422 -- 430. 44. Ochs, S., R. Jersild, Jr., R. Pourmand, and C. Potter, 1994. The beaded form of myelinated nerve fibers. Neuroscience 61:361 -- 372. 45. Bar-Ziv, R., and E. Moses, 1994. Instability and "pearling" states produced in tubular membranes by competition of curvature and tension. Phys. Rev. Lett. 73:1392 -- 1395. 46. Bar-Ziv, R., E. Moses, and P. Nelson, 1998. Dynamic excitations in membranes induced by optical tweez- ers. Biophys.J. 75:294 -- 320. 47. Bernal, R., F. Melo, and P. A. Pullarkat, 2009. Drag force as a tool to test the active mechanical response of PC12 neurites. in press in Biophys. J. . 48. Bar-Ziv, R., and E. Moses, 1999. Pearling in cells: a clue to understanding cell shape. Proc. Natl. Acad. Sci. USA 96:10140 -- 10145. 49. Nelson, P., T. Powers, and U. Seifert, 1995. Dynamic theory of pearling instability in cylindrical vesicles. Phys. Rev. Lett. 74:3384. 50. Rayleigh, L., 1892. On the instability of a cylinder of viscous liquid under capillary force. Philos.Mag. 34:195. 51. Chandrasekhar, S., 1970. Hydrodynamic and hydro- magnetic stability. Dover, New York, 3 edition. 52. Fern´andez, P., and A. Ott, 2008. Single cell mechan- ics: stress stiffening and kinematic hardening. Phys. Rev. Lett. 100:238102. 53. Cornet, M., J. Ubl, and H.-A. Kolb, 1993. Cytoskele- ton and ion movements during Volume regulation in cultured PC12 cells. J. Membrane Biol. 133:161 -- 170. Thecytoskeletoninvolumeandshaperegulation 14 54. Fern´andez, P., L. Heymann, A. Ott, N. Aksel, and P. A. Pullarkat, 2007. Shear rheology of a cell mono- layer. New J. Phys. 9:419.
1108.5390	1	1108	2011-08-26T20:09:20	Thermal and mechanical denaturation properties of a DNA model with three sites per nucleotide	[ "physics.bio-ph", "q-bio.BM" ]	In this paper, we show that the coarse grain model for DNA, which has been proposed recently by Knotts, Rathore, Schwartz and de Pablo (J. Chem. Phys. 126, 084901 (2007)), can be adapted to describe the thermal and mechanical denaturation of long DNA sequences by adjusting slightly the base pairing contribution. The adjusted model leads to (i) critical temperatures for long homogeneous sequences that are in good agreement with both experimental ones and those obtained from statistical models, (ii) a realistic step-like denaturation behaviour for long inhomogeneous sequences, and (iii) critical forces at ambient temperature of the order of 10 pN, close to measured values. The adjusted model furthermore supports the conclusion that the thermal denaturation of long homogeneous sequences corresponds to a first-order phase transition and yields a critical exponent for the critical force equal to sigma=0.70. This model is both geometrically and energetically realistic, in the sense that the helical structure and the grooves, where most proteins bind, are satisfactorily reproduced, while the energy and force required to break a base pair lie in the expected range. It therefore represents a promising tool for studying the dynamics of DNA-protein specific interactions at an unprecedented detail level.	physics.bio-ph	physics	Thermal and mechanical denaturation properties of a DNA model with three sites per nucleotide Ana-Maria FLORESCU and Marc JOYEUX(#) Laboratoire Interdisciplinaire de Physique (CNRS UMR 5588), Université Joseph Fourier - Grenoble 1, BP 87, 38402 St Martin d'Hères, France Abstract - In this paper, we show that the coarse grain model for DNA, which has been proposed recently by Knotts, Rathore, Schwartz and de Pablo (J. Chem. Phys. 126, 084901 (2007)), can be adapted to describe the thermal and mechanical denaturation of long DNA sequences by adjusting slightly the base pairing contribution. The adjusted model leads to (i) critical temperatures for long homogeneous sequences that are in good agreement with both experimental ones and those obtained from statistical models, (ii) a realistic step-like denaturation behaviour for long inhomogeneous sequences, and (iii) critical forces at ambient temperature of the order of 10 pN, close to measured values. The adjusted model furthermore supports the conclusion that the thermal denaturation of long homogeneous sequences corresponds to a first-order phase transition and yields a critical exponent for the critical force equal to 70.0=s . This model is both geometrically and energetically realistic, in the sense that the helical structure and the grooves, where most proteins bind, are satisfactorily reproduced, while the energy and force required to break a base pair lie in the expected range. It therefore represents a promising tool for studying the dynamics of DNA-protein specific interactions at an unprecedented detail level. (#) email : [email protected] 1 I - INTRODUCTION This article deals with the thermal and mechanical denaturation of DNA, that is, the separation of the two strands upon heating [1-6] or application of a force [7-15]. More precisely, we report on our investigations of the denaturation properties of a mesoscopic model, which describes each nucleotide as a set of three interacting sites. The motivations for this study are twofold. First, all of the models that have been developed up to now to investigate theoretically the thermal and mechanical denaturation of long DNA sequences describe each base pair with a very limited number of degrees of freedom. Actually, in most cases a single degree of freedom is used to represent a base pair. This degree of freedom is usually either the distance between paired bases [16-24] or the state of the base pair (open or closed) [25-32]. A few models additionally take into account the rotation of the bases around the strand axes [33-36] or the bending of the chain [37,38]. At some point, the results and predictions obtained from models with such strongly reduced dimensionality must be compared with those obtained from models that describe more accurately the actual structure of DNA sequences, in order to check their robustness and validity. The second motivation deals with the investigation of DNA-protein interactions. As for DNA denaturation, an atomistic description of these problems is prohibitively expensive from the point of view of CPU time requirements, so that mechanical mesoscopic models [39- 45] are the only alternative to kinetic ones [46-58]. For example, we have recently shown that models where 15 DNA base pairs are represented by a single bead are very useful to investigate non-specific DNA-protein interactions [59-61], that is, the alternation of 3- dimensional diffusion in the buffer and 1-dimensional sliding along the DNA sequence, by which the protein scans the DNA sequence while searching for its target [62-70]. Such models 2 are, however, no longer sufficient when it comes to investigate specific DNA-protein interactions, that is, the process by which the protein recognizes its target and binds firmly to it. For this purpose, more accurate models are needed, which describe correctly the helical geometry of DNA, as well as the minor and major grooves where most proteins bind. Moreover, since the binding of the protein to its target may involve the opening of one or several base pairs, the thermal and mechanical denaturation properties of these models should also match real ones. Several mesoscopic models that take the helical structure of DNA and sometimes the grooves into account have been proposed over the past few years [71-80]. These models have however been used uniquely to investigate the denaturation (and in a few cases the renaturation [79,80]) of short oligonucleotides with a few tens of base pairs. It must be emphasized, that this problem differs substantially from the denaturation of long sequences, in the sense that the abrupt denaturation phase transition of long sequences is replaced, in the case of short oligonucleotides, by a much smoother dissociation, due to finite-size effects. The model of Knotts, Rathore, Schwartz and de Pablo (KRSdP) [78] is particularly interesting in the context of the two motivations reported above, because (i) of its precise geometry, and (ii) the authors have already shown that it correctly predicts several physical properties of real sequences, like base specificity, the effect of salt concentration on duplex stability, and the long persistent length of double-stranded DNA. In this model, each nucleotide is mapped onto three interaction sites, so that 9 coordinates are necessary to describe its position. This is about one order of magnitude larger than for most mechanical models aimed at investigating DNA denaturation [16-24], and about two orders of magnitude larger than for "beads and springs" models of DNA and the model we have developed to study DNA-protein interactions [39-45,59-61], but calculations involving a few thousands of base pairs are still affordable with nowadays computers. 3 The remainder of the paper is consequently organized as follows. The model and the evolution equations are described in Section II. Special attention is paid to the description of the modifications we brought to the original KRSdP model in order to adapt it to the investigation of long sequences. Sections III and IV then respectively describe the thermal and mechanical denaturation properties of this model. In Section V, we discuss the effect on denaturation of two terms of the original KRSdP model, namely the excluded volume term and the extension of base pairing to all complementary base pairs, which are neglected in the main body of this work. We finally conclude in Section VI. II - EXPRESSION OF THE MESOSCOPIC MODEL The KRSdP model used in this work has been described in detail in Section II of Ref. [78]. Nevertheless, we provide here a short description thereof for the sake of completeness and in order to point out clearly the modifications we brought to adapt it to the study of the denaturation of long sequences. As already mentioned, each nucleotide is mapped onto three interaction sites, namely one site for the phosphate group, one site for the sugar group, and one site for the base. At equilibrium, the phosphate and sugar sites are placed at the center of mass of the respective moities, while the site is place at the N1 position for A and G purine bases and at the N3 position for C and T pyrimidine bases. Reference coordinates for each site determined from the standard B isoform [81] are provided in Table I of Ref. [78]. The reference geometry is illustrated in Fig. 1. The potential energy potE of the system includes six distinct contributions, E pot = V bond + V angle + V dihedral + V stack + V bp + V qq , (II.1) 4 where bondV , angleV and V dihedral describe the stretch, bend and torsion contributions, respectively, while stackV denotes the stacking interaction between bases belonging to the same strand, bpV the hydrogen bonding between complementary bases, and qqV the Coulomb interaction between the charged phosphate sites. Note that the original model contains an additional term exV , which describes excluded volume interactions between any two sites that do not interact by means of one of the other six contributions. However, this term is quite expensive from the point of view of CPU time requirements, while it influences only moderately the average quantities we are interested in. It was therefore dropped in the calculations presented in Sections III and IV. Its influence will however be discussed in some detail in Section V. The expressions for the six contributions to potE are ( d i k 1 ∑ i - d 20 ) i + ( d i k 2 ∑ i - d 40 ) i = = V bond V angle V dihedral k q 2 = k ∑ i ∑ f i qq - 20 ( ) i i 1[ - ff - 0 cos( i i )] V stack = e ∑ < i j [( 0 r ij r ij 12 ) - (2 6 ) + ]1 0 r ij r ij V bp = e AT (5[ ∑ pairs base AT 12 ) - (6 + e GC ∑ pairs base G C (5[ - 12 ) (6 0 r ij r ij 0 r ij r ij 10 ) + ]1 10 ) + ]1 0 r ij r ij 0 r ij r ij = V qq 2 e pe 4 OH 2 ∑ < j i - exp( r ij k D ) r ij . (II.2) In Eq. (II-2), id represents the distance between two bound sites, q the angle formed by i three consecutively bound sites, and f the dihedral angle formed by four consecutively i bound sites, all of these sites belonging of course to the same strand. In contrast, ijr stands for the distance between two sites that are not directly bound and that do not necessarily belong 5 to the same strand. 0 id , q , 0 i f , and 0 i 0 ijr denote the values of these variables for the reference configuration (for numerical values, see Tables III and IV of Ref. [78]). We used =k 1 26.0 kcal mol-1 Å -2, =k 2 0.26 kcal mol-1 Å -4, =qk 104 0. kcal mol-1 rad-2, and 04.1=fk kcal mol- 1, as proposed in Ref. [78] (it is reminded that 1 kcal mol-1 ≈ 43.4 meV). We also conformed to Ref. [78] in letting stackV couple not only bases i and i+1 of the same strand, but also bases i and i+2, and in using 26.0=e kcal mol-1. In contrast, we modified somewhat the base pairing interaction bpV . Indeed, it was assumed in Ref. [78] that bpV describes hydrogen bonding between any complementary base pair and acts both intra- and interstrand. However, as will be discussed in more detail in Section V, this assumption leads to the unphysical result that many bases form bonds with two complementary bases instead of one at most. Therefore, we assumed in this work that bpV applies only to bases belonging to the same pair i. This modification certainly imposes certain limitations to the model, which will also be discussed further in Section V, but it prevents unphysical multiple base pairings. It also has the consequence that the two DNA strands are less tightly bound and separate at lower temperature compared to the original model. Stated in other words, using =e AT 77.2 kcal mol-1 and =e GC 16.4 kcal mol-1 as proposed in [78] in the modified model with the simple pairing scheme results in critical temperatures that are too low. For example, let us consider a 480 base pairs (bp) sequence, hereafter called the ...AAAAA... sequence, one strand of which is composed of adenine bases and the other one of thymine bases. We checked that simulations performed with the parameters of Ref. [78] (that is, =e AT 77.2 kcal mol-1) and the modified model (that is, assuming that bpV acts only between bases belonging to the same pair i) lead to a critical temperature » cT 230 K for this sequence, while it actually lies around 335 K at =+ ]Na[ 50 mM salt concentration. This 6 e result is illustrated in Fig. S1 of the Supplementary Material [82]. We therefore adjusted AT and e to get correct denaturation temperatures for long ...AAAAA... and ...GGGGG... GC sequences, that is, around 335 and 375 K, respectively. As will be shown below, we found that =e AT 90.3 kcal mol-1 and =e GC 37.4 kcal mol-1 meet this requirement. At last, in the expression for the Coulomb interaction qqV between the charged phosphate sites, e stands for the charge of the electron, which is actually carried by each phosphate group, e = OH 2 78 e 0 for the dielectric constant of water at room temperature, and k D 603.13= Å for the Debye length at =+ 50 ]Na[ mM salt concentration. The dynamics of the model was studied by integrating Langevin's equations m j j 2 r d 2 dt -(cid:209)= E pot - m j r d g + j dt g Tkm j B 2 )( t dW dt . (II.3) where jr and jm denote the position and the mass of site j, g is the dissipation coefficient, and )(tW a Wiener process. The first term in the right-hand side of Eq. (II.3) describes internal forces, while the two last terms model the effects of the buffer, namely friction and thermal noise. For numerical purposes, the derivatives in Langevin equations were replaced by finite differences. The position of site j at time step 1+i , 1+i jr , was consequently obtained from the positions i jr and 1-i jr at the two previous time steps according to m j 1( g D+ t 2 + 1 = r ) i j 2 m r i j j - m j 1( g D- t 2 - 1 r ) i j (cid:209)- E pot +D 2 t g D 2/3 )( twtTkm B j 2 , (II.4) where tD is the time step and )(tw a normally distributed random function with zero mean and unit variance. We used 10=Dt fs and 5=g ns-1. Note that in the thermodynamic limit of infinitely long homogeneous sequences, the averages of thermodynamic observables do not depend on the particular value that is assumed for the dissipation coefficient g [83]. It is unfortunately not feasible to investigate with the model above the melting properties of very 7 long sequences, so that most results presented below were obtained with 480 bp long ones. Still, preceding studies dealing with simpler dynamical models suggest that 480 bp sequences are indeed already rather close to the thermodynamic limit (see for example Ref. [84]). III - THERMAL DENATURATION A. Critical temperatures and denaturation rates It has been recognized quite early [85] that the thermal denaturation of DNA homopolymer pairs (that is, sequences whose complementary strands are composed of a single type of bases), is rather straightforward. These homogeneous sequences are characterized by a well-defined melting (or critical) temperature cT . For temperatures T larger than cT , the two strands separate, leading to single-stranded DNA. Denaturation actually starts at the two extremities of the sequence before spreading towards its middle part. This is clearly seen in Fig. 2 of Ref. [36] and can be derived analytically for 1-dimensional models (see for example Eqs. (18) and (19) of Ref. [84]). This point is also illustrated in the left plot of Fig. 2, which shows an instantaneous snapshot of the 480 bp ...AAAAA... sequence with clearly melted extremities. Typical plots showing the time evolution of the number of open base pairs of the initially closed ...AAAAA... sequence at 340 K, that is, about 5 K above the critical temperature cT = 335 K, are displayed in the lower panel of Fig. 3. Each curve corresponds to a single trajectory integrated for 2.5·108 steps with a different set of random numbers. It is seen that the sequence remains closed for a certain amount of time before the extremities begin to separate. This delay may be quite long and may differ a lot from one trajectory to the 8 other. In contrast, once initiated, denaturation or renaturation (annealing) proceed at comparable rates for all trajectories performed at the same temperature. Instead of starting from a closed sequence, a sensible method for determining the critical temperature of a homogeneous sequence therefore consists in considering a sequence that is half melted, like the one shown in the left panel of Fig. 2, and in monitoring whether it opens or closes completely at a given temperature T. The denaturation (respectively, annealing) rate is then computed by dividing the number of base pairs that opened (respectively, closed) by the time it took for them to open (respectively, close). The critical temperature is the temperature at which the denaturation/annealing rate vanishes. Such a plot of the denaturation rate is provided in Fig. 4 for the 480 bp ...AAAAA... and ...GGGGG... sequences. Negative denaturation rates correspond of course to a positive annealing rate. Each point in this figure was obtained by integrating a single trajectory. The number of required integration steps obviously varied strongly with temperature, typically from 5·106 steps for the ...GGGGG... sequence at 400 K up to 5·108 steps for the ...AAAAA... sequence at 335 K. By using this method, we were able to determine that, for the model described in Section II and the newly adjusted values of e (3.90 kcal mol-1) and AT e (4.37 kcal mol-1), long GC ...AAAAA... sequences melt at = cT 5.00. 335 m K and long ...GGGGG... ones at = cT 374 75. m 25.0 K. These melting temperatures are in good agreement with the corresponding values of 336.3 and 373.1 K that are obtained with the statistical model of Ref. [86] by using the parameters of Blossey and Carlon [32] and a =+ ]Na[ 50 mM salt concentration. Fig. 4 additionally shows that average denaturation and annealing rates increase linearly with Tc - . It must be emphasized that, in contrast with other quantities discussed in T this paper, these rates do depend on the particular value that is assumed for the dissipation 9 coefficient g (we used 5=g ns-1), but the linear dependence should still be observed for different values of g. On the other hand, the fact that the rates reported in Fig. 4 are about two orders of magnitude larger than those estimated from experiments dealing with short oligoribonucleotides [87,88] indicates that one should assume a value of g substantially larger than 5=g ns-1 in order for the model to reproduce the true time scales of denaturation/annealing events. For practical purposes (i.e. CPU time requirements), we however used 5=g ns-1 for all results reported below. Still, for the sake of completeness, we independently estimated the correct value for g by computing the evolution with g of the 3-dimensional diffusion coefficient 3DD of a 367 bp double-stranded ...AAAAA... sequence at 298 K. More precisely, for each value of g, the time evolution of the squared deviation -t )( r 2)0( r of the center of mass of the sequence was averaged over four trajectories integrated for about 108 steps and subsequently fitted against the < - r )( t r )0( 2 >= 6 tD 3D law. Note that for 1310=g s-1 we had to use a time step 2=Dt fs instead of 10=Dt fs. The results shown in Fig. 5 indicate that, in the range extending from 1010 to 1013 s-1, 3DD decreases as the inverse of g according to »D 3D g/94.7 , where g is expressed in s-1 and 3DD in m2 s-1. Since the experimentally determined value of the diffusion coefficient of a 367 bp sequence is =D 3D 58.1 · 10 - 11 m2 s-1 [89], the realistic value for g is close to 5·1011 s-1, that is 500 ns-1. This value is 100 times larger than the value assumed in this paper, as already suggested by the predicted denaturation rates discussed just above. B. Order and width of the transition 10 It can be seen in the bottom plot of Fig. 3 that the process that leads to complete denaturation of a sequence is not uniform. The curves are "noisy", which reflects the fact that, under the antagonistic effects of binding energy and random thermal noise, some parts of the sequence open and close transiently several times before remaining open. The amplitude of oscillations decreases as temperature moves away from the critical temperature. Close to the critical temperature, oscillations are instead very pronounced. For example, the top plot of Fig. 3 displays the time evolution of the number of open base pairs of the initially closed 480 bp ...AAAAA... sequence at 335 K, that is at about the critical temperature. This curve was obtained from a single trajectory integrated for 4.5·108 steps. It is seen that opening/closing oscillations take place at all investigated time scales and can involve several hundreds of base pairs. Obviously, if the length of the sequence is smaller than the amplitude of the oscillations, then the sequence will open even if the temperature is smaller than the critical one. This is the reason, why the denaturation and annealing of short oligonucleotides with a few tens of base pairs are rather smooth transitions, which extend over a substantial temperature range [71-80,84,90,91] and differ drastically from the much narrower melting of long sequences described in the present paper. The melting transition of (infinitely) long homogeneous sequences is indeed essentially characterized by its order or, more finely, by the critical exponent α that describes the behaviour of the singular parts of the free energy per base pair, f , entropy per base sing pair, sings , internal energy per base pair, singu , and specific heat per base pair, ( Vc ) sing , in the neighbourhood of the critical temperature, according to 11 f sing s sing u sing (cid:181) t a - 2 ¶ f sing ¶ T + sing -= = f ( c V ) sing -= T a - 1 (cid:181) t a - 1 (cid:181) t - a (cid:181) t , fT sing ¶ 2 f sing ¶ 2 T (III.1) where = / cTTt - 1 denotes the reduced temperature. If 1=a , then melting corresponds to a first-order phase transition and sings and singu display a discontinuity at cTT = . Therefore, the sequence absorbs a large amount of energy when it is heated, but its temperature remains constant till denaturation is not complete. The sequence is actually in a mixed-phase regime, where some parts of the sequence (bubbles) are already melted, while other ones are still closed (double-stranded). This is equivalent to the turbulent mixture of liquid water and vapor bubbles that arises at the boiling point of water. In contrast, if 1<a , then melting corresponds to a second-order phase transition. sings and singu are continuous at cTT = , so that no heat is absorbed at the critical point, like, for example, in the ferromagnetic transition. The question of the order of the DNA melting transition has been (and it still is) much debated. From the point of view of statistical models, the order of the transition depends on the way the partition function of a loop, and particularly the loop closure exponent c, is calculated [32]. When using self-avoiding walks in 3D space, c is numerically estimated to be close to 75.1»c , which corresponds to a second-order phase transition [92] (the boundary between second- and first-order transitions is 2=c [25,93]). In contrast, when using loops embedded in chains [94,95], which is probably a better approximation, one gets 15.2»c [94- 96], which corresponds to a first-order transition. Unfortunately, experimental results can be equally well reproduced using sets of parameters with 75.1»c and 15.2»c [32], so that statistical models cannot help deciding whether the melting transition is first- or second-order. 12 One-dimensional dynamical models lead to conclusions that are even more ambiguous. Indeed, the easiest way to estimate α consists in considering that the singular part of Vc varies much more rapidly than its non-singular part in the neighbourhood of cT and, consequently, in deducing α from the slope of log-log plots of the temperature evolution of Vc instead of ( Vc ) sing [97-100] (remember also that experimentalists have no means to separate the singular from the non-singular part of the measured specific heat). For realistic one- dimensional DNA models, values of α determined in this way are consistently larger than 1 [90,101,102], which may be interpreted as an indication that DNA denaturation corresponds to a first-order phase transition followed by a crossover to another regime in the last few kelvins below the critical temperature [90]. As far as simulations are concerned, one is instead formally able to separate any quantity into a singular and a non-singular part. For example, it was suggested in Ref. [102] that the free energy per base pair, f, may be written as the sum of a singular part, f sing , and a non-singular one, nsf = f + f ns f sing , (III.2) where nsf is the average free energy per base pair when the two strands are widely (infinitely) separated, that is, when the sequence is single-stranded. This definition leads to well-behaved quantities. Indeed, the singular part is zero above the dissociation temperature, while the non- singular part behaves smoothly when crossing this temperature. It was furthermore shown that the estimation of α from the slope of log-log plots of the temperature evolution of f sing according to Eq. (III.1) leads to values that vary between 0.5 and 1, depending on the explicit expression that is assumed for the non-linear part of the stacking interaction in these one- dimensional models [102]. Stated in other words, the order of the melting transition depends on the shape and strength of the stacking interaction. 13 One might consequently wonder what is the order of the transition predicted by the more realistic model of Section II. While for one-dimensional models all thermodynamics quantities that appear in Eq. (III.1) can be straightforwardly computed by using the Tranfer- Integral operator technique [84,90,101-104], this is, however, unfortunately no longer the case for more complex models. For such models, the only method we could think of consists in extracting the average internal energy per base pair, u, from molecular dynamics simulations, Eqs. (II.3)-(II-.4) [36,105]. Practically, u is obtained by averaging potE over long times and/or many simulations and in dividing the obtained value by the number of base pairs of the sequence. In agreement with Ref. [102], u can then be written as the as the sum of a singular part, singu , and a non-singular one, nsu = u u ns + u sing , (III.3) where nsu is the average internal energy per base pair when the two strands are widely (infinitely) separated, that is, when the sequence is single-stranded. From the practical point of view, nsu is easily obtained by launching additional simulations where initial conditions correspond to two widely separated strands instead of double-stranded DNA. singu is then obtained as the difference between u and nsu . Results are shown in Fig. 6 for the 480 bp ...AAAAA... sequence. Each point in this figure was obtained by averaging the internal energy along a single trajectory integrated for 5·107 steps. It is seen in the bottom plot of Fig. 6 that the evolution of nsu with temperature is perfectly linear in all the investigated range (275-353 K), as is also the case for u between 275 and 334 K. (remind that above the critical temperature cT = 335 K the sequence is melted, so that u = nsu and u sing = 0 ). Moreover, the slopes of the two curves are exactly the same, so that the singular part of the internal energy, singu , remains constant between 275 and 334 K. It can indeed be seen in the top plot of Fig. 6 14 that computed values of singu vary by less than 1% in this temperature range. It is admittedly difficult to determine precisely what happens between 334 and 335 K, because of the large fluctuations with long period that occur in this interval (see the top plot of Fig. 3). However, except for this very narrow temperature interval, the step-like behaviour of singu , which switches abruptly from about -3.30 kcal mol-1 below 335 K to 0 above 335 K (see Fig. 6) indicates that for this model the characteristic exponent α is equal to 1, which unambiguously supports the thesis that DNA denaturation corresponds to a first-order phase transition. Figures 4 and 6 moreover indicate that, in agreement with experimental results, the width of the melting transition predicted by the model of Section II is quite small for long sequences. We indeed estimate that the transition width is smaller than 1 K for the 480 bp ...AAAAA... sequence and than 0.5 K for the 480 bp ...GGGGG... sequence. C. Thermal denaturation of inhomogeneous sequences We next used the model of Section II to compute the thermal denaturation curve of the (inhomogeneous) 1793 bp human β-actin cDNA sequence (NCB entry code NM_001101), which has already been discussed in Refs. [23,36,90,104,106]. It has been known since the early work of Gotoh [6] that the denaturation of sufficiently long inhomogeneous sequences occurs through a series of local openings when temperature is increased and that this multi- step process is clearly reflected in the denaturation curve of inhomogeneous sequences in the 1000-10000 bp range. As can be checked in Fig. 1 of Ref. [104], denaturation of the actin sequence does not start at the extremities of the strands, but rather in narrow AT-rich regions centred around positions n=1300, n=1450 and n=1600, before the sequence abruptly melts for all n>1100 at slightly higher temperatures. There is then a plateau of several kelvins before the lower end of the sequence finally melts. The portion of the sequence that melts at the 15 highest temperature is located around the GC-richest region aroung n=150. The fingerprints of this multi-step process are clearly seen in the dashed curve of Fig. 7, which shows the temperature evolution of the fraction of open base pairs of the actin sequence obtained from the statistical model of Ref. [86] by using the parameters of Blossey and Carlon [32] and a salt concentration =+ ]Na[ 50 mM. Shown on the same figure are the results obtained with the model of Section II (open circles). It is seen that the melting curves obtained with the two models are in qualitative agreement, although the one obtained with the model proposed in this paper appears to encompass a somewhat broader temperature range. At this point we should discuss the convergence of the results. The ones presented in Figs. 3 and 4 were obtained by waiting till the sequence completely opened or closed, which is an unambiguous criterion. Moreover, the integration time of the trajectories used for the results in Figs. 5 and 6 (as well as Figs. 9-12 below) was sufficiently long to enable a careful check of their stationarity properties. For example, we checked that splitting each trajectory into two segments and calculating diffusion coefficients, internal energies and mean forces from each segment leads to results that are essentially undistinguishable from those presented in these figures. However, we could not perform a similar check for the actin sequence, because each integration step for a 1793 bp sequence lasts as long as about 15 steps for a 480 bp sequence. Each point in Fig. 7 was consequently obtained from a single trajectory that was integrated for 5·107 steps and we checked that the fraction of open base pairs did not vary significantly during the last 107 steps. However, it cannot be completely excluded that longer integration times would lead to slightly different results. That is why, we are not absolutely sure of the convergence of the results displayed in this later plot. IV - MECHANICAL DENATURATION 16 While thermal denaturation is achieved by raising the temperature of a sequence up to (or above) its critical temperature cT , mechanical denaturation may instead be achieved at temperatures T < cT by pulling on the extremities of the strands. Mechanical DNA unzipping experiments are usually performed either at constant pulling rate [7,8] or at constant force [12-15]. In the former case, the externally applied force is adjusted to compensate for the action of internal restoring forces exerted by the two strands [7,8]. At ambient temperature, the typical force that must be exerted to keep the two strands open lies in the range 10-15 pN [7-15]. From a practical point of view, constant rate experiments are conveniently modelled by separating slowly the two phosphate groups linked to the first base pair of the sequence and computing the average of the projection along the separation direction of internal forces acting on these phosphate groups. The result of such a simulation performed for the 480 bp ...AAAAA... sequence at 280 K is shown in Fig. 8. Phosphate groups were separated at the speed of 2 cm s-1 and each point represents the value of the projected force F averaged over 0.25 ns, that is 25000 integration steps. The complete curve shown in Fig. 8 corresponds to 108 integration steps. Comparison of Fig. 8 with Figs. 5 and 6 of Ref. [107] indicates that separation-force plots obtained with the model of Section II are qualitatively similar to those obtained with one-dimensional models. They essentially consist of a rather large force barrier at short distances [108] followed by a plateau. It has been known since the pioneering work of Ref. [7] that this plateau is flat in the case of an homogeneous sequence, but that it instead displays fluctuations proportional to the local AT/GC concentration when an inhomogenous sequence is being unzipped. We checked that use of the more realistic value =g 500 ns-1 instead of 5=g ns-1 leads to a plot that does not differ significantly from Fig. 8. The maximum force along the force-displacement curve is known as the threshold force thrF . It may be simply estimated from the maximum of the )(dFF = curve, as 17 illustrated in the inset of Fig. 8. On the other hand, the average asymptotic force at large displacements d represents the critical force cF that must be exerted on the widely separated extremities of a long sequence to prevent it from zipping. From a practical point of view, cF was obtained by averaging F over all separations d comprised between 100 and 200 Å. Fig. 8 indicates that the value of cF obtained in this way is close to 13 pN for the ...AAAAA... sequence at 280 K, in fair agreement with experimentally measured values. We checked that the computed value of cF does not vary when the pulling rate is divided by a factor two and the strands are separated at the speed of 1 cm s-1 instead of 2 cm s-1. Figure 9 further shows the computed temperature evolution of the threshold force thrF and the critical force cF for the 480 bp ...AAAAA... sequence. Each point in this figure was obtained from a curve similar to that in Fig. 8 but integrated at a different temperature. It appears that thrF decreases slowly (with a slope of about -1.5 pN K-1) in the investigated temperature range, from 250 pN at 275 K down to 150 pN at the melting temperature = cT 335 K. In contrast, cF obviously decreases down to zero at the critical temperature, since at T = cT thermal denaturation occurs spontaneously, i.e. without application of an external force. The bottom plot of Fig. 9 furthermore indicates that computed values of cF follow rather precisely a power law of the form = F c (77.0 T c - T 70.0) . Stated in other words, the critical exponent of the critical force is equal to 70.0=s . At this point, it is worth mentioning that, for the several one-dimensional models that were considered, the critical exponents of the specific heat, α, and the critical force, σ, were shown to be related through the linear relation [102] - = sa 2 . 2 (IV.1) 18 Since 1=a for the model used here (see Section IIIB), Eq. (IV.1) would suggest that 1=s 2 instead of 70.0=s . Conclusion therefore is that the scaling law of Eq. (IV.1) no longer holds for the more complex model analysed in this paper. On the other hand, it has also been shown that, for the Poland-Scheraga model where self-avoiding interactions are taken into account, the critical force scales like (cid:181) ( T c - n) T F c , where ν is the correlation length exponent of a self-avoiding random walk [109]. Numerically, ν is close to 0.588 for a 3-dimensional walk. This value again differs somewhat from the computed exponent, 70.0=s . Last but not least, it should be reminded that experiments actually point out that the phase diagram of mechanical denaturation in the temperature-force plane may be more complex than a simple law of the form (cid:181) ( T c - s) T F c anyway [13]. V - EFFECT OF EXCLUDED VOLUME AND GENERALIZED BASE PAIRING In this section, we discuss the effect of two terms, which are included in the original KRSdP model [78] but were discarded in the computations that led to the results presented in Sections III and IV, namely the excluded volume interaction term and the generalization of base pairing to all complementary base pairs. The original KRSdP model indeed contains a seventh term in addition to the six terms of Eq. (II.2). This additional term is an excluded volume interaction term, which insures that sites i and j do not overlap spatially. It writes V ex = e ∑ < j i cut ( rH ij - r ij )[( cut r ij r ij 12 ) - (2 cut r ij r ij + 6 ) ]1 , (V.1) where )( xH is the Heaviside step function, which is equal to 0 if 0<x and to 1 if 0‡x . cut ijr is the threshold at which exV starts repelling site i away from site j. It was assumed in Ref. 19 [78] that cut = ijr 00.1 Å if sites i and j are two mismatched bases and cut = ijr 86.6 Å if sites i and j are any other pair of sites. Included in the sum of Eq. (V.1) are all couples of sites i and j, which do not belong to the same strand and to the same or nearest-neighbour base pairs and, additionally, which do not interact through the pairing, stacking or electrostatic interactions (since these later interactions also have a repulsive core). The essential reason, why exV was excluded from the calculations discussed in Sections III and IV, is that this approximation reduces the required CPU time by a factor 2, while results are expected to remain essentially unchanged at physiological temperatures. Indeed, at these temperatures the sequence is almost entirely double-stranded, so that neighbouring sites are not likely to overlap spatially, while electrostatic repulsion between charged phosphate sites insures that widely separated portions of the sequence do not cross. This is, however, no longer the case close to the critical temperature, where large bubbles are observed. We consequently discuss below to which extent the excluded volume interaction term affects the results presented in the previous sections. Fig. 10 shows the computed temperature evolution of the denaturation/annealing rate for a 480 bp homogeneous sequence with pairing energy of 4.37 kcal mol-1 (like ...GGGGG... sequence in Sections III and IV) when exV is taken into account. It may be seen by comparing this plot to Fig. 4 that exV has two main effects. First, the critical temperature is displaced by about 40 K to lower temperatures, that is, from 375 K down to 335 K. Moreover, denaturation (but not annealing) rates appear to saturate at rather low values, smaller than 0.5 base pairs per nanosecond, when exV is taken into account (compare Fig. 10 with Fig. 4 and Supplementary figure S2 [110]). This is most probably due to the fact that denaturation pathways become more complex when excluded volume repulsion is taken into account. This is a point, which would certainly deserve more attention. 20 Finally, we checked that homogenous sequences with base pairing energies of 4.68 kcal mol-1 denaturate at about 360 K when exV is taken into account (see Fig. 10). Linear extrapolation consequently indicates that the values of pairing energies that lead to correct melting temperatures at =+ ]Na[ 50 mM salt concentration are =e AT 37.4 kcal mol-1 and =e GC =e GC 87.4 kcal mol-1 when exV is taken into account, instead of =e AT 90.3 kcal mol-1 and 37.4 kcal mol-1 when it is neglected. In contrast, we checked that taking exV into account does not modify the value of the 3-dimensional diffusion coefficient of a sequence (see Fig. 5). Therefore, the realistic value for g is still close to 500 ns-1. Similarly, we checked that the temperature evolution of the singular part of the internal energy, singu , still displays a clear step at the critical temperature when exV is taken into account (see Fig. 11). This indicates that exV does not change the predicted order of the denaturation transition, which remains first-order. At last, let us mention that we were unable to check whether the excluded volume term affects the critical exponent of the critical force. The reason is that pulling rates of the order of 2 cm s-1, like that used in Section IV, lead to extremely large computed forces when exV is taken into account. As for denaturation rates, the reason is probably that pathways for denaturation become more complex. The pulling rate should consequently be sufficiently low to provide the system with sufficient time to follow these complex pathways. From a practical point of view, this however means that simulations become much too long to be feasible. The last difference between the model of Section II and the original KRSdP model consists in the fact that it was assumed in Ref. [78] that the base pairing term bpV "describes hydrogen bonding between any complementary base pair and acts both intra- and interstrand" [78], while we instead assumed up to now that bpV connects only bases belonging to the same pair. We launched several simulations to check the influence of such a generalized base 21 pairing scheme. The essential result is that generalized base pairing increases the melting temperature of a sequence by several tens of kelvins. For example, simulations showed that a 480 bp ...GGGGG... sequence is still zipped at 390 K when both the excluded volume term and generalized base pairing are taken into account, while it melts at about 335 K when our simpler base pairing scheme is used. The reason for this discrepancy can be understood by computing the distribution of the number n of bonds that is formed by each base. For a given base i, n can be estimated at a given time t according to = n ∑ j - (5( H 0 r ij r ij 12 ) + (6 0 r ij r ij 10 ) - (5)[ 0 r ij r ij 12 ) + (6 0 r ij r ij 10 ) ] , (V.2) where )( xH is again the Heaviside step function and the sum runs over all bases j that are connected to base i through bpV . For the simple pairing scheme, there is of course only one base j that may contribute to the sum, while for the generalized base pairing scheme any base j that is complementary to base i can potentially contribute. Moreover, only attractive interactions contribute to n (this is the role of the Heaviside step function), with a weight that varies between 1 (when the distance ijr between the two bases is equal to the equilibrium distance 0 ijr ) and 0 (when ijr becomes very large or, conversely, sufficiently small for the interaction to become repulsive). The normalized distributions )(nP are shown in Fig. 12 for a 480 bp homogeneous sequence with a pairing energy of 4.37 kcal mol-1 (like ...GGGGG... sequence in Sections III and IV). Temperature is 330 K for the simple base pairing scheme (dashed blue line) and 390 K for the generalized base pairing scheme (solid red line). Each distribution was obtained by computing n according to Eq. (V.2) at each step and for each base pair of the sequence along a single trajectory integrated for 1.5·106 steps. Not surprisingly, the )(nP distribution exhibits a single peak culminating at 1=n for the simple base pairing scheme. Less expected 22 is the fact that, for the generalized pairing scheme, )(nP additionally exhibits a second peak culminating at 2=n , which is much higher than the peak at 1=n . This indicates that the strands deform in such a way that many bases are able to form two pairing bonds with successive bases of the opposite strand. These double bonds are of course more difficult to break than a single one, which explains why the melting temperature is higher for the generalized base pairing scheme than for the simple one. Needless to say that such pairing of a base with two complementary ones is not physically correct. In real DNA, a base can pair with at most one complementary base, although this base is not necessarily the complementary one located at the same position on the opposite strand, as is assumed by the simple base pairing scheme. We built several other pairing schemes based on the expression of bpV in Eq. (II.2), where each base is allowed to pair with at most one complementary base, typically the nearest one. However, these schemes turned out to be numerically unstable, because rapid switching of the bonds due to the nearest neighbour criterion led to an artificial increase in kinetic energy and hence of the temperature of the sequence. Following these unsuccessful trials, our feeling is that any improvement in the description of base pairing should take the relative orientations of the sugar-base bonds into account, which would "naturally" select the correct complementary base, if any, to pair with. We leave such an improvement for an eventual future work. Here we instead choose to use the simple base pairing scheme, although we are aware of the limitations it brings to the model. The main limitation is probably that transient pairing of partially matched parts of the sequence cannot take place close to the critical temperature, but we are convinced that this neglect has little consequence on the results presented above. As already emphasized in Section II, use of the simple base pairing scheme instead of the generalized one also has the practical consequence that the values of base pairing energies must be increased compared to the original KRSdP model [78]. 23 VI - CONCLUSION In this paper, we have shown that the KRSdP model can be adapted to describe the thermal and mechanical denaturation of long homogeneous and inhomogeneous DNA sequences. This was achieved by using the simple base pairing scheme instead of the generalized originally proposed in Ref. [78] and in consequently refining the two parameters for base pairing interactions, namely =e AT 90.3 kcal mol-1 and =e GC 37.4 kcal mol-1 when the excluded volume interaction term is neglected, and =e AT 37.4 kcal mol-1 and =e GC 87.4 kcal mol-1 when exV is taken into account. These modifications lead to () critical temperatures for long ...AAAAA... and ...GGGGG... sequences, namely 335 and 375 K, which are in good agreement with both experimental ones and those obtained from statistical models, () a realistic step-like denaturation behaviour for inhomogeneous sequences, (*) a critical force at ambient temperature of the order of 10 pN, which is also close to measured values. The modified model furthermore supports the thesis that the thermal denaturation of long homogeneous sequences corresponds to a first-order phase transition. In contrast, it yields a critical exponent for the critical force equal to 70.0=s , which suggests that the scaling law relating the characteristic exponents for specific heat and critical force that was derived for one-dimensional models no longer holds for this more complex model. The KRSdP model with refined base pairing energies therefore represents a good compromise for studying the dynamics of DNA-protein specific interactions at an unprecedented detail level. It is indeed both geometrically and energetically realistic, in the sense that (i) the helical structure and the grooves, where most proteins bind, are satisfactorily 24 reproduced, and (ii) the energy and force required to break a base pair lie in the expected range. We are therefore confident that the combination of this model with a mesoscopic model describing proteins at a comparable level of accuracy will provide new and important insights into this fundamental but very complex problem. 25 REFERENCES [1] R. Thomas, Bull. Sté. Chim. Biol. 35, 609 (1953) [2] R. Thomas, Biochim. Biophys. Acta 14, 231 (1954) [3] J. Marmur, R. Rownd and C.L. Schildkraut, Prog. Nucleic Acid Res. 1, 231 (1963) [4] D. Poland and H.A. Scheraga, Theory of Helix-Coil Transitions in Biopolymers (Academic Press, New York, 1970) [5] R.M. Wartell and A.S. Benight, Phys. Rep. 126, 67 (1985) [6] O. Gotoh, Adv. Biophys. 16, iii (1983) [7] B. Essevaz-Roulet, U. Bockelmann and F. Heslot, Proc. Natl. Acad. Sci. USA 94, 11935 (1997) [8] U. Bockelmann, B. Essevaz-Roulet and F. Heslot, Phys. Rev. E 58, 2386 (1998) [9] G.U. Lee, L.A. Chrisey and R.J. Colton, Science 266, 771 (1994) [10] M. Rief, H. Clausen-Schaumann and H.E. Gaub, Nat. Struct. Biol. 6, 346 (1999) [11] T. Strunz, K. Oroszlan, R. Sch äfer and H.J. Gü ntherodt, Proc. Natl. Acad. Sci. USA 96, 11277 (1999) [12] C. Danilowicz, V.W. Coljee, C. Bouzigues, D.K. Lubensky, D.R. Nelson and M. Prentiss, Proc. Natl. Acad. Sci. USA 100, 1694 (2002) [13] C. Danilowicz, Y. Kafri, R.S. Conroy, V.W. Coljee, J. Weeks and M. Prentiss, Phys. Rev. Lett. 93, 078101 (2004) [14] J.D. Weeks, J.B. Lucks, Y. Kafri, C. Danilowicz, D.R. Nelson and M. Prentiss, Biophys. J. 88, 2752 (2005) [15] K. Hatch, C. Danilowicz, V. Coljee and M. Prentiss, Phys. Rev. E 75, 051908 (2007) [16] Y. Gao and E.W. Prohofsky, J. Chem. Phys. 80, 2242 (1984) [17] M. Techera, L.L. Daemen and E.W. Prohofsky, Phys. Rev. A 40, 6636 (1989) 26 [18] M. Peyrard and A.R. Bishop, Phys. Rev. Lett. 62, 2755 (1989) [19] T. Dauxois, M. Peyrard and A.R. Bishop, Phys. Rev. E 47, 684 (1993) [20] T. Dauxois, M. Peyrard and A.R. Bishop, Phys. Rev. E 47, R44 (1993) [21] T. Dauxois and M. Peyrard, Phys. Rev. E 51, 4027 (1995) [22] N. Theodorakopoulos, T. Dauxois and M. Peyrard, Phys. Rev. Lett. 85, 6 (2000) [23] M. Joyeux and S. Buyukdagli, Phys. Rev. E 72, 051902 (2005) [24] R. Tapia-Rojo, J.J. Mazo and F. Falo, Phys. Rev. E 82, 031916 (2010) [25] D. Poland and H.A. Scheraga, J. Chem. Phys. 45, 1456 (1966) [26] D. Poland and H.A. Scheraga, J. Chem. Phys. 45, 1463 (1966) [27] D. Poland, Biopolymers 13, 1859 (1974) [28] M. Fixman and J.J. Freire, Biopolymers 16, 2693 (1977) [29] J. SantaLucia, Proc. Natl. Acad. Sci. USA 95, 1460 (1998) [30] R.D. Blake and S.G. Delcourt, Nucleic Acids Res. 26, 3323 (1998) [31] R.D. Blake, J.W. Bizzaro, J.D. Blake, G.R. Day, S.G. Delcourt, J. Knowles, K.A. Marx and J. SantaLucia, Bioinformatics 15, 370 (1999) [32] R. Blossey and E. Carlon, Phys. Rev. E 68, 061911 (2003) [33] M. Barbi, S. Cocco and M. Peyrard, Phys. Lett. A 253, 358 (1999) [34] M. Barbi, S. Lepri, M. Peyrard and N. Theodorakopoulos, Phys. Rev. E 68, 061909 (2003) [35] T. Michoel and Y. van de Peer, Phys. Rev. E 73, 011908 (2006) [36] S. Buyukdagli, M. Sanrey and M. Joyeux, Chem. Phys. Lett. 419, 434 (2006) [37] J. Palmeri, M. Manghi and N. Destainville, Phys. Rev. Lett. 99, 088103 (2007) [38] J. Palmeri, M. Manghi and N. Destainville, Phys. Rev. E 77, 011913 (2008) [39] H.M. Jian, A.V. Vologodskii and T. Schlick, Journal of Computational Physics 136, 168 (1997) 27 [40] K. Klenin, H. Merlitz and J. Langowski, Biophys. J. 75, 780 (1998) [41] J. Huang and T. Schlick, J. Chem. Phys 117, 8573 (2002) [42] G. Arya, Q. Zhang and T. Schlick, Biophys. J. 91, 133 (2006) [43] J.D. Dwyer and V.A. Bloomfield, Biophys. J. 65, 1810 (1993) [44] J.D. Dwyer and V.A. Bloomfield, Biophysical Chemistry 57, 55 (1995) [45] T. Wocjan, K. Klenin and J. Langowski, J. Phys. Chem. B 113, 2639 (2009) [46] O.G. Berg, R.B. Winter and P.H. von Hippel, Biochemistry 20, 6929 (1981) [47] R.B. Winter, O.G. Berg, and P.H. von Hippel, Biochemistry 20, 6961 (1981) [48] O.G. Berg and M. Ehrenberg, Biophys. Chem. 15, 41 (1982) [49] P.H. von Hippel and O.G. Berg, J. Biol. Chem. 264, 675 (1989) [50] S.E. Halford and J.F. Marko, Nucleic Acids Res. 32, 3040 (2004) [51] M. Slutsky and L.A. Mirny, Biophys. J. 87, 4021 (2004) [52] T. Hu, A.Y. Grosberg, and B.I. Shklovskii, Biophys. J. 90, 2731 (2006) [53] K.V. Klenin, H. Merlitz, J. Langowski, and C.-X. Wu, Phys. Rev. Lett. 96, 018104 (2006) [54] T. Ambj örnsson and R. Metzler, J. Phys. : Cond ens. Matter 17, S1841 (2005) [55] M.A. Lomholt, T. Ambj örnsson and R. Metzler, P hys. Rev. Lett. 95, 260603 (2005) [56] M.A. Lomholt, B. van den Broek, S.-M. J. Kalisch, G.J.L. Wuite and R. Metzler, Proc. Natl. Acad. Sci. USA 106, 8204 (2009) [57] O. Bénichou, Y. Kafri, M. Sheinman and R. Voituriez, Phys. Rev. Lett. 103, 138102 (2009) [58] O. Bénichou, C. Chevalier, J. Klafter, B. Meyer and R. Voituriez, Nature Chemistry 2, 472 (2010) [59] A.-M. Florescu and M. Joyeux, J. Chem. Phys. 130, 015103 (2009) [60] A.-M. Florescu and M. Joyeux, J. Chem. Phys. 131, 105102 (2009) 28 [61] A.-M. Florescu and M. Joyeux, J. Phys. Chem. A 114, 9662 (2010) [62] P.C. Blainey, A.M. van Oijen, A. Banerjee, G.L. Verdine, and X.S. Xie, Proc. Natl. Acad. Sci. USA 103, 5752 (2006) [63] I. Bonnet, A. Biebricher, P.-L. Porté, C. Loverdo, O. Bénichou, R. Voituriez, C. Escudé, W. Wende, A. Pnigoud et P. Desboilles, Nucleic Acids Res. 36, 4118 (2008) [64] B. Bagchi, P.C. Blainey and X.S. Xie, J. Phys. Chem. B 112, 6282 (2008) [65] J. Elf, G.W. Li and X.S. Xie, Science 316, 1191 (2007) [66] Y.M. Wang, R.H. Austin and E.C. Cox, Phys. Rev. Lett. 97, 048302 (2006) [67] A. Tafvizi, F. Huang, J.S. Leith, A.R. Fersht, L.A. Mirny and A.M. van Oijen, Biophys. J. 95, L1 (2008) [68] J.H. Kim and R.G. Larson, Nucleic Acids. Res. 35, 3848 (2007) [69] A. Granéli, C.C. Yeykal, R.B. Robertson and E.C. Greene, Proc. Nat. Acad. Sci. USA 103, 1221 (2006) [70] J. Gorman, A. Chowdhury, J.A. Surtees, J. Shimada, D.R. Reichman, E. Alani, and E.C. Greene, Mol. Cell 28, 359 (2007) [71] N. Bruant, D. Flatters, R. Lavery and D. Genest, Biophys. J. 77, 2366 (1999) [72] H.L. Tepper and G.A. Voth, J. Chem. Phys. 122, 124906 (2005) [73] R.K.Z. Tan and S.C. Harvey, J. Mol. Biol. 205, 573 (1989) [74] M.A. El Hassan and C.R. Calladine, J. Mol. Biol. J. Mol. Biol. 251, 648 (1995) [75] K. Drukker and G.C. Schatz, J. Phys. Chem. B 104, 6108 (2000) [76] K. Drukker, G. Wu and G.C. Schatz, J. Chem. Phys. 114, 579 (2001) [77] M. Maciejczyk, W.R. Rudnicki and B. Lesyng, J. Biomol. Struct. Dyn. 17, 1109 (2000) [78] T.A. Knotts, N. Rathore, D.C. Schwartz and J.J. de Pablo, J. Chem. Phys. 126, 084901 (2007) [79] E.J. Sambriski, V. Ortiz and J.J. de Pablo, J. Phys. : Condens. Matter 21, 034105 (2009) 29 [80] E.J. Sambriski, D.C. Schwartz and J.J. de Pablo, Biophys. J. 96, 1675 (2009) [81] P.J. Struther Arnott, C. Smith and R. Chandrasekaran, Atomic coordinates and molecular conformations for DNA-DNA, RNA-RNA and DNA-RNA helices, vol. 2 of CRC Handbook of biochemistry and molecular biology (CRC Press, Cleveland, 1976), pp. 411-422 [82] See Supplementary Material Document No._________ for Figure S1, which shows the temperature evolution of denaturation/annealing rates for a 480 bp ...AAAAA... sequence when =e AT 77.2 kcal mol-1. [83] R. Balian, From microphysics to macrophysics - Methods and applications of statistical physics (Springer-Verlag, Berlin, 1991) [84] S. Buyukdagli and M. Joyeux, Phys. Rev. E 76, 021917 (2007) [85] R.B. Inman and R.L. Baldwin, J. Mol. Biol. 8, 452 (1964) [86] E. Tøstesen, F. Liu, T.-K. Jenssen and E. Hovi g, Biopolymers 70, 364 (2003) [87] D. P örschke and M. Eigen, J. Mol. Biol. 62, 36 1 (1971) [88] M.E. Craig, D. Crother and P. Doty, J. Mol. Biol. 62, 383 (1971) [89] R. Pecora, Science 251, 893 (1991) [90] M. Joyeux and A.-M. Florescu, J. Phys.: Condens. Matter 21, 034101 (2009) [91] A. Campa and A. Giansanti, Phys. Rev. E 58, 3585 (1998) [92] C. Vanderzande, Lattice Models of Polymers (Cambridge University Press, Cambridge, 1998) [93] M.E. Fisher, J. Chem. Phys. 44, 616 (1966) [94] Y. Kafri, D. Mukamel and L. Peliti, Phys. Rev. Lett. 85, 4988 (2000) [95] E. Carlon, E. Orlandini and A.L. Stella, Phys. Rev. Lett. 88, 198101 (2002) [96] M. Baiesi, E. Carlon and A.L. Stella, Phys. Rev. E 66, 021804 (2002) [97] L.P. Kadanoff, W. Götze, D. Hamblen, R. Hecht, E.A.S. Lewis, V.V. Palciauskas, M. Rayl, J. Swift, D. Aspnes and J. Kane, Rev. Mod. Phys. 39, 395 (1967) 30 [98] H.B. Weber, T. Werner, J. Wosnitza, H. v. Löhn eysen and U. Schotte, Phys. Rev. B 54, 15924 (1996) [99] J.A. Souza, Y.K. Yu, J.J. Neumeier, H. Terashita and R.F. Jardim, Phys. Rev. Lett. 94, 207209 (2005) [100] M. Massot, A. Oleaga, A. Salazar, D. Prabhakaran, M. Martin, P. Berthet and G. Dhalenne, Phys. Rev. B 77, 134438 (2008) [101] S. Buyukdagli and M. Joyeux, Phys. Rev. E 73, 051910 (2006) [102] S. Buyukdagli and M. Joyeux, Chem. Phys. Lett. 484, 315 (2010) [103] Y.-L. Zhang, W.-M. Zheng, J.-X. Liu and Y.Z. Chen, Phys. Rev. E 56, 7100 (1997) [104] S. Buyukdagli and M. Joyeux, Phys. Rev. E 77, 031903 (2008) [105] M. Joyeux, S. Buyukdagli and M. Sanrey, Phys. Rev. E 75, 061914 (2007) [106] E. Carlon, M.L. Malki and R. Blossey, Phys. Rev. Lett. 94, 178101 (2005) [107] N. Singh and Y. Singh, Eur. Phys. J. E 17, 7 (2005) [108] S. Cocco, R. Monasson and J.F. Marko, Phys. Rev. E 65, 041907 (2002) [109] Y. Kafri, D. Mukamel and L. Peliti, Eur. Phys. J. B 27, 135 (2002) [110] See Supplementary Material Document No._________ for Figure S2, which shows the results of Fig. 4 on broader temperature and rate scales. 31 FIGURE CAPTIONS Figure 1 : Schematic representation of the model, showing the positions at equilibrium of the various sites for a short stretch of the actin sequence discussed in Section III. Letter P indicates a phosphate group (red), S a sugar group (green), and B a base (blue). Figure 2 : Snapshots of the denaturation of the 480 bp ...AAAAA... homogeneous sequence (left) and the 1793 bp inhomogeneous actin sequence (NCB entry code NM_001101) (right). Figure 3 : Time evolution of the number of open base pairs for the 480 bp ...AAAAA... sequences at 335 K (top plot) and 340 K (bottom plot). In the bottom plot, each curve corresponds to a single trajectory obtained with a different set of random numbers. A given base pair is considered as open if its pairing energy is smaller than e 10/AT . The critical temperature of this sequence is very close to 335 K. Figure 4 : Denaturation/annealing rates for 480 bp ...AAAAA... (blue ·) and ...GGGGG... (red +) sequences as a function of temperature, expressed in base pairs per nanosecond. Positive (respectively, negative) rates correspond to denaturation (respectively, annealing). The vertical dash-dotted lines indicate the positions of the critical temperatures (335.0 and 374.75 K, respectively) where the rates are zero. These rates were obtained by starting simulations with half-open sequences like the one shown in the left side of Fig. 2 and in checking how long it takes for these sequences to open or close completely. Figure 5 : Evolution with g of the 3-dimensional diffusion coefficient 3DD of a 367 bp double-stranded ...AAAAA... sequence at 298 K. g is expressed in s-1 and 3DD in m2 s-1. Red 32 circles (respectively, blue squares) denote results obtained without (respectively, with) the excluded volume term discussed in Section V. Values of 3DD were computed by adjusting the mean squared deviation of the center of mass of the sequence against the < - r )( t r )0( 2 >= 6 tD 3D law. The solid line shows the »D 3D g/94.7 law, which best interpolates between the computed points. The horizontal dot-dashed line shows the experimentally determined value of the diffusion coefficient of a 367 bp sequence, that is =D 3D 58.1 · 10 - 11 m2 s-1 [89]. Figure 6 : Temperature evolution of u and nsu (bottom plot) and singu- (top plot) for the 480 bp ...AAAAA... sequence. u (respectively, singu ) denotes the average internal energy per base pair for the double-stranded (respectively, single-stranded) sequence, while u The vertical dash-dotted lines indicate the position of the critical temperature -= uu . ns sing = cT 335 K. Remind that u = nsu and u sing = 0 above cT . Figure 7 : Denaturation curve of the 1793 bp human β-actin cDNA sequence (NCB entry code NM_001101) obtained from the statistical model of Ref. [86] by using the parameters of Blossey and Carlon [32] and a salt concentration =+ ]Na[ 50 mM (dashed line), and from molecular dynamics simulations performed with the model described in Section II (empty circles). A base pair is considered as open if its pairing energy is on average smaller than e 10/AT or e 10/GC . Figure 8 : Plot, as a function of d, of the average force F that must be exerted to maintain the two phosphate groups linked to the first base pair of the 480 bp ...AAAAA... sequence separated by a distance d at =T 280 K. 0=d corresponds to the equilibrium distance of 33 the two phosphates. F is the force projected on the separation vector and averaged over 0.25 ns. The insert shows in more detail the large force barrier thrF that occurs at short distances. Experimentally measured critical forces cF correspond to the asymptotic value of F for large values of d, that is, to the average force that must be exerted on the widely separated extremities of long sequences to prevent them from closing. Figure 9 : Temperature evolution of the threshold force thrF (top plot) and the critical force cF (bottom plot) for the 480 bp ...AAAAA... sequence. thrF was obtained as the maximum of F in force-separation curves like that shown in Fig. 8 (averaging time for each point is 0.25 ns). These values are represented as brown · in the top plot. cF was obtained from force- separation curves as the average of F for all separations comprised between 100 and 200 Å (averaging time for each point is consequently 0.5 ms). These values are represented as blue crosses in the bottom plot. Solid lines show the best adjustment of a linear law (top plot) and a power law (bottom plot) against computed forces. Figure 10 : Denaturation/annealing rates for 480 bp homogeneous sequences with pairing energies of 4.37 kcal mol-1 (blue ·) and 4.68 kcal mol-1 (red +) when the excluded volume term exV is taken into account. Positive (respectively, negative) rates correspond to denaturation (respectively, annealing). The vertical dash-dotted lines indicate the positions of the critical temperatures (about 335 and 360 K, respectively). These rates were obtained by starting simulations with half-open sequences, like the one shown in the left side of Fig. 2, and in checking how long it takes for these sequences to open or close completely. 34 Figure 11 : Same as Fig. 6, except that the excluded volume interaction term of Eq. (V.1) is taken into account in the expression of the Hamiltonian and the investigated sequence has a base pairing energy of 4.37 kcal mol-1, as ...GGGGG... sequences in Sections III and IV. Critical temperature is close to 335 K. Figure 12 : Distribution of the number of bonds n formed by each base for a 480 bp homogeneous sequence with a pairing energy of 4.37 kcal mol-1, as ...GGGGG... sequences in Sections III and IV. Temperature is =T 330 K for the simple base pairing scheme (dashed blue line) and =T 390 K for the generalized base pairing scheme (solid red line). Unpaired bases are not taken into account in the statistics. 35 FIGURE 1 36 FIGURE 2 37 FIGURE 3 38 FIGURE 4 39 FIGURE 5 40 FIGURE 6 41 FIGURE 7 42 FIGURE 8 43 FIGURE 9 44 FIGURE 10 45 FIGURE 11 46 FIGURE 12 47 SUPPLEMENTARY FIGURE S1 Figure S1 : Denaturation/annealing rates for a 480 bp ...AAAAA... sequence as a function of =e kcal mol-1) is assumed 77.2 temperature, when the base pairing energy of Ref. [78] ( AT =e kcal mol-1). Positive (respectively, negative) 90.3 instead of that derived in Section II ( AT rates correspond to denaturation (respectively, annealing). The vertical dash-dotted line indicates the position of the critical temperature (about 230 K) where the rate is zero. These rates were obtained by starting simulations with half-open sequences like the one shown in the left side of Fig. 2 and in checking how long it takes for these sequences to open or close bpV acts only between bases belonging to the completely. It was furthermore assumed that same pair i, as for the results displayed in Fig. 4. 48 SUPPLEMENTARY FIGURE S2 Figure S2 : Same as Fig. 4, but on broader temperature and rate scales. 49
1306.1207	2	1306	2013-06-06T13:53:01	Neuronal Alignment On Asymmetric Textured Surfaces	[ "physics.bio-ph", "cond-mat.mtrl-sci", "q-bio.NC" ]	Axonal growth and the formation of synaptic connections are key steps in the development of the nervous system. Here we present experimental and theoretical results on axonal growth and interconnectivity in order to elucidate some of the basic rules that neuronal cells use for functional connections with one another. We demonstrate that a unidirectional nanotextured surface can bias axonal growth. We perform a systematic investigation of neuronal processes on asymmetric surfaces and quantify the role that biomechanical surface cues play in neuronal growth. These results represent an important step towards engineering directed axonal growth for neuro-regeneration studies.	physics.bio-ph	physics	NEURONAL ALIGNMENT ON ASYMMETRIC TEXTURED SURFACES Ross Beighley1,&, Elise M. Spedden1,&, Koray Sekeroglu2,&, Timothy Atherton1, Melik C. Demirel2,, Cristian Staii1, 1. Department of Physics and Astronomy and Center for Nanoscopic Physics, Tufts University, Medford, MA 02155 2. Materials Research Institute and Department of Engineering Science, Pennsylvania State University, University Park, PA, 16802 [*] Corresponding Authors: Prof. M. C. Demirel, E-mail: [email protected] and Prof. C. Staii, E-mail: [email protected] [&] These authors have contributed equally to the paper Keywords: Neuron, Asymmetry, Texture, Poly(p-xylylene), Fokker-Planck Abstract Axonal growth and the formation of synaptic connections are key steps in the development of the nervous system. Here we present experimental and theoretical results on axonal growth and interconnectivity in order to elucidate some of the basic rules that neuronal cells use for functional connections with one another. We demonstrate that a unidirectional nanotextured surface can bias axonal growth. We perform a systematic investigation of neuronal processes on asymmetric surfaces and quantify the role that biomechanical surface cues play in neuronal growth. These results represent an important step towards engineering directed axonal growth for neuro-regeneration studies. 2 Artificial growth of neurons on various substrates is of great interest for brain tissue engineering 1,2. Neuronal cells in the brain develop two types of processes: a single, long axon that transmits information to other cells and multiple, shorter dendrites that receive electrical impulses from the axons of other neurons. Neuronal cells have been cultured on a variety of scaffolds including biopolymers, silk, and hydrogels 3 as well as grown in alignment on patterned surfaces 4,5. These surfaces also provide physical guidance, and chemical support for neuronal cell adherence, axonal extension, network formation, and function. The axons, and in particular their dynamic unit known as the growth cone are able to detect and respond to environmental signals such as functionalization of surfaces with extracellular matrix proteins, biomolecules released by neighboring neurons at extremely low concentrations (molecular level), substrate stiffness and topographical and geometrical cues 6. Over the past decade, there has been rapid progress in our understanding of the role played by chemical signaling and surface-based biochemical guidance on the growth cone dynamics and axonal elongation. For example, it is known that axonal navigation to their target depends on the precise arrangement of extracellular proteins on the growth surfaces 2,6,7. It is also now recognized that mechanical interactions between neurons and their environment are playing an essential role in neuronal growth and development 5,8. However, the neuronal response to mechanical and topographical stimuli, and the details of cell-surface interactions such as adhesion forces and traction stress generated during growth are currently poorly understood 9,10. Directional surfaces composed of asymmetric structures are widely used in nature for wet and dry adhesion 11. Inspired by these surfaces, Demirel et al. synthesized asymmetric textured surface 12 and reported an engineered nanotextured surface deriving its anisotropic adhesive wetting directly from its asymmetric nanoscale roughness 13. In an earlier study, Demirel et al. studied the fibroblast adhesion and removal on directional nanofilms 14, using a fluidic shear stress to remove cells from a microfluidic channel. It has been shown that cells were removed with lower shear stresses when the flow was in the direction of nanorod tilt, compared to flow against the tilt 14. Adhesion and retraction under asymmetric mechanical cues demonstrated unique properties 15. Cell polarization (i.e. response to external cues such as chemical gradients and mechanical deformation) has been studied extensively on textured surfaces to understand cell fate 16. However, unidirectional polarization in response to surface mechanical cues has not been demonstrated earlier. Here, we report axonal extension and network formation on asymmetric nanotextured surfaces. We demonstrate that axons preferentially extend along the asymmetry of the surface texture, and display an angular distribution that broadens with the increase in the surface density of the cells. We also show that a simple theoretical model, based on Brownian motion in constant field, can help to interpret the experimental results. This opens up the possibility of performing systematic studies in which the influence of different types of mechanical and topographical guidance cues could be precisely quantified. Our results could also lead to creating neural networks where signaling pathways could be studied in detail. Figure 1 shows the schematic of growth cone adhesion and axon alignment on the directional nanotextured surface. We have been exploring nanotextured poly(chloro-p-xylylene) surfaces, which formed through vapor-phase polymerization and directed deposition of [2.2] paracyclophane derivatives as biocompatible templates. Figure 1a (top) is a schematic representation of our approach for unidirectional axon growth. Figure 1a also shows experimental results from fluorescence (middle) and bright field (bottom) imaging, which 3 support the unidirectional model. Figure 1b shows cross sectional Scanning Electron Microscope (SEM) and Atomic Force Microscope (AFM) images of a nanotextured surface. (a) (b) axon under tension new actin new adhesion !" Tilt Direction 80 60 40 20 m µ 10 µm 20 40 0 0 10 µm 200 100 0 m n -100 -200 60 80 µm Figure 1. (a) top: Schematics of the growth cone adhesion to the asymmetric nanotextured surface, where β is the angle of the tilted nanorods. The axons extend preferentially in a direction opposite to the direction of the tilted nanorods. The direction of growth is defined as the reference direction (i.e. the 0 radians direction) for the analysis presented in the paper. Fluorescence (middle) and bright field (bottom) images of two different cells growing on the surface are also shown. (b) Scanning Electron Microscopy (top) and Atomic Force Microscope (bottom) image of the surface. Polymer deposition was performed using a modified Deposition System PDS 2010 (SCS, Indianapolis, IN, USA) as described earlier 12. Glass substrates are cleaned with ethanol and acetone, and subsequently treated with a silane layer to improve the adhesion of polymer to the substrate. The film is deposited using the source material, (2,2)-dichloroparacyclophane in powder form (Uniglobe-Kisco, Whiteplains, NY, USA) to produce a directed vapor flux onto a substrate through the nozzle by tilting the substrate 10 degrees. Conventional poly(chloro-p- xylylene) deposition parameters are adopted for sublimation (175ºC) and pyrolysis (690 C and 32 Torr) of the monomer. These nanotextured surfaces were affixed to 3.5 cm glass disks using silicone glue and allowed to dry. Each surface was rinsed with sterile water, then spin-coated with 3 mL of Poly-D-lysine (PDL) (Sigma-Aldrich, St. Louis, MO) solution (0.1 mg/mL) at 1000 RPM for 10 minutes conformally. The plates were then sterilized using ultraviolet light for 4 ≥30 minutes. The surface roughness of the nanotextured surface was measured via AFM topography before and after the application of PDL and no significant difference was observed. !/2 30 (a) ! axon "# 0 (b) t n u o C n o x A 20 10 -!/2 -!/2 0 !/2 25 (c) 50 t n u o C n o x A (d) 0 90 0 -!/2 0 !/2 60 30 t n u o C n o x A (e) 50 t n u o C n o x A 25 0 !" !" !" 50 25 0 90 60 30 0 50 25 0 0 -! 0 ! 50 # 25 !/2 ! -!/2 0 0 -! 0 ! ! Π Π 90 60 30 # !/2 ! -!/2 0 0 -! 0 ! ! Π Π 50 25 # !/2 ! -!/2 "# 0 0 ! Π Π -! 0 ! "# -!/2 0 !/2 Angular distribution, "# Figure 2. (a) The direction of axon is shown in a schematic as the direction opposite to the direction of the nanotextured surface. Angular distributions for neural growth at (b) glass surface (control) (c) 2000 cells/cm2, (d) 6000 cells/cm2 and (e) 25000 cells/cm2. All angles are measured with respect to this direction. All plots on (c)-(e) are on the same vertical scale. The bins at π radians for each histogram shown in the middle column collect axon counts from the bins at π and -π of the corresponding histogram shown in third column. The data shows weaker preferential growth in the direction of the nanotextured surface (θ=π) than in the opposite direction (θ=0). Solid lines in the right hand column histograms represent theoretical fits using a model of Brownian motion in constant field. 5 Figure 2 shows the histograms of axon orientation with respect to the tilting direction of asymmetric nanorods. The bright field images are collected from a Nikon Eclipse ME 600 microscope, and the fluorescent images are collected on the inverted stage (Nikon Eclipse Ti) of an Asylum Research MFP3D Atomic Force Microscope. All images are analyzed using the Image J (http://rsweb.nih.gov/ij) computer software. The cells were grown in three different cell densities. Rat cortices were obtained from embryonic day 18 rats (Tufts Medical School). The corticies were incubated in 5 mL of trypsin at 37ºC for 20 minutes, then the trypsin was inhibited with 10 mL of neurobasal medium (Life Technologies) supplemented with GlutaMAX, b27 (Life Technologies), and antibotics (penicillin/streptomycin), containing 10 mg of soybean trypsin inhibitor (Life Technologies). The neurons were then mechanically dissociated, centrifuged, the supernatant removed, and the cells were resuspended in 20 mL of neurobasal medium containing l-glutamate (Sigma-Aldrich, St. Louis, MO). The cells were re-dispersed with a pipette, counted, and plated on the nanotextured surface at three different densities of 2000, 6000 and 25000 cells/cm2. For fluorescence imaging, the live cortical samples were incubated for 30 minutes at 37ºC with 100 nM Tubulin Tracker Green (Oregon Green 488 Taxol, bis-Acetate) (Life Technologies, Grand Island, NY) in PBS. Fluorescence images were taken using a standard Fluorescein isothiocyanate (FITC) filter with excitation and emission of 495 nm and 521 nm respectively. Figure 2 shows histograms for all experiments as well as theoretical fits from the Fokker- Plank model as explained in the next paragraph. Figure 2a show schematic of axon orientation as a function of angular distribution, q. Figure 2b shows angular distribution of neuronal growth on PDL/glass surface (i.e. control study). As expected the data for these samples shows non- polarized neuronal growth, with uniform angular distributions for axons. Figure 2c, 2d, and 2e show cell density histograms for 2000, 6000, and 25000 cells/cm2, respectively. Axon orientation in Figure 2c shows a clear peak at 0 radians, which is defined as the direction opposite to the directionality of the nanotextured surface (see Figure 1). The data for the two low densities (2000, 6000 cells/cm2, respectively) also shows a peak at π radians, that is in the direction of the nanotextured surface. As the cell density increases, the standard deviation of the corresponding histogram increases, which reflects the fact that the axons are making more connections at higher densities, therefore deviating from the surface tilting direction. This is consistent with the fact that the neuron-neuron chemical signaling becomes more important as the cell density increases 9,17; therefore inducing more turns in the growth cone. To understand the axonal growth on asymmetric surfaces, we developed a simplified model based on Brownian motion in constant field. Similar models have been adopted for axon following Langevin equation, !! = −!! + ! + ! ! , whre ! is an effective mass, t is the time, growth in the literature for symmetric surfaces 18,19. To interpret the observed alignment of the ! is a Stokes drag coefficient, ! is a constant force and !(!) is a random force which has zero axonal growth along the asymmetric surfaces, we derived the expected distribution of the tangent mean ! = 0 and is Markovian !(!) ∙ !(!′) = !!Γ!(! − ! ! ) where Γ represents the strength of angles from a biased random walk model. The motion of the growth cone is described by the environment. In the absence of such a force, ! ! = 0, the model is deterministic and the equilibrium solution is motion with a constant velocity, !! = !/! , where the reduced drag the noise and d is the Dirac delta function 18. The random force in the model causes the ! = !/! is introduced. stochastic motion of the growth cone observed experimentally as it explores the local 6 force ! and may also cause the distribution of the random forces to be uniaxial rather than i.e. when the angle ! between the posts and surface normal is ! = 0, then ! = 0 and ! ! is The directed surface enters the model in two ways: it is assumed to provide a constant distributed symmetrically. While it is clear on symmetry grounds that if ! = !/2 then ! = 0 the precise dependence of these quantities on ! requires a more detailed model of the interaction of azimuthally symmetric. Hence, if the alignment of the posts were parallel to the surface normal, model, which correspond to observables in the experiment: 1) ! = !! !! Γ , which quantifies the strength of the effective force imparted by the surface onto the growth cone ( ! corresponds to the growth cone with the posts than pursued here. the bias parameter in a random walk); and 2) d, satisfying the requirement −1 ≤ ! ≤ +1 (see To describe the growth anisotropy we introduce two dimensionless parameters in our distribution of the random force field ! ! . Figure 3 provides ensembles of trajectories for equation 1 below), which is a parameter characterizing the degree of anisotropy of the various values of these two parameters. κ 0.3 0.2 0.1 0 0 0.25 δ 0.5 0.9 Figure 3. Ensembles of 100 trajectories generated from a biased, symmetric (δ = κ = 0) and asymmetric random walk for various values of bias parameter κ and azimuthal anisotropy parameter δ. Starting points are indicated with white circles. !"(!, !) ! !, ! + 1 + ! ! !!! ! + 1 − ! ! !!! ! ! !, ! (1) !" = ∇ ! ! − !! The motion of a dilute (non-interacting) ensemble of growth cones is described by the following Fokker-Planck equation: where −1 ≤ ! ≤ +1. For d equals to zero, the cylindrically symmetric distribution is recovered force !, and consequently !! , is directed along the !-axis, the equilibrium solution of (1) is: (see Figure 3 for d = k = 0). For limiting values of d = ±1, the noise is purely along the x (d = +1) or y (d = -1) axes exclusively and for other values the distribution is ellipsoidal. The factors (1 + d) and (1 - d) in equation (1) are required to preserve the overall normalization of the noise. If the 7 1 + ! + !! !1 − ! !Γ 1 − ! ! Exp − !Γ ! !! − !! ! (2) ! ! = Switching to polar coordinates ! = !(cos ! , sin !) and integrating over ! , we obtain the ! ! !!!! 1 − ! ! + !" 1 − ! ! !! !"#! ! 12!" ! 1 − ! !! ! 1 + Erf ! ! = following functional form for the distribution of the tangent angles in an ensemble of axons: ! ! 1 + ! ! ! (3) where the two functions ! ! = 1 − ! cos 2! and ! ! = ! ! / cos ! have been defined together with the parameter ! = !! !! Γ. It is instructive to examine two limiting cases for k, when ! = 0. For small force where ! ≪ 1 we have that: ! ! ≈ 12! + 12 !! cos ! + ⋯ , (4) while for large force, i.e. ! !! ≪ 1, the distribution approaches a Gaussian: ! ! ≈ !! ! !!!! . (5) Figure 2 shows experimental distributions obtained for various cell densities N, as well as fits of these distributions with the theoretical model (p(q) given by eqn. 3). Theoretical fitted values for k and d are: 0.17±0.05 and 0.44±0.08 for 2000 cells/cm2, 0.15±0.03 and 0.06±0.06 for 6000 cells/cm2, and 0.04±0.01 and 0.17±0.03 for 25000 cells/cm2 respectively. The fitted value of k clearly decreases proportionally with increasing cell seeding density N. This is expected as k represents a balance between the applied force on the growth cones, which is a property of the surface and hence should be identical for each experiment, and the random forces from chemotactic signaling described by the reduced drag term g. These fitted values are consistent with the physical expectation that g ~ N. The asymmetry parameter d was found to vary between 0.06 and 0.44, which indeed confirms that the presence of the surface induces an anisotropy in the chemotactic signaling. We note that the asymmetry parameter, d, should be only considered as a theoretical qualitative estimate, which is needed in order to introduce anisotropy in the distribution of the random force field. However, our simple model captures the main features of axonal elongation on these asymmetric surfaces. In conclusion, we have demonstrated that a unidirectional surface can bias axonal growth. By varying the density of neuron cells on asymmetric textured surfaces, we showed the competition between mechanical and chemical (neuronal signaling) on these surfaces. The unidirectional mechanical cues dominate cell growth at low cell densities, while the spatially symmetric chemical cues start to play an increasingly important role at higher densities. We also note that directional axonal growth with mechanical cues has potential applications in peripheral 8 3 4 and spinal cord injuries. We plan to extend our studies to high-throughput assays for neural cue studies in the near future. Author Contributions: MCD and CS planned and supervised the research. KS prepared the nanotextured surfaces, EMS and RB studied the neural growth, TA developed the theoretical model for axonal growth. All authors contributed to writing and revising the manuscript, and agreed on its final contents. References 1 Z. Wen and J. Q. Zheng, Curr. Opin. Neurobiol. 16 (1), 52 (2006). 2 A. B. Huber, A. L. Kolodkin, D. D. Ginty, and J. F. Cloutier, Annu. Rev. Neurosci. 26, 509 (2003). A. R. Nectow, K. G. Marra, and D. L. Kaplan, Tissue engineering. Part B, Reviews 18 (1), 40 (2012); M. Hronik-Tupaj and D. L. Kaplan, Tissue engineering. Part B, Reviews 18 (3), 167 (2012). C. Staii, C. Viesselmann, J. Ballweg, L. Shi, G.-Y. Liu, J.C. Williams, E.W. Dent, S.N. Coppersmith, and M.A. Eriksson, Biomaterials 30, 3397 (2009); C. Staii, C. Viesselmann, J. Ballweg, L. Shi, G.-Y. Liu, J.C. Williams, E.W. Dent, S.N. Coppersmith, and M.A. Eriksson, Langmuir 27, 233 (2011); G. S. Withers, C. D. James, C. E. Kingman, H. G. Craighead, and G. A. Banker, J Neurobiol 66 (11), 1183 (2006); A. A. Oliva, Jr., C. D. James, C. E. Kingman, H. G. Craighead, and G. A. Banker, Neurochem. Res. 28 (11), 1639 (2003); Y. Nam, J. Chang, D. Khatami, G. J. Brewer, and B. C. Wheeler, IEE Proc. Nanobiotechnol. 151 (3), 109 (2004). H. Francisco, B. B. Yellen, D. S. Halverson, G. Friedman, and G. Gallo, Biomaterials 28 (23), 3398 (2007). L.A. Lowery and D. van Vactor, Nature Rev. Mol. Cell Biol. 10, 332 (2009). B. J. Dickson, Science 298 (5600), 1959 (2002). P. Lamoureaux, G. Ruthel, R. E. Buxbaum, and S. R. Heidemann, Journal Cell Biology 159, 499 (2002); K. Franze, J. Gerdelmann, M. Weick, T. Betz, S. Pawlizak, M. Lakadamyali, J. Bayer, K. Rillich, M. Gogler, Y. B. Lu, A. Reichenbach, P. Janmey, and J. Kas, Biophys. J. 97 (7), 1883 (2009). K. Franze and J. Guck, Rep. Progr. Phys. 73 (9), 094601 (2010). D. Koch, W. J. Rosoff, J. Jiang, H. M. Geller, and J. S. Urbach, Biophys. J. 102 (3), 452 (2012). M. J. Hancock, K. Sekeroglu, and M. C. Demirel, Adv. Funct. Mat. 22 (11), 2223 (2012). M. Cetinkaya, S. Boduroglu, and M. C. Demirel, Polymer 48 (14), 4130 (2007); M. Cetinkaya, N. Malvadkar, and M. C. Demirel, J. Polymer Science Part B-Polymer Physics 46 (6), 640 (2008); M. C. Demirel, Colloids and Surfaces a-Physicochemical and Engineering Aspects 321 (1-3), 121 (2008); M. Cetinkaya and M. C. Demirel, Chemical Vapor Deposition 15 (4-6), 101 (2009). N. A. Malvadkar, M. J. Hancock, K. Sekeroglu, W. J. Dressick, and M. C. Demirel, Nature Materials 9 (12), 1023 (2010). C. Christophis, K. Sekeroglu, G. Demirel, I. Thome, M. Grunze, M. C. Demirel, and A. 9 Rosenhahn, Biointerphases 6 (4), 158 (2011). 13 14 9 10 5 6 7 8 11 12 15 16 17 18 19 E. So, M. C. Demirel, and K. J. Wahl, J. Phys. D-Appl. Phys. 43 (4) (2010). C. S. Chen, M. Mrksich, S. Huang, G. M. Whitesides, and D. E. Ingber, Science 276 (5317), 1425 (1997); V. Vogel and M. Sheetz, Nature Rev. Mol. Cell Biol. 7 (4), 265 (2006). J. M. Corey, D. Y. Lin, K. B. Mycek, Q. Chen, S. Samuel, E. L. Feldman, and D. C. Martin, J. Biomed. Mat. Res. Part A 83 (3), 636 (2007). T. Betz, D. Lim, and J. A. Kas, Phys. Rev. Lett. 96 (9), 098103 (2006). Y. E. Pearson, E. Castronovo, T. A. Lindsley, and D. A. Drew, Bull. Math. Biol. 73 (12), 2837 (2011). 10
1807.00764	2	1807	2018-08-06T10:32:14	A Nonhyperbolic Toy Model of Cochlear Dynamics	[ "physics.bio-ph", "q-bio.NC" ]	Cochlea displays complex and highly nonlinear behavior in response to wide-ranging auditory stimuli. While there have been many recent advancements in the modeling of cochlear dynamics, it remains unclear what mathematical structures underlie the essential features of the extended cochlea. We construct a dynamical system consisting of a series of strongly coupled critical oscillators to show that high-dimensional nonhyperbolic dynamics can account for high-order compressive nonlinearities, amplification of weak input, frequency selectivity, and traveling waves of activity. As a single Hopf bifurcation generically gives rise to features of cochlea at a local level, the nonhyperbolicity mechanism proposed in this paper can be seen as a higher-dimensional analogue for the entire extended cochlea.	physics.bio-ph	physics	A Nonhyperbolic Toy Model of Cochlear Dynamics Keith Hayton,∗ Dimitrios Moirogiannis,† and Marcelo Magnasco Center for Study in Physics and Biology, The Rockefeller University (Dated: August 7, 2018) Cochlea displays complex and highly nonlinear behavior in response to wide-ranging auditory stimuli. While there have been many recent advancements in the modeling of cochlear dynamics, it remains unclear what mathematical structures underlie the essential features of the extended cochlea. We construct a dynamical system consisting of a series of strongly coupled critical oscilla- tors to show that high-dimensional nonhyperbolic dynamics can account for high-order compressive nonlinearities, amplification of weak input, frequency selectivity, and traveling waves of activity. As a single Hopf bifurcation generically gives rise to features of cochlea at a local level, the nonhyper- bolicity mechanism proposed in this paper can be seen as a higher-dimensional analogue for the entire extended cochlea. I. INTRODUCTION The nature of signal processing in the cochlea has been a focus of interest in the study of sound perception. Gold was first to posit that the cochlea does not operate as a passive Fourier transformer but instead utilises an ac- tive regenerative system which amplifies incoming sig- nals [1, 2]. Experiments on living specimens have con- firmed Gold's hypothesis [3, 4] despite initial studies on cadavers showing that cochlea acts as a simple passive spatial frequency analyser [5]. Indeed, live, reasonably intact cochlea exhibits an active nonlinear process with three key characteristic properties: high gain amplifica- tion, sharp frequency tuning, and nonlinear compression of the dynamic range (Fig. 1) [3, 6]. FIG. 1. Cochlear Velocimetric Data. [4] Cochlear ve- locimetry data taken by laser interferometry at a spot in the basilar membrane of a living chinchilla. Basilar membrane speed vs frequency for varying intensities is plotted. Adja- cent curves are separated by 10 dB. To understand the origin of the active process, one first needs to examine the physiological and anatomical prop- erties of the cochlea. As sound waves enter the inner ∗ [email protected] ; author contributed equally † [email protected]; author contributed equally ear, they set into vibration the cochlear partition, which results in traveling waves of activity that propagate uni- directionally from base to apex. The basilar membrane is the main compliance of the cochlear partition, whose stiffness increases, by two orders of magnitude, from apex to base. This stiffness gradient results in each point on the basilar membrane responding maximally to a sin- gle characteristic frequency with high frequencies at the base and exponentially decreasing frequencies towards the apex [5]. The cochlear partition also contains the organ of Corti, where the hair cells, the sound sensitive cells, are located. The hair cells play an essential role in generating the features of the active process through hair-bundle motility in nonmammals [7 -- 11] a combina- tion of hair-bundle motility [12 -- 15] and membrane-based electromotility in mammals [16 -- 18]. It is now known [19] that at a local level, at a spe- cific location on the basilar membrane, the characteris- tic features of the active process generically arise from a Hopf bifurcation [20, 21], a form of structural instability in dynamical systems theory which gives rise to critical oscillations. Several theoretical studies have extensively investigated this relationship between local cochlear dy- namics and a single Hopf bifurcation [22 -- 24]. At the level of the extended cochlea, encompassing the full basilar membrane, models relying on multiple Hopf bifurcations, biophysical details of basilar membrane me- chanics, and surrounding fluid hydrodyanmics have suc- cessfully reproduced a number of key features of cochlear dynamics [25 -- 27]. However, in contrast to a single Hopf bifurcation in the local case, it is still unclear what un- derlying high-dimensional mathematical structures can generically give rise to the complex, nonlinear responses of the extended cochlea. In this paper, we propose such a mathematical structure; we show that a high-dimensional nonhyperbolic system residing on a full-dimensional cen- ter manifold can exhibit the key nonlinear features of the active process as described above. In dynamical systems theory, the classical approach to studying behavior in the neighborhood of an equilibrium point is to examine the eigenvalues of the Jacobian of the system at this point. If all eigenvalues have nonzero real part, then the equilibrium point is called hyperbolic 8 1 0 2 g u A 6 ] h p - o i b . s c i s y h p [ 2 v 4 6 7 0 0 . 7 0 8 1 : v i X r a and the dynamics around the point is topologically conju- gate to the linearized system determined by the Jacobian [28 -- 30]. On the other hand, if there exists at least one eigenvalue with zero real part, then the equilibrium point is called nonhyperbolic and the linearization does not de- termine the qualitative dynamics around the point. Dynamics around nonhyperbolic fixed points are com- plex and give rise to a number of interesting features. First, since the dynamics are not enslaved by the ex- ponent of the Jacobian, nonlinearities and input pa- rameters play a crucial role in determining dynamical properties such as relaxation timescales and correlations [31, 32]. Nonhyperbolic points are also not structurally stable, meaning that small perturbations of the vector field can lead to substantial topological changes of the orbits around the point. Examples of these topological changes include the appearance of new invariant sets such as periodic orbits and tori, and if the dimension is high enough, chaotic dimensions can arise. The standard technique in studying dynamics around a nonhyperbolic point is to investigate the reduced dy- namics on an invariant subspace called a center manifold. The local existence of center manifolds has been rigor- ously proven [33, 34], and its dimension is equal to the number of eigenvalues on the imaginary axis. If there are no eigenvalues with positive real part (unstable modes), then the center manifold is attracting [33], and instead of studying the full system, we can study the reduced dynamics on the center manifold. The approximation theorem for center manifolds provides us with the tool to calculate the center manifold and the reduced dynamics up to any degree of accuracy [33]. In this paper, we present a dynamical system poised at a nonhyperbolic point with all eigenvalues of the lin- earization being purely imaginary; the dynamics reside on a full dimensional center manifold. In general, the greater the number of eigenvalues on the imaginary axis, the more complex the dynamics could be [20]. Thus, the rich and complex behavior we will be discussing in our paper is not surprising. We posit that the nervous system utilizes nonhyperbolic equilibrium points and the corre- sponding unique dynamical properties on center mani- folds to flexibly respond to a wide range of input pa- rameters and exhibit complex nonlinear behavior. Our aim in this paper is not to provide a detailed anatomi- cal and biophysical model of cochlea, but rather to con- struct a toy model which provides an existence proof that center manifold dynamics can account for and connect key characteristic properties of cochlea: high-order com- pressive nonlinearities, amplification of weak input, fre- quency selectivity and traveling waves of activity. We do not suggest that a simple 1-D line topology is nec- essarily present in cochlear anatomy; although, if the cochlea does indeed utilize center manifolds in the pro- cessing of sound, it might be the case that the full high dimensional phase space of cochlear dynamics could be reduced to simple, low dimensional structures on the cen- ter manifold. This approach of constructing a toy model 2 to explain how a given mechanism can lead to a par- ticular set of properties is common practice in theoreti- cal physics and is the underlying philosophical approach of several well known theoretical neuroscience models, e.g. Wilson-Cowan equations, Hopfield networks, and Kuramoto models [35, 36]. There are a number of studies regarding nonhyperbolic dynamics in neural systems, including entire hemisphere ECoG recordings [37], experimental studies in premotor and motor cortex [38], theoretical [39] and experimental studies [40] of slow manifolds (a specific case of center manifolds) in oculomotor control, slow manifolds in deci- sion making [41], Hopf bifurcation in the olfactory system [42], and theoretical work on regulated criticality [43]. Nonhyperbolic dynamics, also commonly referred to as dynamical criticality, is distinct from statistical criti- cality [44], which is related to the statistical mechanics of second-order phase transitions. It has been proposed that neural systems [45], and more generally biological systems [46], are statistically critical in the sense that they are poised near the critical point of a phase tran- sitions [47, 48]. Statistical criticality is characterized by power law behavior such as avalanches [49 -- 51] and long- range spatiotemporal correlations [52, 53]. While both dynamical criticality and statistical criticality have had success in neuroscience, their relation is still far from clear [46, 54, 55]. II. METHODS A. Mechanism for Higher-Order Compression We first show that nonlinear compression arises natu- rally from a system of strongly coupled critical oscillators, on an order exponentially larger than one would obtain from a single critical oscillator alone. It's well known that the response of a single Hopf oscillator to exponentially distributed periodic input is given by the curves in Fig. 2. At the center of resonance, the response R scales as the cubic root of the forcing strength F , R ∝ F 1/3. Away from resonance, the response scales linearly for small forc- ing and as a cubic root for large forcing. Now we look at the more interesting case of a series of unidirectionally coupled Hopf oscillators (Fig. 3); the output of one oscillator acts as input to the next oscillator downstream. For sake of exposition, let us consider the case where all connections have strength of magnitude 1, coupling nonlinearities are discarded, and we force, with periodic input, only the last cell on the top row: xi = −yi − xi2xi + xi+1 + F δi,N eiωt yi = xi − yi2yi (1) where i ∈ {1, 2, ..., N}, xi and yi ∈ C. At resonance, when ω = 1, the response xN−d at a distance d from the input scales as xN−d ∝ F 1/3d+1 for all forcing strengths F . Away from resonance, the response scales linearly for small forcing and as xN−d ∝ F 1/3d+1 for large forcing. This scaling behavior is illustrated in Fig. 4 where we plot the response of the last cell we force as well as the next two oscillators downstream. By construction, the linear connectivity matrix de- scribing the network in Fig. 3 is non-normal and has purely imaginary eigenvalues. Both features are not nec- essary to produce the high order power law scaling seen in Fig. 4; however, at resonance, purely imaginary eigen- values are necessary to give rise to nonlinear compression across all forcing strengths. FIG. 2. Response of a Single Critical Oscillator. The response of a single critical oscillator to a range of forcing fre- quencies and forcing strengths is shown. Each colored curve corresponds to a different forcing strength and the strength increases from bottom to top (direction of arrow). At res- onance, when the forcing frequency is 1.0, the response in- creases as the cubic root of the forcing strength. Away from resonance the response is linear for small forcing and a cubic root for larger forcing. FIG. 3. Strongly Coupled Critical Oscillators. Net- work of N critical oscillators with excitatory cells on top and inhibitory cells on the bottom. The oscillators are coupled together with connections along the excitatory layer. All con- nections have strength of magnitude equal to 1. Input to the network is denoted by the arrow. 3 B. Main Toy Model We have shown that a series of coupled critical oscil- lators can generate responses which are compressed to an arbitrarily large degree. We will now incorporate this mechanism into an extended model of cochlea that is se- lective to frequencies over an exponential range. Let us consider a network of 2N cells paired together to form N strongly coupled oscillators as shown in Fig. 5. The activity of the cells on the top row are given by xi ∈ C and cells on the bottom by yi ∈ C. The ith oscillator in the network, consisting of cells xi and yi, has character- istic frequency ωi and is coupled to the oscillator on its left via a unidirectional connection of strength ωi from xi to xi−1. In agreement with frequency selectivity in the basilar membrane, the characteristic frequencies are exponentially distributed with ω1 < ω2 < ω3 < ... < ωN. The cells evolve in time, under the influence of forcings Ixi(t) and Iyi (t), according to the equations: xi = −ωiyi − xi2xi + ωi+1xi+1 + Ixi(t) yi = ωixi − yi2yi + Iyi(t) (2) where i ∈ {1, 2, 3, ..., N}, xi and yi ∈ C (for formality let us define: xN +1 = ωN +1 = 0). It should be noted that in these equations, nonlinearities are confined to the cubic order terms and do not couple distinct cells in the network. Now let us concatenate the vectors (cid:126)x and (cid:126)y into a single vector X ∈ C2N such that Xi = xi and Xi+N = yi for i = 1, ..., N. In a similar way, we form I(t) ∈ C2N from Ixi(t) and Iyi (t). We can then rewrite (2) as: X = AX − X2X + I(t) (3) where the X2 and X2X are element wise, and the matrix A describes the connectivity in Fig. 5. It is easy to check that A has a purely imaginary spectrum. Thus, the center manifold is full dimensional and we should expect to see nonlinear behavior even for the entire range of forcings. We are specifically interested in the time-asymptotic response of the system to periodic input I(t) = F eiωt. The high dimensionality and nonhyperbolicity of the dy- namics makes the ODE in (3) difficult to integrate since the step size necessary to numerically integrate the sys- tem decreases as the forcing strength is increased. For- tunately, we can bypass numerical integration methods by looking for asymptotic solutions X(t) = Zeiωt, where Z ∈ C2N. Substituting these into (3), we find that: 0 = (A − iω)Z − Z2Z + F (4) and define g(Z) to be equal to the right hand side of (4). The solution of (4) can be found numerically by using the multivariable Newton-Raphson method in C4N: 4 FIG. 4. Scaling Behavior Along a Series of Coupled Identical Critical Oscillators. We force the last oscillator in a series of coupled identical critical oscillators with periodic input and plot the response as a function of forcing strength at the three different network locations (vertical arrow). The top and bottom rows in the figure correspond to the forcing frequency exactly at and away from resonance, respectively. At resonance, the response xN−d at a distance d from the input scales as xN−d ∝ F 1/3d+1 for the entire range of forcing strengths. Away from resonance, the response scales linearly for small forcing and as xN−d ∝ F 1/3d+1 for large forcing. FIG. 5. Network of Coupled Oscillators. A network of 2N cells, with top and bottom cell activities labeled as xi and yi, respectively, form N oscillators with exponentially distributed characteristic frequencies ω1 < ω2 < ω3 < ... < ωN . The oscillators are coupled to one another through unidirectional connections along the top. (cid:101)Z → (cid:101)Z − J((cid:101)Z)−1(cid:101)g(Z) where (cid:101)Z := (z1, z2) = (Re(Z), Im(Z)), (cid:101)g(Z) := (g1(Z), g2(Z)) = (Re(g(Z)), Im(g(Z))), and J is the Jacobian of(cid:101)g with respect to (cid:101)Z: (5) Jij(z) = ∂gi ∂zj This Newton-Raphson algorithm can fail to converge for randomly chosen initial points. Furthermore, the al- gorithm's trajectory can get trapped in periodic orbits for step sizes not adequately small. To make simulations possible, we incorporate adaptive step sizes and initial points into the algorithm. We define the response of the system as Z, the element-wise complex modulus of Z. We define the re- sponse of a single oscillator to be the l2-norm of the oscil- i+N for i ∈ 1, ..., N, which makes lator Ri = sense since the response of top and bottom cells are pro- portional element wise. (cid:113)Z2 i + Z2 III. RESULTS A. Frequency Tuning, Compression, and Amplification We consider the case of an N = 60 oscillator network with characteristic frequencies ωi = 2(i−1)/10 where i ∈ Z, 1 (cid:53) i (cid:53) N. We force every cell in the network with uniform input of equal strength and plot the response of a single oscillator as a function of forcing frequency ω. We do this for a range of exponentially distributed forcings 2−45+k where k ∈ Z, 0 (cid:53) k (cid:53) 63. The results are shown in Fig. 6, where each curve corresponds to a different forcing strength and the strength increase in the direction of the arrow from bottom to top. It is clear from Fig. 6 that the oscillator's response is selective for a particular forcing frequency. For weak forc- ing, there is a single peak in the response curves, which corresponds to an effective characteristic frequency dif- ferent from the oscillator's characteristic frequency. This shift in frequency selectivity for weak forcing is a result of the oscillator integrating both direct input and input from other upstream oscillators with higher characteris- tic frequencies. For our specific model parameters, the effective characteristic frequency ωef f is given by 2 7 10 ωc, where ωc is the characteristic frequency of the oscillator. Note that ωef f is exponentially distributed. As the forc- ing increases, the system reaches a regime in which the peaks of the response curves shift left towards the true characteristic frequency ωc, in agreement with previous studies [4] and which can be seen in Fig. 1. Finally, for high forcing the curves flatten out across all frequencies, lose selectivity and compress by a factor of 1/3. We now examine the three frequency domains demar- cated by the vertical dashed lines ω = ωc and ω = ωef f in Fig. 6. For ω < ωc, the response scales linearly up un- til the point where the curves flatten out and compress by a factor of 1/3. For ωc < ω < ωef f , we find a sharp transition to cubic root scaling for small forcing. This sharp transition is an artifact of the unidirectional cou- pling between oscillators; for forcing frequencies greater than the characteristic frequency, the oscillator integrates both direct input and input from oscillators upstream, while for forcing frequencies less than the characteristic frequency, the input from upstream oscillators is negligi- 5 ble compared to the direct input. For large forcing in this regime, (ωc < ω < ωef f ) we also find cubic compression. Between the two cubic scaling regimes, we observe the existence of higher-order compressive nonlinearities, with the response becoming nearly invariant as ω approaches ωef f . This invariance for medium forcing persists for the frequency domain ω > ωef f but is now accompanied by a large forcing regime in which the response scales linearly up until the curves flatten and exhibit cubic compression. The scaling behavior in the different frequency domains is depicted in Fig. 7. Our results generally agree with Fig. 1 except for the sharp transition from linear to cu- bic scaling on the left flank of the response curves as the forcing frequency ω passes through ωc. Finally, we note that weak input is preferentially am- plified in the model. We define the amplification of an oscillator in our network as R/F where R is the oscillator response and F the input amplitude. For F << 1 and forcing frequencies above ωc, R ∝ F 1/3, which implies that the amplification is F −2/3 >> 1. To the left of ωc, there is no amplification of weak input. B. Traveling Waves of Activity In our toy model simulations, we observe traveling waves of activity, which propagate along the network of oscillators in the direction of decreasing characteris- tic frequency. Since the waves arise from the linear part of (3), we can study the waves by examining the eigen- vectors of the connectivity matrix A. In Fig. 8, we plot the real (red dashed) and imaginary (blue) parts of the eigenvector corresponding to the eigenfrequency of me- dian absolute value. The envelope or complex modulus of the eigenvector is also depicted. We omit the final N components of the eigenvector as they are a rescaled copy of the first N components. It is clear that the real and complex parts of the eigen- vector are out of phase by π/2. This phase difference is the source of traveling wave behaviour in our model. The envelope shape of the eigenvector is nearly symmet- ric and sharply peaks at a particular location, while the spatial frequency of the eigenvector is approximately uni- form. These features are in disagreement with basilar membrane recordings, which predict a gradual increase in wave amplitude and spatial frequency in the direction of decreasing characteristic frequencies (basilar membrane base to apex). C. Long-Range Connections The addition of long-range connections between oscil- lators helps to resolve this disagreement by improving both wave shape and spatial frequency variation. We incorporate this into our toy model by including linear, long-range, skew-symmetric connections that are propor- tional in strength to the characteristic frequencies and 6 FIG. 6. Response Curves of a Single Oscillator. We plot the response of a single oscillator in the coupled network (Fig. 5) in response to a range of forcing frequencies and 64 exponentially distributed input strengths (each colored curve corresponds to a different strength), F = 2−45 to F = 218, which increase in the direction of the arrow from bottom to top. The two vertical dashed lines correspond to the true and effective characteristic frequencies. decay as an exponential with distance. We define the strength of the connections between two oscillators sepa- rated a distance d apart as Dωe−d where ω is the charac- teristic frequency of the oscillator at the right side (higher frequency) of the connection and D is a scaling parame- ter. The skew-symmetric connectivity is depicted in Fig. 9. These connections sit on top of the original connec- tions from Fig. 5. We plot the response curves and eigenvectors of the system in Fig. 10(a) is just the orignal response curves without any skew-symmetric connections. Fig. 10(b), corresponding to D = 0.25, exhibits a shift in the shape of the eigenvectors with the amplitude gradually increas- ing in the direction of decreasing characteristic frequen- cies. The spatial frequency of the eigenvector also in- creases in the direction of decreasing characteristic fre- quency. The case of relatively strong skew-symmetric connections, D = 0.75, is plotted in Fig. 10(c). The ini- tial tail and spatial frequency variation are exaggerated in comparison to (a) and (b) and generally in agreement with basilar membrane studies, but this comes at a cost. As we increase D, the response curves lose their shape and no longer exhibit higher-order compression over a wide range of forcings; this is a direct result of the long range connections pushing the eigenvalues away from the imaginary axis and preventing the system from taking full advantage of the unique properties of center man- ifold dynamics. Therefore, to achieve response curves and waves that agree with experiments, a balance be- tween the strength of the skew-symmetric and original network connections is needed. IV. CONCLUSION We have shown that high-dimensional nonhyperbolic- ity with dynamics residing on a full-dimensional center manifold can exhibit the key characteristic nonlinearities of the active process in cochlea: frequency selectivity, amplification of weak input, and higher-order nonlinear compression. The toy model presented in this paper also gives rise to traveling waves of activity that propagate unidirectionally across the system. In order to better approximate the experimentally measured shape of trav- eling waves on the basilar membrane, we have included long-range, skew-symmetric connections between oscilla- tors; however, these connections negatively impact the shape of the response curves by pushing the dynamics away from nonhyperbolicity. While we have been able to find a reasonable balance between the shape of traveling waves and response curves (Fig. 10(b)), we believe this conflict could be better solved by either finding an appro- 7 FIG. 7. Scaling Regimes in the Response of a Single Oscillator. We plot the response of a single oscillator (Fig. 6) against the forcing strength for three different forcing frequencies ω. This oscillator has characteristic frequency ωc = 8.00 and effective frequency ωef f = 12.13. These define the boundaries of three different scaling regimes. In the plots, the red triangles represent 1/3 scaling, yellow dots represent linear scaling, and the blue curve is the oscillator response. In the first plot, ω = 4.0, a linear oscillator response turns into 1/3 compression past a certain forcing strength. In the second regime, ω ≈ 11.3, the response first scales as the cubic root of the input, then approaches a region of near invariance, whereafter it settles back down to a cubic root. Finally, in the last regime, represented by ω ≈ 29.9, we find the following scaling pattern: 1/3, invariance, linear, 1/3. FIG. 9. Long-Range, Skew Symmetric Connectivity. We illustrate the additional skew-symmetric connections to and from a single cell in the network. We omit showing the other connectivity (Fig. 5) to avoid clutter. Arrows denote excitation, while bars denote inhibition. This pattern is re- peated for every cell in the network. ity of the original system unchanged. As our model is just a toy model of center manifold dynamics, we do not suggest that our abstract critical oscillator network corresponds to actual anatomical and biophysical structures in the cochlea; although, if the cochlea does indeed utilize center manifolds in the pro- cessing of auditory stimuli, it might well be the case that the full high-dimensional phase space of cochlear dynam- ics could be reduced to a more simple structure on the center manifold. One possible simple structure is the toy model presented in this paper. FIG. 8. An Example Eigenvector of the Network Con- nectivity Matrix A. The real (red dashed) and imaginary (blue) parts of an eigenvector of the network connectivity ma- trix A is plotted above along with the eigenvector's complex modulus. As waves in the model are determined by the linear part of (3), wave behavior and shape can be studied in the context of eigenvectors. priate class of long range linear connections that preserve the high dimensionality of the center manifold or by in- cluding nonlinear couplings between the critical oscilla- tors. Nonlinear couplings won't affect the linearization of the system, leaving the high-dimensional nonhyperbolic- ACKNOWLEDGEMENTS We would like to thank Alex Katsov for useful discus- sions. 8 FIG. 10. Coupled Critical Oscillators with Long-Range Connections. The response of a single oscillator in the network with long-range connections is plotted along with a representative eigenvector for three different scaling parameters D of the skew-symmetric connections. Columns (a), (b), (c) correspond to D = 0, 0.25, 0.75, respectively. K.H. and D.M. contributed equally to this work. [1] T. Gold and R.J. Pumphrey, Proc. R. Soc. London B. 7, 19 (2006). 135, 462 (1948). [2] T. Gold, Proc. R. Soc. London B. 135, 492 (1948). [3] M.A. Ruggero, Curr. Opin. Neurobiol. 2, 449 (1992). [4] M.A. Ruggero et al, Proc. Natl. Acad. Sci. U.S.A. 22, [19] A.J. Hudspeth et al., J. Neurophysiol. 104, 1219 (2010). [20] S. Wiggins, Introduction to Applied Nonlinear Dynami- cal Systems and Chaos (Springer, 1990). [21] Y.A. Kuznetsov, Elements of applied bifurcation theory. [5] G. Von Békésy and E.G. Wever, Experiments in Hearing [22] Y. Choe, M.O. Magnasco, and A.J. Hudspeth, Proc. 11744 (2000). (1997). 11765 (2000). (McGraw-Hill Book Co., New York, 1960). [6] M.A. Ruggero et al., J. Acoust. Soc. Am. 101, 2151 [7] A.J. Hudspeth, Neuron 4, 530 (2008). [8] A.J. Hudspeth et al., Proc. Natl. Acad. Sci. U.S.A. 97, (2000). [9] P. Martin, Active Processes and Otoacoustic Emissions (2003). in Hearing (Springer, 2008), pp. 93-143. [10] P. Martin and A.J. Hudspeth, Proc. Natl. Acad. Sci. U.S.A. 96, 14306 (1999). U.S.A. 98, 14386 (2001). (2005). [11] P. Martin and A.J. Hudspeth, Proc. Natl. Acad. Sci. [12] D.K. Chan and A.J. Hudspeth, Biophys. J. 89, 4382 [13] D.K. Chan and A.J. Hudspeth, Nat. Neurosci. 8, (2005). [14] H.J. Kennedy et al., Nature 433, (2005). [15] H.J. Kennedy et al., J. Neurosci. 26, 2757 (2006). [16] J. Ashmore, Physiol. Rev. 88, 173 (2008). [17] P. Dallos et al., J. Physiol. 576, 37 (2006). [18] R. Fettiplace and C.M. Hackne, Nature Rev. Neurosci. (Springer, 2013). Natl. Acad. Sci. U.S.A. 95, 15321 (1998). [23] V.M. Eguíluz et al., Phys. Rev. Lett. 22, 5232 (2000). [24] S. Camalet et al.,Proc. Natl. Acad. Sci. U.S.A.. 97, 3183 [25] M.O. Magnasco, Phys. Rev. Lett. 90, 058101 (2003). [26] T. Duke and F. Jülicher, Phys. Rev. Lett. 90, 158101 [27] A. Kern and R. Stoop, Phys. Rev. Lett. 91, 128101 (2003). (1959). (2012). 2012). [28] D.M. Grobman, Dokl. Akad. Nauk SSSR. 128, 880 [29] P. Hartman, Proc. Am. Math. Soc. 11, 610 (1960). [30] P. Hartman, Bol. Soc. Mat. Mex. 5, 220 (1960). [31] X.H. Yan and M.O. Magnasco, PloS ONE. 7, e41419 [32] K. Hayton, D. Moirogiannis, and M. Magnasco, PLoS ONE 13, e0196566 (2018). [33] J. Carr, Applications of centre manifold theory (Springer, [34] A. Kelley, Journal of Differential Equations. 3, 546 (1967). [35] G.B. Ermentrout and D.H. Terman, Mathematical foun- dations of neuroscience (Springer, 2010). [36] F.C. Hoppensteadt and E.M. Izhikevich, Weakly con- 242, 343 (1998). nected neural networks (Springer, 2012). [37] G. Solovey et al., J. Neurosci. 35, 10866 (2015). [38] M.M. Churchland et al., Nature. 487, 51 (2012). [39] H.S. Seung, Neural Networks. 11, 1253 (1998). [40] H.S. Seung et al., Neuron. 26, 259 (2000). [41] C.K. Machens, R. Romo, and C.D. Brody, Science 307, [42] W.J. Freeman and M.D. Holmes, Neural Networks. 18, 9 [45] D.R. Chialvo, Nat. Phys. 6, 744 (2010). [46] T. Mora and W. Bialek, J. Stat. Phys. 144, 268 (2011). [47] L. da Silva, A.R. Papa, and A.C. de Souza, Phy. Rev. A. [48] D. Fraiman et al., Phy. Rev. E. 79, 061922 (2009). [49] J.M. Beggs and D. Plenz, J. Neurosci. 23, 11167 (2003). [50] A. Levina, J.M. Herrmann, and T. Geisel, Nat. Phys. 3, 857 (2007). 105, 7576 (2008). [51] E.D. Gireesh and D. Plenz, Proc. Natl. Acad. Sci. U.S.A. [52] V.M. Eguiluz et al., Phys. Rev. Lett. 94, 018102 (2005). [53] M.G. Kitzbichler et al., PLOS Comput. Biol. 5, e1000314 1121 (2005). 497 (2005). 361 (1998). 163 (2012). [43] E. Bienenstock and D. Lehmann, Adv. Complex Syst. 1, [54] M.O. Magnasco, O. Piro, and G.A. Cecchi, Phys. Rev. [44] J.M. Beggs and N. Timme, Frontiers in physiology. 3, [55] K. Kanders, T. Lorimer, and R. Stoop, Chaos 27, 047408 Lett. 102, 258102 (2009). (2009). (2017).
1107.4510	1	1107	2011-07-22T13:04:55	Analysis of noise-induced bistability in Michaelis Menten single-step enzymatic cycle	[ "physics.bio-ph", "q-bio.MN" ]	In this paper we study noise-induced bistability in a specific circuit with many biological implications, namely a single-step enzymatic cycle described by Michaelis Menten equations with quasi-steady state assumption. We study the system both with a Master Equation formalism, and with the Fokker-Planck continuous approximation, characterizing the conditions in which the continuous approach is a good approximation of the exact discrete model. An analysis of the stationary distribution in both cases shows that bimodality can not occur in such a system. We discuss which additional requirements can generate stochastic bimodality, by coupling the system with a chemical reaction involving enzyme production and turnover. This extended system shows a bistable behaviour only in specific parameter windows depending on the number of molecules involved, providing hints about which should be a feasible system size in order that such a phenomenon could be exploited in real biological systems.	physics.bio-ph	physics	Analysis of noise-induced bistability in Michaelis Menten single-step enzymatic cycle Daniel Remondini,∗ Enrico Giampieri, Armando Bazzani, and Gastone Castellani Physics Dept. of Bologna University and INFN Bologna Amos Maritan Physics Dept. of Padova University (Dated: October 3, 2018) In this paper we study noise-induced bistability in a specific circuit with many biological im- plications, namely a single-step enzymatic cycle described by Michaelis Menten equations with quasi-steady state assumption. We study the system both with a Master Equation formalism, and with the Fokker-Planck continuous approximation, characterizing the conditions in which the con- tinuous approach is a good approximation of the exact discrete model. An analysis of the stationary distribution in both cases shows that bimodality can not occur in such a system. We discuss which additional requirements can generate stochastic bimodality, by coupling the system with a chem- ical reaction involving enzyme production and turnover. This extended system shows a bistable behaviour only in specific parameter windows depending on the number of molecules involved, pro- viding hints about which should be a feasible system size in order that such a phenomenon could be exploited in real biological systems. PACS numbers: I. INTRODUCTION eling. Many biological phenomena (e.g. memory induction, chromatin remodeling, cell-fate determination) are re- ceiving a great deal of attention in recent years, due to an increasing interest in the description of their com- plex behaviour by means of basic biochemical circuitry [1, 6, 9]. The signal transduction machinery is mainly based on enzymatic reactions, whose average kinetics can be described within the framework initally proposed by Michaelis and Menten (MM). The steady state ap- proximation of MM model accounts for the majority of known enzymatic reactions, and can be adjusted for the description of regulatory properties such as cooperativ- ity, allostericity and activation/inhibition [10]. The MM equations are still valid at small molecule numbers (as it frequently happens in real cells) if the microscopic inter- pretation is changed correspondingly [5, 7, 12] but the discrete stochastic aspects become predominant, and a deterministic or a stochastic continuous model can not describe the system in sufficient detail [4, 13, 15]. A large class of enzymatic reactions controls the re- versible addition and removal of phosphoric groups, phosphorylation/dephosphorylation reactions catalyzed by kinases and phosphatases respectively. The phos- pho/dephosphorylation cycle (PdPC) is thus a post- translational substrate modification that is central for the regulation of several biological processes [8, 14]. How these processes can show a bistable behaviour [3, 11] in the presence of fluctuations [2], reflected by a bimodal stationary distribution of protein num- ber/concentration, is a crucial question for their mod- ∗Electronic address: [email protected] The deterministic version of a single PdPC is not bistable in general, but it is hypothesized that external noise can trigger such a behavour [14]. We study this cy- cle by a Chemical Master Equation approach, and also (in the limit of large molecule number) the related Fokker- Planck equation. A closed form for the stationary distri- bution of the system is obtained for both approaches: we show that the system can not have a bimodal stationary distribution due to intrinsic fluctuations only, but an ad- ditional external noise obtained by a plausible biological mechanism (i.e. the coupling of the system with an en- zyme production/activation reaction) can produce such feature. We define analytically the conditions in which bimodality occurs, as a function of the reaction param- eters (kinetic constants) and system size (number of en- zyme and substrate molecules), and we verify our results by numerical simulations with a Gillespie algorithm. II. THE MODEL The PdPC (also referred to as the futile cycle) is com- posed by one phosphorylation and one dephosphorylation reaction, catalyzed respectively by enzymes E1 and E2: E1 + A ⇋ E1A → AP + E1 E2 + AP ⇋ E2AP → A + E2 (1) The deterministic dynamics of this cycle can be described via the MM formalism. Assuming a steady-state approxi- mation for both enzymatic reactions A and AP , we obtain the following equations: AP = v1 − v2 A = v2 − v1 where B. Fokker-Planck approximation 2 v1 = KC1 · E1 A KM1 + A = VM1 AP A KM1 + A AP v2 = KC2 · E2 KM2 + AP = VM2 KM2 + AP (2) Imposing the conservation of the total substrate concen- tration, let x be the A molecule concentration, we obtain: x = VM2 1 − x KM2 + 1 − x − VM1 x KM1 + x , (3) that can be easily shown to have only one solution inside the substrate domain (see [14]). A. The CME approach Starting from the previous equations, a Chemical Mas- ter Equation (CME) approach [16] is introduced to ac- count for intrinsic noise (pn is the A-molecule distri- bution function over the possible states n ∈ [0 : N ], D+f (n) = f (n + 1) − f (n)): pn = D+J; J = rnpn − gn−1pn−1 (4) where rn = V ′ M1 n K ′ M1 + n gn = V ′ M2 N − n M2 + N − n K ′ N is the total number of the substrate molecules, n is the the number of A molecules and the MM constants have been accordingly scaled: K ′ M = N·VM . In the hypothesis of fast relaxation times, the stationary solution of eq. (4) describes the statistical properties of the reaction in Fig. 1. The stationary distribution ps n is derived by imposing pn(t) = 0; excluding the existence of a constant current in the system, we get the condition M = N·KM and V ′ ps n ps n−1 = gn−1 rn ⇒ D+ ln ps(n) = ln gn rn+1 If we define a potential V (n), such that D+V (n) = − ln(gn/rn+1), the stationary solution has the Boltzmann form n = F · e−V (n), ps (5) where F is a normalizing constant. According to (5), the maxima and minima of the distribution are obtained by imposing D+V (n) = 0, extending the n domain to the set of real numbers. This leads to gn/rn+1 = 1, similar to (3) and with an unique solution inside the [0 : N ] domain. This result is also confirmed by Gillespie simulations of the dynamical process. CME dynamics can be more easily studied by consid- ering a continuous approximation by means of a Fokker- Planck (FP) equation, with N is sufficiently large to get the natural boundary conditions ps(1) = ps(0) ≃ 0. Ex- panding in power series the D+ operator up to second order, from eq. (4) we get: ∂p ∂t (x, t) = − ∂ ∂x C(x)p(x, t) + 1 2 ∂2 ∂x2 D(x)p(x, t) (6) where the drift and diffusion coefficients are defined as C(x) = g(x) − r(x), D(x) = [r(x) + g(x)] /N , x = n/N (0 ≤ x ≤ 1), pn = p(x)/N , and g(x) = r(x) = ; VM2 · (1 − x) KM2 + 1 − x VM1 · x KM1 + x up to an error O(N −1). We remark that the diffusion coefficient becomes negligible for large N with respect to the drift term: the fluctuations scale as 1/√N according to the law of large numbers. The stationary solution ps(x) can be written explicitly: ps(x) = F D(x) exp(cid:18)2Z x C(y) D(y) dy(cid:19) that for the symmetric case VM1 = VM2, KM1 = KM2 = KM reduces to ps(x) = F (KM + 1 − x) (KM + x)(cid:0)KM + 2x − 2x2(cid:1)KM −1 where F is a normalizing constant. (7) The FP stationary solution is a good approximation of CME when the drift coefficient is of the same order of the diffusion coefficient (i.e. C(x) ≃ O(1/N ), see Ap- pendix). Therefore, we can apply the FP approximation nearby the critical points x∗ where C(x) ≃ 0. By us- ing a Gaussian approximation, in the neighborhood of a critical point x∗ we get the distribution (C ′(x) = dC/dx) ps(x) ∝ exp(cid:18)−C ′(x∗) D(x∗) (x − x∗)2 2 (cid:19) (8) In the symmetric case, for generic enzyme concentration, we explicitly compute C(x) = KC(cid:18)E1 1 − x KM + 1 − x − E2 x KM + x(cid:19) so that C ′(x) D(x) = −KM N(cid:18) E1 (KM + 1 − x)2 + (cid:18) E1(1 − x) KM + 1 − x + E2 (KM + x)2(cid:19) KM + x(cid:19)−1 (9) E2x The variance D(x∗)/C ′(x∗) scales as (KM N )−1 and, in general, is weakly dependent on the enzyme concentra- tion. Therefore, in the continuous limit N ≫ 1, we get a finite variance only for KM ≪ 1. In Fig. 1 we compare the exact CME stationary distribution, the exact solu- tion of the FP equation and the approximation of Eq. (8) in the symmetric case, where the condition C(x) ≪ 1 is satisfied. In the FP equation (6) two minima may ap- pear in the physical domain of x for specific values of KM only when N is sufficiently low, but these critical points have no meaning being due to a bad approximation of the CME. Thus intrinsic noise cannot induce stochastic bifurcation in system (1). 0.04 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0 0 10 20 30 40 50 60 70 80 90 100 FIG. 1: Solutions for the futile cycle: N = 100, VM = 1, KM = 0.1, E1 = E2. Squares: CME stationary solution; line: FP exact solution; circles: FP approximate solution (8). 3 When γ = 1 (i.e. E1 = E2) we have the trivial solu- tion u = 0 (an unique maximum with n = (N + 1)/2, x = 1/2), whereas for γ − 1 > 0 (resp. < 0) u shifts towards −a (resp. a). Supposing that enzyme concentration can fluctuate around the average value, given ξ = (γ − 1)/KM and p(ξ) the corresponding probability distribution, we have du = p(ξ(u))(cid:18) 1 (a + u)2 + 1 (a − u)2(cid:19) du p(u)du = p(ξ(u))(cid:12)(cid:12)(cid:12)(cid:12) dξ du(cid:12)(cid:12)(cid:12)(cid:12) (13) Under the hypotheses that ξ fluctuates around zero and p(ξ) tends sufficiently fast to zero at the boundaries (natural boundary condition), we study the conditions for bimodality of p(u). The critical points of p(u) must satisfy dp(u) du = d2ξ du2 p(ξ(u)) +(cid:18) dξ du(cid:19)2 dp dξ = 0 (14) If we approximate p(ξ) with a Gaussian distribution (jus- tified for a sufficiently large enzyme molecule number and K sufficiently small, see Fig. 2) so that p(u) =(cid:18) 1 (a + u)2 + 1 (a − u)2(cid:19) e − 1 2σ2 ξ ( 1 a+u − 1 a−u )2 the equation (14) reads C. Bimodality induced by enzyme noise 1 σ2 ξ (cid:18) Relaxing the assumption of fixed enzyme concentra- tion, we characterize the effect of enzyme fluctuations on the substrate stationary distribution. In the symmetric case K ′ C2, the equilib- rium points of the average equation (2) corresponding to maxima of ps n can be calculated explicitly as a function of the ratio between enzymes γ: M and K ′ M1 = K ′ M2 = K ′ C1 = K ′ γ = E1 E2 = N − n + 1 n K ′ M + n K ′ M + N − n + 1 (10) If one introduces the variable u: u = n N − N + 1 2N = n N − a u ∈(cid:20)−a, +a + 1 N(cid:21) (11) where a ≃ 1/2 for N ≫ 1, the condition (10) reads γ = a − u a + u · KM + a + u KM + a − u , where KM = K ′ M /N . Assuming that KM ≪ 1, so that the critical point is quite sensitive to the enzyme con- centration, and performing a perturbative approach over KM , the previous equation can be rewritten as (cid:18) 1 a + u − 1 a − u(cid:19) = γ − 1 KM (12) 1 (a + u)2 + 1 1 (a − u)2(cid:19)2(cid:18) 1 −2(cid:18) (a + u)3 − a − u(cid:19) a + u − (a − u)3(cid:19) = 0 1 1 (15) If we exclude the symmetric solution u = 0, we get the following condition for bimodality (recalling that σξ = σγ/KM ) σ2 ξ > 2 3a2 ⇒ σ2 γ > 2K 2 M 3a2 (16) The condition (16) can be realized if the enzymatic con- centrations fluctuate largely enough around the mean, but for the system (1) this fluctuation cannot be achieved by means of intrinsic noise only. III. STOCHASTIC HYPERSENSITIVITY INDUCES BISTABILITY As we have shown in previous sections, intrinsic noise cannot induce bimodality in PdPC, but large enzyme fluctuations can produce this effect for KM sufficiently small. This can be achieved by coupling the initial sys- tem (1) with a further reaction involving enzyme activa- tion (e.g. by a second messenger like Calcium) or local- ization (e.g. inside a delimited region like an organelle). Then, in the hypothesis that the activator is in low abun- dance (or that the reaction region is small) a situation of competition between the two enzymes for the reaction is obtained. Under suitable kinetic parameters, the extra fluctuation may lead the cyclic reaction alternately to- wards one of the two products, producing a bimodal dis- tribution function for the substrates. Defining as E ∗ 1 and E ∗ 2 the inactive (external to the reaction region) enzyme concentrations, and EA the concentration of activating molecules, the full kinetic reaction scheme becomes: EA + E ∗ EA + E ∗ 1 ⇋ E1 2 ⇋ E2 E1 + A ⇋ E1A → AP + E1 E2 + AP ⇋ E2AP → A + E2 These equations are identical to those of the enzymatic cycle in (1) as long as there is a sufficiently large number of enzyme molecules, so that the fluctuations in enzyme concentration become negligible. A simplified version of the enzyme competition can be obtained by considering a direct interchange between the two active enzymes: E1 ⇋ E2 (E2 = ET − E1) with ET total enzyme concentration. The resulting FP equation of the system is two-dimensional, and the sta- tionary distribution can be expressed as a function of en- zyme ratio γ and substrate concentration x: remember- ing that ξ = (γ − 1)/KM we have ps(x, ξ) = Π(xξ)ps(ξ), in which Π(xξ) is the conditional distribution of sub- strate given the enzyme concentration. Let us consider 4 the case when the PdPC kinetics is much faster than the enzyme (achieved by choosing the KC parameters in equations (2) sufficiently large) so that one can apply an adiabatic approach to the global system evolution, ap- proximating Π(xξ) with the stationary solution ps(x) as in (7). The substrate distribution ps(x) is obtained as the marginal distribution of ps(x, ξ): ps(x, ξ) =Z ps(xξ)ps(ξ)dξ =Z ps(xx∗)ps(x∗)dx∗, (17) where we use the univocal relationship between the sta- tionary distribution of the enzyme ratio and the equilib- rium concentration of the substrate. Applying the calcu- lation performed previously, we have shown that ps(x∗) is a bimodal distribution if the E1/E2 distribution can be well approximated by a Gaussian-like function, and if the condition (16) is satisfied. In Fig. 2 we show that the Gaussian approximation for the enzyme ratio is valid even for a moderate number of enzymes. For the simplified model in the symmetric case we get a modified version of the previous MM equation (with the steady-state approximation and E1 + E2 = 1) x = KC (1 − E) (1 − x) KM + 1 − x KC E x KM + x − and we have an explicit dependence of the critical point x∗ on enzyme concentration E: x∗(E) = 2E + KM − 1 −p1 + 2KM − 4E + K 2 2 (2E − 1) M − 8EKM + 4E2 + 8KM E2 (18) In Fig. 3 we show the plot of x∗(E) with KM = 0.1 to point out the sigmoidal behaviour due to the extreme sensitivity of the solution to enzyme concentration. Ac- cording to (17) the substrate distribution ps(x) can be computed: ps(x) ∝Z 1 0 exp(cid:18)−C ′(x∗) D(x∗) (x − x∗)2 2 (cid:19) p(x∗)dx∗ (19) where p(x∗) is the critical point distribution defined by (13) and we use the Gaussian approximation for the FP stationary distribution. If p(x∗) is a bimodal distribu- tion (i.e. the condition (16) is satisfied), the final dis- tribution ps(x) will also be bimodal when the variance D(x∗)/C ′(x∗) is sufficiently small: indeed smaller than the distance of the critical points of p(x∗). In general, this condition is satisfied if (KM N )−1 is sufficiently small, In Figure which is consistent with the choice N ≫ 1. 4, we perform numerical simulations of the CME, using a low number of enzyme and substrate molecules. We 3.5x 105 3 2.5 2 1.5 1 0.5 0 0 0,5 1 1,5 2 2,5 FIG. 2: Gaussian approximation of p(γ) (γ = E1/E2) for E1 + E2 = 100. Bars: empirical distribution of γ; continuous line: gaussian distribution with same mean and variance. observe that bimodality occurs only under suitable con- ditions depending on the system size: fixing the chem- ical reaction constants and the ratio between enzyme ∗ x 1 0.8 0.6 0.4 0.2 0 0 0.1 0.2 0.3 0.4 0.5 E 0.6 0.7 0.8 0.9 1 FIG. 3: Plot of x∗(E) solution (18), KM = 0.1. x 104 x 105 5 4 3 2 1 0 0 200 400 x 104 12 10 8 6 4 2 0 0 8 7 6 5 4 3 2 1 5 10 0 0 20 40 FIG. 4: Histograms of x for N = 10, 50, 500, obtained by numerical simulations run for 5 · 106 iterations. The ra- tio between M (enzyme molecule number) and N (substrate molecule number) is set to 0.15; the constants for enzyme pro- duction reactions k1, k−1 are set to 2, KC = 100, KM = 0.1. and substrate total molecule number ET /XT , only in a specific interval of molecule number (say XT , the total number of substrate molecules) we have such behaviour. The condition (16) can be applied only in the last case (N = 500) and it is consistent with numerical simulations (loss of bimodality). This phenomenon can be explained as follows: in- creasing the molecule number, we recover the Michaelis- Menten deterministic limit, losing the noise-induced bistability; on the other side, if the molecule number is too low, even if bistability is not lost from an analytical point of view, the separation between the two maxima becomes indistinguishable. IV. CONCLUSIONS In this paper we study the conditions that result in a bimodal stationary distribution of a single phosphory- lation/dephosphorylation cycle described by Michaelis- Menten equations in the quasi-steady state assumption. The simple addition of noise (as produced by a Master Equation approach) is not enough to achieve bimodal- ity in this system that has a unique deterministic stable state, and the same result can be stated for the Fokker- Planck description of the system obtained as a limit of the Master Equation. Since the Fokker-Planck approach 5 is often used instead of the exact Master Equation ap- proach, we characterize the conditions in which it ap- proximates correctly the system. We propose a different approach to generate stochas- tic bimodality (that does not change the number of de- terministic stable states, thus a purely noise-driven phe- nomenon). With an additional chemical reaction we in- troduce competition between enzymes, for binding to an activating molecule or for reaching a specific site in which reaction can occur (e.g. an organelle), that can be biolog- ically feasible. For this system we clarify the conditions for which stochastic fluctuations in enzyme concentration can lead to bimodality in substrate concentration, and we show that it depends on the time scales involved in the two reactions (related to the kinetic constants) and also on the size of the system (i.e. the number of enzyme and substrate molecules involved). In particular, if we fix the kinetic constants and the ratio between enzyme and substrate molecule numbers, bimodality is observed only in specific parameter s, re- lated to system size in terms of total number of molecules involved. This results define in more detail the feasibil- ity of this phenomenon in real biological systems, stating that if the biological systems have to exploit noise to achieve a bimodal behaviour there must be a relation- ship between the chemical parameters of the system and its size. V. APPENDIX The FP approximation (7) for the stationary solution (5) requires that the condition D+V (n) = − ln gn/rn+1 reduces to dV dx = −2 C(x) D(x) + ∂ ∂x ln D(x) where in the thermodynamics limit N → ∞ where x = n/N and rn+1/2 = N r(x), gn+1/2 = N g(x). This is the case when gn− rn+1 is small so that using a perturbative expansion we have ln gn rn+1 − gn rn+1 + gn ≃ rn+1 ≃ 2 # + O(cid:18) [r(x) − g(x)]2 (cid:19) N 2 # 2 1 r(x) + g(x) ∂ ∂x (r(x) + g(x)) N " N (r(x) − g(x)) + 1 Therefore if N (r(x) − g(x)) is finite we can approximate dx ≃ N D+V (n) ≃" N (r(x) − g(x)) + 1 ∂ ∂x (r(x) + g(x)) dV with an error of order O([r(x) − g(x)]2). Finally we get r(x) + g(x) dV dx ≃ −2N g(x) − r(x) r(x) + g(x) + = −2 ∂ ∂x C(x) D(x) ln(r(x) + g(x)) + ∂ ∂x ln D(x) The condition gn − rn+1 ≪ 1 (i.e. the generation and re- combination rates of the enzymatic cycle have almost the same value) means that the FP approximation is justified only nearby the critical points when gn/rn+1 ≃ 1. 6 [1] Batchelor, E., Loewer, A., Mock, C., and Lahav, G. (2011). Stimulus-dependent dynamics of p53 in single cells. Molecular Systems Biology, 7; 488. [2] Berg, O., Paulsson, J., and Ehrenberg, M. (2000). Fluc- tuations and quality of control in biological cells: zero- order ultrasensitivity reinvestigated. Biophysical Jour- nal, 79(3):1228 -- 1236. [3] Castellani, G., Bazzani, A., and Cooper, L. (2009). To- ward a microscopic model of bidirectional synaptic plas- ticity. Proceedings of the National Academy of Sciences, 106(33):14091. [4] D.Levens and Gupta, A. (2010). Reliable noise. Science, 370:1088 -- 1089. [5] English, B., Min, W., van Oijen, A., Lee, K., Luo, G., Sun, H., Cherayil, B., Kou, S., and Xie, X. (2006). Ever- fluctuating single enzyme molecules: Michaelis-Menten equation revisited. Nat Chem Biol, 2(2):87 -- 94. [6] Giampieri, E., Remondini, D., de Oliveira, L., Castellani, G., and Li´o, P. (2011). Stochastic analysis of a miRNA- protein toggle switch. Molecular Biosystems, Epub June 30:PMID: 21717010. [7] Gillespie, D. (2002). The Chemical Langevin and Fokker- Planck Equations for the Reversible Isomerization Reac- tion. J. Phys. Chem. A, 106 (20):5063 -- 5071. [8] Krebs, E., Kent, A., and Fischer, E. (1958). The muscle phosphorylase b kinase reaction. J Biol Chem, 231(1):73 -- 83. [9] Lim, H. and van Oudenaarden, A. (2007). A multistep epigenetic switch enables the stable inheritance of dna methylation states. Nature Genetics, 39(2):269 -- 275. [10] Min, W., Gopich, I., English, B., Kou, S., Xie, X., and Szabo, A. (2006). When does the Michaelis-Menten equa- tion hold for fluctuating enzymes? J Phys Chem B, 110(41):20093 -- 7. [11] Ortega, F., Garc´es, J., Mas, F., Kholodenko, B., and Cascante, M. (2006). Bistability from double phos- phorylation in signal transduction. FEBS JOURNAL, 273(17):3915. [12] Qian, H. and Elson, E. (2002). Single-molecule enzy- mology: stochastic Michaelis-Menten kinetics. Biophys Chem, 101-102:565 -- 76. [13] Raj, A. and van Oudenaarden, A. (2008). Nature, nur- ture, or chance: stochastic gene expression and its con- sequences. Cell, 135(2):216 -- 26. [14] Samoilov, M., Plyasunov, S., and Arkin, A. (2005). Stochastic amplification and signaling in enzymatic futile cycles through noise-induced bistability with oscillations. Proc Natl Acad Sci U S A, 102(7):2310 -- 5. [15] To, T. and Maheshri, N. (2010). Noise can induce bi- modality in positive transcriptional feedback loops with- out bistability. Science, 327:1142 -- 1145. [16] van Kampen, N. G. (2007). Stochastic Processes in Physics and Chemistry. North Holland, third edition.
1606.00657	1	1606	2016-06-02T13:01:43	Fighting the flow: the stability of model flocks in a vortical flow	[ "physics.bio-ph" ]	We investigate the stability of self-propelled particle flocks in the Taylor-Green vortex, a steady vortical flow. We consider a model where particles align themselves to a combination of the orientation and the acceleration of particles within a critical radius. We identify two distinct regimes, if alignment with orientation is dominant the particles tend to be expelled from regions of high vorticity. In contrast if anticipation is dominant the particles accumulate in areas of large vorticity. In both regimes the relative order of the flock is reduced. However we show that there can be a critical balance of the two effects which stabilises the flock in the presence of external fluid forcing. This strategy could provide a mechanism for animal flocks to remain globally ordered in the presence of fluid forcing, and may also have applications in the design of flocking autonomous drones and artificial microswimmers.	physics.bio-ph	physics	Fighting the flow: the stability of model flocks in a vortical flow A. W. Baggaley1, 2 1School of Mathematics and Statistics, Newcastle University, Newcastle upon Tyne, NE1 7RU, UK 2Joint Quantum Centre Durham-Newcastle We investigate the stability of self-propelled particle flocks in the Taylor-Green vortex, a steady vortical flow. We consider a model where particles align themselves to a combination of the orien- tation and the acceleration of particles within a critical radius. We identify two distinct regimes, if alignment with orientation is dominant the particles tend to be expelled from regions of high vor- ticity. In contrast if anticipation is dominant the particles accumulate in areas of large vorticity. In both regimes the relative order of the flock is reduced. However we show that there can be a critical balance of the two effects which stabilises the flock in the presence of external fluid forcing. This strategy could provide a mechanism for animal flocks to remain globally ordered in the presence of fluid forcing, and may also have applications in the design of flocking autonomous drones and artificial microswimmers. PACS numbers: 47.32.Ef,47.63.-b,87.18.-h Introduction In a vast range of biological systems, from bird flocks to fish schools to insect swarms, collective behaviour is ob- served. Studying why and how such collective behaviour arises can be important to first understand and then ad- dress a number of ecological issues, mainly due to human impact on the environment. In addition there are also im- portant technological applications, collective robot mo- tion for example [1]. In this paper we investigate one of the most important and interesting examples of collective behaviour, collec- tive motion. Whilst various modelling approaches have been suggested in the literature, one of the most popu- lar is based on self-propelled particles (SPPs), building on the seminal Vicsek model [2]. In this numerical ap- proach N particles move in a two dimensional domain (extension to higher dimensions is straighforward) with a constant velocity V . A particles direction of motion is instantaneously updated at every numerical time-step to align with neighbouring particles within some fixed critical radius, R. Noise is introduced in the system by applying a random rotation of a given size to each par- ticle after the alignment step. This is to model intrinsic noise, due to the fact that animals will never perfectly align, and extrinsic noise, i.e. forcing from the external environment. The number of subsequent variants of the Vicsek model is far too great to list here and we recommend the inter- ested reader consult [3] and references therein. Whilst it has been shown that the behaviour of marching locusts could be modelled using an SPP approach [4], Khurana & Ouellette [5] showed that Vicsek flocks were particu- larly sensitive to spatio-temporally correlated noise. In particular flocks were more easily destabilised when the extrinsic noise consisted of a model of a turbulent flow, in contrast to the case where a random (delta-correlated in space and time) field forced the system. Furthermore we recently showed [6] that Vicsek flocks in a steady vortical flow are concentrated into areas of high vorticity. This has a profound effect on the morphology of the flock, with a dramatic increase in the filamentarity, i.e. the perimeter of the flock is increased for a given area. One reasons animals exhibit collective motion is it gives them a better chance of avoiding predation [7]. If one assumes a predator generally will attack the closest individual, an animal can reduce the area (volume) of the region in which it is the closest prey to a predator by joining a 'flock' [8]. Of course the size of this 'domain of danger' is also dependent on the shape of the flock, with safety re- ducing if the perimeter (surface area) of a flock increases for a given area (volume). Hence our earlier findings [6] could have profound implications for animals flocking in a turbulent environment, or more likely animals have de- veloped strategies to counteract this effect. Finding such a strategy is the goal of this paper, in particular (moti- vated by the recent study of Morin et al. [9]), we wish to understand if both alignment and anticipation can sta- bilise model flocks in the presence of spatially correlated extrinsic noise. Modelling and Computational Methods We consider an extension to the self-propelled parti- cle (SPP) model presented in [9], taking N = 500 self- propelled particles in a two-dimensional square periodic domain with sides of size L = 2π. Each particle has a position xi(t) and an intrinsic, self-driven, velocity vi(t). As is typical in SPP models, all particles are assumed to move with the same speed, V = 1, and a particles intrinsic velocity is determined by vi = (V cos θi, V sin θi), (1) where θi determines the direction the particle moves in. In the Vicsek model [2] θi is periodically (at each time increment) determined from the average of the particle's own direction, plus the directions of its neighbours within 6 1 0 2 n u J 2 ] h p - o i b . s c i s y h p [ 1 v 7 5 6 0 0 . 6 0 6 1 : v i X r a 2 FIG. 1: Typical structure and trajectories of the flocks with no external flow. (Left) Snapshots of the system in a statistically steady-state with radius of interaction R = 1.0 ((a) α = 0, (c) α = π/2); arrows indicate the particles direction of motion. (Right) corresponding particle trajectories(for 50 particles), dark to light indicates the direction of time. a critical radius, R, such that θi =(cid:10)θj (cid:11) xi−xj<R + ηξi, (2) where angled brackets denote suitable averaging of the orientation of neighbours within the critical radius. The final term in Eq. (2) is a noise term; specifically ξi is a uniformly distributed random variable on the interval [−1, 1] and η is the intensity of the noise. Morin et al. [9] proposed an extension to the model by including both alignment and anticipation such that the rate of change of orientation is given by θi(t) = − 1 τ (cid:10) sin [θi − (θj + αχj)](cid:11) xi−xj<R + ηξi, (3) where α is a parameter we shall discuss shortly, χj ≡ θj/ θj is the sign of the angular velocity (the particles spin) and τ is an orientation rate. This is more easily un- derstood if we expand the sine function and recast Eq. (3) as: θi(t) = − 1 τ − 1 τ cos α(cid:10) sin (θi − θj)(cid:11) sin α(cid:10) sin θi − (θj + χj xi−xj<R + ηξi. xi−xj<R (cid:105)(cid:11) π 2 ) (cid:104) (4) It is then clear that the first term is the standard Vicsek interaction which acts to promote alignment with ori- entations, whereas the second term promotes alignment with the acceleration of particles within the critical ra- −3−2−10123−3−2−10123xy(a)−3−2−10123−3−2−10123xy(b)−3−2−10123−3−2−10123xy(c)−3−2−10123−3−2−10123xy(d)dius. The relative contribution of these two terms is de- termined by α; in the limit τ → 0 with α = 0 we recover In contrast with α = π/2 particles the Vicsek model. align purely with neighbouring particles' acceleration. It is worth noting that a model which included both align- ment and anticipation of others motion was earlier con- sidered in Szab´o et al. [10] and we shall discuss their findings alongside our own later in the article. Morin et al. [9] showed that (contrary to what one might expect) including anticipation in the model does not enhance the stability of the flock. Indeed they found with increasing values of α there was a transition from a flocking state to a spinning state. This can be seen in Fig. 1 where we plot snapshots of the system and particle trajectories. In this paper we shall investigate if including antici- pation can stabilise the flock in the presence of exter- nal noise which exhibits complex spatiotemporal correla- tions, such as one would expect flocks forming in a turbu- lent fluid environment would experience. Based on our previous arguments [6] we forgo the computational ex- pense and complexity of a direct numerical simulation (DNS) of the Navier-Stokes equations. Instead we turn to a well studied and widely used [11 -- 14] 'toy' flow, the Taylor Green (TG) vortex [15], defined as vf (x) = (uf , vf ) = Vf (sin(x) cos(y),− cos(x) sin(y)), (5) where x = (x, y). The vorticity field is given by ω = ∇ × vf = 2Vf sin(x) sin(y), (6) the flow is incompressible (∇ · vf = 0), and consists of cells of counter-rotating vortices as seen in Fig. 3. Vf is a scaling parameter which can be adjusted to modify the relative intrinsic particle speed to that of the background flow. The equation of motion for the SPPs is modified to dxi dt = vi = (V cos(θi) + uf (xi), V sin(θi) + vf (xi)). (7) We follow [5, 6] and assume that particles orient them- selves to the direction of motion of nearby particles, re- placing θj in Eq. (3) with [16] θj = atan2(V sin(θj) + vf (xj)), V cos(θj) + uf (xj)). (8) We retain the intrinsic noise (η) in Eq. (3), to model the fact that it is unlikely real animals will perfectly align themselves with neighbours within the critical radius and fix η = 0.2/τ . Particles are evolved according to an explicit Euler scheme such that xi(t + ∆t) = xi(t) + ∆tvi(t) θi(t + ∆t) = θi(t) + ∆t θi(t) (9) where at each timestep vi(t) is updated according to Eqns. (7), (8) & (3). Note ξi is drawn randomly at each timestep, we take ∆t = 0.05 and in each simulation evolve the system for 2 × 103 timesteps. We fix τ = ∆t, such that if α = 0 we recover the standard Vicsek model. The global order of the system can be characterised by computing, (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) N(cid:88) i=1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . vi ψ(t) = 1 N V 3 (10) In a typical simulation ψ grows from 0 until it saturates and fluctuates around some mean value, which depends on R, α and Vf , hence it is a convenient measure to estab- lish if the system has reached a statistically steady state. As we are interested in systems which exhibit flocking behaviour in the limit of small intrinsic noise and no ex- trinsic noise, we set R = 1.0 for all simulations reported here. We then perform a suite of numerical simulations to thoroughly investigate a two dimensional (α, Vf )- parameter space, with α ∈ [0, π/4] and Vf ∈ [0.1, 1.25]. For each point in (α, Vf )-space we perform i = 1, . . . , 10 simulations, computing the mean value of ψα,Vf ,i and its variance σ2 α,Vf ,i in each simulation (once it has reached a statistically steady value). We report the ensemble av- eraged mean (weighted by the inverse of the variance) value over the 10 simulations. We denote this value (cid:104)ψ(cid:105), where the angled brackets indicate the use of temporal and ensemble averaging, by taking a weighted mean the standard deviation of (cid:104)ψ(cid:105) (for a given α and Vf ) is [17] (cid:118)(cid:117)(cid:117)(cid:116)(cid:32) n(cid:88) σ−2 i (cid:33)−1 σ(cid:104)ψ(cid:105) = . (11) i=1 FIG. 2: Flock stability. Temporally and ensemble averaged (denoted by angled brackets) values of the order parameter ψ, plotted as a function α (see Eq. (3) & (4)), for varying flow speed Vf . Errorbars are given by ±2σ(cid:104)ψ(cid:105), see Eq. (11). αhψi00.10.20.30.40.50.60.70.40.50.60.70.80.91Vf=0.1Vf=0.25Vf=0.5Vf=0.75Vf=1.0Vf=1.254 FIG. 3: Typical structure and trajectories of the flocks with in the Taylor-Green Vortex. (Left) Snapshots of the system in a statistically steady-state with radius of interaction R = 1.0 ((a) α = 0; (c) α = 0.4; (e) α = π/4); Vf = 0.75; arrows indicate the particles direction of motion. The magnitude of the flow vorticity is indicated by the pseudocolour plot, with light (yellow) corresponding to regions of large positive vorticity, and dark (blue) negative vorticity. (Right) corresponding particle trajectories (for 50 particles), dark to light indicates the direction of time. −202−3−2−10123xy(a)−202−3−2−10123xy(b)−202−3−2−10123xy(c)−202−3−2−10123xy(d)−202−3−2−10123xy(e)−202−3−2−10123xy(f)5 the destabilising effect of the imposed flow field. We note that in the large α limit, where particles move into areas of high vorticity there is a 'matching' between the par- ticles spin, i.e. the sign of its angular velocity, and the sign of the vorticity, as is clear in Fig.4. Interestingly in Szab´o et al. [10] they found that the information exchange between particles was maximised at a critical balance between alignment and anticipation. They conjectured (due to the importance of information exchange in animal societies) that such a critical bal- ance may provide an optimal behavioural strategy. Here we show that it also provides a method to overcome the destabilising effects of spatially correlated noise. In order to quantify the results presented in Fig. 3 we define two relevant statistics. Firstly we quantify the 'patchiness' of the spatial distribution of particles in the domain. Following [11, 18] we course-graining the particles onto a 16 by 16 regular array of boxes. Within each box we compute the particle density (based on the number of particles lying within the box) and denote this quantity n(x, t). As the particle density in each box is Poisson distributed this has a mean value E[n] = λ = N/4π2 (cid:39) 12.6. If particles preferentially ac- cumulate in certain regions of the domain the standard deviation of n, σn, increases relative to its initial value, σP = λ1/2. Hence σn can be appropriately normalised to give the accumulation index [18] D = (σn−σP )/λ, which is a measure of the spatial distribution of the points in the domain. Large values of D indicate patchiness, i.e. the particles are concentrated in smaller subdomain(s), D = 0 indicates a random distribution of particles, and D < 0 indicates segregation of particles, relative to a random distribution. To extract the regions the particles are located, we use define ζ to be ζ = (cid:90) A nωdA, (12) the integral of the product of the particle density field and the modulus of the flow's vorticity field. For a ran- dom distribution of particles we would expect ζ = ζ0 = ¯ω(N/4π2) (cid:39) 10, where the overbar denotes the spatial mean of the modulus of the vorticity field. If the particles are concentrated in regions of vanishing vorticity then we would expect ζ (cid:39) 0. Conversely if ζ > ζ0 then particles are concentrated in regions of high vorticity. Figure 5 shows the temporally and ensemble averaged (as described above) values of D (left panel) and ζ (right). We see without any anticipation, as the flow speed in- creases particles are confined into the regions of low vor- ticity, hence increasing values of (cid:104)D(cid:105), and decreasing values of (cid:104)ζ(cid:105). However some anticipation (the optimal amount depending on the flow speed as one may expect) is seen to lead to values of ζ ∼ ζ0. Finally we see for large values of α the particles tend to collect in regions of high vorticity consistent with our earlier discussion of Fig. 3. FIG. 4: Spin-vorticity matching. Snapshots of the system in a statistically steady-state with R = 1.0, α = π/4 and Vf = 0.75. Particles with positive spin ( χi ≡ θi/ θi = 1 ) are plotted as black circles, those with negative spin (χi = −1) as white circles. The magnitude of the flow vorticity is indicated by the pseudocolour plot, with light (yellow) corresponding to regions of large positive vorticity, and dark (blue) negative vorticity. Results Our main results are presented in Fig. 2 where we plot (cid:104)ψ(cid:105) vs. α for varying Vf . For all values of Vf a moderate value of α is seen to enhance the global alignment of the the flock, at larger values of α the stability breaks down, as particles form smaller clusters which follow tight spiral trajectories. However what is striking is that as the flow speed increases anticipation is seen to have a profound stabilising effect. Note also that there is a reduction in the value of σ(cid:104)ψ(cid:105) for moderate values of α, at least for Vf < 1.0, which indicates a reduction in the magnitude of the fluctuations of ψ. One would imagine that this is also advantageous allowing information (e.g. changes in direction, arrival of a predator) to propagate more efficiently through the flock. In order to understand this phenomena in Fig. 3 we plot particle trajectories and snapshots of the system for Vf = 0.75 with varying α. For α = 0, i.e. the Vic- sek model, we see the particles are expelled from regions of high vorticity, and form filamentary structures as re- ported in our earlier work [6]. In contrast for large val- ues of α, where anticipation becomes dominant particles move into the areas of high vorticity and form small clus- ters where each particle follows a tight spiral trajectory. However at the interface of these two regimes we find that the 'correct' amount of anticipation can counteract −202−3−2−10123xy6 FIG. 5: Spatial distribution of the particles: (a) The 'patchiness' of the spatial distribution of points. Large values of D indicate the particles are confined to a small region of the domain, D = 0 indicates a random distribution of particles, and D < 0 indicates segregation of particles. The inset shows a zoom in of the region where α < 0.4. (b) ζ (E.q. 12) plotted as a function of α, the dashed line indicates a random distribution of points; ζ < 0 particles confined to regions of low vorticity; ζ > 0 particles confined to regions of high vorticity. This also ties into our earlier discussion about the mo- tivation for collective motion, in terms of safety in num- bers to minimise the 'domain of danger'. Clearly with too little anticipation (where the particles in our model are forced into thin filamentary structures) or too much (where particles concentrate in dense patches) the mor- phology of the flock is not optimal for providing increased safety in numbers. However a balance between these two competing affects does appear at least one viable strategy for the flock's morphology to not be strongly influenced by the underlying structure of the external fluid forcing. Summary To summarise we have investigated an extension to the widely used Vicsek model in which collective mo- tion emerges due to alignment with neighbouring parti- cles and anticipation of their motion. With the addition of extrinsic noise in the form of a steady vortical flow we find the global order of the flock is significantly re- duced and particles are confined to regions of low vortic- ity. In contrast in a model based purely on anticipation we find particles concentrate in regions of high vorticity. Most strikingly we find particles with a critical balance of alignment and anticipation are no longer slave to the flow, and global coherence emerges. At this critical bal- ance (for Vf < 1.0) we also see a reduction in the mag- nitude of the fluctuations of ψ, which surely would also be advantageous to members of the flock. Hence one strategy for animals flocking in a complex (i.e. turbulent) flow could be not only align with neigh- bours but also to anticipate their motion, which seems entirely plausible. In addition our findings could have implications for flocking autonomous drones (unmanned ariel vehicles) and artificial microswimmers [19]. By varying the amount of alignment and anticipation dif- ferent regions of a fluid could be probed, or by tun- ing their relative contributions the separation between devices could be maximised, i.e. to prevent collisions. Whilst this clearly does not mark the end of the story, particularly in biological systems, we strongly believe that by studying how flocks react to external perturba- tions (fluid motion, predatory threats etc.) and compar- ing to the dynamics of models will enhance our under- standing of collective motion in biological systems. [1] D. Floreano and R. J. Wood, Nature 521, 460 (2015). [2] T. Vicsek, A. Czir´ok, E. Ben-Jacob, I. Cohen, and O. Shochet, Phys. Rev. Lett. 75, 1226 (1995). [3] T. Vicsek and A. Zafeiris, Phys. Rep. 517, 71 (2012), Collective motion. [4] J. Buhl et al., Science 312, 1402 (2006). [5] N. Khurana and N. T. Ouellette, New J. Phys. 15, 095015 (2013). [6] A. W. Baggaley, Phys. Rev. E 91, 053019 (2015). [7] J. Krause and G. Ruxton, Living in Groups Oxford Se- ries in Ecology and Evolution (Oxford University Press, Oxford, 2002). αhDi00.10.20.30.40.50.60.7−10123456Vf=0.1Vf=0.25Vf=0.5Vf=0.75Vf=1.0Vf=1.25αhDi00.20.4−0.200.20.40.6(a)αhζi00.10.20.30.40.50.60.74681012141618202224ζ0(b)7 [8] W. Hamilton, Journal of Theoretical Biology 31, 295 Phys. Plasmas 4, 1 (1997). (1971). [9] A. Morin, J.-B. Caussin, C. Eloy, and D. Bartolo, Phys. Rev. E 91, 012134 (2015). [10] P. Szab´o, M. Nagy, and T. Vicsek, Phys. Rev. E 79, 021908 (2009). [15] G. I. Taylor, Philos. Mag. 46, 671 (1923). [16] Note the use of the atan2 function, the four-quadrant inverse tangent tan−1(y, x). Two arguments are required, so that the signs of the inputs is not lost, essential to return the correct quadrant of the computed angle. [11] W. M. Durham, E. Climent, and R. Stocker, Phys. Rev. [17] W. Eadie, Statistical methods in experimental physics Lett. 106, 238102 (2011). (North-Holland Pub. Co., 1971). [12] M. R. Maxey and S. Corrsin, J. Atmos. Sci. 43, 1112 [18] J. R. Fessler, J. D. Kulick, and J. K. Eaton, Physics of (1986). Fluids 6, 3742 (1994). [13] L. Bergougnoux, G. Bouchet, D. Lopez, and E. Guazzelli, [19] R. Dreyfus et al., Nature 437, 862 (2005). Phys. Fluids 26 (2014). [14] C. Nore, M. E. Brachet, H. Politano, and A. Pouquet,
1910.14645	1	1910	2019-10-31T17:35:00	Non-uniform growth and surface friction determine bacterial biofilm morphology on soft substrates	[ "physics.bio-ph", "cond-mat.soft" ]	During development, organisms acquire three-dimensional shapes with important physiological consequences. While the basic mechanisms underlying morphogenesis are known in eukaryotes, it is often difficult to manipulate them in vivo. To circumvent this issue, here we present a study of developing Vibrio cholerae biofilms grown on agar substrates in which the spatiotemporal morphological patterns were altered by varying the agar concentration. Expanding biofilms are initially flat, but later experience a mechanical instability and become wrinkled. Whereas the peripheral region develops ordered radial stripes, the central region acquires a zigzag herringbone-like wrinkle pattern. Depending on the agar concentration, the wrinkles initially appear either in the peripheral region and propagate inward (low agar concentration) or in the central region and propagate outward (high agar concentration). To understand these experimental observations, we developed a model that considers diffusion of nutrients and their uptake by bacteria, bacterial growth/biofilm matrix production, mechanical deformation of both the biofilm and the agar, and the friction between them. Our model demonstrates that depletion of nutrients beneath the central region of the biofilm results in radially-dependent growth profiles, which in turn, produce anisotropic stresses that dictate the morphology of wrinkles. Furthermore, we predict that increasing surface friction (agar concentration) reduces stress anisotropy and shifts the location of the maximum compressive stress, where the wrinkling instability first occurs, toward the center of the biofilm, in agreement with our experimental observations. Our results are broadly applicable to bacterial biofilms with similar morphologies and also provide insight into how other bacterial biofilms form distinct wrinkle patterns.	physics.bio-ph	physics	Non-uniform growth and surface friction determine bacterial biofilm morphology on soft substrates Chenyi Fei,1, 2 Sheng Mao,3 Jing Yan,1, 3 Ricard Alert,2, 4 Howard A. Stone,3 Bonnie L. Bassler,1, 5 Ned S. Wingreen,1, 2, 4, ∗ and Andrej Košmrlj3, 6, † 1Department of Molecular Biology, Princeton University, Princeton, New Jersey 08544, USA 2Lewis-Sigler Institute for Integrative Genomics, Princeton University, Princeton, New Jersey 08544, USA 3Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, New Jersey 08544, USA 4Princeton Center for Theoretical Science, Princeton University, Princeton, New Jersey 08544, USA 5The Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA 6Princeton Institute for the Science and Technology of Materials (PRISM), Princeton University, Princeton, New Jersey 08544, USA During development, organisms acquire three-dimensional shapes with important physiological consequences. While the basic mechanisms underlying morphogenesis are known in eukaryotes, it is often difficult to manipulate them in vivo. To circumvent this issue, here we present a study of developing Vibrio cholerae biofilms grown on agar substrates in which the spatiotemporal morpho- logical patterns were altered by varying the agar concentration. Expanding biofilms are initially flat, but later experience a mechanical instability and become wrinkled. Whereas the peripheral region develops ordered radial stripes, the central region acquires a zigzag herringbone-like wrinkle pattern. Depending on the agar concentration, the wrinkles initially appear either in the periph- eral region and propagate inward (low agar concentration) or in the central region and propagate outward (high agar concentration). To understand these experimental observations, we developed a model that considers diffusion of nutrients and their uptake by bacteria, bacterial growth/biofilm matrix production, mechanical deformation of both the biofilm and the agar, and the friction be- tween them. Our model demonstrates that depletion of nutrients beneath the central region of the biofilm results in radially-dependent growth profiles, which in turn, produce anisotropic stresses that dictate the morphology of wrinkles. Furthermore, we predict that increasing surface friction (agar concentration) reduces stress anisotropy and shifts the location of the maximum compressive stress, where the wrinkling instability first occurs, toward the center of the biofilm, in agreement with our experimental observations. Our results are broadly applicable to bacterial biofilms with similar morphologies and also provide insight into how other bacterial biofilms form distinct wrinkle patterns. The intricate shapes of organisms are determined by the spatiotemporal patterns of growth as well as the me- chanical properties of their underlying biological com- ponents [1 -- 3]. Three-dimensional (3D) shape transfor- mations in developing organisms often arise via differen- tial growth of connected tissues [1, 4]. Such asymmetric growth patterns generate compressive stresses within the faster growing tissues, which may cause mechanical in- stabilities [5 -- 7]. Growth-induced mechanical instabilities drive the formation of many convoluted morphologies, such as the gyrification of brains [2, 8, 9], the vilifica- tion and looping of guts [10, 11], and the branching of lungs [12] as well as 3D structures of synthetic systems with patterned swelling [5, 13 -- 16]. Biofilms, which are surface-associated bacterial com- munities encapsulated by a self-produced extracellular matrix [17, 18], also display a variety of 3D developmen- tal morphologies ranging from radial stripes, to concen- tric rings, to disordered labyrinth and herringbone pat- terns [19 -- 23]. In the case of Vibrio cholerae, a model ∗ [email protected] † [email protected] biofilm former, quantitative imaging revealed a 3D un- dulating topography with an intrinsic wavelength that depends on the stiffnesses of both the substrate and the biofilm [24]. Over the course of growth on an agar sub- strate, an initially flat V. cholerae biofilm expands and forms a 3D pattern in which a disordered core is sur- rounded by radial stripes extending to the edge [25]. These morphological transitions in V. cholerae biofilms are proposed to be caused by mechanical instabilities. The major components of the V. cholerae biofilm matrix and their roles in defining the biofilm's bulk and interfacial mechanical properties have been well ex- plored [18, 26 -- 30]. V. cholerae biofilms behave as soft viscoelastic solids similar to hydrogels, and possess finite adhesion to the agar surface on which they are grown [31]. Thus, as the biofilm expands, it is mechanically con- strained by the friction with the agar substrate. Mechan- ical compression due to constrained biofilm expansion ul- timately triggers instabilities that result in out-of-plane deformation and the 3D biofilm morphology [24, 32]. A key to understanding the full 3D morphodynamics of V. cholerae biofilms involves the cells' spatially het- erogeneous physiology [33]. Soon after the initial expan- sion of the biofilm, growth occurs primarily at the edge of the biofilm due to nutrient limitation near the center arXiv:1910.14645v1 [physics.bio-ph] 31 Oct 2019 [23, 24, 34, 35]. However, little is known about how this non-uniform growth profile, combined with the mechani- cal interaction between the biofilm and substrate, lead to the observed morphodynamics. While a mechanical basis for instability-induced pattern formation in biofilms has been suggested previously [32, 36], the dynamics of stress accumulation during biofilm expansion and the conse- quences for global pattern formation remain largely un- known. Here, we determine the biophysical mechanisms con- trolling V. cholerae biofilm expansion and pattern forma- tion. We show that the observed kinematic and morpho- dynamic features of growing biofilms are well captured by a reduced two-dimensional (2D) chemo-mechanical model. Consistent with experimentally measured veloc- ity profiles, our model predicts three distinct kinematic stages of biofilm expansion, even before the formation of wrinkles. We also demonstrate that non-uniform growth due to nutrient depletion generates anisotropic compres- sive stresses in the outer biofilm region leading to radial stripes; by contrast, in the interior of the biofilm, the compressive stresses are more isotropic, leading to zigzag herringbone-like patterns. We conclude that the spa- tiotemporal distribution of mechanical stresses dictates the morphodynamics of experimental biofilms grown on substrates of different agar concentrations. Our model thus illustrates the mechanical principles underlying how growth drives the emergent 3D morphologies of biofilms. RESULTS Biofilm morphodynamics depends on substrate stiffness After a liquid drop inoculates V. cholerae on an agar substrate, a biofilm initially expands radially and remains flat with no recognizable morphological features except at the center where inoculation occured (Fig. 1A and B). Expansion occurs because bacteria consume nutri- ents from the agar substrate, proliferate, and produce extracellular matrix. Growing biofilms adhere to the non-growing agar substrate, and the sliding friction be- tween biofilm and agar mechanically constrains biofilm expansion. Thus, growing biofilms become compressed and build up mechanical stresses. When the compressive stress reaches a critical value, a mechanical instability generates wrinkles (Fig. 1A). Wrinkles are vertical de- formations of the biofilm together with the adhered sub- strate with a characteristic wavelength (Fig. 1C) that depends on the thickness of the biofilm and on the mechanical properties of the biofilm and the agar sub- strate [24, 37, 38]. Subsequently, as compressive stresses continue to build up, a biofilm can partially detach from the agar substrate, forming delaminated blisters [24]. In this manuscript, however, we restrict our focus to explor- ing the original wrinkle patterns outside the inoculation core -- localized cell death has been shown to facilitate 2 pattern formation inside the inoculation core [39]. Notably, the development of wrinkle patterns depends on the stiffness of the agar substrate. For V. cholerae, after about 30 h of growth on soft substrates (low agar concentration), a pattern of radial wrinkles initially ap- pears at the outer edge of the biofilm and subsequently propagates towards the center (Fig. 1A and B, top). By contrast, on stiff substrates (high agar concentration), radial wrinkles initially form near the center and prop- agate outward (Fig. 1A and B, bottom). After about 40 h of growth, herringbone-like zigzag patterns emerge in the central region, surrounded by the outer region of radial stripes. Both of these regions expand outward at approximately the same speed as the expanding edge of the biofilm (Fig. 1B). In this steadily expanding state, surface profiling by confocal microscopy reveals a wedge- shaped rim (∼ 200 µm in width) with a constant leading angle φ, followed by a narrow region (∼ 500 µm in width) of nearly constant height, followed, in turn, by the region of radial stripe patterns (Fig. 1C). Chemo-mechanical model of biofilm development To understand the observations described in the pre- vious section, we developed a chemo-mechanical model of biofilm development that takes into account the diffu- sion of nutrients and their uptake by bacteria (Fig. 2A), growth of the biofilm, mechanical deformation of the biofilm and the agar substrate, and the friction between them (Fig. 2B). In this section, we focus on the early stage of development, when the biofilm surface is still flat. We denote by superscript 0 the deformations of the flat biofilm. The modifications of the model required to describe the wrinkled morphologies are discussed in a later section. The kinematics of biofilm development are described by a time-varying mapping between an internal mate- rial coordinate system X0 and the laboratory frame x, i.e. x = x(X0, t). Following the finite-strain formalism [40], we define the deformation gradient F = ∂x/∂X0, which captures the local change in shape and volume of a biofilm relative to its initial configuration. The overall change in shape arises from both growth and mechani- cal deformation. Accordingly, we follow the convention of multiplicative decomposition [41 -- 43], and decompose F = FeFg into a contribution Fg due to growth (which results in a post-growth intermediate configuration X, where neighboring regions may overlap creating incom- patibility; Fig. 2B) and a contribution Fe due to elastic deformation, characterizing the reorganization required to ensure compatibility (deformed contours in Fig. 2B). Using this theoretical framework, we next specify our model of biofilm growth and mechanics (see also Sup- plementary Notes I and II for details). During development, V. cholerae biofilms on agar sta- bilize at a thickness of roughly 100 µm, which is set by the penetration depth of oxygen [44], and subsequently 3 Fig. 1: Morphogenesis of V. cholerae biofilms grown at an air-solid interface. (A) Transmission images of biofilms grown on 0.4% (top) and 0.7% (bottom) agar substrates at the designated times, where time is measured relative to the time when a biofilm starts expanding radially. Black solid circles mark the boundaries of entire biofilms, blue dotted circles mark the boundaries of regions with radial patterns, and red dotted circles mark the boundaries of regions with zigzag patterns. Blue single-headed arrows indicate radial morphological features near the edge. Blue double-headed arrows span the regions with radial patterns. Red double-headed arrows span the regions with zigzag patterns. Scale bars: 3 mm. (B) Kymograph representation of the pattern formation dynamics of experimental biofilms grown on 0.4% (top) and 0.7% (bottom) agar substrates, where r measures the distance from the center of the biofilm and t is time. Gray, blue, and magenta colors indicate, respectively, regions without patterns, with radial patterns and with zigzag patterns. The internal white zone indicates the region possessing patterns related to the initial biofilm at t = 0. Biofilms shown in A are marked by horizontal double-headed arrows arrows at the designated times. The boundaries of different regions (as outlined in A) were obtained based on the intensity of transmitted light (see Methods). (C) Left: height map of a 3.2 mm × 2.4 mm region of the edge of a biofilm grown on 0.6% agar (t = 22 h). Middle and right: height profiles corresponding to the positions spanned by the yellow (denoted by 1) and brown (denoted by 2) arrows in the left panel. Intersecting dashed lines denote the biofilm leading angle φ. The zero value for z was chosen to coincide with the average height of a line profile on the agar surface. extend primarily in 2D along the substrate. Experimen- tation shows that bacterial cells in V. cholerae biofilms grown on agar do not locally order in the horizontal di- rections [31]. Therefore, we model the growth part of the deformation gradient as Fg = λgIk ⊕ 1, reflecting an isotropic increase in size in the planar direction by a factor λg, and neglecting growth in the vertical direction (the thickness H of the undeformed biofilm is assumed to be constant). The thin-film geometry also permits a simplified 2D representation of a biofilm in which phys- ical quantities are expressed as functions of an in-plane coordinate xk. In order to account for nutrient-dependent biofilm growth, we consider the kinematics of a 2D nutrient field ck(xk, t): ∂ck ∂t = D∇2 kck − Q0J−1 e,k ck (K + ck) . (1) Here, D is the diffusion constant, spatial derivatives are taken with respect to xk, and the final term describes the uptake of nutrients by bacteria according to the Monod 0 15 10 5 0 5 10 15 15 10 5 5 10 15 0 r (mm) 10 h 20 h 30 h 20 h 33 h 40 h B 0 20 20 40 40 t (h) 60 60 15 10 5 0 5 10 15 15 10 5 5 10 15 0 r (mm) 0.5 0.5 Δr2 (mm) 1 1 20 20 40 40 60 60 t (h) 150 150 60 100 100 30 50 50 0 0 0 0 0 0 1.5 20 h 33 h 40 h 10 h 20 h 30 h .4% agar A 0 0.7% agar C φ 1 0.5 0.5 1 Δr1 (mm) 60 30 0 1.5 0 z (µm) 12 z (µm) 60 01 4 Fig. 2: A chemo-mechanical growth model captures the kinematics of biofilm expansion. (A) Schematic of nutrient diffusion-uptake dynamics. Nutrients diffuse through the agar substrate (gray) and are taken up by the bacterial biofilm (red), where blue arrows indicate the magnitudes of nutrient fluxes. Bacterial growth rate is proportional to nutrient uptake, which in turn depends on the local nutrient concentration. The established nutrient concentration profile (see C) sets a nutrient-rich annular periphery (its width denoted by ac) where cells actively grow. Lighter (darker) red color indicates slower (faster) growth. r indicates the lateral distance from the center of the biofilm. (B) Schematic of the plane-stress elasto-growth model (color code as in A). Starting from an initial stress-free configuration (left), local growth of the biofilm Fg creates a virtual stress-free intermediate state (middle), which is further deformed by elastic deformation F0 compatibility (no overlap between marked regions), into a stressed current configuration (right). The elastic deformation F0 e,k, and a stretch γ0 of the film thickness H (bottom right). As the decomposed into an in-plane compression, denoted by F0 biofilm expands and moves relative to the substrate, it experiences a surface friction (black arrows) f = −ηv, where η is the friction coefficient and v is the expansion velocity. In the bulk, friction impedes biofilm expansion and is balanced by internal stresses; at the rim, friction increases the biofilm leading angle from φ0 to φ (bottom middle and bottom right). (C) Nutrient concentration ck and (D) radial expansion velocity v versus the radial coordinate r at the designated times. ck is normalized by the concentration at the edge of the biofilm. Shaded gray area in C indicates the active growth zone where ck > 0.5. Solid curves with shaded error bands in D represent experimental data (mean ± std) for a biofilm grown on a 0.7% agar substrate. The radial velocity was extracted by averaging over a ring of the biofilm at radius r from the center. Dotted curves represent simulation results for the parameters chosen by fitting the simulation velocity profiles to the experimental data (see Methods and Fig. S1). (E) Theoretical predictions for the biofilm leading angle φ (Eq. (3), dotted curves) as a function of the dimensionless friction ηvb/Gb and initial angle φ0 (colorbar), where vb is the expansion velocity at the biofilm's edge and Gb is the biofilm shear modulus. Theory curves are computed with the circumferential compression of the biofilm edge set to e,θθ = 0.8, but the results depend only minimally on this choice (Fig. S9). Colored circles show experimental data (mean ± F0 std, n = 3) at the designated agar concentrations. φ is measured at t = 36 h. Horizontal error bars are dashed because the friction coefficient η is not directly measured, but rather is inferred by fitting (see Methods for details). e, to ensure its e is law [45], where K is the concentration of nutrients at the half-maximal uptake rate, Q0 is the maximum up- take rate per unit area in the intermediate grown con- figuration, and the Je,k = det(Fe,k) factor is included to account for the change in areal density of bacteria due to elastic deformation (see Methods and description be- low). The growth field λg(xk, t) evolves in time accord- ing to the consumption of nutrients ∂λg/∂t = kg(ck)λg, where the growth rate kg(ck) is related to the Monod law described above (see Supplementary Note II for details). This reduced 2D model provides a reasonable approxi- mation to the full 3D diffusion dynamics of nutrients in the agar (see Supplementary Note IIE and Figs. S2 and S3). Importantly, the 2D model is sufficient to capture the spatially non-uniform growth that plays an essential role in biofilm morphodynamics. Mechanically, we model the biofilm as a plane-stress thin film, where it is assumed that the stress compo- B Fg ⋅ dX0 = dX F0e ⋅ dX = dx dX0 dX F0e = F0 e,∥ ⊕ γ0 dx H φ0 h = γ0H φ f = − ηv 0123 20 10 φ0 (°) 0 0 0 0.5 100 100 Friction ηvb/Gb 1 01 0 0-1L 10-1 1.5% 1.0% 0.8% 0.6%A 1 gar conc. 0.5 60 60 30 30 eading angle φ (°) E 6 h 1 h 16 h 2 2 r (mm) r (mm) 4 4 3 2 1 0 0123 0 0 v (µm/min) m/min) vr ( D A z Growth zone ac Nutrient uptake Diffusion ac r 6 h 0 0 0 0 16 h 4 4 2 2 r (mm) r (mm) 1 1 1 h 0.5 0.5 C Nutrient c vr ( ~ m/min) 5 e = F0 nents perpendicular to the biofilm surface are negligi- ble. The plane-stress simplification allows for the elas- e,k ⊕ γ0 to be decomposed into tic deformation F0 the in-plane compression F0 and the vertical stretch γ0 e,k (see Supplementary Note IIB and Fig. 2B), leading to a quasi-2D description of a biofilm with varying thickness h(xk, t) = γ0(xk, t)H. In our model, mechanical stresses in the biofilm arise from elastic deformation, and are specified by the consti- tutive relation σk(Fe), where σk is the in-plane stress tensor. Biofilms are complex hydrogel-like materials, whose constitutive relations are well approximated by nearly incompressible neo-Hookean elasticity (see Sup- plementary Note IIF and Fig. S4). Here, we modeled the biofilm as an incompressible neo-Hookean elastic mate- rial [8, 10, 46], but our results are largely insensitive to any plausible choice of rheological model for the biofilm (see Supplementary Note VII and Fig. S14). We obtain the expansion velocity v(xk) of a growing biofilm from a differential equation for local force balance, g g with a steady velocity (Fig. S6). These three kinematic stages are also observed in experiments and our model predictions closely match the measured velocity profiles (Fig. 2D). where kmax The three kinematic stages can be understood in the following way: During the first stage, which corresponds is the maximal to early times (t (cid:28) 1/kmax growth rate of a biofilm), friction with the agar sub- strate prevents the growing biofilm from expanding radi- ally in the central region. As stresses gradually build up (Fig. 3A and B), the width of the mobile annular zone at the biofilm edge increases in proportion to (t/η)1/2 (see Supplementary Note III and Fig. S7). The second stage ensues once this mobile zone spreads through the entire biofilm, so that the radial velocity becomes approx- imately a linear function of the distance from the center (see Supplementary Note III and Figs. 2D and S8). The third stage follows once the nutrients in the central re- gion are depleted, which slows down biofilm growth and reduces the radial velocity in that region (Fig. 2C and D). ∇k · (hσk) − ηv = 0. (2) Here, we assume that friction between the growing biofilm and the agar arises from binding and unbind- ing of biofilm matrix polymers with the adhesive biofilm proteins that have been secreted onto the agar surface [29]. In particular, we model the friction as viscous fric- tion, and we assume the friction coefficient η to be pro- portional to the shear modulus Gs of the agar substrate [47 -- 49] (see Supplementary Note IIG and Fig. S5). The partial differential equations for this model were solved numerically in the Lagrangian coordinate system with the open source computing platform FEniCS (see Meth- ods for details). Next, we present model results for the early stages of biofilm development, prior to wrinkling. Biofilm expansion has three kinematic stages In the model, nutrients are gradually depleted under- neath the growing biofilm (Fig. 2C). Once a steadily expanding state is achieved after about 15 h, most of the growth is restricted to the narrow nutrient-rich zone ac ≈ 1 mm near the rim of the biofilm (Figs. 2C and S3), which is consistent with experiment [24]. Our model predicts three stages of biofilm expansion with distinct radial velocity profiles v(r, t) (Figs. 2D and S6), where r is the distance from the center of the biofilm. In the initial stage, the interior of the biofilm is station- ary while the biofilm edge moves slowly outward. The magnitude of radial velocity and the size of the mov- ing region gradually increase until, at the second stage, the entire biofilm undergoes uniform expansion with a radial velocity v that is linearly proportional to r. At later times, in the third stage, the biofilm expansion in the central region slows due to the depletion of nutrients while the edge of the biofilm continues to move outward Higher friction increases the biofilm leading angle We next investigate how friction shapes the edge of an expanding biofilm. Specifically, we consider the lo- cal deformation of the wedge-shaped edge of a biofilm when sliding on a surface with velocity vb. The surface provides a frictional shear force of magnitude ηvb acting on the bottom of the biofilm edge, and thus generates a simple shear parallel to the horizontal plane. This shear deformation increases the leading angle from φ0 in the rest state to φ in the deformed state (Fig. 2B). To quantify how the biofilm leading angle increases with friction, we decomposed the elastic part of the de- formation gradient Fe into the product of rotations R and principal stretches U to connect the geometry, char- acterized by the angles φ and φ0, to the stress state of the biofilm edge (see Supplementary Note IV and Fig. S9). This analysis yields the following relation tan2 φ = tan φ0 + ζ 1/ tan φ0 − ζ (3) between the leading angle φ and friction, where ζ = (Fe,θθ)ηvb/Gb denotes the scaled friction normalized by the biofilm shear modulus Gb, and Fe,θθ describes the circumferential compression at the biofilm edge. In the absence of friction, i.e., when ζ = 0, Eq. (3) reduces to φ = φ0. In the presence of friction, our analy- sis predicts that the leading angle φ increases with ζ when ζ < 1/ tan φ0, while the biofilm edge bulges out and constantly tumbles (no steady-state translation) if ζ > 1/ tan φ0. The experimental difficulty in measuring the friction coefficient η and the circumferential compression Fe,θθ precludes a direct quantitative comparison with theory; nevertheless, it is clear from the measured φ of exper- imental biofilms that higher friction (i.e., higher con- centration agar) increases the biofilm leading angle (see Methods and Figs. 2E and S9). Non-uniform biofilm growth results in anisotropic stress The evolution of mechanical stresses during the early stages of biofilm growth dictates the onset of mechanical instability and the consequent morphology of the wrin- kles. Thus, we investigate the evolution of the magni- tude of radial stress σrr, the magnitude of circumferen- tial/hoop stress σθθ, and the stress anisotropy defined as ασ = (σθθ − σrr)/(σθθ + σrr) [50]. The isotropic stress state corresponds to ασ = 0, while pure hoop stress cor- responds to ασ = +1 and pure radial stress corresponds to ασ = −1. The spatial distributions of stresses have distinctive characteristics during each of the three kinematic stages Initially, the inner core of the of biofilm expansion. biofilm only minimally expands (F0 Fg ≈ Ik), which, e,k given the material growth of the biofilm, Fg = λgIk, must result in a compensating isotropic in-plane deformation (F0 g Ik), and thus an isotropic compressive stress state with σrr = σθθ (Figs. 3A-C). Moreover, stresses are approximately uniform in magnitude throughout the im- mobile core region of the biofilm, but decline in the outer mobile region. Note that the value of radial stress σrr necessarily decreases to zero at the edge of the biofilm, while the hoop stress σθθ can be nonzero. Therefore, the stress anisotropy is initially localized to the outer mobile region. e,k ≈ λ−1 As the biofilm continues to grow, internal stresses in- crease exponentially in time, and eventually overcome friction, enabling the entire biofilm to expand uniformly (Figs. 2D and 3A-C). During this second stage, mechan- ical stresses continue to increase exponentially and ac- quire a characteristic parabolic profile (see Supplemen- tary Note III and Fig. 3). During the third stage when nutrients become depleted, stresses increase more slowly near the center of the biofilm due to the reduced rate of biofilm growth, while the magnitude of hoop stress near the edge still increases exponentially due to continuous biofilm growth in this nutrient-rich region (Fig. 2C). As a result, when friction is low, the location of the maximum hoop stress shifts away from the center of the biofilm during the later stages (Fig. 3A). Note that the stress anisotropy is always positive (Fig. 3C), meaning that the compressive hoop stress is always larger in magnitude than the radial stress. We found that the region of the biofilm under anisotropic stress becomes larger during the third stage of development (Fig. 3C). Thus, we hypothesized that the non-uniform growth pattern due to depletion of nu- trients plays an important role in generating anisotropic stresses. To quantify the extent of stress anisotropy for 6 the entire biofilm, we computed the normalized range of anisotropy ∆rα/rb, defined as the radial range of the area where ασ > 0.1 relative to the biofilm radius rb, as a function of time. According to our model, the increase of ∆rα/rb is accompanied by a narrowing of the nutrient- rich zone (Fig. S10). To further explore this connection, we also computed the normalized anisotropy range for a uniformly growing biofilm, which we found to be close to zero (Fig. S10). We conclude that the faster growth at the biofilm edge promotes predominantly circumferential stress (see Supplementary Note V), which explains the appearance of radial wrinkles in the peripheral regions in experiments (Fig. 1A). Friction favors isotropic stress and shifts the position of maximal circumferential stress How does friction ξ affect the distribution of mechan- ical stresses in a growing biofilm? To address this ques- tion, we compared the distribution of circumferential stress σθθ and stress anisotropy ασ for a series of sim- ulations with different friction coefficients (see Fig. 3D). Notably, at a typical time when biofilms start to form patterns in experiments, our simulations show that the radial position r∗, corresponding to the maximal circum- ferential stress, varies with the magnitude of friction: r∗ is near the biofilm edge when friction is small, while r∗ is near the biofilm center when friction is large (Figs. 3D and E). Moreover, the stress anisotropy α∗σ at r∗ de- creases towards zero (isotropic stress state) with increas- ing friction (Fig. 3D and E). Intuitively, these differences in stress distribution re- sult from the counteracting effects of friction and non- uniform growth. Friction impedes biofilm expansion (Fk → Ik when η → ∞), retards the relaxation of growth-induced isotropic compression (F0 e,k → λ−1 g Ik when η → ∞), and thus favors isotropic stress in the biofilm center. By contrast, non-uniform growth favors peripheral circumferential stress due to the mismatch be- tween the biofilm perimeter that increases only linearly in time, and the exponential material growth of the biofilm at the edge. The fact that when friction is small the cir- cumferential stress close to the biofilm rim is larger than that at the center (Fig. 3D and F) explains why, in experi- ments, the wrinkle pattern emerges from the outer region (Fig. 1A and B). In contrast, in experiments with high concentration agar, the wrinkle pattern first appears in the center of the biofilm because the large friction results in strong isotropic compression in that region. The in-plane stress field determines the morphology of biofilm wrinkle patterns Lastly, we address how the stress profiles discussed above dictate the morphology of biofilm wrinkles. As the biofilm grows, the magnitude of compressive stresses 7 g Rb0) η(kmax Gb(H/Rb0) , where η denotes the friction coefficient, kmax Fig. 3: Spatiotemporal evolution of the stress field in a growing biofilm. (A-C) Magnitude of (A) the circumferential stress σθθ, (B) the radial stress σrr, and (C) the stress anisotropy ασ = (σθθ − σrr)/(σθθ + σrr), plotted against the radial coordinate r at the designated times for high friction (left, ξ = 20) and low friction (right, ξ = 4). All the stresses are normalized by the biofilm shear modulus Gb, i.e., σ = σ/Gb. Identical simulation parameters were used as in Figs. 2C and D. (D) Circumferential stress σθθ (left, normalized by the stress at the center of the biofilm σ0 θθ) and stress anisotropy ασ (right) versus the radial coordinate r normalized by the biofilm radius rb plotted for different dimensionless friction parameters, defined as ξ = denotes the maximum growth rate of the biofilm, and Rb0 and H denote, respectively, the initial biofilm radius and thickness. Color scale indicates the values of ξ on a logarithmic scale. The five curves (colored from blue to cyan) correspond to ξ = e0.7, e1.7, e2.7, e3.7, and e4.7, respectively. r∗ denotes the radial position where the circumferential compressive stress reaches a maximum (black dotted curves). (E) The normalized radial coordinate r∗/rb and the stress anisotropy α∗σ at the position of maximum circumferential compressive stress, plotted as functions of the dimensionless friction parameter ξ. (F) The magnitude of the largest circumferential compressive stress σ∗θθ (r = r∗, solid curve) compared to the circumferential stress at the edge (r = rb, dotted curve) and at the center (r = 0, dashed curve) for different dimensionless friction parameters ξ. When the circumferential stress at the edge of the biofilm is larger (smaller) than the circumferential stress at the center, radial patterns start forming near the edge (near the center) as indicated by the inset figures. In experiments, the transition between two different morphologies occurs at 0.7% agar concentration (gray dashed line). Panels D-F show simulation results at time t = 30 h, which is roughly when the experimental biofilms start to form periodic wrinkles. g s G1/3 increases (Fig. 3). Once compressive stresses reach a critical value σc, the flat state becomes unstable to the formation of wrinkles [24, 51, 52]. The critical com- pressive stress σc increases with Gs(∝ ξ), and scales as σc ∼ G2/3 in the asymptotic limit where the shear modulus of the biofilm Gb is much larger than that of the agar substrate Gs (see Methods and [24] for details). It was previously shown that, for highly anisotropic stresses, wrinkles are oriented orthogonal to the direction of maximum compressive stress, whereas for isotropic stresses, wrinkles form zigzag herringbone- b like patterns [51, 52] (Fig. S11). The stress profiles in Fig. 3 are thus consistent with the experimental observa- tions in Fig. 1A that radial wrinkles form in the outer re- gions, where the stress is predominantly circumferential, whereas zigzag wrinkles form in the core region where the stress is largely isotropic. In order to more quantitatively understand the spa- tiotemporal evolution of biofilm wrinkle patterns, we de- veloped a 2D coarse-grained model that employs two scalar order parameter fields A and S to describe, re- spectively, the amplitude and the shape of the wrinkle 1 1 Anisotropy ασ 0.8 0.8 .4 0.4 00 0 ξ 1 1 0.5 0.5 0 0 0 0 1 2 0 1 2 200 0.5 0.5 r / rb r/rb ασ r c 1 1 0.5 0.5 r / rb r/rb 1 1 0 0 101 102 101 101 Friction ξ 102 102 101 101 100 100 Stress σ θθ ~ 2 r (mm) 0 2 4 r (mm) 1 0.5 m/min) 6 h 0.5 0.5 1 h 0 0 0 2 4 2 4 r (mm) r (mm) 16 h 4 4 vr ( 2 2 r (mm) r (mm) 0 0 0 0 Anisotropy ασ vr ( m/min) 2 1 0 012 0 0 D σθθ /σ0 θθ ~ ~ c 0.5 0.5 position r/rb Maximum stress E F 6 6 6 6 Low friction ξ = 4 2 4 r (mm) 0 024 024 m/min) 16 h vr ( 4 m/min) 16 h vr ( 4 0 6 h 1 h 2 r (mm) 6 h 1 h 0 High friction ξ = 20 2 0 024 4 024 2 0 4 1 1 m/min)A vr ( Stress σ θθ ~ B vr ( m/min) Stress σ rr ~ C 8 e , maps a pre-stressed, flat biofilm (top) to a wrinkled biofilm Fig. 4: Morphology and spatiotemporal dynamics of biofilm wrinkle patterns. (A,B) Schematic of the wrinkling model. The wrinkling deformation tensor, denoted by Fw (bottom). Color code as in Fig. 2B. The wrinkling pattern is characterized by two dimensionless scalar fields: the normalized amplitude A and the shape factor S. A is defined as the product of the wave number (denoted by kw) and the amplitude (denoted by Aw) of periodic wrinkles. S = kpAp/√6 where kp and Ap denote, respectively, the wave number and the amplitude of in-plane wiggles. The shape parameter S was normalized such that its values are restricted to the interval [0, 1]. Right: Panels in A show close-up cross-sections of a flat ( A = 0) and a wrinkled ( A > 0) biofilm. uz denotes the out-of-plane displacement. B shows top view schematics of the herringbone ansatz, uz = Aw cos[kw(xθ − Ap cos kpxr)], for a straight striped pattern ( S = 0) and a zigzag pattern ( S > 0). Here, xθ = rθ and xr = r denote, respectively, the linear coordinates along the circumferential and radial directions. (C, D) Kymograph representation of the evolution of patterns of the modeled biofilm for designated parameters (σc denotes the critical stress, normalized by the biofilm modulus Gb, for the onset of the wrinkling instability). The top and bottom kymographs can be interpreted as biofilms grown on different agar concentrations (see Fig. 1B). σc in C is chosen such that the wrinkling instability occurs at a time similar to that in experiment. σc in D is then inferred according to the dependence of critical stress on Gs/Gb ∝ ξ. See Methods for details. Solid, dotted, and dashed curves, respectively, denote the boundaries of the entire biofilm, regions with a radial stripe pattern ( A > 0, S = 0), and regions with a zigzag pattern ( A > 0, S > 0). Gray denotes the region without any pattern. For the patterned regions, the color of each spatiotemporal bin indicates the local amplitude A(r, t) and shape S(r, t) of the pattern. The color code is shown in the inset of D. e F0 patterns. Specifically, A = 0 ( A > 0) corresponds to the flat (wrinkled) state, and S = 0 ( S > 0) corresponds to striped (zigzag) wrinkles. The total elastic deformation e is decomposed as the superposi- of a biofilm Fe = Fw e , which also deforms tion of the wrinkling deformation Fw the agar, and the planar compression F0 e (Fig. 4A). We follow previous work [51, 52] to describe wrinkles of dif- ferent morphologies, and we use the herringbone ansatz to approximate the out-of-plane displacement in coarse- grained patches of the biofilm (see Supplementary Note VI and Fig. 4B). The primary sinusoidal wrinkling with amplitude Aw (in the vertical z direction) and wavelength λw = 2π/kw occurs along the direction that corresponds to the maximum compressive stress (circumferential di- rection eθ in our case). The secondary sinusoidal wig- gles with amplitude Ap (in the horizontal direction) and wavelength λp = 2π/kp appear in the orthogonal di- rection (Fig. 4B). In terms of these quantities, the rel- evant dimensionless order parameters are A = kwAw and S = kpAp/√6. The formation of wrinkles relaxes the elastic compres- sional energy of the biofilm, but at the expense of the bending energy of the biofilm and the elastic deforma- tion energy of the agar. By taking into account these energy contributions and using the above ansatz for the shape of wrinkles, we derived the following total free- energy density per unit area (see Supplementary Note 20 20 1 1 Stripe 10 10 1 ~ 0.5 S 0 1 0 0 0 ~ A No pattern Zigzag 0 0 0 0 r (mm) 10 10 20 20 Low friction ξ = 4 ~ σc = 3.6 0 20 20 40 40 60 60 20 10 20 10 0 High friction ξ = 20 20 20 ~ σc = 6.0 40 40 60 60 20 20 10 10 t (h) t (h) Section view C uz 2Aw 2π/kw D 1 0 uz /Aw -1 Top view 2π/kw 2π/kw 2Ap 2π/kp ~ S = 0: Stripe ~ S > 0: Zigzag A Fw e eθ B e r 9 (Figs. 1B and 4D). According to our model, compared to the case where the wrinkling instability is prevented, the expansion of the wrinkled biofilm is slowed, the stress anisotropy is reduced, and the magnitude of compressive stress is reduced as well (Figs. S11 and S12). Thus, our model suggests that wrinkling due to a growth-induced mechanical instability feeds back and further influences biofilm expansion and pattern formation by modifying the distribution of internal stress. DISCUSSION Our experimental and modeling results highlight the connections between nutrient supply, bacterial growth, biofilm and substrate mechanics, and friction in shap- ing the morphology of developing bacterial biofilms on soft substrates. The depletion of nutrients beneath the center of the biofilm leads to localized growth primar- ily near the biofilm edge, consistent with previous ex- periments [24, 53, 54]. This uneven growth profile, in turn, produces anisotropic compressive stresses, which are predominantly circumferential at the periphery of the biofilm, but are largely isotropic in the central region. The consequence of such a stress profile is the formation of radial wrinkles in the outer region of the biofilm and a zigzag herringbone-like pattern in the central region. Moreover, the location of the maximum circumferential stress -- where wrinkles first appear once the magnitude of the stress reaches a critical value -- varies with the mag- nitude of friction, from near the outer edge when friction is small to near the center when friction is large. As a result, for biofilms grown on soft agar substrates with low friction, wrinkles first appear in the peripheral region and propagate inward. In contrast, for biofilms grown on stiff agar substrates with high friction, wrinkles first appear in the central region and propagate outward. What are the biological implications of forming 3D biofilm structures? One possibility is that the wrin- kled thin film structure provides a larger surface area-to- volume ratio compared to a flat film, thereby enhancing access to nutrients and conferring growth advantages to the bacterial population [33, 55]. Furthermore, under ad- verse nutrient conditions, biofilms disassemble through a process called dispersal, and dispersing cells primarily de- part from the biofilm's outer surface [56]. Therefore, the large surface area of wrinkled biofilms may facilitate dis- persal when submerged. The convoluted 3D structure of biofilms also reduces the average distances between cells compared to a flat film of the same area, which might enhance communication between bacterial cells, e.g. via quorum-sensing signaling [57, 58]. Finally, the 3D biofilm structure positions biofilm cells at different heights, po- tentially generating a "bet hedging" strategy under par- ticular conditions. For example, the rough surfaces of wrinkled biofilms exposed to external flows will alter the flow field, forming large (small) shear stress zone near the peaks (valleys) of wrinkles. Consequently, the cells (cid:16)(σ0 θθ − σc)(1 + 3b S2) + 3(σ0 (cid:16)1 + (6b + 3) S2(cid:17) A4, rr − σc) S2(cid:17) A2 (4) 1 4 1 8 Ψtotal k GbH ≈ − + VI) θθ < 0 and σ0 which is valid for stresses near the critical stress σc = (3Gs/Gb)2/3 of the wrinkling instability. Here, stresses σ ≡ σ/Gb are normalized by the biofilm shear modu- lus Gb. σ0 rr < 0, respectively, denote the circumferential and radial pre-stress due to the planar compression F0 e. We set b ≈ 2/3 to ensure that the re- laxed stresses due to wrinkling remain isotropic when the imposed pre-stress is isotropic (see Supplementary Note VI). Here, we assume that the dynamics of wrinkling is determined by the slower dynamics of the stress field. Under this approximation, the predicted wrinkled mor- phology corresponds to the minimum of the free en- ergy in Eq. (4). The wrinkling instability occurs via two successive continuous phase transitions controlled by the magnitude and by the anisotropy of the pre- stress (see Supplementary Note VI and Fig. S11). The primary bifurcation from the planar state ( A = 0) to the wrinkled state ( A > 0) occurs if the magnitude of the maximum compressive stress σ0 θθ exceeds the criti- cal value σc, while a secondary bifurcation from striped wrinkles ( S = 0) to zigzag wrinkles ( S > 0) occurs only when the stress anisotropy is sufficiently small, θθ + σc). Note that for isotropic σ < (σ0 α0 rr) the free-energy den- compressive pre-stresses (σ0 sity is minimized by S = 1. θθ − σc)/(3σ0 θθ = σ0 The evolution of the stress field determines the biofilm wrinkling morphodynamics Wrinkling relaxes the mechanical stresses in the biofilm by releasing the in-plane compressive strain through out- of-plane deformation. Once wrinkling occurs, this relax- ation mechanism prevents the magnitude of compressive stresses from increasing beyond the critical stress (see Methods, Supplementary Note VI and Fig. S11). We incorporated the above mean-field description of the wrinkling instability and consequent stress relaxation into our chemo-mechanical model (see Methods and Sup- plementary Note VI for details). Consistent with the experimental observations in Fig. 1B, we find that for biofilms grown on low concentration agar (low shear mod- ulus of the substrate and small friction since we assume η ∝ Gs), radial wrinkles initiate near the outer edge, then propagate inward and once they reach the center, zigzag wrinkles form in the core region (Fig. 4C). On the other hand, for biofilms grown on high concentration agar (high shear modulus of the substrate and large friction) radial wrinkles initiate in the center and expand outward, while zigzag wrinkles simultaneously appear in the core region near the peaks will exhibit a larger probability to detach while the cells near the valleys will tend to stay attached to the surfaces. [59]. Our results provide insight into the spatiotemporal de- velopment of V. cholerae biofilm morphology, but it re- mains to be explored whether and how the increasing mechanical stress and/or the formation of the 3D biofilm structure affect the proliferation rate of bacteria, alter biofilm matrix production, or promote survival success of cells in particular biofilm regions. Additional experimen- tal studies will be required to complete our understand- ing of how growth, mechanical stresses, and morphologi- cal transitions are coupled to drive biofilm development. Furthermore, our study focused only on the initial stages of wrinkling, during which the amplitudes of wrinkles are small and biofilms remain in contact with the agar sub- strate. We previously demonstrated that at later stages of development, biofilms can locally delaminate from the agar substrate, which significantly influences the subse- quent development of morphological patterns [24]. In this work, we modeled the rheology of growing V. cholerae biofilms as that of a hyperelastic material. How- ever, our previous measurements show that V. cholerae biofilms actually behave as more complex viscoelastic media that yield upon large shear deformation [31]. In- deed, our elastic material model leads to stresses that are larger than the measured yield stress. We thus suspect that yielding constantly occurs during V. cholerae biofilm growth. More generally, reorganization and yielding of growing biological materials are commonly observed dur- ing morphogenesis, for example in plants [60], fruit flies [61, 62] and brain tissues [63]. Thus, the effects of vis- coelasticity [64] and elastoplasticity (Figs. S13 and S14) of biofilms on their morphological development will be an important topic for future studies. The concepts we presented here to analyze the devel- opment of V. cholerae biofilms should also be applicable to biofilms of many other bacterial species that form sim- ilar morphological patterns [65 -- 67]. However, there are also examples of biofilms with distinct morphologies, such as the distorted concentric rings observed in wild-type Pseudomonas aeruginosa PA14 biofilms [68] and biofilms formed by Escherichia coli K-12 strain W3110 [19]. Our model suggests that if biofilm growth and/or matrix pro- duction is faster in the central region than the outer re- gion, one expects a region in which the radial compres- sive stress surpasses the circumferential stress. In this case, our model predicts that wrinkles will form as con- centric (possibly distorted) rings (see Fig. S11). Such a pattern of matrix production is indeed reported in the two biofilm formers mentioned above. For example, in wild-type P. aeruginosa biofilms, cells at the biofilm cen- ter display upregulated matrix production due to oxygen limitation, whereas cells located in the oxygen rich pe- riphery downregulate matrix production [68]. In biofilms formed by E. coli K-12 strain W3110, cells generate ma- trix components (amyloid curli fibers) only upon entry into stationary phase when nutrients are depleted, which 10 typically occurs first at the biofilm center [19]. Thus, we expect that similar physical mechanisms to those un- derlying the dynamics of expansion and pattern forma- tion of V. cholerae biofilms may be widely applicable to other bacterial biofilms, including those with distinct morphologies. MATERIALS AND METHODS Growing and imaging experimental biofilms Bacterial strain and biofilm growth The V. cholerae strain used in this study is a deriva- tive of the V. cholerae O1 biovar El Tor strain C6706str2 [69] that harbors a missense mutation in the vpvC gene (VpvC W240R), which elevates the levels of c-di-GMP and confers a rugose biofilm phenotype [25]. Standard lysogeny broth (LB) medium solidified with different per- centages of agar was used as the solid support on which biofilms were grown. Initially, V. cholerae was streaked onto LB plates containing 1.5% agar and grown at 37◦C overnight. Individual colonies were selected and inocu- lated into 3 mL of LB liquid medium containing ∼10 glass beads (MP Biomedicals Roll and Grow Plating Beads, 4 mm diameter) and these cultures were then grown at 37◦C with shaking to mid-exponential phase (about 5 h). Subsequently, the cultures of bacteria were mixed by vor- tex to break changeclumps up into individual cells, the OD600 was measured with a cell density meter (Amer- sham Biosciences Ultrospec 10), and then the cultures were diluted back to an OD600 value of 0.5. 1 µL of these preparations were spotted onto pre-warmed agar plates made with different concentrations of agar. Sub- sequently, the plates were incubated at 37◦C. During the first 10 h after inoculation, bacterial colonies formed the initial biofilms without extending beyond the inoculated circle (radius R0 ≈ 2 mm). These biofilms were used as the initial/reference configurations for modeling, with t = 0 in our simulations corresponding to the time when a biofilm starts expanding radially. Four biofilms were grown per agar plate for the surface topography mea- surements, while for the time-lapse imaging, one biofilm was grown on each agar plate. Time-lapse transmission imaging The imaging system has been described previously [24]. Briefly, an agar plate containing the inoculum was placed on an LED illumination pad (Huion L4S Light Box) and imaged with a Nikon D3300 SLR camera equipped with a Sigma 105 mm F2.8 Macro Lens. The entire setup was placed in a 37◦C environmental room and was covered to exclude light. The camera was controlled with DigiCam- Control software. Imaging started 5 h after inoculation when the camera was capable of focusing on the grow- ing biofilms. Image snapshots were taken automatically every 15 min for 3 days. Image processing The protocol and Matlab codes used to analyze the morphological features of biofilms have been reported previously [24]. In brief, an intensity-based threshold- ing method was used to binarize the pre-processed trans- mission images (using a built-in thresholding function in Matlab) and to separate the biofilm region from the back- ground. For each biofilm, the transmission image taken 12 h after inoculation was used to define the biofilm cen- ter for the entire time course. The biofilm radius rb was computed by averaging the distance between each point on the circumference and the center. Regions of the biofilm were binned into rings of width 0.2 mm for further analysis. First, we took the Fourier transform of the image intensity in the circumferential direction for each separate ring at distance r from the center of the biofilm. Radial stripes appear in the re- sulting power spectrum as a sharp maximum at non-zero spatial frequency f (r, t). The radial coordinate, at which the peak power instead appears at zero spatial frequency, was defined as the boundary of the region with a radial stripe pattern. Next, the radial intensity distribution I(r) was obtained by averaging the intensity values over the circumferential direction for each ring. The intensity I(r) for the disordered core is distinctly different (darker) from that of the outer region of the biofilm. Thus, we set a threshold intensity value for each biofilm to identify the central region with a disordered zigzag pattern, enabling us to measure this central region's radius as a function of time. The velocity field of an expanding biofilm was mea- sured by particle image velocimetry (PIV) performed with the open source tool PIVlab [70]. The 2D displace- ment field between two successive frames (separated by 30 min) was computed via a Fourier transform correlation with three passes. The sizes of interrogation windows for the three passes were chosen to be 128 pixels, 64 pixels, and 32 pixels, respectively. By averaging the radial com- ponents of the coarse-grained velocity vectors over the circumferential direction for each ring, we obtained the radial velocity field in Fig. 2D. Error bands correspond to standard deviations of the means. 3D confocal profiling and leading angle measurement The surface profiles of biofilms grown for different times were analyzed with a Leica DCM 3D Micro-optical System. A 10× objective was used to image a roughly 3 mm × 3 mm region of the biofilm, with a step size of 2 µm in the z direction. Subsequent processing and analy- ses were performed using Leica Map software. First, the 11 three-point flattening procedure was performed on the agar surface to level the image. 3D views of biofilms were then rendered with a built-in function in the software. To measure the leading angle of an expanding biofilm, line profiles perpendicular to the biofilm periphery and spanning the region from the agar surface to the top sur- face of the biofilm were generated at five different loca- tions. For each line profile, two points were manually selected on the biofilm edge, which were used to obtain a sloped line. The leading angle was then extracted with a built-in function in the software from the slope of this line. The measurements of the initial angle were per- formed 12 h after inoculation. The steady-state leading angles were measured every 4 h from 24 h to 48 h after inoculation. Modeling biofilms Continuum modeling To model the combined role of growth and mechanics in the morphological transition of V. cholerae biofilms, we adopted the formulation of elastic growth [41, 42], where the total geometric stretches F = ∂x/∂X0, defined as the deformation gradient from the initial configura- tion X0 to the current configuration x, are decomposed into stretches Fg due to growth and stretches Fe due to elastic deformation. Below, we describe a 2D chemo- mechanical model of biofilm development that includes diffusion of nutrients and their uptake by bacteria, bac- terial growth/extracellular matrix production, and me- chanical deformation. The subscript k is used to denote the in-plane components of 3D vectors/tensors and re- duced 2D variables, while a tilde ∼ above a variable is used to denote a dimensionless variable. Growth: Local growth of the V. cholerae biofilm i.e., Fg = is treated as horizontal isotropic growth, (cid:18)λg(t)Ik 0 1(cid:19), where Ik denotes the 2D identity matrix. 0 λg(t) describes the stretch in the horizontal direction due to growth, while there is no growth in the z di- rection because biofilms maintain approximately con- stant thickness. The stretch due to growth evolves as ∂λg/∂t = kgλg, where kg denotes the local growth rate. To capture the nutrient-dependent spatially non-uniform growth, kg(ck) is assumed to be a function of the nor- malized 2D nutrient field ck(xk, t) (normalized by the concentration of nutrients c0 at the edge of the biofilm, i.e., ck(xk, t) ≡ ck(xk, t)/c0), which reflects the nutri- ent availability in the agar medium. The equation that describes diffusion and uptake of nutrients is ∂tck = φ(ck)), where D∇kck − Q(Fe)φ(ck) = Q0(a2 D is the nutrient diffusion coefficient, Q0 characterizes the maximum nutrient uptake rate by bacteria in the undeformed grown configuration, and Je,k = det(Fe,k) is introduced to account for the increased areal den- c∇kck − J−1 e,k sity of bacterial cells upon elastic deformation of the biofilm. The characteristic width of the nutrient-rich an- nulus near the biofilm edge is given by ac = (D/Q0)1/2. We assume that the uptake of nutrients depends on the local availability of nutrients via the Monod law, i.e., φ(ck) = ck/(ck + K), where K = 0.5 is the concentra- tion of nutrients at the half-maximal uptake rate [45] (Note that our model results are insensitive to the specific choice of K: for a different K, a similar ck profile can be obtained by adjusting the uptake rate Q0; see below for the fitting procedure). Finally, the growth rate is speci- fied as kg(ck) = k0φ(ck) + kr, where k0 is the maximum rate of nutrient-dependent growth and a small, constant nutrient-independent growth rate kr is added to account for the residual biofilm growth due to the vertical diffu- sion of nutrients (see Supplementary Note II and Figs. S2 and S3). Mechanics: The growth of the biofilm drives its expan- sion. As the biofilm moves relative to the agar, the fric- tion f between the biofilm and the agar impedes biofilm expansion and induces internal mechanical stresses σ. Friction is modeled as a viscous drag, i.e., f = −ηvk, which is proportional to the expansion velocity of the biofilm vk = (∂xk/∂t)X0, and the drag coefficient η is assumed to be proportional to the agar shear modulus Gs [47 -- 49, 71, 72] (see Supplementary Note IIG and Fig. S5). In order to relate stress σ to elastic deformation Fe, we leverage the fact that the thickness of the biofilm (∼ 100 µm) is always about 10 to 100 times smaller than its radius (' 5 mm), and we treat the biofilm as a plane stress thin film made from nearly incompress- ible hyperelastic material. Thus, the thin-film defor- mation Fe = Fw e is decomposed into the product of e and a planar deformation a wrinkling deformation Fw denotes the in-plane com- F0 e =(cid:18)F0 pression and γ0 denotes the resulting vertical stretch. Lateral force balance yields ηqvk = H∇k · (γ0σk), where H denotes biofilm thickness in the undeformed configu- e ) is a factor that accounts for the ration, and q = q(Fw increase of the contact area between the biofilm and the agar due to the wrinkling profile. γ0(cid:19), where F0 e F0 e,k 0 e,k 0 (F0 e,k = Gb[F0 e,k Prior to wrinkling, the biofilm is flat and thus Fw = ∂xk/∂X0,k, and from γ0 = 1/det(F0 e,k e = I e can be obtained from and q = 1. The deformation F0 ) due to F0 e,k incompressibility. The in-plane stresses are calculated to )T − (γ0)2Ik], where Gb denotes be σ0 k the biofilm shear modulus (see Supplementary Notes I,II for details). A wrinkling instability occurs once com- pressive stresses reach the critical value. To describe the wrinkling deformation Fw e we use two coarse-grained scalar fields, the amplitude A and the shape S (see Fig. 4 and Supplementary Note VI). The fields A and S are computed from a Landau-Ginzburg-type free en- ergy density Eq. (4) and the factor q is approximated as q( A) = 1+ 1 A2[1+3(b+1) S2], where b ≈ 2/3 (see Supple- 4 mentary Note VI for details). The stress relaxation from the pre-stress σ0 k is described by (see Supplementary Note VI) to the true stress σk due to wrinkling 12 σθθ ≈ σ0 σrr ≈ σ0 θθ + A2h1 + (3b + 3/2) S2iGb, rr + A2h1/2 + (3b/2 + 3) S2iGb. (5) Note that for isotropic compressive pre-stresses (σ0 θθ = rr), the relaxed stresses remain isotropic ( S = 1, b = σ0 2/3). Dimensionless governing equations: We define dimen- sionless variables σ = σ/Gb, x = xk/Rb0, and τ = t/τ0, where the shear modulus of the biofilm Gb was chosen as the scale for stresses, the initial biofilm radius Rb0 as the characteristic length scale, and the inverse of the growth rate at the edge of the biofilm τ0 = (kmax )−1 = [kg(r = rb)]−1 as the characteristic time scale associated with biofilm expansion. Upon non-dimensionalizing the equa- tions describing biofilm growth and mechanics discussed above, we obtain the following equations g Nutrient diffusion and uptake: Nutrient limited growth: Force balance: Constitutive relation: kck − J−1 ∇2 e,k c ∂τ ck = Q0ha2 ∂τ λg =h(1 − kr) ∇k · (γ0 σk) = ξq( A)vk, σ0 k = F0 φ(ck) φ(1) e,k(F0 e,k)T − (γ0)2Ik, φ(ck)i, + kriλg, (6a) (6b) (6c) (6d) where Q0 = Q0τ0, kr = krτ0, and the dimensionless fric- tion ξ = η(Rb0/τ0) Gb(H/Rb0) is identified to be a control param- eter of the model. Before wrinkling occurs, σk = σ0 . k After wrinkling occurs, Eq. (6d) describes the pre-stress , and the actual stress σk is computed from Eq. (5). σ0 k Taken together, the set of dimensionless governing equa- tions is able to describe both the planar expansion of the biofilm ( A = 0) and the 3D biofilm wrinkling morphol- ogy ( A > 0). The parameters in our model are either estimated directly from experiment or are obtained by fitting to experimental data (see Fig. S1 and Table S2). Numerical simulations The numerical solutions of Eq. (6) were obtained by performing finite element simulations. Rather than solv- ing Eqs. (6a) and (6c) in the Eulerian frame, these equa- tions were rewritten and solved in the Lagrangian frame of reference (see Supplementary Note IID for details). We further assumed axisymmetric solutions and expressed the governing equations in polar coordinates to numer- ically solve for six scalar fields r, λg, ck, γ, A, and S as functions of the dimensionless initial radial coordinate R0. The validity of the axisymmetry assumption was verified by comparing to simulations on 2D circular do- mains with no assumption of symmetry. We used a fixed 1D domain of R0 ∈ [0, 1] that was discretized and gener- ated by Gmsh [73]. The geometric stretch near the edge R0 = 1 is larger than that near the center R0 = 0 due to the non-uniform growth. Therefore, we used a finer discretization of the domain near R0 = 1 to ensure high precision numerical solutions. The initial conditions are r( R0, τ = 0) = R0, ck( R0, τ = 0) ≡ 1, λg( R0, τ = 0) ≡ 1, γ0( R0, τ = 0) ≡ 1, A( R0, τ = 0) = S( R0, τ = 0) ≡ 0 , and boundary con- ditions ck( R0 = 1, τ ) = 1, σrr( R0 = 1, τ ) = 0. Partial differential equations were then converted to their equiva- lent weak forms and computationally discretized by first- order (2 noded) linear elements [74], and implemented in the open-source computing platform FEniCS [75]. The time increment was set to be ∆τ = 0.01. At each time step, we used the standard Crank-Nicolson method to perform the numerical integration [76]. To ensure nu- merical convergence, we checked explicitly whether the wrinkling instability occurred by evaluating the differ- ence between the circumferential stress and the critical σ > 0 at stress δσ = σ0 τ = n∆τ for the wrinkling to occur at τ = (n + 1)∆τ. θθ − σc. We required that δ(n) Choice of parameters Critical stress for wrinkling: Our previous study re- vealed that a trilayer model quantitatively captures the biofilm wrinkle wavelength [24]. The trilayer theory also predicts how the critical stress varies with the stiffness contrast between the biofilm and the substrate Gs/Gb ∝ ξ [77] (see also [24] for the calculated values of criti- cal stress and Gs/Gb for different agar concentrations). However, our chemo-mechanical model (Eq. (6)) reaches the theoretical critical stress earlier than the time when wrinkling occurs in the experiments because we model biofilms as elastic materials and do not consider viscoelas- ticity and plasticity (see Discussion and Supplementary Note VII). In practice, we rescale the critical stress σc in Fig. 4C and D such that wrinkling instability in the sim- ulations occurs at a time similar to that in experiment. Fitting parameters from the velocity profiles: The di- mensionless friction parameter ξ and the dimensionless maximum rate of nutrient uptake Q0 were determined by fitting the radial velocity profiles of the modeled biofilm to those extracted from experiments at different times. The similarity between the radial velocity profiles was assessed in terms of the normalized mean squared dis- tance (MSD). In experiments, we measured the radial velocity profiles for a biofilm grown on 0.7% agar con- centration at 40 different time points tj separated by 30 min from t = 0 h to t = 20 h (before the wrinkling instability occurs) as described above in the Image Pro- 13 cessing section. At each time tj, the experimental data were represented as (ri,j, vi,j) (i = 1, . . . , Nj; vi,j aver- aged over the circumferential direction). The number of data points Nj at each time point is equal to the ratio of the biofilm radius to the width of radial bins. For a particular set of parameters (ξ, Q0), we first numeri- cally computed the velocity profiles vr(r, tj) of the mod- eled biofilm. For each time point tj we computed the normalized squared distance (SD) ∆s2 i,j between the ex- perimental data points (ri,j, vi,j) and the simulated pro- file vr(r, tj) as ∆s2 where we used a characteristic length scale L0 = 5 mm and a characteristic velocity V0 = 3 µm/min. The nor- malized SD between the radius and edge velocity for the experimental biofilm and those of the modeled biofilm was used as one additional data point ∆s2 Nj +1,j associ- ated with time tj. Finally, the normalized MSD was cal- i,j. We searched the culated as parameter space to find the optimal parameter values ξ∗ and Q0∗ that minimize the normalized MSD (Fig. S1). For simulations with different friction, we varied the pa- rameter ξ keeping all the other parameters fixed. (Nj+1)NtPNt j=1PNj +1 r (cid:8)( ri,j−r L0 )2(cid:9) i,j = min )2 + ( vi,j−vr(r,tj ) V0 1 i=1 ∆s2 Analysis of the biofilm leading angle To compare the biofilm leading angles in experiments with theoretical predictions, we inferred the value of η/Gb for biofilms grown on 0.7% agar by fitting the veloc- ity profiles as described above, i.e., (η/Gb)∗ = ξ∗(H/Rb0) . Rb0/τ0 Next, we inferred the normalized friction (η/Gb)vb for biofilms grown on different agar concentrations (agar shear modulus denoted by Gs) by making the assump- tion that (η/Gb) ∝ Gs/Gb, i.e., (η/Gb)vb = (η/Gb)∗ × (Gs/Gb)agar conc. = 0.7% × vb. The uncertainty of these val- ues (Fig. 2E horizontal error bars) was estimated by tak- ing into account the measurement errors of Gs, Gb, and vb. The value of the circumferential compression Fe,θθ in Eq. (3) remains undetermined. Nevertheless, we can estimate Fe,θθ ∈ (0.7, 0.9) from the wrinkling instability analysis [24]. The specific choice of Fe,θθ in this range only minimally affects the results (Fig. S9). (Gs/Gb) Data and software availability Code availability Matlab codes for the image processing have been described in a previous publication [24]. The sim- ulation codes used to model the biofilm are avail- able on GitHub (https://github.com/f-chenyi/biofilm- mechanics-theory). Data availability The data are available upon request. ACKNOWLEDGEMENTS This work was supported by the Howard Hughes Med- ical Institute (B.L.B., etc.), National Science Foundation Grants MCB-1713731 (B.L.B.), MCB-1853602 (B.L.B., H.A.S., and N.S.W.), NIH Grant 1R21AI144223 (B.L.B., 14 H.A.S., and N.S.W.), NIH Grant 2R37GM065859 (B.L.B.), NIH Grant GM082938 (N.S.W.), the NSF through the Princeton University Materials Research Science and Engineering Center DMR-1420541 (B.L.B., H.A.S., A.K.), and the Max Planck Society-Alexander von Humboldt Foundation (B.L.B.). J.Y. holds a Ca- reer Award at the Scientific Interface from the Burroughs Wellcome Fund. R.A. acknowledges support from the Human Frontiers of Science Program (LT-000475/2018- C). We thank Dr. Maria Holland for helpful discussions. [1] A. Goriely, The Mathematics and Mechanics of Biological Growth, vol. 45. Springer, 2017. [2] R. de Rooij and E. Kuhl, "A physical multifield model predicts the development of volume and structure in the human brain," J Mech Phys Solids, vol. 112, pp. 563 -- 576, 2018. [3] M. Genet, M. Rausch, L. C. Lee, S. Choy, X. Zhao, G. S. Kassab, S. Kozerke, J. M. Guccione, and E. Kuhl, "Heterogeneous growth-induced prestrain in the heart," J Biomech, vol. 48, no. 10, pp. 2080 -- 2089, 2015. [4] D. W. Thompson, On Growth and Form. Cambridge Univ. Press, 1917. [5] Y. Klein, E. Efrati, and E. Sharon, "Shaping of elastic sheets by prescription of non-euclidean metrics," Science, vol. 315, no. 5815, pp. 1116 -- 1120, 2007. [6] J. Huxley, F. Churchill, and R. Strauss, Problems of Rel- ative Growth. Johns Hopkins University Press, 1993. [7] J. Sachs, Text-book of Botany: Morphological and Physi- ological. Clarendon Press, 1875. [8] T. Tallinen, J. Y. Chung, F. Rousseau, N. Girard, J. Lefèvre, and L. Mahadevan, "On the growth and form of cortical convolutions," Nat Phys, vol. 12, no. 6, p. 588, 2016. [9] S. Budday, P. Steinmann, and E. Kuhl, "The role of me- chanics during brain development," J Mech Phys Solids, vol. 72, pp. 75 -- 92, 2014. [10] A. E. Shyer, T. Tallinen, N. L. Nerurkar, Z. Wei, E. S. Gil, D. L. Kaplan, C. J. Tabin, and L. Mahadevan, "Vil- lification: How the gut gets its villi," Science, vol. 342, no. 6155, pp. 212 -- 218, 2013. [11] T. Savin, N. A. Kurpios, A. E. Shyer, P. Florescu, H. Liang, L. Mahadevan, and C. J. Tabin, "On the growth and form of the gut," Nature, vol. 476, no. 7358, pp. 57 -- 62, 2011. [12] H. Y. Kim, M.-F. Pang, V. D. Varner, L. Kojima, E. Miller, D. C. Radisky, and C. M. Nelson, "Localized smooth muscle differentiation is essential for epithelial bifurcation during branching morphogenesis of the mam- malian lung," Dev Cell, vol. 34, no. 6, pp. 719 -- 726, 2015. [13] J. Kim, J. A. Hanna, M. Byun, C. D. Santangelo, and R. C. Hayward, "Designing responsive buckled surfaces by halftone gel lithography," Science, vol. 335, no. 6073, pp. 1201 -- 1205, 2012. [14] C. Modes and M. Warner, "Shape-programmable mate- rials," Physics Today, vol. 69, no. 1, p. 32, 2016. [15] A. S. Gladman, E. A. Matsumoto, R. G. Nuzzo, L. Ma- hadevan, and J. A. Lewis, "Biomimetic 4D printing," Nat Mater, vol. 15, no. 4, pp. 413 -- 418, 2016. [16] S. Yang, K. Khare, and P.-C. Lin, "Harnessing sur- face wrinkle patterns in soft matter," Adv Funct Mater, vol. 20, no. 16, pp. 2550 -- 2564, 2010. [17] G. O'Toole, H. B. Kaplan, and R. Kolter, "Biofilm for- mation as microbial development," Annu Rev Microbiol, vol. 54, no. 1, pp. 49 -- 79, 2000. [18] J. K. Teschler, D. Zamorano-Sánchez, A. S. Utada, C. J. Warner, G. C. Wong, R. G. Linington, and F. H. Yildiz, "Living in the matrix: assembly and control of Vibrio cholerae biofilms," Nat Rev Microbiol, vol. 13, no. 5, pp. 255 -- 268, 2015. [19] D. O. Serra, A. M. Richter, G. Klauck, F. Mika, and R. Hengge, "Microanatomy at cellular resolution and spa- tial order of physiological differentiation in a bacterial biofilm," mBio, vol. 4, no. 2, pp. e00103 -- 13, 2013. [20] D. O. Serra, A. M. Richter, and R. Hengge, "Cellulose as an architectural element in spatially structured Es- cherichia coli biofilms," J Bacteriol, vol. 195, no. 24, pp. 5540 -- 5554, 2013. [21] L. E. Dietrich, C. Okegbe, A. Price-Whelan, H. Sakhtah, R. C. Hunter, and D. K. Newman, "Bacterial community morphogenesis is intimately linked to the intracellular re- dox state," J Bacteriol, vol. 195, no. 7, pp. 1371 -- 1380, 2013. [22] D. Romero, C. Aguilar, R. Losick, and R. Kolter, "Amy- loid fibers provide structural integrity to Bacillus sub- tilis biofilms," Proc Natl Acad Sci USA, vol. 107, no. 5, pp. 2230 -- 2234, 2010. [23] J. Yan, C. D. Nadell, H. A. Stone, N. S. Wingreen, and B. L. Bassler, "Extracellular-matrix-mediated osmotic pressure drives Vibrio cholerae biofilm expansion and cheater exclusion," Nat Commun, vol. 8, no. 1, p. 327, 2017. [24] J. Yan, C. Fei, S. Mao, A. Moreau, N. S. Wingreen, A. Košmrlj, H. A. Stone, and B. L. Bassler, "Mechanical instability and interfacial energy drive biofilm morpho- genesis," eLife, vol. 8, p. e43920, 2019. [25] S. Beyhan and F. H. Yildiz, "Smooth to rugose phase variation in Vibrio cholerae can be mediated by a sin- gle nucleotide change that targets c-di-GMP signalling pathway," Mol Microbiol, vol. 63, no. 4, pp. 995 -- 1007, 2007. [26] F. H. Yildiz and G. K. Schoolnik, "Vibrio cholerae O1 El Tor: Identification of a gene cluster required for the rugose colony type, exopolysaccharide production, chlo- rine resistance, and biofilm formation," Proc Natl Acad Sci USA, vol. 96, no. 7, pp. 4028 -- 4033, 1999. [27] F. Yildiz, J. Fong, I. Sadovskaya, T. Grard, and E. Vino- gradov, "Structural characterization of the extracellular polysaccharide from Vibrio cholerae O1 El-Tor," PLoS One, vol. 9, no. 1, p. e86751, 2014. [28] J. C. Fong and F. H. Yildiz, "The rbmBCDEF gene clus- ter modulates development of rugose colony morphology and biofilm formation in Vibrio cholerae," J Bacteriol, vol. 189, no. 6, pp. 2319 -- 2330, 2007. [29] V. Berk, J. C. Fong, G. T. Dempsey, O. N. Develi- oglu, X. Zhuang, J. Liphardt, F. H. Yildiz, and S. Chu, "Molecular architecture and assembly principles of Vibrio cholerae biofilms," Science, vol. 337, no. 6091, pp. 236 -- 239, 2012. [30] J. C. Fong, A. Rogers, A. K. Michael, N. C. Parsley, W.-C. Cornell, Y.-C. Lin, P. K. Singh, R. Hartmann, K. Drescher, E. Vinogradov, et al., "Structural dynamics of RbmA governs plasticity of Vibrio cholerae biofilms," eLife, vol. 6, p. e26163, 2017. [31] J. Yan, A. Moreau, S. Khodaparast, A. Perazzo, J. Feng, C. Fei, S. Mao, S. Mukherjee, A. Košmrlj, N. S. Wingreen, et al., "Bacterial biofilm material properties enable removal and transfer by capillary peeling," Adv Mater, vol. 30, no. 46, p. 1804153, 2018. [32] C. Zhang, B. Li, X. Huang, Y. Ni, and X.-Q. Feng, "Morphomechanics of bacterial biofilms undergoing anisotropic differential growth," Appl Phys Lett, vol. 109, no. 14, p. 143701, 2016. [33] P. S. Stewart and M. J. Franklin, "Physiological hetero- geneity in biofilms," Nat Rev Microbiol, vol. 6, no. 3, pp. 199 -- 210, 2008. [34] J. N. Anderl, J. Zahller, F. Roe, and P. S. Stewart, "Role of nutrient limitation and stationary-phase existence in Klebsiella pneumoniae biofilm resistance to ampi- cillin and ciprofloxacin," Antimicrob Agents Chemother, vol. 47, no. 4, pp. 1251 -- 1256, 2003. [35] A. Seminara, T. E. Angelini, J. N. Wilking, H. Vlamakis, S. Ebrahim, R. Kolter, D. A. Weitz, and M. P. Bren- ner, "Osmotic spreading of bacillus subtilis biofilms driven by an extracellular matrix," Proceedings of the National Academy of Sciences, vol. 109, no. 4, pp. 1116 -- 1121, 2012. [36] C. Zhang, B. Li, J.-Y. Tang, X.-L. Wang, Z. Qin, and X.-Q. Feng, "Experimental and theoretical studies on the morphogenesis of bacterial biofilms," Soft Matter, vol. 13, no. 40, pp. 7389 -- 7397, 2017. [37] X. Chen and J. W. Hutchinson, "Herringbone buckling patterns of compressed thin films on compliant sub- strates," J Appl Mech, vol. 71, no. 5, pp. 597 -- 603, 2004. [38] Z. Huang, W. Hong, and Z. Suo, "Nonlinear analyses of wrinkles in a film bonded to a compliant substrate," J Mech Phys Solids, vol. 53, no. 9, pp. 2101 -- 2118, 2005. [39] M. Asally, M. Kittisopikul, P. Rué, Y. Du, Z. Hu, T. Çağatay, A. B. Robinson, H. Lu, J. Garcia-Ojalvo, and G. M. Süel, "Localized cell death focuses mechani- cal forces during 3D patterning in a biofilm," Proceedings of the National Academy of Sciences, vol. 109, no. 46, pp. 18891 -- 18896, 2012. [40] R. W. Ogden, Nonlinear Elastic Deformations. Courier Corporation, 1997. [41] E. K. Rodriguez, A. Hoger, and A. D. McCulloch, "Stress-dependent finite growth in soft elastic tissues," J Biomech, vol. 27, no. 4, pp. 455 -- 467, 1994. 15 [42] A. Goriely and M. B. Amar, "Differential growth and instability in elastic shells," Phys Rev Lett, vol. 94, no. 19, p. 198103, 2005. [43] J. Dervaux and M. B. Amar, "Morphogenesis of growing soft tissues," Phys Rev Lett, vol. 101, no. 6, p. 068101, 2008. [44] J. W. Costerton, Z. Lewandowski, D. E. Caldwell, D. R. Korber, and H. M. Lappin-Scott, "Microbial biofilms," Annu Rev Microbiol, vol. 49, no. 1, pp. 711 -- 745, 1995. [45] J. Monod, "The growth of bacterial cultures," Annu Rev Microbiol, vol. 3, no. 1, pp. 371 -- 394, 1949. [46] S. Budday, C. Raybaud, and E. Kuhl, "A mechanical model predicts morphological abnormalities in the de- veloping human brain," Sci Rep, vol. 4, p. 5644, 2014. [47] S. Walcott and S. X. Sun, "A mechanical model of actin stress fiber formation and substrate elasticity sensing in adherent cells," Proc Natl Acad Sci USA, vol. 107, no. 17, pp. 7757 -- 7762, 2010. [48] P. Sens, "Rigidity sensing by stochastic sliding friction," EPL, vol. 104, no. 3, p. 38003, 2013. [49] U. S. Schwarz and S. A. Safran, "Physics of adherent cells," Rev Mod Phys, vol. 85, pp. 1327 -- 1381, Aug 2013. [50] S. Chen, T. Bertrand, W. Jin, M. D. Shattuck, and C. S. O'Hern, "Stress anisotropy in shear-jammed packings of frictionless disks," Phys Rev E, vol. 98, p. 042906, Oct 2018. [51] S. Cai, D. Breid, A. J. Crosby, Z. Suo, and J. W. Hutchin- son, "Periodic patterns and energy states of buckled films on compliant substrates," J Mech Phys Solids, vol. 59, no. 5, pp. 1094 -- 1114, 2011. [52] B. Audoly and A. Boudaoud, "Buckling of a stiff film bound to a compliant substrate - Part I: Formulation, linear stability of cylindrical patterns, secondary bifurca- tions," J Mech Phys Solids, vol. 56, no. 7, pp. 2401 -- 2421, 2008. [53] S. Srinivasan, C. N. Kaplan, and L. Mahadevan, "A mul- tiphase theory for spreading microbial swarms and films," eLife, vol. 8, p. e42697, 2019. [54] J. Liu, A. Prindle, J. Humphries, M. Gabalda-Sagarra, M. Asally, D. L. Dong-yeon, S. Ly, J. Garcia-Ojalvo, and G. M. Süel, "Metabolic co-dependence gives rise to collective oscillations within biofilms," Nature, vol. 523, no. 7562, pp. 550 -- 554, 2015. [55] C. P. Kempes, C. Okegbe, Z. Mears-Clarke, M. J. Fol- lows, and L. E. Dietrich, "Morphological optimization for access to dual oxidants in biofilms," Proc Natl Acad Sci USA, vol. 111, no. 1, pp. 208 -- 213, 2014. [56] P. K. Singh, S. Bartalomej, R. Hartmann, H. Jeckel, L. Vidakovic, C. D. Nadell, and K. Drescher, "Vibrio cholerae combines individual and collective sensing to trigger biofilm dispersal," Curr Biol, vol. 27, no. 21, pp. 3359 -- 3366, 2017. [57] M. B. Miller and B. L. Bassler, "Quorum sensing in bac- teria," Annu Rev Microbiol, vol. 55, no. 1, pp. 165 -- 199, 2001. [58] B. K. Hammer and B. L. Bassler, "Quorum sensing con- trols biofilm formation in Vibrio cholerae," Mol Micro- biol, vol. 50, no. 1, pp. 101 -- 104, 2003. [59] Y. Shen, G. L. Monroy, N. Derlon, D. Janjaroen, C. Huang, E. Morgenroth, S. A. Boppart, N. J. Ash- bolt, W.-T. Liu, and T. H. Nguyen, "Role of biofilm roughness and hydrodynamic conditions in Legionella pneumophila adhesion to and detachment from simulated drinking water biofilms," Environ Sci Technol, vol. 49, no. 7, pp. 4274 -- 4282, 2015. [60] J. Dumais, S. L. Shaw, C. R. Steele, S. R. Long, and P. M. Ray, "An anisotropic-viscoplastic model of plant cell morphogenesis by tip growth," Int J Dev Biol, vol. 50, pp. 209 -- 222, 2006. [61] B. He, K. Doubrovinski, O. Polyakov, and E. Wi- eschaus, "Apical constriction drives tissue-scale hydrody- namic flow to mediate cell elongation," Nature, vol. 508, no. 7496, pp. 392 -- 396, 2014. [62] B. Guirao and Y. Bellaïche, "Biomechanics of cell rear- rangements in Drosophila," Curr Opin Cell Biol, vol. 48, pp. 113 -- 124, 2017. [63] O. Foubet, M. Trejo, and R. Toro, "Mechanical morpho- genesis and the development of neocortical organisation," Cortex, 2018. [64] D. Matoz-Fernandez, F. A. Davidson, N. R. Stanley- Wall, and R. Sknepnek, "Wrinkle patterns in active vis- coelastic thin sheets." preprint, arXiv:1904.08872, 2019. [65] J. N. Wilking, V. Zaburdaev, M. De Volder, R. Losick, M. P. Brenner, and D. A. Weitz, "Liquid transport facili- tated by channels in Bacillus subtilis biofilms," Proc Natl Acad Sci USA, vol. 110, no. 3, pp. 848 -- 852, 2013. [66] J. Dervaux, J. C. Magniez, and A. Libchaber, "On growth and form of Bacillus subtilis biofilms," Interface Focus, vol. 4, no. 6, p. 20130051, 2014. [67] S. Haussler and C. Fuqua, "Biofilms 2012: New discover- ies and significant wrinkles in a dynamic field," J Bacte- riol, vol. 195, no. 13, pp. 2947 -- 2958, 2013. [68] J. S. Madsen, Y.-C. Lin, G. R. Squyres, A. Price-Whelan, A. de Santiago Torio, A. Song, W. C. Cornell, S. J. Sørensen, J. B. Xavier, and L. E. Dietrich, "Facultative control of matrix production optimizes competitive fit- ness in Pseudomonas aeruginosa PA14 biofilm models," Appl Environ Microbiol, vol. 81, no. 24, pp. 8414 -- 8426, 2015. [69] K. H. Thelin and R. K. Taylor, "Toxin-coregulated pilus, but not mannose-sensitive hemagglutinin, is required for 16 colonization by Vibrio cholerae O1 El Tor biotype and O139 strains.," Infection and Immunity, vol. 64, no. 7, pp. 2853 -- 2856, 1996. [70] W. Thielicke and E. Stamhuis, "PIVlab - towards user- friendly, affordable and accurate digital particle image ve- locimetry in MATLAB," J Open Res Softw, vol. 2, no. 1, p. e30, 2014. [71] A. Zemel, F. Rehfeldt, A. Brown, D. Discher, and S. Safran, "Optimal matrix rigidity for stress-fibre polar- ization in stem cells," Nat Phys, vol. 6, no. 6, pp. 468 -- 473, 2010. [72] P. Marcq, N. Yoshinaga, and J. Prost, "Rigidity sensing explained by active matter theory," Biophys J, vol. 101, no. 6, pp. L33 -- L35, 2011. [73] C. Geuzaine and J.-F. Remacle, "Gmsh: A 3-D fi- nite element mesh generator with built-in pre-and post- processing facilities," Int J Numer Meth Eng, vol. 79, no. 11, pp. 1309 -- 1331, 2009. [74] H. Langtangen and K. Mardal, Introduction to Numeri- cal Methods for Variational Problems. Texts in Compu- tational Science and Engineering, Springer International Publishing, 2019. [75] M. Alnaes, J. Blechta, J. Hake, A. Johansson, B. Kehlet, A. Logg, C. Richardson, J. Ring, M. E. Rognes, and G. N. Wells, "The FEniCS project version 1.5," Arch Numer Software, vol. 3, no. 100, pp. 9 -- 23, 2015. [76] J. Crank and P. Nicolson, "A practical method for numer- ical evaluation of solutions of partial differential equa- tions of the heat-conduction type," in Mathematical Pro- ceedings of the Cambridge Philosophical Society, vol. 43, pp. 50 -- 67, Cambridge University Press, 1947. [77] E. Lejeune, A. Javili, and C. Linder, "Understanding geo- metric instabilities in thin films via a multi-layer model," Soft Matter, vol. 12, no. 3, pp. 806 -- 816, 2016. Supplementary Information I. GENERAL FORMULATION OF GROWING ELASTIC TISSUES A general formulation that takes into account the combined effect of mechanics and growth, was originally proposed by Rodriguez et al. [1], and has since then been widely used for modeling different biological systems, such as in [2 -- 5]. Briefly, the model considers three configurations (see Fig. 2 in main text): a stress-free initial configuration, a stress- free virtual configuration after growth, and a deformed current configuration. In a fixed orthonormal Cartesian basis {e1, e2, e3}, the coordinates of material points in the three configurations are denoted by X0, X, and x, respectively. A complete kinematic description of deformed shapes for growing tissues is given by the mapping x = x(X0, t) from the initial configuration (X0) to the current deformed configuration (x). Equivalently, one can define the total deformation gradient F = ∂x/∂X0 to describe local changes in shape and density. The fundamental idea of Rodriguez et al. [1] is that the total deformation F can be decomposed into a deformation Fg due to growth to the virtual stress- free reference state, and an elastic deformation Fe that ensures compatibility of the growing tissue. Formally, the multiplicative decomposition reads F = Fe · Fg = ∂x ∂X · ∂X ∂X0 , (S1) where A · B denotes the matrix multiplication (A · B)ij = AikBkj and summation over repeated indices is implied. Besides the assumption above, the model of growing elastic tissues generally consists of the following three ingredients: • A governing equation for growth: The rate of deformation dFg/dt due to growth can in general be a function of many physical fields. For example, the growth rate dFg/dt may depend on external fields, such as a nutrient concentration field. It may also depend on internal stresses, if the organisms can sense and respond to their mechanical environment (see for example [6, 7]). • A stress-strain constitutive relation: A stress-strain constitutive relation relates tissue deformation to stresses. Here, we assume that the internal Cauchy stress σ arises solely from the elastic deformation, i.e. σ = σ(Fe). For a hyperelastic material, this relation can be determined from the elastic free energy density Ψ = Ψ(Fe). Specifically, for isotropic materials the energy density can be expressed in terms of the two invariants Ψ = E(IC, Je), where IC = tr(FT e Fe) is the trace (first invariant) of the right Cauchy-Green deformation tensor, and Je = det(Fe) denotes the volumetric change due to elastic deformation [8]. Different stress measures can be defined based on whether they report the force in an undeformed or a deformed configuration. For example, the first Piola-Kirchhoff (PK) stress P measures the force per area in the undeformed configuration, which is evaluated as P = ∂Ψ ∂Fe = 2 ∂E ∂IC Fe + ∂E ∂Je(cid:0)adj(Fe)(cid:1)T , (S2) where adj denotes the adjugate of a matrix [9]. The PK stress PiJ is asymmetric, because it has one index i attached to the current configuration, and one index J attached to the undeformed virtual configuration. In contrast, the Cauchy stress σij measures the force per area in the current deformed configuration, and it is symmetric due to conservation of angular momentum [10]. The Cauchy and PK stresses are related by σ = J−1 e PFT e = 2J−1 e (cid:18) ∂E ∂IC(cid:19) B +(cid:18) ∂E ∂Je(cid:19) I, e denotes the left Cauchy-Green deformation tensor [8]. where B = FeFT • A force-balance equation: The balance of acceleration and forces is given by ρm v − ∇ · σ − ρmb = 0, (S3) (S4) where ρm(x, t) denotes the mass density, v(x, t) the acceleration ( v denotes total time derivative of velocity v), ∇ the derivative with respect to the coordinates x, σ(x, t) is the Cauchy stress, and ρmb the body force density (per volume). The acceleration term can often be neglected in small systems and it is negligible for the biofilms considered here (see Sec. II B). arXiv:1910.14645v1 [physics.bio-ph] 31 Oct 2019 In the next section, we describe a chemo-mechanical model of V. cholerae biofilm development and present the corresponding three key relations discussed above. Unless otherwise specified, we use the notation where the lower case letters, the upper case letters, and the upper case letters with the subscript 0 denote quantities in the current, the virtual, and the initial configurations, respectively. In addition, we use the subscript k for reduced 2D quantities. Greek letter indices are used to denote the components of in-plane 2D vectors/tensors and Latin letters are used to denote the components of 3D vectors/tensors. Summation over repeated indices is to be assumed unless otherwise specified. Table S1 lists all symbols and describes their meaning. 2 Table S1: Summary of the used symbols and their descriptions. General notations (Secs. I -- VII) Symbols Descriptions Symbols Descriptions k Grad0, Grad, ∇k, ∇k Subscript: reduced 2D variables or the in-plane components of a 3D vector/tensor Spatial gradient taken in the initial, the virtual, the current, and the de- formed flat configuration ∼ Normalized variales or dimension- less parameters adj, sym The adjugate or the symmetric part of a matrix General formulation of growing elastic biofilms (Secs. I, II B and II C) Symbols X0, X, x S0, P, σ B λg γ η, f , ξ Descriptions Coordinates of the initial, virtual, and current configuration Stress measured in the initial, vir- tual, and current configuration The left Cauchy-Green deformation tensor Geometric stretch due to growth Vertical stretch of biofilm Drag coefficient, dimensional and dimensionless friction Symbols F, Fe, Fg C, IC J, Je kg, kr Gb, Gs rb Descriptions Total, elastic, and growth deforma- tion gradients The right Cauchy-Green deforma- tion tensor and its first invariant The determinant of F and Fe Total and residual growth rate Biofilm and agar shear modulus Biofilm radius Nutrient diffusion-uptake dynamics (Secs. II A and II E) Symbols Descriptions Symbols Descriptions c, ck Q0, J0 K Nutrient concentration in a 2D or a 3D model Maximum nutrient uptake rate in a 2D or a 3D model Half-rate concentration of nutrient uptake D ac Nutrient diffusion constant Width of the nutrient-rich annulus Hs, Rs Radius and thicknesss of the agar substrate Model discussion: Lagrangian specification (Sec. II D) Symbols Descriptions Symbols Descriptions ρ, ρ0 Material density in the deformed and the undeformed configuration χ The reference vector denoting the initial coordinate of a material point Continued on next page Model discussion: Constitutive model (Sec. II F) Symbols Descriptions Π vw, µw F0, λ0 Osmotic pressure Volume and chemical potential of water molecule Equilibrium swelling χ Flory interaction parameter Model discussion: Friction (Sec. II G) Descriptions Symbols Kp, Ks, K0 τon, τoff Eadh, Spring constant of the polymer spring, the substrate, and them connected in series Time scales associated with the binding/unbinding processes Dimensional and dimensionless ad- hesion energy parameter Coarse-grained wrinkling model (Sec. VI) Symbols φ0 Xd, Fd Jd ΨJ Symbols kon, koff ρp ξ0T 3 Descriptions Equilibrium dextran concentration Coordinate and deformation gradi- ent associated with the dry state Volumetric change of biofilm matrix compared to its dry state Free with volumetric change energy increase associated Descriptions The binding/unbinding rate be- tween biofilm polymers and biofilm proteins Average density of bound polymers Typical the polymer springs activated by thermal energy stretch of Symbols F0 e, Fw e x0 Π, Ψ Descriptions Horizontal compression and wrin- kling deformation Coordinate in the deformed flat configuration Symbols u, w Descriptions In-plane displacement and out-of-plane Superscript 0 Quantities without wrinkling Elastic energy Superscript c A, β1β2 Amplitude and shape of wrinkles kw, λw a0 11, a0 22 A, S Principal stretches of F0 e Dimensionless order parameter of wrinkles σc q Model discussion: Elasto-plasticity (Sec. VII) wavenumber in the coarse-grained Quantities description Wrinkle wavelength Critical instability Area factor accounting for the area increase in the wrinkled state and stress of the wrinkling Symbols E, s E, σe Descriptions Symbols Descriptions Deviatoric deformation and stress Scalar measures of the deviatoric deformation and stress σY G0 Yielding stress Normalized plastic modulus 4 II. A CHEMO-MECHANICAL MODEL FOR BACTERIAL BIOFILMS To investigate how mechanical stresses build up in V. cholerae biofilms when growing on agar, we model the biofilm as a thin elastic film that consumes nutrients, proliferates, and expands. The effective surface friction between the expanding biofilm and the agar generates compressive stresses inside the biofilm. Although viscoelastic relaxation and plastic deformations are quite likely to occur during biofilm development [11], we neglect all sources of internal dissipation in our model (see Sec. VII for further discussion). This section is organized as follows: in Sec. II A we begin by modeling the nutrient-dependent biofilm growth; in Sec. II B we propose a material constitutive model for the biofilm and derive the force-balance equation; in Secs. II C we summarize the governing equations and relevant model parameters, which are converted to dimensionless form; in Secs. II D -- II G we discuss different aspects of the model, including the general formulation, the nutrient diffusion-uptake dynamics, the constitutive model, and the friction between biofilm and agar. A. Nutrient-limited biofilm growth In experiments we previously found that there is an annulus of width ∼1 mm at the (outer) edge of the biofilm, in which the fraction of dead cells is significantly lower compared to the interior part of the biofilm [12]. This indicates that cell growth occurs primarily at the edge of the biofilm, presumably due to nutrient limitation elsewhere. To account for the non-uniform growth pattern observed in experiments, we model nutrient-dependent growth as follows. Nutrient uptake-diffusion dynamics: The diffusion of nutrients in the agar substrate, denoted by concentration c(xk, z, t), is described by Fick's law ∂c/∂t = D∇2c, where D is the nutrient diffusion constant and xk ≡ (x, y). As shown in Fig. 2 in the main text, biofilms are located at the top surface z = 0, and consume nutrients from the agar. This sets the boundary condition −D∂zc = J−1 J0φ(c) at z = 0 and xk < rb, where J0 is the maximum nutrient e,k uptake rate per unit area of cells in the undeformed configuration, J−1 accounts for the increase of biofilm thickness e,k due to compression, and rb denotes the radius of the biofilm in the current configuration. The rate of nutrient uptake is assumed to depend on local nutrient availability via the Monod law φ(c) = c/(K +c) where K is the concentration of nutrients at the half-maximal uptake rate [13]. Zero flux boundary conditions are imposed elsewhere on the boundary of the agar substrate, i.e. n · ∇c = 0, where n is the normal vector to the agar boundary. Simulating the full diffusion dynamics of nutrients requires tracking the concentration field in the three-dimensional (3D) space of the agar medium in the Eulerian formulation, while the mechanical stresses in the biofilm are more readily computed in a two-dimensional (2D) plane in the Lagrangian formulation (see Sec. II B and II E). However, to capture the nutrient-limited non-uniform growth of the biofilm, it is sufficient to consider only the lateral diffusion of nutrients. Therefore, we introduced a reduced 2D model to simplify the full 3D dynamics of nutrients and to make the computations more tractable. The reduced 2D model is given by ∂ck ∂t = D∇2 kck − J−1 e,k Q0φ(ck), (S5) where ck(xk, t) denotes the reduced 2D nutrient concentration and Q0 denotes the maximum uptake rate. The factor J−1 is introduced to account for the change of areal density of the biofilm upon deformation. The concentration of e,k nutrients at the edge of the biofilm (xk = rb(t)) is set to a fixed value c0,k. As the biofilm grows and expands, nutrients get depleted underneath the biofilm (see Fig. 2C in the main text). The concentration profile of nutrients has a maximum at the edge of the biofilm and decreases toward the interior of the biofilm with a penetration depth ac ∼pDc0,k/Q0. To verify that the reduced 2D model provides a reasonably accurate approximation of the full 3D dynamics of nutrients, we compared the concentration profiles of nutrients for both models in Sec. II E. Nutrient-dependent growth: During development V. cholerae biofilms expand primarily in the horizontal plane [12, 14], while the biofilm thickness is approximately constant, being set by the oxygen penetration depth [15]. Thus, we consider only horizontal (in-plane) growth in our model. Growth is assumed to be isotropic because our experiments reveal that the orientation of bacteria inside the biofilm is disordered and so the growth, division, and matrix production by the bacteria should result in isotropic expansion. The deformation due to growth is thus 1(cid:19), where λg(xk, t) denotes the local stretch due to growth, and Ik is the 2D identity matrix. Nutrient Fg =(cid:18)λgIk 0 dependent growth is described by ∂λg/∂t = kg(ck)λg, where the growth rate is kg(ck) = (kmax g − kr)φ(ck) + kr, where kmax is the maximum growth rate and we added a small nutrient-independent growth rate kr to account for the fact g that in 3D simulations the concentration of nutrients does not drop to zero beneath the center of the biofilm (see Sec. II E for details). This nutrient-dependent growth combined with the diffusion-uptake dynamics of nutrients, captures 0 our experimental finding that the active growth zone is restricted to a finite annulus near the biofilm edge (see Fig. 2 in the main text and Fig. S3). 5 B. Thin plate mechanics of biofilms Plane stress assumption: The V. cholerae biofilms in our experiments can be considered as thin-film structures (∼ 100 µm in height and 5 − 15 mm in diameter). In continuum mechanics, a plane stress assumption is often used for the analysis of thin films, such that the stress components perpendicular to the biofilm surface are negligible throughout the whole biofilm. Thus, we use the plane stress model to describe the mechanics of our biofilms. In this section, we consider only a flat biofilm configuration, before wrinkling occurs. The out-of-plane deformation induced by the wrinkling instability is discussed in Sec. VI. Consider the full 3D configuration x(X0, t) of a thin, flat, elastic film. The elastic deformation tensor Fe describes deformation from the virtual stress-free grown configuration X to the final deformed configuration x (see Fig. 2 in the main text), which can be written explicitly as Fe = Fe,k (Gradk z)T ∂Zxk ∂Zz! , (S6) where Fe,k = Gradk xk is the in-plane deformation tensor, and Gradk denotes gradient with respect to the spatial coordinates Xk in the intermediate virtual stress-free configuration. For typical biofilms in our experiments, we estimate that Gradk z ∼ ∆hb/rb . 100µm/5mm = 0.02, where ∆hb denotes the height difference between the biofilm center and the biofilm edge. Thus, we assume that Gradk z is negligible compared to the in-plane deformation Fe,k and to the elastic change of thickness ∂Zz. The plane stress assumption also implies that ∂Zxk must be small (see for example, Föppl-von Karman plate theory or Kirchhoff-Love plate theory [10, 16, 17]). For simplicity, we set Gradk z = ∂Zxk = 0 and define the vertical stretch γ = ∂Zz ≡ γ(Xk), which describes the relative change of thickness. This yields the plane stress elastic deformation tensor Fe ≈ Fe,k 0 γ! 0T (S7) for a thin, flat biofilm. Finally, the relation between γ and Fe,k can be derived from the plane stress condition σzz = 0 (see Eq. (S3) in Sec. I for the expression for σ), which yields: 2J−1 e (cid:18) ∂E ∂IC(cid:19) γ2 +(cid:18) ∂E ∂Je(cid:19) = 0, (S8) e Fe) = γ2 + IC,k is the trace (first invariant) of the right Cauchy-Green deformation tensor and where IC = tr(FT Je = det(Fe) = γJe,k denotes the volumetric change due to elastic deformation. Here we have introduced the invariants IC,k = tr(FT Fe,k) and Je,k = det(Fe,k) that are related to the in-plane part of the elastic deformation e,k Fe,k. Constitutive relation: Next, we specify the constitutive model that relates the stress to the deformation of a biofilm. Here, we use a common neo-Hookean material model [18 -- 20] to approximate biofilm elasticity; this model will be motivated and discussed in more detail in Sec. II F. The strain energy density for the neo-Hookean model is given by E(IC, Je) = , (S9) Gb 2 hIC − 3 − 2 ln Jei + λb 2 (cid:0) ln Je(cid:1)2 where Gb and λb are the Lamé elastic constants. Note that Ψ in Eq. (S9) above is the energy density per unit volume in the virtual stress-free configuration X. For an incompressible neo-Hookean material, the energy density reads E(IC, Je) = (Gb/2)(IC − 3) − p(Je − 1), where a pressure p is introduced as the Lagrange multiplier to ensure Je = 1. For a given in-plane deformation Fe,k, the vertical stretch γ is obtained by inserting Eq. (S9) into Eq. (S8) and solving for γ. The in-plane components of the Cauchy stress σk are obtained from Eq. (S3). Results for the compressible and incompressible case are: • Compressible case: γ is obtained by solving the nonlinear equation Gb(γ2 − 1) + λb ln(γJe,k) = 0. The in-plane 6 (S10) components of the Cauchy stress are σk = (γJe,k)−1(cid:2)GbFe,kFT e,k + λb ln(γJe,k)Ik − GbIk(cid:3). • Incompressible case: Constraint Je ≡ 1 = γJe,k specifies the relation γ = 1/Je,k = 1/ det(Fe,k). The pressure p can be calculated as p = Gbγ2, and therefore the in-plane components of the Cauchy stress are σk = Gb(Fe,kFT e,k − γ2Ik). (S11) Force-balance equation: A growing biofilm moves relative to the substrate and experiences an effective friction. We model this friction as a viscous drag [21, 22], i.e., we assume that the friction f (force per unit area in the current deformed configuration) takes the form f = −ηvk, where vk is the expansion velocity of the biofilm, and η the drag coefficient. The justification for the assumption of viscous friction is discussed in Sec. II G where we present a microscopic model for the origin of this friction. The general force-balance relation is presented in Eq. (S4) in Sec. I. Here we first discuss the relative importance of inertial effects. For biofilms the mass density is estimated to be ρm ∼ 103 kg/m3 since they contain mostly water. The typical velocity and time scales associated with biofilm growth and expansion are U0 ∼ 3 µm/min and τ0 ∼ 1 h. The shear modulus of the biofilm is Gb ∼ 103 Pa, and the typical radius of biofilm is rb ∼ 10 mm. The magnitude of inertial forces ρm v can be estimated as ρmU0/τ0, while the magnitude of forces related to internal stress ∇ · σ is estimated as Gb/rb. Comparing these estimates we find ρmU0/τ0 Gb/rb ∼ 10−15 and thus the inertial forces are negligible compared to the forces resulting from internal stresses. A similar argument can be made concerning the gravitational body force, which can be neglected as well. Thus the 3D force-balance equation for the biofilm simplifies to ∇ · σ = 0 with the traction-free boundary condition (σ · ez = 0) at the top of the biofilm and with the frictional traction (σ · (−ez) = f) at the biofilm-agar interface. Note that ∇ are derivatives with respect to coordinates in the deformed configuration. By integrating the force-balance equation in the z direction over the deformed biofilm thickness h = γH, where H is thickness of the undeformed biofilm, we obtain the horizontal force-balance relation ∇k · (γHσk) + f = 0. C. Summary (S12) Governing equations, nondimensionalization, and control parameters: We define dimensionless variables σ = σ/Gb, x = xk/Rb0, τ = t/τ0, and ck = ck/c0,k, where the shear modulus of the biofilm Gb was chosen as the scale for stresses, the initial biofilm radius Rb0 as the characteristic length scale, the inverse of the growth rate at = [kg(r = rb)]−1 as the characteristic time scale associated with biofilm expansion, and the concentration c0,k of nutrients at the edge of the biofilm as the relevant concentration scale. Upon nondimensionalizing the equations describing biofilm growth and mechanics discussed above, we obtain the following equations the edge of the biofilm τ0 = (cid:0)kmax (cid:1)−1 g c Nutrient diffusion-uptake: ∂τ ck = Q0(cid:2)a2 kck − J−1 ∇2 e,k Nutrient limited growth: ∂τ λg = kg(ck)λg, Force balance: ∇k · (γ σk) = ξ vk, Constitutive relation: σk =(Fe,kFT e,k − γ2Ik (γJe,k)−1hFe,kFT φ(ck)(cid:3), (incompressible) + λb Gb(cid:0) ln(γJe,k) − 1(cid:1)Iki e,k (S13) (S14) (S15) (S16) (compressible), where Q0 = Q0/(c0,k/τ0) characterizes the rate of nutrient uptake, ac = R−1 b0 (Dc0,k/Q0)1/2 denotes the relative width of the nutrient-rich zone, and φ(ck) = ck/( K + ck). The nutrient-dependent growth rate takes the form kg(ck) = φ(1)−1(1 − kr)φ(ck) + kr. We define a dimensionless friction parameter ξ = η(Rb0/τ0) Gb(H/Rb0), which serves as a control parameter of the model and quantifies the relative ratio of the typical frictional forces to the characteristic forces resulting from internal stresses in the biofilm. In Sec. II G, we show that ξ increases monotonically with the substrate modulus Gs. Therefore, increasing the value of ξ in the model corresponds to increasing the agar concentration and thus the substrate stiffness in experiments. We assume axial symmetry and thus all the fields only depend on the distance r from the center of the biofilm. Thus, the dimensionless governing equations are solved with the following initial conditions and boundary conditions: 7 • Initial conditions: r( R0, τ = 0) = R0, ck( R0, τ = 0) ≡ 1, λg( R0, τ = 0) ≡ 1 and γ( R0, τ = 0) ≡ 1. • Boundary conditions: σk,rr( R0, τ ) = 0 and ck( R0, τ ) = 1 at the edge of the biofilm. g Further details about the numerical simulation of the above equations can be found in the Methods section of the main text. Parameters of the model (see Table S2): The initial radii of biofilms in experiments are measured to be Rb0 ≈ 2.0 mm. The width of the active growth zone can be estimated from experiments to be ac ≈ 1.0 mm (see experiments in [12]). Thus, we fix ac = 0.5 in our simulations. The growth rate at the edge of the biofilm kg(r = rb) = kmax , which sets the time scale τ0, can be estimated from the steady-state expansion velocity of the biofilm grown on a soft substrate (0.4% agar), where friction is small. When biofilms in experiments enter the third kinematic stage, the radii of biofilms increase linearly in time with the radial velocity vb ≈ 2ackmax g = 3 µm/min. From this we estimate the characteristic time scale as τ0 = 1/kmax g ≈ 2ac/vb ≈ 10 h. The thickness of a biofilm in the undeformed configuration H ≈ 50 µm can be estimated from experiments with a soft substrate (For two-day-old biofilms grown on 0.4% agar, the thickness is measured to be h = 55 ± 4 µm [12]). The concentration of nutrients at the half-maximal uptake rate is set to K = 0.5. Note that our model results are insensitive to the specific choice of K: for a different K, a similar ck profile can be obtained by adjusting the free parameter Q0. The two remaining parameters, the friction parameter ξ and the nutrient uptake rate Q0, are difficult to probe experimentally. Thus, we treat them as fitting parameters and their values were determined by fitting the radial velocity profiles vr(r) from the model to the experimental data for a biofilm grown on 0.7% agar (see Fig. 2D in the main text, Fig. S1, and the Methods section in the main text). From the optimal fitting values, we obtain Q0 = 2, and ξ = 20. When numerically simulating the growth of biofilms on agar substrates with different concentrations, we varied the value of the friction parameter ξ, with the values of other parameters fixed. Assuming that the friction coefficient η increases with the substrate shear modulus Gs as η ∝ Gs, we find that the dimensionless friction parameter scales as ξ ∝ Gs/Gb. Then we compute the friction parameter as ξ = 20 Gs/Gb(agar conc.=0.7%) from the measured values of the shear modulus of the substrate Gs and of the biofilm Gb [12]. Gs/Gb Table S2: Summary of the measured/estimated quantities for biofilms in experiments and of the numerical values of parameters used in the chemo-mechanical model of biofilm growth. Measured/estimated quantities for biofilms in experiments Values 2 mm Descriptions Symbols Rb0 rb H hb ac Gb νb τ0 vb Rs Hs Initial biofilm radius Typical biofilm radius Undeformed biofilm thickness Typical biofilm thickness Width of nutrient-rich annulus Biofilm shear modulus Biofilm Poisson's ratio Biofilm development time scale Typical expansion velocity Agar substrate radius Agar substrate thickness 5 − 15 mm 50 µm 100 µm 1 mm ≈ 1 kPa 0.37 ∼ 0.46 10 h 3 µm/min 45 mm 6 mm Numerical values for parameters used in the model ac kr Q0 ξ Normalized width of the nutrient- rich zone Normalized residual growth rate Normalized 2D nutrient uptake Dimensionless friction parameter 0.5 0.15 2.0 2 ∼ 200 8 Fig. S1: Fitting of biofilm model parameters. (A) Time averaged normalized MSD values (see Methods for details) between the time series of velocity profiles for the modeled biofilm and the experimental biofilm grown on 0.7 % agar upon variation of the dimensionless friction ξ and the maximal nutrient uptake Q0. The red star denotes the optimal parameter values, ξ = 20 and Q0 = 2.0, which minimize the normalized MSD. (B) Normalized MSD values plotted versus time for ξ = 20 and Q0 = 2.0. D. Model discussion: Eulerian versus Lagrangian description In Sec. II C, the governing equations (S13) -- (S16) for the chemo-mechanical model mix the Eulerian and the Lagrangian frames of reference. The nutrient diffusion-uptake dynamics in Eq. (S13) and the force-balance condition in Eq. (S15) are stated in the current coordinates xk of the deformed biofilm, i.e. in the Eulerian frame of reference. On the other hand the nutrient limited growth in Eq. (S14) and the constitutive relation in Eq. (S16) are stated in the initial coordinates X0,k of biofilm, i.e. in the Lagrangian frame of reference. In order to numerically solve the governing equations we have to rewrite them all either in the Lagrangian or the Eulerian frame of reference. For numerical simulations we used the Lagrangian frame of reference, as described below. We start with the 3D force-balance equation. In the absence of body forces, the force-balance condition for Lagrangian formulation of the chemo-mechanical model: Since the nutrient limited growth in Eq. (S14) and the constitutive relation in Eq. (S16) are already stated in the Lagrangian frame of reference, what remains is to rewrite the nutrient diffusion-uptake dynamics in Eq. (S13) and the force-balance condition in Eq. (S15) in terms of the initial coordinates X0,k of biofilm. an arbitrary element with volume ω of an elastic material in the current configuration yields: Rω(d3x)∇ · σ = H∂ω σ · n da = 0, where ∂ω is the boundary surface of that element and n is the unit vector normal to the boundary. The coordinate transformation must guarantee that the force acting on a particular area element is the same regardless of the choice of coordinate system. The corresponding stress measure S0 in the initial configuration must thus satisfy S0 · N0 dA0 = σ · n da for any area element, where N0 is the unit vector normal to the area element dA0 in the initial configuration. Consequently, the force-balance condition for the same volume element in the initial configuration reads, H∂ω σ · nda =H∂Ω0 (d3X0)Grad0 S0 = 0, or equivalently, Grad0 S0 = 0. Here Grad0 represents derivatives with respect to the initial coordinates X0. The Nanson's relation, (da)n = (JdA0)F−T · N0, [23] can be used to relate the stress measure S0 to the Cauchy stress σ as S0 = σF−T J, where F = ∂x/∂X0 is the total deformation gradient and J = det F is the Jacobian due to total deformation. The stress measure can also be expressed in terms of the PK stress P as S0 = PF−T g Jg, where Fg = ∂X/∂X0 is the deformation due to growth and Jg = det Fg is the Jacobian due to growth. S0 · N0dA0 =RΩ0 Similar to our derivation of the 2D force balance in Eq. (S12) in the current coordinates x, we integrate the 3D equation Grad0 S0 = 0 in the z direction over the biofilm thickness H in the reference initial coordinates to obtain the dimensionless 2D force-balance equation where ^Grad0,k refers to the derivative with respect to nondimenzionalized initial coordinates X0,k, λg is the local ^ Grad0,k S0,k = ^Grad0,k(λg Pk) = Jkξ vk, (S17) 0.02 0.02 0.01 0.01 0 0 0 0 10 10 t (h) 20 20 Normalized MSDB 0.07 0.04 .02 10 10 Friction ξ 50 50 5 5 1 1 Nutrient uptake Q0 ~ A Normalized MSD 0 0 0 9 k · nk dl −Rs J−1 e,k stretch due to growth, and the Jacobian factor Jk = det(Fk) on the right side results from the transformation of the area elements. In the Lagrangian description, the radial velocity of the biofilm is vk = (∂txk)X0,k , where the subscript means that the time derivative is taken at fixed X0,k, and the elastic part of the deformation gradient Fe,k = λ−1 g Fk is used to calculate the PK stress Pk(Fe,k) (see Eq. (S2)). The same method as above is applied to transform the diffusion-uptake equation (S5) for the nutrient field Q0φ(ck) da, where we used the divergence theorem to transform the diffusion term ck. The integral form of the equation for any given area element s in the current configuration is Rs ∂tck da = H∂s D∇kck Rs D∇2 ck · ds =H∂s D∇kck · nk dl. Here, nk is the unit normal to the boundary element dl. To transform the first and the third terms in the above equation, we use the Jacobian Jk to transform the area elements as da = JkdA0. In order to transform the time derivative of the concentration field we note that ck(X0,k, t) = ck(xk(X0,k, t), t). The time derivative of the concentration field is thus transformed as (∂tck)X0,k + (Grad0,k ck)t · F−1 k To transform the term with the diffusion constant, we first again employ the Nanson's relation (dl)nk = (JkdL0)F−T k bining all of the results above, we derive the transformed equation for the diffusion and uptake of nutrients in terms of the initial coordinates as · N0,k and the chain rule to derive ∇kck · nkdl = JkdL0(Grad0,kck) ·(cid:0)F−1 )(cid:3) − J−1 · vk. In the last step we used the chain rule ∂ck/∂xk = (∂ck/∂X0,k) · (∂X0,k/∂xk). Grad0,k ·(cid:2)JkD(Grad0,kck) · (F−1 k (cid:1) · N0,k. By com- (∂tck)X0,k − (Grad0,k ck)t · F−1 k + (∇kck)t · vk = (∂tck)xk The dimensionless equation is then easily obtained from the equation above. Q0φ(ck). (S18) · vk = J−1 k = (∂tck)xk Eulerian formulation of the chemo-mechanical model: Finally, we give the equivalent formulation of the chemo-mechanical model using the Eulerian formulation, where the variables of interest are the velocity profile of the biofilm vk(cid:0)xk, t(cid:1), the areal mass density of the biofilm ρk(cid:0)xk, t(cid:1), and the concentration of nutrients ck(cid:0)xk, t(cid:1). Below we discuss how to rewrite the nutrient limited growth in Eq. (S14) and the constitutive relation in Eq. (S16) in the Eulerian frame of reference. First, we show that the nutrient limited growth equation in the Lagrangian formulation ∂tλg = kgλg is equivalent F−T k k F−T k e,k to the mass conservation equation ∂tρk + ∇k · (ρkvk) = 2kgρk (S19) in the Eulerian formulation. We will do so by rewriting the mass conservation equation above in the Lagrangian formulation. The factor of 2 is related to the assumption that the growth of the biofilm is two dimensional, while the biofilm thickness remains unchanged. We assume that the areal mass density of the undeformed biofilm is constant ρ0,k. Elastic deformation of the biofilm then changes the density to ρk = ρ0,k/Je,k, which follows from the fact that the mass of a small element in the current coordinates ρk d2xk must be equal to the mass of the element in the virtual coordinates ρ0,k d2Xk. Here we used the Jacobian Je,k = det(Fe,k). The time derivative of the areal mass density of the biofilm can be transformed from the current to the initial coordinates as (∂tρk)X0,k + (∇kρk)t · vk, which is identical to the transformation for the concentration of nutrients described in the previous section. By combining this equation with the mass conservation equation (S19) we obtain (∂tρk)X0,k + ρk(∇k · vk) = 2kgρk. By inserting the relation ρk = ρ0,k/Je,k in the above equation we obtain (S20) = (∂tρk)xk . (∂tJ−1 e,k )X0,k + J−1 e,k (∇k · vk) = 2kgJ−1 e,k The divergence term in the equation above can be transformed to the Lagrangian formulation as ∇k · vk = ∂vα ∂xα = ∂vα ∂X0,β ∂X0,β ∂xα = ( Fk)αβ(F−1 k )βα = tr( Fk · F−1 k ) = Fk : F−T k , (S21) where we used the definition of the deformation gradient Fk = ∂xk/∂X0,k and we defined the tensor contraction A : B = tr(A · BT ) = AαβBαβ. Note that the Greek indices in above equations denote the components of in-plane 2D vectors and tensors. By combining Eqs. (S20) and (S21) we obtain −J−1 e,k (∂tJe,k)X0,k + Fk : F−T k = 2kg. The time derivative term of the Jacobian Je,k can be expressed using Jacobi's formula [9] as ). Fe,k) = tr(Je,kF−1 e,k · Fe,k) = Je,k( Fe,k : F−T e,k ∂tJe,k = tr(adj(Fe,k) · (S22) (S23) 10 Next, we leverage the fact that the total in-plane deformation can be decomposed as Fk = (Fe · Fg)k = λgFe,k. Thus we obtain Fk = (∂tλg)Fe,k + λg Fe,k and F−T . By combining these equations with Eqs. (S22) and (S23) k we obtain g F−T e,k = λ−1 g (∂tλg)(Fe,k : F−T λ−1 e,k ) = 2kg. (S24) = tr(Fe,k · F−1 e,k Note that Fe,k : F−T e,k growth equation ∂tλg = kgλg, which is thus indeed equivalent to the mass conservation law in Eq. (S19). ) = tr(Ik) = 2. Therefore, the equation above is equal to the nutrient limited To complete the Eulerian formulation of the chemo-mechanical model we also need to express the constitutive relation σk(Fe,k) in the current coordinates xk. We start by defining a map χ(xk, t) = X0,k from the current coordinates xk to the initial coordinates X0,k. By definition the material derivative of this map is zero, i.e. ∂tχ + vk · ∇kχ = 0. (S25) The gradient of the map χ is related to the total deformation gradient as ∇kχ = ∂X0,k/∂xk = F−1 . The elastic part k of the deformation gradient can then be written as Fe,k = Fk · F−1 g Fk. Note that the determinant of the elastic gk g Jk. The final step is to express the growth factor λg in the Eulerian deformation gradient is Je,k = Jk/Jg,k = λ−2 formulation. We recall that the areal mass density of the biofilm is ρk = ρ0,k/Je,k and thus we obtain the relation for g = Je,k/Jk = (ρ0,k/ρk) det(∇kχ). By combining these relations we can express the elastic part the growth factor λ−2 of the deformation gradient in the Eulerian formulation = λ−1 Fe,k = λ−1 g Fk =qρ−1 k ρ0,k det(cid:0)∇kχ(cid:1) (∇kχ)−1. (S26) Using the above expression we can then calculate the Cauchy stress σk(Fe,k) from the constitutive relation in Eq. (S16). In the Eulerian formulation of the chemo-mechanical model the unknowns are χ, vk, ρk, and ck. They can be obtained by solving Eqs. (S13), (S15), (S19), and (S25) together with the constitutive relation in Eq. (S16) and the elastic part of the deformation gradient in Eq. (S26). Solving these equations is complicated, because of the moving boundary conditions at the biofilm edge, and therefore requires special numerical treatment, such as the phase field approach [24 -- 27]. E. Model discussion: Comparison of 3D and 2D nutrient dynamics To compare the full 3D dynamics of nutrients with the reduced 2D model, we simulate the dynamics of nutrients in the 3D substrate, in the presence of a growing biofilm on top of the substrate. For simplicity, we set the friction coefficient to zero, i.e. η = 0. We consider dimensionless variables x = x/Rb0, τ = t/τ0, and c = c/c0, where c0 is the initial concentrations in the 3D substrate, and the length scale Rb0 = 2 mm and the time scale τ0 = 1/kg(c = c0) = 10 h are the same as in the 2D model. The nondimensionalized equation for the diffusion of nutrients is ∂τ c = D ∇2c, where D = Dτ0/R2 b0 is fixed to be the same value as in the 2D model. A biofilm of radius rb(τ ) is located at the top surface z = 0 and it consumes nutrients from the substrate, which sets the boundary condition −∂z c = J0φ(c) at z = 0 and xk < rb(τ ), where we used the Monod law φ(c) = c/( K + c) as described previously. Zero flux boundary conditions are imposed elsewhere on the boundary of the substrate, i.e. n · ∇c = 0, where n is the vector normal to the substrate boundary. To simplify the treatment of biofilm expansion we consider a 2D incompressible biofilm with an initial radius rb(τ = 0) = 1. Due to the consumption of nutrients the total area of the biofilm A(τ ) = π(rb(τ ))2 increases at a ka(r) 2πr dr, where ka(r) denotes the 2D areal growth rate and is related to the linear growth rate kg by ka = 2kg. The growth rate ka(r) was chosen to be proportional to the local nutrient uptake rate, specifically rate ∂τ A =R rb ka(r) = 2φ(cid:0)c(r, z = 0)(cid:1). The radius of the biofilm thus expands with velocity 0 (S27) ∂τ rb = r−1 ka(r) r dr. b Z rb 0 The 3D geometry of the substrate was chosen to be a cylinder with radius Rs and thickness Hs, and the initial concentration of nutrients was set to c(r, z, τ = 0) ≡ 1. Note that the 3D model of nutrient dynamics depends on two parameters K and J0 related to the uptake of nutrients; the corresponding parameters in the reduced 2D model are 11 Fig. S2: Dynamics of nutrient diffusion in the agar substrate and uptake by biofilm cells. (A) Snapshot of the nutrient concentration field in the substrate, denoted by c(r, z), computed from the full 3D nutrient diffusion model (see Secs. II A and II E for details) at time t = 20 h. The characteristic length scale Rb0 = 2 mm and the time scale τ0 = 10 h were used to convert nondimensional coordinates and time. Color scale indicates the dimensionless nutrient concentration c. Dashed curves are the contours for c = 0.2, 0.4, 0.6, and 0.8. Red bar on top of the substrate indicates the extent of the biofilm at the corresponding time, within which a non-zero flux boundary condition is applied. (B) The nutrient field at the center of the biofilm/substrate shown in A, defined as cz = c(r = 0, z, t), versus z coordinate at time t = 20 h (blue), 30 h (green), 40 h (yellow), and 50 h (red). The nutrient concentration at r = 0 and z = 0, denoted by c0, shows a slow decay over time. Inset: the evolution of the nutrient field cz can be collapsed onto a master curve cz = c0(t) + (1 − c0)f (τ−1/2 z), where τ and z denote the dimensionless time and z-coordinate, respectively. (C) The nutrient-dependent Monod factor at r = 0 and z = 0, expressed as c0/(c0 + K), versus time t, for the full 3D diffusion model (blue circles) and a 1D vertical diffusion model (red circles). The black line indicates a slope of -1/2 on a log-log scale. This value of the slope indicates that c0 scales as t−1/2 in the long time limit, and the slow decay arises from the vertical diffusion dynamics. Simulation model parameters are: K = 0.4 and J = 2.2. K and Q. velocity of biofilm: In simulations we compared the following three scenarios, where we used the same Eq. (S27) for the expansion (A) the reduced 2D model of nutrient dynamics in Eq. (S13); (B) the full 3D model of nutrient dynamics with a thin substrate Rs = 22.5, and Hs = 0.05 (thin substrate); and (C) the full 3D model of nutrients with a thick substrate Rs = 22.5, and Hs = 3 (these dimensions are comparable to the experimental agar substrates, Rs ≈ 45 mm and Hs ≈ 6 mm). −Hs Fig. S3 shows the profile of nutrients c(r, z = 0, τ ) and the radius of the biofilm rb(τ ) for the three scenarios described above. Note that if the substrate is sufficiently thin, i.e. Hs (cid:28) 1, the diffusion of nutrients across the substrate thickness is very fast and therefore the concentration of nutrients c becomes nearly uniform across the thickness and thus independent of coordinate z. In this case the reduced 2D model of nutrient dynamics approximates the averaged nutrient concentration ck = R 0 c dz with the maximum uptake rate Q0 = J0/ Hs (see for example, [28]). Fig. S3 shows that the results for models A and B are in fact nearly identical except at very early times τ (cid:28) H 2 related to the characteristic time for diffusion of nutrients across the substrate thickness. For the thick substrate (model C) we find a nutrient rich annulus at the periphery of the biofilm, similar to the other two models. However, for model C the concentration of nutrients directly beneath the center of the biofilm decreases very slowly (see Fig. S2A), due to the vertical diffusion of nutrients. To quantify this effect, we consider z c with the same uptake of nutrients −∂z c = J0c/(c + K) at a 1D diffusion problem in the z direction, ∂τ c = D∂2 z = 0. The asymptotic solution for this 1D problem is c(τ, z) = c0(t) + (1− c0)f (t−1/2z), where c0(t) ∼ t−1/2 denotes the concentration at z = 0, and the function f (x) satisfies the differential equation f00(x) + (1/2)xf0(x) = 0 with boundary conditions f (0) = 0 and f (∞) = 1. This scaling is consistent with the results of simulations of the 1D model described above and for the 3D simulations of model C (see Fig. S2). Thus the concentration of nutrients underneath the core of biofilm evolves in time as c(r = 0, z = 0, τ ) ∼ τ−1/2. To account for this slowly vanishing concentration, in the reduced 2D model we introduced a small residual growth rate in Eq. (S14) that is independent of the reduced nutrient concentration ck. s 1D vertical diffusion 3D diffusion ~ t -1/2 101 101 t (h) 100 100 0.6C 0.6 0.4 0.4 0.2 0.2 c0 / (c0+K) ~~ ~ 5 5 6 10 6 012 2 t (h) 50 0.5 0 0 𝞽1/2z~ 0 0 3 5 3 z (mm) 1 01 (cz - c0)/(1-c0) ~~ ~ 0.5 1 0.5 0 0 0 0 0 0 ~ c0 0 0 1 20 h 30 h 40 h 50 h 0.5 B 1 1 0.5 0.5 1 Nutrient cz ~ 5 r (mm) 10 01 ~ c 0.2 0.4 0.6 0.8 0 -3 -6 0 A z (mm) 12 Fig. S3: Validation of the reduced 2D model for nutrient diffusion-uptake dynamics. (A-C) Model schematic (top) and the spatiotemporal evolution of the nutrient field (bottom) are shown for (A) the reduced 2D diffusion-uptake model (dotted curves), (B) the full 3D diffusion model with a thin substrate (substrate thickness Hs = 0.1 mm, dashed curves) and (C) the full 3D diffusion model with a thick substrate (experimental substrate thickness Hs = 6 mm, solid curves). For each model, the nutrient concentration at the biofilm-substrate interface, defined as ck = c(r, z = 0), versus radial coordinate r is plotted within the biofilm regime (r < rb, where rb denotes the biofilm radius) for different times. ck is normalized by the half-rate constant K to visualize the spatial differences in growth/uptake rates. Color scale indicates time. The characteristic length scale Rb0 = 2 mm and the time scale τ0 = 10 h were used to convert nondimensional coordinates and time. See text for details about the implementation of biofilm expansion dynamics. (D) Biofilm radius rb plotted against time t for the three models. Line styles are the same as in A-C. Simulation parameters were chosen such that the three models result in similar biofilm expansion dynamics as follows: (A) K = 0.5, a−2 c = 3.0; (B) K = 0.4, J = 0.26; and (C) K = 0.4 and J = 2.2 as in Fig. S2. F. Model discussion: Constitutive model In our chemo-mechanical model, we employ a phenomenological hyperelastic material model to describe biofilm mechanics. It is not known whether growing V. cholerae biofilms are effectively compressible or not, since they are physically swelled, and can, in principle, undergo volumetric change by shifting water to the agar substrate. Our goals here are (i) to derive the constitutive relation from a model of the microscopic interactions, and discuss the use of the simple neo-Hookean model, and (ii) to estimate the effective compressibility of biofilms grown on agar. Previous works [11, 29] showed that the mechanical behavior as well as the matrix composition of V. cholerae biofilms resemble those of hydrogels. Thus, we use a hydrogel-like constitutive model to approximate biofilm elasticity. We caution that the microscopic model is still highly simplified, and cannot capture all the complex rheological behaviors observed in the experiments [11]. A swollen hydrogel model: We model biofilms grown on agar as cross-linked polymer networks in contact with a reservoir of solvent (water) molecules. Similar models have been proposed before to study polymeric gels [30, 31]. Since the agar substrate contains a much larger total amount of water than the biofilms, we assume that the presence of agar provides a constant chemical potential of water molecules µw. Experimentally, µw can be determined from the 0/3, where the chemical potential equilibrium concentration φ0 of a dextran droplet atop the agar [11] as µw = −kBT φ3 for a theta solvent [32] was used, because in aqueous solution dextran assumes a conformation close to that of an ideal coil [33, 34]. 3D thin substrate 0 15 03 012 t (h) 0.5 5 5 r (mm) 1 10 10 1 0.5 0 0 0 0 012 012 10 10 B c / K ~~ Nutrient 2D reduction 0 5 10 3D thick substrate D 2D 3D thin 3D thick 20 2 30 t (h) 10 1 05 05 0 0 rb (mm) Biofilm radius 0 0 5 5 r (mm) 10 10 012 012 012 012 A c / K ~~ Nutrient C Nutrient c / K ~ ~ 13 Fig. S4: Compressibility of a biofilm in contact with an agar substrate that provides a water reservoir. (A) Schematic of the cross-linked polymer network model for the biofilm matrix. When biofilms are grown on agar substrates, they can exchange water with the agar (denoted by black double-headed arrows). We assume the total water content of the agar is much higher than that of the biofilm. Thus, we treat the agar as a reservoir of water molecules with a constant chemical potential µw. (B) Schematic representation of the multiplicative decomposition of the deformation of a cross-linked polymer network. The polymer network is cross-linked in the dry state (left). In the presence of a water reservoir, it swells to a minimum free energy equilibrium state (middle). The equilibrium swelling is denoted by F0. External mechanical forces (represented by the gray arrows) will further deform the network (denoted by Fe, right), which could accumulate or lose water accordingly. The total deformation is Fd = FeF0. (C) The free-energy density increase ΨJ due to the volumetric change of the biofilm, defined as all the terms containing Je = det(Fe) in the total free-energy density, normalized by the biofilm shear modulus Gb, plotted versus the volumetric strain Je − 1 for the neo-Hookean constitutive model (solid curves, Eq. (S9)) and the cross-linked polymer network constitutive model (dashed curves, Eq. (S33)). Poisson's ratio ν = 0.37 (blue) and ν = 0.46 (red) correspond to biofilms grown on 0.6% and 1.5% agar, respectively. Next, we consider the free energy of the system when a biofilm is in contact with agar. We take the stress-free dry cross-linked network of biofilm matrix as the reference state (coordinate denoted by Xd, not to be confused with the initial coordinates in the chemo-mechanical model). The dry network can swell and deform due to the presence of osmotic pressure or mechanical loads (Fig. S4). The deformation gradient tensor can be defined as Fd,ij = ∂xi/∂Xd,j where we use our standard notation x to specify the current deformed coordinates. It is commonly assumed that the volume of the swollen polymer network equals the volume of the dry network plus the volume of the absorbed solvent, i.e. both the polymers and the water molecules are incompressible and the volumetric change of the matrix network is solely due to absorbing water from (or losing water to) the agar. We adopt the same assumption and write Jd = det(Fd) = 1 + vwC, where C denotes the number of water molecules per unit reference volume. The volume fraction of polymers in a swollen network is thus φ = J−1 d . 1 Following the equilibrium swelling theory of the Flory and Rehner model [35, 36] and previous works on swollen 1 Here, we assume the polymer network is cross-linked in its dry state. But this assumption is not essential for the results derived below. d where φd denotes the polymer volume fraction in the state where they are cross-linked. One can In fact, we can assume φ = φdJ−1 perform the same calculations outlined here and obtain the same results. 𝝼 = 0 . 37 0.8 -0.2 0 0 0.7 -0.3 0.9 -0.1 Volumetric strain 1 0 Je - 1 Neo-Hookean Polymer network = 0.46 𝝼 1 1 1.5 1.5 0.5 0.5 ΨJ/ Gb C Energy density μw Water molecule exchange between biofilm matrix and agar substrate F0 Fe Dry Swollen （equilibrium） Stressed A B 14 µw vw (Jd − 1), (S28) where N denotes the number of polymer strands per unit reference volume, and the long chain limit is assumed [32]. The first term arises from stretching of the polymer network, and has an entropic origin [35]. The second term is the standard mixing free energy of polymer solutions, in which the Flory interaction parameter χ characterizes the two-body interaction between the polymer and the solvent. Note that the usual expression f (φ) ≈ v−1 w kBT [(1 − φ) ln(1− φ) + χφ(1− φ)] measures the mixing free energy per current volume, so Wpoly-sol(Jd) = Jdf (φ) = Jdf (J−1 d ). Finally, the third term accounts for the free energy cost of absorbing water from the agar. d . Using the relation The results above allow us to calculate the Cauchy stress σ by σ = J−1 d (δW/δFd) · FT δJd = JdF−T d : δFd, we derive d ) + χ(1 − J−1 d )i − kBT vw h(Jd − 1) ln(1 − J−1 N kBT(cid:2)tr(FT d Fd) − 3 − 2 ln Jd(cid:3) + 1 2 = W = Welastic(Fd) + Wpoly-sol(Jd) + ∆Wagar hydrogels [30], we obtain the total free energy density (per unit reference volume) σ = J−1 d N kBT (FdFT d − I) + kBT vw h ln(1 − J−1 d ) + J−1 d + χJ−2 µw vw I. (S29) d iI − Equilibrium condition: Consider the mechanical and chemical equilibrium condition of the swollen biofilm on agar. We assume the swelling is isotropic, and denote by F0 = λ0I the equilibrium deformation-gradient tensor, and by J0 = λ3 0 its determinant. The equilibrium condition σ = 0 yields λ−1 0 N kBT (1 − λ−2 0 ) + kBT vw h ln(1 − λ−3 0 ) + λ−3 0 + λ−6 0 χi − µw vw = 0, (S30) 0 0 + (φ3 0 − λ−9 0 )/3 + O(λ−12 0 (cid:28) 1 to the second term. The resulting equation becomes Gb(1 − λ−2 from which the initial swelling can be determined. We assume large equilibrium swelling [37] so that we can apply a series expansion in λ−3 1/2)λ−6 is exactly the Lamé's second parameter (shear modulus) of the biofilm material. 0 ) + (kBT /vw)(cid:2)(χ − 0 N kBT and we shall see in the following paragraph that Gb For V. cholerae biofilms, we argue that χ ≈ 1/2. In the absence of protein cross-linkers (i.e. the ∆ABC strain [38]), a submerged biofilm behaves like a polymer solution in contact with pure solvent. One can readily calculate the osmotic pressure in this case to be Π(φ) ≈ −(χ − 1/2)φ2 + φ3/3 + O(φ4). If χ < 1/2 (good solvent), then there is a tendency for the biofilm to swell indefinitely under water; if χ > 1/2, there exists a stable gel fraction φeq = 3(χ− 1/2) for which Π = 0. The experiments [38] suggest that φeq (cid:28) 1 and thus χ is close to (slightly greater than) 1/2, and for finite osmotic pressure, it is reasonable to neglect the order φ2 term. )(cid:3) = 0, where Gb = λ−1 With this approximation, we finally estimate, to the lowest order, the equilibrium swelling ratio to be J−1 0 = λ−3 0 ≈(cid:16)φ3 0 + Gb kBT /3vw(cid:17)1/3 (S31) Lamé parameters: To probe the mechanical properties of the swollen polymer network, we can introduce addi- tional deformations Fe on top of F0 and calculate the resulting Cauchy stress. The total deformation is denoted by e ≈ I + 2, where is the symmetric strain tensor defined in linear elasticity Fd = FeF0. For small deformations, FeFT theories [10]. In the small deformation limit, the Cauchy stress can be specified by two Lamé parameters λb, Gb, according to σ = 2Gb + λbtr()I. Thus, by comparing the Cauchy stress calculated from the polymer network model to the linear elasticity result, one can determine the equivalent Lamé parameters for the swollen biofilm matrix, and consequently the Poisson's ratio νb. First, consider a shear deformation Fe with Je = 1. An example of this is the shear test of biofilm material on a 0 N kBT from rheometer2. [11] Substituting Fd = λ0Fe and Jd = J0 into Eq. (S29), one immediately finds Gb = λ−1 the anisotropic part of the Cauchy stress. The experimentally measured Poisson's ratio of V. cholerae biofilms scraped off a substrate is near νb = 0.5 [11, 12], indicating that the biofilm is a nearly incompressible material. However, when the biofilms are grown on agar, the water molecules retained by the biofilm matrix might be redistributed into the agar upon mechanical loading and/or elastic deformation. In the present paper, since we are modeling biofilm growth and development, we are interested in the possible compressibility of a biofilm in contact with agar. 2 For rheological measurement, the µw term does not exist in Eq. (S29), and instead a Lagrange multiplier should be introduced to ensure that Je = 1 is satisfied. 15 We therefore consider a small isotropic compression/extension, i.e. Fe = (1+/3)I. From linear elasticity theory, the Cauchy stress is given by σ = (λb + 2Gb/3)I. On the other hand, if we set F = λ0(1 + /3)I in Eq. (S29), and expand 0 kBT /vw. 0 kBT /vw(cid:3)I. Assuming λ0 (cid:29) 1, we obtain λb ≈ −Gb + λ−9 in series of , we derive σ =(cid:2)Gb(λ−2 0 with experimentally measurable quantities using (S31), we finally obtain 0 − 1/3) + λ−9 Replacing λ−3 λb Gb = 2νb (1 − 2νb) = 2 + kBT /vw Gb φ3 0. (S32) In our chemo-mechanical model, we proposed to use a neo-Hookean constitutive model Ψ(Fe) = Gb Estimation of νb and comparison to the neo-Hookean model: Note that the right hand side of Eq. (S32) is always larger than 2, which corresponds to a Poisson's ratio larger than 1/3. Our modeling of the biofilm matrix as a swollen polymer network thus leads to the conclusion that biofilms growing on agar behave as almost incompressible hyperelastic materials. Using the experimentally measured values of biofilm shear modulus Gb and the equilibrium dextran concentration φ0 [11], along with kBT = 4 × 10−21J, and vw = 3 × 10−29m3, we can calculate νb ≈ 0.37 for low agar concentration (0.6% agar, φ0 = 0.02, Gb = 1.1kPa) and νb ≈ 0.46 for high agar concentration (1.5% agar, φ0 = 0.045, Gb = 1.4kPa). e Fe) − 3 − 2 ln Je) + λb 2 (ln Je)2 for biofilm material, whereas here we presented a model motivated by the thermodynamics of swelling. To end the discussion on the constitutive relation, we shall compare the resulting strain energy density of the two models. Note that Eq. (S28) gives the strain energy density per unit volume in the dry polymer network configuration. Thus, to compare with Eq. (S9), we rewrite Eq. (S28) to give the strain energy density per unit volume in the equilibrium swollen configuration, i.e. Ψ(Fe) = J−1 e Fe) + ΨJ (Je) + constant. e Fe)/2 term in Here, we have explicitly separated Ψ(Fe) into three parts: the first term is identical to the Gbtr(FT Ψ of the neo-Hookean model; the third term is a constant independent of Fe, and does not contribute to the stress strain relation; the second term is only a function of Je, and reads 0 W (λ0Fe) = 1 0 N kBT tr(FT 2 (tr(FT 2 λ−1 ΨJ (Je) = −J−1 0 N kBT ln(Je) + 0 ) ln(1 − J−1 0 J−1 e ) − J−2 0 J−1 = Gbn − λ−2 0 ln(Je) + kBT vw h(Je − J−1 h(Je − J−1 Gb kBT /vw 0 ) ln(1 − J−1 0 J−1 e ) − J−2 0 J−1 e /2 + φ3 (S33) e /2i − µw vw (Je − J−1 0 ) 0(Je − J−1 0 )/3io. Similarly, we obtain, for the neo-Hookean model, ΨJ = Gb{− ln Je + 0.5(λb/Gb)(ln Je)2} where λb/Gb can be related to the polymer network model via Eq. (S32). We plot ΨJ (Je) − ΨJ (1) for the two models in Fig. S4. We find that the two models relate isotropic compression to energy density in a similar way. G. Model discussion: A microscopic model for surface friction Stick-slip mechanism of surface friction: Adhesion and friction are closely related in many natural processes [39, 40]. We assume that the friction between biofilm and agar arises from the binding and unbinding of biofilm matrix polymers and/or cell surface polymers with the adhesive biofilm proteins that bacteria deposit on the substrate surface (Fig. S5). Specifically, we model polymers as "springs" with spring constant Kp. For polymer blobs of radius Rp, p [32]. For simplicity, we only consider the horizontal stretch of these springs. the spring constant is Kp ∼ kBT /R2 The rates of binding (kon) and unbinding (koff) at equilibrium are related by k0 off = kon exp(−Eadh/kBT ) where Eadh denotes the energy difference of the bound and unbound state for a single unstretched polymer spring. When the biofilm moves relative to the agar substrate at a velocity v, the adhered polymer springs become stretched. The total displacement of the anchor point of a single spring on the biofilm is ∆x = vt at time t after binding. The resulting elastic force on this spring is fel = K0∆x where K0 = KpKs/(Kp +Ks) accounts for the elasticity of both the substrate and the polymer (Fig. S5). The effective spring constant of the elastic substrate is given by the Boussinesq solution Ks = πEsa/((1 + νs)(2 − νs)) [10], in which Es and νs denote, respectively, the Young's modulus and the Poisson's ratio of the substrate, and a denotes the radius of the interaction area. The bound polymers ultimately detach from the substrate and this stick-slip process gives rise to the kinetic friction when the biofilm moves relative to agar. Consider an area element ∆A in which Nb polymer springs are attached to the substrate. The kinetic friction per unit area of biofilm can be calculated as f = (∆A)−1PNb i=1 K0vti = ρpK0vτoff where ρp = Nb/∆A is the average density of polymers bound to the substrate and τoff = k−1 off is the average lifetime of bonds between the polymers and the substrate. At steady state, ρp = ρp0τoff /(τoff + τon), where ρp0 ≈ 1/a2 denotes the density of all possible binding sites, τon = k−1 on is the average time needed for a free floating polymer to bind to the surface again after unbinding. 16 Fig. S5: Microscopic model for surface friction between a biofilm and the agar. (A) Close-up schematic of the mechanical interactions between a biofilm and the substrate surface. Bacterial cells deposit biofilm proteins (yellow Y-forks) on the substrate surface, to which extracellular polysaccharides (EPS, black curvy lines) can adhere [11, 29, 41]. (B) A two-state toy model for the binding/unbinding kinetics of biofilm polymers and proteins. Polymers, represented by springs (black), can be either unbound (left) or bound to surface-adhered biofilm proteins (right). Binding and unbinding of the polymer springs occur at rates kon and koff, respectively. (C) Schematic of the stick-slip process underlying surface friction between a biofilm and the agar substrate. As the biofilm expands relative to the substrate, polymer springs that are bound to the substrate surface (via biofilm-derived proteins) become stretched and they deform the substrate. This activity gives rise to the frictional forces that resist biofilm expansion. Inset: a simplified representation of a polymer bound to the elastic substrate showing a polymer spring (stiffness Kp) connected in series with a substrate spring (stiffness Ks). Substituting the steady state ρp into the above expression for f, the kinetic friction reads f = a−2K0v τ 2 off (τoff + τon) . (S34) Estimation of η: The thermodynamic picture of reversible binding implies that the stretching of a polymer would favor unbinding since the free energy of the bound state is increased by Eel = (K0vt)2/(2Kp), and the unbinding rate of a stretched polymer becomes koff = k0 off exp(Eel/kBT ) = kon exp(cid:16) K02/Kp 2kBT (vt)2(cid:17), (S35) 0 P (t)dt. where we define = exp(−Eadh/kBT ). We next calculate the average lifetime for bound polymers τoff. Let P (t) be the probability of remaining bound at time t after initial binding. It is straightforward to show that P follows 2kBT (vt0)2(cid:17)dt0(cid:17), and τoff can be readily calculated by dP/dt = −koff (t)P . Thus, P (t) = exp(cid:16) −R t τoff =R ∞ one can expand the exponential terms in a series, and obtain τoff /τon = R ∞ Let ξ0T = (2kBT Kp)1/2/K0 be the characteristic stretch of the polymer springs activated by thermal energy. We consider a dimensionless parameter α = vτon/ξ0T and a dimensionless variable x = t/τon. To calculate τoff explicitly, (2n+1)n! x2n+1)dx ≡ φ(α/). Substituting this relation into (S34), we obtain 0 kon exp(cid:16) K02/Kp 0 exp(−x −P∞n=1 (α/)2n f = ρp0K0ξ0T αφ2(α/) ( + φ(α/)) . (S36) In general, f has a complex dependence on v (or α). However, in the limit α/ (cid:28) 1, i.e. when the binding/unbinding events happen so frequently that the change of activation energy, and its effect on the unbinding rate, can be neglected, A B C kon koff v Kp Ks 17 Fig. S6: Spatiotemporal evolution of multiple fields in the chemo-mechanical model for biofilm expan- sion. Time evolution of (A) the reduced nutrient field ck, (B) the radial velocity field v, (C) the vertical stretch γ, (D) the magnitude of the circumferential stress σθθ, (E) the magnitude of the radial stress σrr, and (F) the stress anisotropy ασ = (σθθ − σrr)/(σθθ + σrr) in the chemo-mechanical model for biofilm expansion. Simulation parameters are the same as in Figs. 2C and D. Color scale indicates time t. we find that φ(α/) → 1, and we find that the friction force takes the form f = ηv with the drag coefficient η = ρp0K0τon ( + 1) ≈ −1ρp0K0τon, for (cid:28) 1. (S37) Parameter estimation: Finally, we estimate all the parameters in the above analysis based on experimental results. First, for the effective spring constant of the polymer, Rp ∼ 10 nm, then Kp ∼ 0.1 pN·nm−1. For a substrate with shear modulus Gs = 103 Pa, if a ∼ 3 nm, then Ks ∼ 10−2 pN·nm−1; thus, for the agar concentration we are interested in (Gs < 3 kPa), we expect Ks (cid:28) Kp and K0 is dominated by Ks only. In this regime, Gs is related to η by η ∝ Gs. To estimate τon, we follow [42], and assume each polymer blob thermally behaves as a hard sphere of radius Rp p/kBT = (10−3 Pa · s × 103 nm3)/(4 pN · nm) ∼ 10−6s, fluctuating around its equilibrium position, so τon ≈ µR3 where µ denotes the viscosity of water. Next, the energetic factor can be estimated from the measurements of adhesion energy. From the microscopic picture, the energy required to separate the biofilm from the substrate is 1+exp(−Eadh/kBT ) . Experimentally, the measured Γ value is about 5 mN · m−1 Γ = ρb0Eadhτ 0 [11], which leads to (Eadh/kBT ) ≈ 10. However, because an order 1 difference in (Eadh/kBT ) can lead to 10-fold change in , we estimate the range of to be 10−6 ∼ 10−4 . off + τon) = a−2kBT off /(τ 0 (Eadh/kBT ) We then check whether our approximation in deriving the linear dependence on v, namely α (cid:28) , holds true. The thermal length scale we defined, ξ0T can be estimated to be ξ0T ∼ 100 nm for Gs = 103 Pa. For a typical biofilm expansion velocity v ∼ 3 µm/min = 50 nm/s, one finds α = vτon/ξ0T = 5× 10−7 < . Thus, the approximation is valid according to the estimated values. Finally, we can calculate the drag coefficient η and compute the order of magnitude of the dimensionless control parameter ξ = η(Rb0/τ0) Gb(H/Rb0). Assuming the shear modulus of the film is Gb = 103 Pa, we obtain η ≈ −1ρb0K0τon = (104 ∼ 106)(0.1 nm−2)(10−2 pN/nm)(10−6s) = 104 ∼ 106 Pa · s · µm−1, and ξ = η(Rb0/τ0) extracted by fitting the experimental velocity profile. ≈ 1 ∼ 100. This estimation is also consistent with the value of ξ we Gb(H/Rb0) ≈ (104∼106 Pa·s·µm−1)×(3 µm/min) 103 Pa×(100µm/3mm) 0 10 02 012 t (h) 0.5 0.5 0.5 0 4 0 0 0 0 2 0 0 2 1 0.5 2 2 4 4 r (mm) 4 4 0 0 2 4 2 2 4 4 4 123 123 2 1 1 123 z-stretch γ 0 C 1 Anisotropy ασ 0.5 0.5 0 0 0 4 0 F 1 1 123 0 2 2 4 2 4 4 0 2 2 4 2 r (mm) 4 4 024 2 024 2 024 024 024 024 0 0 0 0 024 v (µm/min) 4 0 024 Stress σrr ~ 4 0 E B 0 2 2 0 0 2 4 4 2 4 4 4 2 4 024 0 2 2 2 r (mm) 0 0 024 0 0 0 0 024 024 0 0 1 1 1 1 0.5 0.5 0.5 0.5 A c ~ Nutrient Stress σθθ ~ D III. ANALYSIS OF BIOFILM EXPANSION DYNAMICS In both the simulations and the experiments, the expansion velocity of a biofilm shows three distinct stages (Figs. 2D and S6). In the first stage, the interior remains immobile and only the edge of the biofilm moves outward. In the second stage, the entire biofilm undergoes uniform expansion and the radial velocity scales (almost) linearly with the radial coordinate, i.e. vr ∝ r. In the third stage, the expansion at the center slows down again due to nutrient limitation [28, 43]. To understand the expansion kinematics in the first two stages, we hypothesize that nutrient limitation and non-uniform growth play a minor role in these early stages of biofilm development. Thus, we focus on analyzing the expansion dynamics of a biofilm that is undergoing uniform isotropic growth in the presence of surface friction. 18 A. Uniform biofilm growth Due to rotational symmetry, the force-balance equation (S17) can be expressed in polar coordinates as ∂τ(cid:17) R0 Jkξ(cid:16) ∂ r = ∂ R0 ( S0,k)rr + ( S0,k)rr − ( S0,k)θθ R0 , (S38) where Jk = ∂(r2)/∂( R2 0) measures the change of area from the initial to the current configuration and ( S0,k)rθ = 0 due to symmetry. Note that due to the symmetry of the model, we do not explicitly distinguish between the radial and circumferential directions in the initial coordinates and those in the current coordinates. Consider uniform and isotropic growth Fg,k ≡ eτ Ik. For simplicity, we focus on discussing the behavior of an e,k − γ2Ik, where γ = 1/Je,k and the stress Jkγ. Axisymmetry yields the following expressions for Fk and incompressible material whose dimensionless Cauchy stress is σk = Fe,kFT measure in the initial configuration is S0,k = σk · F−T k Fe,k in polar coordinates: Fk =(cid:18)∂ R0 0 r 0 r/ R0(cid:19) , and Fe,k = Fk · F−1 g,k =(cid:18)e−τ ∂ R0 0 r 0 e−τ r/ R0(cid:19) , (S39) where the first (second) row and column denote the radial (circumferential) component. Substituting these expressions into the constitutive relation, the force-balance equation then becomes a partial differential equation (PDE) for the unknown function r( R0, τ ). It is difficult to analytically solve this equation. Nevertheless, to gain insight into the first two stages of biofilm expansion, we analyze the short time and long time expansion kinematics of Eq. (S38). B. Stage I: Stress accumulation For the initial stage of biofilm growth, when the displacement u is small compared to the biofilm radius, we can rewrite r( R0, τ ) as R0 + u( R0, τ ) and expand the force-balance equation in powers of δ = u( R0, τ )/ R0 (cid:28) 1. This yields, to the leading order, ∂ R0 ( S0,k)rr = (3e6τ + 1)u00 + 2e6τ (u/ R0)0, R−1 0 (cid:2)( S0,k)rr − ( S0,k)θθ(cid:3) = (e6τ + 1)(u0/ R0 − u/ R2 ξ u = (3e6τ + 1)(cid:0)u00 + R−1 0 u(cid:1). 0 u0 − R−2 0), and (S40) (S41) The boundary conditions are u R0=0 ≡ 0 and (2u0 + u/ R0) R0=1 ≈ 3τ for τ (cid:28) 1, the latter deriving from the stress-free boundary condition at the edge (( S0,k)rr R0=1 = 0). In the early stage of biofilm expansion, the radial expansion velocity is negligible in the interior of the biofilm (denoted by zone I) because the internal stress is insufficient to drive the bulk of the biofilm to expand against friction. Clearly, u = 0 is one possible solution for the interior displacement field. By contrast, the edge of the biofilm is able to move outward (denoted by zone II) from the beginning due to the existence of a stress-free boundary. Thus, we seek for another solution in zone II (similar to the boundary layer theory in fluid mechanics) to connect the immobile core to the moving edge. Analytically, in the short time limit, τ (cid:28) 1, the equation becomes ξ u = (4 + 18τ )(cid:0)u00 + R−1 0 u0 − R−2 boundary at the edge R0 = 1 is (2u0 + u) R0=1 = 3τ. We seek for solution in the form u = τ nf ( 1− R0 0 u(cid:1), and the τ m ) where the 19 Fig. S7: Biofilm expansion dynamics in the initial stage. (A) The radial displacement field u plotted against the radial coordinate in the initial reference configuration, denoted by R0, at dimensionless time τ = 0.1 (blue), 0.2 (green), and 0.3 (red). Rb0 = 1 denotes the initial biofilm radius. The dimensionless friction parameter ξ = 50. Inset: the displacement fields u computed for ξ = 50 (solid curves), 100 (dotted curves), and 150 (dashed curves) at time τ = 0.1 (blue), 0.2 (green), and 0.3 (red) all collapsed onto a master curve corresponding to a self-similar solution RB0− R0 u( R0) ≈ ξ−1/2τ 3/2g( (τ /ξ)1/2 ), where g is a function given by Eq. (S44). (B) The dimensionless time duration of the first stage of biofilm expansion T1 (gray region), which ends when the material point at R0 = Rb0/2 = 1/2 starts to move outwards, i.e., the time at which u( R0 = 1) > 0 (illustrated in the inset for ξ = 100), versus the dimensionless friction parameter ξ. The black line indicates a slope of 1 on a log-log scale. (S43) df dx − nf = 0. x 2 d2f dx2 + 4 ξ , and 1 2 m = shape function f (x) is defined on [0, +∞) with f (x → ∞) ≡ 0 to describe the velocity profile in the transition region. Let x = (1 − R0)/τ m. The time derivative of u is given by u = nτ n−1f (x) − mτ n−1xf0(x) ∼ τ n−1. (S42) Similarly, we obtain that the R.H.S. of Eq. (S41) is given by 4u00 = 4τ n−2mf00 to leading order in τ. Comparing the exponents of τ we find The same analyses can be applied to the boundary condition at R0 = 1, which yields n = 3/2, and df dxx=0 = −3/2. Let ¯x = ξ1/2x, and g(¯x) = ξ1/2f (x). We find that the scaling function g follows an ordinary differential equation independent of ξ, i.e. 8 d2g d¯x2 + ¯x dg d¯x¯x=0 = −3/2, g(¯x → ∞) = 0. (S44) dg d¯x − 3g = 0, and  the definition f (x) = ξ−1/2g(cid:0)ξ1/2x(cid:1), we can rewrite u( R0) as u( R0) ≈ ξ−1/2τ 3/2g( 1− R0 The exact solution of the ODE above contains an error function and is not of particular interest. However, from (τ /ξ)1/2 ). This means that the displacement/velocity profile assumes a self-similar solution. Furthermore, since the velocity field depends on τ only through (τ /ξ)1/2, this suggests that the time a biofilm spends in the first stage T1 is proportional to ξ. Both the self-similar ansatz and the dependence of T1 on ξ are verified in the simulation, where T1 is defined as the time when u(1/2) becomes larger than 10−12 (see Fig. S7). To calculate the stress accumulation in stage I, we note that the inner part of the biofilm does not move, which, given the isotropic growth Fg = λgIk, implies a compensating isotropic in-plane deformation, i.e. (Fe,k)rr = (Fe,k)θθ ≈ e−t. The Cauchy stress accumulates as σrr = σθθ ∼ 6t when t (cid:28) 1. 0 0 -10 -20 -20 -10 0 ~ ~ (R0-Rb0)/(𝞽/ξ)1/2 0 0 1 0.5 0.5 Initial coordinate ~ R0 0 1 0 10-2 10-2 101 101 0 0 0.1 0.1 𝞽 1 ~ ξ 1 102 102 Friction ξ 10-9 T1 05 0 10-8 u(R0=0.5) ~ ~ 10-1 10-1 Duration T1 0.02 Edge: ~ B Rb0 0.04 𝞽 = 0.3 𝞽 = 0.2 𝞽 = 0.1 24 024 uξ1/2𝞽-3/2 ~ 0.02 A Center 0.04 ~ u Displacement C. Stage II: Isotropic expansion 20 As the biofilm continues to grow, internal stresses build up, and the mobile region becomes larger. Ultimately, the entire biofilm expands outward. In this stage, r > R0 and both Fk,rr and Fk,θθ become larger than 1. Eq. (S39) implies that the accumulation of elastic strain is partially relaxed by biofilm expansion. Therefore, we expect that the stress will increase at a rate slower than material production. To understand the biofilm expansion kinematics in this regime, we focus on the long time asympoptic behavior of Eq. (S38), i.e. τ (cid:29) T1 and τ (cid:29) 1. Since (Fe,k)rr/θθ = λ−1 g Fk,rr/θθ, we expect that the isotropic stress γ2 = [(Fe,k)rr(Fe,k)θθ]−2 dominates over the Fe,kFT term in Eq. (S16) for the bulk of the biofilm for τ (cid:29) 1 provided that e,k g (cid:28) 1. Neglecting the contribution from Fe,kFT λ−1 e,k , we derive the approximate force-balance equation to be ξ r = R−1 0 e6τ k,rrF 3 F 3 k,θθ (F −1 k,θθ − F −1 k,rr) − F −1 k,rrF −1 k,θθ∂ R0(cid:16) e6τ k,rrF 2 F 3 k,θθ(cid:17), (S45) 0)eωτ for the bulk part of the biofilm, assuming β2 R2 r = r0 and Fk,θθ = r/ R0. Eq. (S45) is to be solved with the boundary conditions at the center where Fk,rr = ∂ R0 r R0=0 = 0 and at the edge r(r0)2 R0=1 = e3τ . Note that the boundary condition at the edge follows from ( Sk)rr = 0, where the Fe,kFT term is kept. Motivated by the observations in the simulations, we test a uniform expansion e,k ansatz rB( R0, τ ) = (β1 R0 + β2 R3 0 term is absent, because it would produce a constant term on the R.H.S. of Eq. (S45)). Note that the form of rB naturally gives rise to the relation vr = ∂τ rB ∝ rB. Inserting the ansatz for rB into Eq. (S45), we find, to leading order in R0, 1 = ξβ9 ξωβ1 R0eωτ = 24β2β−8 32 . However, the ansatz rB does not satisfy the boundary condition at the edge R0 = 1, so we need to find another test solution near the edge. Consider an annulus 1 − ∆ R ≤ R0 ≤ 1, where ∆ R (cid:28) 1 denotes the width of the transition zone (to the bulk solution) at the edge. Since the radius of the biofilm is largely determined by the displacement in the bulk we expect that the edge solution will have the same exponent as the bulk solution. Specifically, we use the ansatz rE = ¯β1e3τ /4 and (r0)E = ¯β1−1/2 e9τ /8 at the edge, the latter derived from the boundary condition. Here, the superscript E denotes functions evaluated at the boundary R0 = Rb0 = 1. R0e6−7ωτ . Comparing the coefficient and the exponent leads to ω = 3/4 and β2 = ξω 0 (cid:28) 1 (the R2 24 β9 1 1 To connect the two solutions, we leverage different continuity constraints at R0 = 1 − ∆ R to compute the undeter- mined coefficients. Assume ∆ R takes the form ∆ R = ∆ R exp(ωEτ ). First, the two solutions should give the same value of r at the intersection, i.e. rB R0=1−∆ R = rE − (r0)E∆ R, from which one can obtain the edge time exponent 1 ∆ R ≈ β1 + β2 = β1(1 + ξβ8 ωE = −3/8 and the prefactor ∆ R satisfies equation ¯β1 − ¯β−1/2 32 ). Additional equations for the undetermined coefficients ¯β1, β1, ∆ R are obtained by considering the continuity of the radial components of stress ( Sk)rr, i.e. ( SB )rr R0=1. Using the ansatz rB for the bulk we obtain the radial stress component ( SB k )rr R0=1 = 0. The derivative of the radial stress component at the edge ∂ R0 e21τ /8 is obtained by using the ansatz for the edge rE in Eq. (S38) at R0 = 1. For scaling analysis, we assume that all the prefactors have the power-law dependence on the control parameter ξ , which yields ¯β1 ∼ β1 ∼ ξ−1/8 and ∆R ∼ ξ−3/16. In summary, we obtain the following scaling formulas, 1 e9τ /4. The boundary condition at the edge sets ( SE k )rr R0=1−∆ R ≈ −β−5 k )rr R0=1−∆ R ≈ ( SE k )rr R0=1 − ∆ R ∂ R0 )rr R0=1 = 3 (SE k (SE k 4 ξβ1 3/2 1 Bulk: Edge: rB( R0) ∼ ξ−1/8e3τ /4 R0, rE ∼ ξ−1/8e3τ /4, Transition layer: ∆ R ∼ ξ−3/16e−(3/8)τ . (r0)B ∼ ξ−1/8e3τ t/4, (r0)E ∼ ξ1/16e9τ /8, (S46) (S47) (S48) By substituting the relations above into the expressions for Cauchy stresses σrr, σθθ and the stretch factor γ, we find the following asymptotic solution: σB rr ≈ σB θθ ∼ ξ1/8eτ /4 σE γB ∼ ξ1/4eτ /2 θθ ∼ ξ1/2eτ in the bulk of the biofilm, near the edge of the biofilm, and for the increase of the biofilm thickness. (S49) (S50) (S51) All of the scaling relations derived above are verified by the simulations (Fig. S8). 21 Fig. S8: Validation of the scaling predictions for the second stage of biofilm expansion.(A) The biofilm radius rb, (B) the vertical stretch at the center of the biofilm γ(r = 0), and (C) the magnitude of the circumferential stress σθθ at the center (black) and edge (magenta) of the biofilm plotted versus the dimensionless time τ (left) and the dimensionless friction parameter ξ (right). The black lines indicate the predicted slopes on log-lin (left) or log-log (right) scales. ξ = 20 in left and τ = 2 in right. Simulation results with uniform growth rates are used to approximate the second stage of biofilm expansion, in which nutrient-dependent non-uniform growth plays a minor role. D. Stage III: Nutrient-limited growth In the final stage of biofilm expansion, growth slows down in the central part of the biofilm due to nutrient limitation. We first argue that the resulting edge growth mode results in an approximately linear increase of biofilm radius with time. To see this, we consider the kinematics of growth of an incompressible flat biofilm, and neglect any nutrient-independent growth. Integrating the material conservation relation λ2 gR0dR0 = γrdr, we calculate the radial coordinate of a material point, labeled R0 in the initial configuration, from r2/2 =Z R0 0 γ−1λ2 gR0dR0. (S52) ξ = 20 ~ e3𝞽/4 101 101 101 101 101 101 𝞽 = 2 ~ ξ -1/8 1 1 1 1 1 1 2 2 2 2 2 2 3 3 3 3 3 3 1 1 1 1 1 1 Center Edge ~ e𝞽/2 2 2 2 2 2 2 3 3 3 3 3 3 ~ e𝞽 ~ e𝞽/4 3 2 2 3 2 2 3 3 3 3 3 2 𝞽 2 2 1 1 1 1 1 1 1 100 100 100 100 100 100 101 101 101 100 100 100 100 100 101 100 101 101 100 100 100 100 100 100 102 102 102 100 100 100 100 100 102 100 102 102 100 100 100 100 100 100 100 100 100 100 100 100 100 102 102 102 102 102 102 ~ ξ 1/4 102 102 102 102 102 102 ~ ξ 1/2 101 Friction ξ ~ ξ 1/8 102 102 102 102 102 102 102 A 101 101 101 101 101 101 101 Biofilm radius rb~ z-stretch γ(r=0) 100 100 0 0 100 100 0 0 100 100 100 101 101 0 0 101 101 101 101 101 B 100 100 0 0 100 100 100 0 0 100 100 102 102 0 0 102 102 102 102 102 101 100 100 100 100 100 100 100 10-1 0 0 0 0 0 0 0 Stress σθθ ~ C Taking the derivative of Eq. (S52) with respect to time, and using Eq. (S14) and the material conservation relation to rewrite the integral in the current coordinates, we obtain r r = 2Z r 0 φ(ck(ρ))ρdρ −Z r 0 γ−1(∂τ γ)ρdρ, (S53) 22 which yields the velocity field in the growing biofilm. In particular, setting r = rb(τ ), one finds how the biofilm radius rb increases over time. In the edge growth regime, φ ≈ 1 for rb − ac < r < rb and φ ≈ 0 for r < rb − ac, so the first term of the R.H.S. of the equation can be estimated to be 2acrb. To estimate the upper bound on the second term of the R.H.S. of the equation, we consider a scenario where the local biofilm growth contributes solely to the increase of γ with no in-plane expansion. In this limiting case, γ−1(∂tγ) = ∂t(ln γ) = const. in the active growing annulus. Therefore, the second integral should scale at most linearly with rb. Combining these results, we obtain rb rb ∼ constant × rb, or rb ∼ constant. We conclude that edge growth yields a linear increase of biofilm radius with time.3 Next, we calculate the accumulation of circumferential stress at the edge in stage III. For simplicity, we consider e,k − γ2Ik. The stress-free boundary condition σrr = 0 yields rr(Fe,k)θθ = 1, where we invoked the incompressibility condition γ = (Fe,k)−1 θθ . Inserting this relation θθ − θθ . Since nutrients are assumed to be always sufficient near the edge in our model, we have λgr=rB = eτ , and g r=rb rb/ Rb0 ∼ τ e−τ , where we have used the fact that rb increases linearly with time. In θθ at the edge, and the circumferential stress increases as σθθr=rb ∼ eτ /τ over an incompressible hyperelastic material, i.e. σk = Fe,kFT (Fe,k)2 between (Fe,k)rr and (Fe,k)θθ at the edge into the expression for circumferential stress, we find σθθ = (Fe,k)2 (Fe,k)−1 hence Fe,kθθr=rb = λ−1 the long time limit, (Fe,k)2 time. This scaling relation was verified in simulations. θθ (cid:28) (Fe,k)−1 rr (Fe,k)−1 IV. ANALYSIS OF THE LEADING ANGLE In our chemo-mechanical model, we highlight the surface friction as a key factor in determining biofilm expansion kinematics and the spatiotemporal stress distribution. The shear force applied to the bottom of the biofilm also suggests that the leading angle increases with increasing friction/agar concentration. This is indeed the case in the experiments. To understand the quantitative relation between the biofilm leading angle and the kinematic friction, we focus on the growth and deformation of the wedge-shaped edge of a biofilm (whose width ∼ 100 µm is much smaller than the biofilm radius rb ∼ 5 mm). We start with a 2D case, where an elastic wedge with initial angle φ0 is placed in the x-z plane. The angle remains unchanged in the virtual configuration after isotropic growth.4 A friction force is applied to the bottom surface z = 0. The analyses below is restricted to the tip region of the wedge so that we can ignore spatial variations.5 The shear stress in the biofilm can be determined from the boundary condition at the bottom, i.e. σxzz=0 = ηvb where vb is the velocity at the edge of the biofilm. Using the stress-free boundary condition σ · n = 0 at the upper surface of the wedge, where n = (sin φ, cos φ) is the normal vector to the upper surface, we derive σxx = −(tan φ)−1ηvb and σzz = −(tan φ)ηvb, where φ is the leading angle measured in the current configuration. On the other hand, the stress should also be related to the deformation of the biofilm via the constitutive relation (incompressible) e is positive definite, and hence can be written σ = GbFeFT 1 (cid:19) RT (β1), where λ1 denotes the principal stretch, R denotes the 2D rotation parameterized as Be = R(β1)(cid:18)λ2 sin β cos β (cid:19). Therefore, the using a single variable β denoting the rotation angle in the x-z plane, i.e. R(β) =(cid:18)cos β − sin β deformation tensor Fe can be decomposed into Fe = R(β1)(cid:18)λ1 1 (cid:19) R(β2). Expressing Be and σ in terms of β1 e −pI. The left Cauchy-Green deformation tensor B = FeFT and λ1, and comparing them with the expression of σ derived from the force-balance equation, we obtain 0 1 0 λ−2 0 0 λ−1 σxz = Gb(λ2 1 − λ−2 1 ) cos β1 sin β1 = ηvb, and σxx − σzz = Gb(λ2 1 − λ−2 1 )(cos2 β1 − sin2 β1) = ηvb(tan φ − 1/ tan φ). (S54a) (S54b) 3 One can apply a similar analysis to a model with a small amount of residual growth. This yields r ∼ exp(krτ ) where kr (cid:28) 1. 4 Since here we focus on the very edge of the biofilm where the thickness of the biofilm has not yet reached a plateau, we consider the isotropic growth in the full space and ignore the oxygen/nutrient limitation. 5 Although the analyses hold true up to a linear gradient. 23 Fig. S9: Higher surface friction increases the biofilm leading angle. (A) 2D schematic of the geometric interpretation for the decomposition of Fe. Surface friction leads to shear deformation and increases the leading angle of the biofilm from φ0 to φ (upper row, solid black arrow). This deformation, denoted by Fe can be decomposed into the product of three simple operations (gray dashed arrows): R(φ0) and R(−φ) describing rigid body rotations, and diag{λi} describing the principal shear deformation. (B) Theoretical predictions for the biofilm leading angle φ as a function of the dimensionless friction ηvb/Gb and initial angle φ0 (colorbar), where vb is the expansion velocity at the edge of the biofilm and Gb is the biofilm shear modulus. This is the same plot as in Fig. 2E but with error estimation for the theory curves (shaded color bands). Theory predicts the relation between φ, φ0, and ηvb/Gb, with an undetermined parameter Fe,θθ denoting the circumferential compression at the leading front of an expanding biofilm. The circumferential strain Fe,θθ−1 will first become increasingly negative due to growth-induced compression, and ultimately it will level off due to the wrinkling instability. We estimate the steady-state Fe,θθ to be 0.7 ∼ 0.9 [12], which corresponds to the upper and lower boundaries of the theory color bands. Dividing Eq. (S54b) by Eq. (S54a), we obtain (tan β1)−1 − tan β1 = tan φ − (tan φ)−1, from which we calculate β1 = −φ. Notice that the decomposition of Fe has a simple geometric interpretation (Fig. S9) -- a rotation of β2, followed by a principal stretch, followed by another rotation of β1. Relating different geometric configurations yields β2 = φ0 and λ−2 1 = tan φ/ tan φ0. Substituting for β1 and λ1 in terms of φ and φ0 in Eq. (S54a), we finally derive tan2 φ = tan φ0 + ζ 1/ tan φ0 − ζ , where ζ ≡ ηvb/Gb. (S55) In the absence of friction, i.e. ζ = 0, this relation reduces to φ = φ0. In the presence of friction, φ increases with φ0 for ζ ≤ (tan φ0)−1; if the friction is further increased the wedge will bulge out and tumble. We can apply similar analyses to a 3D wedge in cylindrical coordinates. A correction from the principal deformation in the θ direction, λθ = Fe,θθ, will be introduced, and ζ should be replaced by λθζ in Eq. (S55) to give the final result in 3D. Although the specific value of λθ can only be obtained by solving the coupled mechanical growth equations, we note that λθ is bounded by the critical strain of mechanical instability, so we estimate 1 > λθ & 0.7. Within this regime, the effect of λθ on the prediction of φ is relatively small (Fig. S9). V. ORIGIN OF STRESS ANISOTROPY Stress anisotropy plays an important role in determining the morphology of wrinkles. It was previously shown that, for anisotropic stresses, wrinkles are oriented orthogonal to the direction of maximum compressive stress, while for isotropic stresses, wrinkles form zigzag herringbone-like pattern [44 -- 46]. In this section, we discuss how nonhomo- geneous isotropic growth could result in stress anisotropy. As already discussed in Sec. III, if the biofilm undergoes homogeneous isotropic growth, the isotropic stress will dominate in the bulk part of the biofilm, i.e. (S0,k)rr ≈ (S0,k)θθ. In contrast, non-homogeneous growth could give rise to anisotropic stress in a much larger region of biofilm. We also verify this argument by quantifying the stress anisotropy in the simulation (see Fig. S10). To see why non-homogeneous growth leads to anisotropic stress, one can simply check the stress state of a biofilm 0 λg(R00)dR00 where λg is the growth factor. We confirm that g ∂r/∂R0 = 1 in this state. The circumferential deformation in this state can be calculated by 0 λg(R00)dR00 denotes the average λg up to some if there is no radial elastic deformation, i.e. r = R R0 indeed (Fe,k)rr = λ−1 0 R R0 (Fe,k)θθ = (r/R0)/λg(R0) = hλgiR0/λg(R0) where hλgiR0 = R−1 20 10 0123 φ0 (°) 0 1 0.5 0 0 0.5 100 100 Friction ηvb/Gb 1 60 60 30 30 0 10-1 0-1 01 B Leading angle φ (°) A Fe = R(−ϕ) ⋅ diag{λi} ⋅ R(ϕ0) φ0 φ Rotation R(ϕ0) Rotation R(−ϕ) Principal stretches λi (incompressible) 24 Fig. S10: Non-uniform biofilm growth results in anisotropic stress. (A) Stress anisotropy ασ, defined as the difference divided by the sum of the two principal stresses, versus the radial coordinate r normalized by the biofilm radius rb, and (B) anisotropy range ∆rα/rb, defined as the normalized range of the biofilm region where ασ > 0.1 (illustrated by gray shaded area in A), versus the dimensionless friction parameter ξ, plotted for the model with non-uniform nutrient-limited growth (solid curve) and uniform growth (dotted curve). Simulation results are shown at a specific time t = 30 h. (C) Anisotropy range ∆rα/rb, and the width of the nutrient rich zone (c > 0.5) ∆rc, plotted versus time t. ξ = 20 was used for the simulations in A and C. reference radius R0. Thus, if λg in an increasing function of R0, i.e. the production of biofilm material is faster at the edge than at the center (as is the case in V. cholerae), then hλgiR0 < λg(R0) and the biofilm will be under circumferential compression. Although the true stress state is also affected by the boundary condition and force balance as a whole, this simple analysis captures the basic physical picture of how stress anisotropy is generated: compression preferentially in one direction ensures the compatibility of the material that would otherwise break apart due to non-homogeneous growth. This analysis also immediately points out another possibility of anisotropic stress if the non-homogeneous growth pattern is reversed. If λg is an decreasing function of R0, i.e. matrix production is faster in the center of the biofilm than at the rim, then we expect a region where the radial compressive stress is larger than the circumferential stress. Such matrix production pattern is indeed found in other bacterial biofilm formers. See the main text for further discussion of this point. VI. COARSE-GRAINED MODEL OF WRINKLE FORMATION A. Plane-stress wrinkling model To investigate how internal mechanical stress shapes the 3D pattern of wrinkles, we develop a plane-stress wrinkling model. Following [47], we decompose the elastic deformation tensor Fe of a wrinkled biofilm into two parts -- a planar compression F0 e that describes the out-of-plane undulation. Specifically, e and a small wrinkling deformation Fw Fe = Fw e F0 e =(cid:18)I + ∇ku ( ∇kw)T 1 + w,z(cid:19)(cid:18)F0 u,z e,k 0 0 γ0(cid:19) , (S56) ∂2 ∂x0i∂x0j (i, j = 1, 2) and subscript ", z" denoting ∂ where u and w denote, respectively, the in-plane and out-of-plane displacement due to wrinkling. Here ∇k denotes the in-plane derivatives with respect to the spatial coordinates in the deformed flat-film configuration (after F0 e), which we refer to as x0 in the following. In this section, we use simplified notation for derivatives of u and w -- subscript ", ij" denoting ∂z . For small deformations and moderate rotations of thin plates it is known that derivatives scale as ∇kw ∼ u,z ∼ δh and ∇ku ∼ w,z ∼ δ2 h, [17] where δh = h/rb (cid:28) 1 is the aspect ratio of thin biofilm. In our derivations below, we keep all the terms to leading orders in δh. First, according to the plane-stress assumption, we set the normal components of the Cauchy stress to vanish at h), where n ≈ ( ∇kw, 1) denotes the normal to the wrinkled configuration. e + (λb ln Je − Gb)I, we find, to the lowest the lowest order, i.e. σ · n = 0 + O(δ3 Substituting Eq. (S56) into the expression for Cauchy stress Jeσ = GbFeFT Nutrient rich zone ∆rc (mm) .5 .5 11 11 .5 .5 00 00 30 30 10 10 20 20 t (h) 0.4 0.4 0.2 0.2 0 0 0 0 C Aniso. range ∆rα/rb Non-uniform growth Uniform growth 1 1 0.5 0.5 0 0 B Aniso. range ∆rα/rb log 1 1 0.5 0.5 ξ = 20 Anisotropy range ∆rα/rb 1 1 0.5 0.5 r/rb r / rb 101 101 Friction ξ log 102 102 0.1 0 0 0 0 A Anisotropy ασ c order in δh, w,z = − u,z = − ∇kw, λb + 2Gb(γ0)2 tr( ∇ku) − λb λb + Gb(γ0)2 λb + 2Gb(γ0)2k ∇kwk2. 25 (S57) (S58) Eq. (S57) implies that the in-plane displacement takes the form u = um − z0 ∇kw where um denotes the in-plane displacement of the mid-plane of the biofilm. Note that um and w are functions of the two in-plane coordinates x01 and x02, and hence ∇ku = ∇kum − z0 ∇k the biofilm can be calculated from Next, we calculate the total elastic energy of the system in the wrinkled configuration. The elastic energy stored in ∇kw. Πbiofilm =Z ΨdV = J−1 e,0Z ds0Z γ0H/2 −γ0H/2 Ψdz0, (S59) where Je,0 = det(F0 e). We shall leave out the full details of derivation here, but add several comments. All the terms in Ψ (which do not average to zero over z0) can be classified into three categories: (1) terms independent of um, w; (2) terms depending on um, w, but independent of z0; (3) terms proportional to (z0)2. The first terms correspond to the elastic energy stored in the pre-strained configuration, the second terms describe the release of compression energy ∝ H, and the integration of the third terms give rise to the bending energy ∝ H 3. To facilitate the discussion, we focus on incompressible material, i.e. λb → ∞. In this case, the elastic energy Πbiofilm =R Ψkds0, with the areal energy density Ψk given by Ψk = Gbγ0H(fpre−stress + fstrain,lin + fstrain,nlin) + Gb(γ0)3H 3 12 fbending, where (S60) (S61a) (S61b) (S61c) fpre−stress = tr(B0 k) +(cid:0)γ0(cid:1)2 − 3, k · ∇kw − (γ0)2k ∇kwk2(cid:17), (cid:16) ∇kw · B0 k · ( ∇kum)T(cid:17) + (γ0)2( ∇kw)T · ∇kum · ∇kw, tr(cid:16)( ∇k tr(cid:16) ∇kum · B0 (γ0)2(cid:16)tr(cid:0) ∇kum(cid:1)(cid:17)2 ∇kw(cid:17) + ∇kw) · B0 k · ( ∇k 1 2 1 2 1 2 fstrain,lin = tr(cid:16)sym(cid:2)( ∇kum)B0 k(cid:3)(cid:17) − (γ0)2tr(cid:16) ∇kum(cid:17) + + 1 2 (γ0)2(cid:16)tr( ∇kum) + k ∇kwk2(cid:17)2 (γ0)2tr(cid:16) ∇kum · ∇kum(cid:17) + 1 2 + 1 2 fstrain,nlin = ∇kw · ∇k (γ0)2tr(cid:16) ∇k and fbending = (γ0)2htr( ∇k ∇kw)i2 Here, we separated the energy costs of stretching according to the scaling with the aspect ratio δh as fpre−stress ∼ δ0 h, )T denotes h, and fstrain,nlin ∼ δ4 h. We denote by sym[·] the symmetric part of a tensor. B0 fstrain,lin ∼ δ2 k the in-plane components of the left Cauchy-Green deformation tensor. The elastic energy of the deformed elastic substrate can be formally written as Πsub =R 1 2 Ksw2ds0, where Ks is an effective spring constant, and could be a function of the wavenumber kw of the wrinkling profile depending on the substrate model. Some examples are as follows [48, 49]: (1)Winkler foundation: Ks = constant; (2) infinitely large elastic substrate (incompressible): Ks = 2Gskw, where Gs is the shear modulus of the substrate; (3) composite substrate in the tri-layer model (incompressible): Ks = 2+kwhi(Gs/Gi−1), where hi is the thickness of the intermediate layer and Gs/Gi denotes the stiffness ratio between the substrate and the intermediate layer. ∇kw)(cid:17). = F0 e,k (F0 e,k (S61d) 4Gskw 1 2 + One direct effect of forming wrinkles is the relaxation of the compressive stress in the biofilm. To see this, we calculate the change of the in-plane components of the Cauchy stress before and after wrinkling occurs. Substituting B. Stress relaxation Eq. (S56) into the general expression of Cauchy stress σ = (FeFT e ) − pIk, we obtain, to order δ2 h, + (γ0)2 ∇kw(cid:16) ∇kw(cid:17)T − pIk, k ·(cid:16) ∇kum(cid:17)T k + ∇kum · B0 k + B0 k · ∇kw − (γ0)2 ∇kw, σk = B0 σk,z = B0 σz,k = ( ∇kw)T · B0 σzz = (γ0)2 + 2(γ0)2w,z + ( ∇kw)T · B0 k − (γ0)2( ∇kw)T , k · ∇kw − p, 26 (S62a) (S62b) (S62c) (S62d) where stresses are evaluated at the midline of the biofilm. The isotropic stress p can be determined again from the plane-stress assumption σ · n = 0 as before, which yields Taken together, we rewrite the stress in the wrinkling configuration as σk =hB0 k − (γ0)2Iki {z } σ0 k +(cid:20)(γ0)2(cid:16) ∇kw ·(cid:0) ∇kw(cid:1)T p = (γ0)2(cid:2)1 − k ∇kwk2 − 2tr( ∇kum)(cid:3). + k ∇kwk2Ik(cid:17) + ∇kum · B0 k + B0 k ·(cid:16) ∇kum(cid:17)T ∆ σk {z (S63) + 2(γ0)2tr( ∇kum)Ik(cid:21) } (S64) . Clearly, the first term recovers the initial pre-stress applied to the biofilm in the deformed flat configuration, and the second term denotes the stress relaxation due to in-plane stretch (um terms) and out-of-plane displacement (w terms). C. A coarse-grained description of post-wrinkling evolution So far, our model only considers a flat circular biofilm expanding on agar. The solutions are axisymmetric, and therefore the problem is essentially 1D under the plane-stress assumption. On the other hand, the formation of wrinkles breaks the symmetry. To study the effect of mechanical instability, one could use a vectorial displacement field x − x0 = (u, w) to describe the wrinkling profile and perform direct numerical simulations to solve the full growth-instability problem. However, this would be computationally expensive. Here, we propose to study the wrinkling dynamics by means of a coarse-grained model, which maintains axisymmetry and, at the same time, captures the most important features of the wrinkling instability. The idea is to conceptually cut the film into patches, which are small compared to the biofilm radius rb, but larger than the wavelength of wrinkles. The amplitude and shape of wrinkles are assumed to be constant within patches and to vary slowly over distances much larger than the wavelength of wrinkles, which slow variation is described with respect to the coarse- grained coordinates xc ≡ (rc, θc). Below, we start by coarse-graining the elastic energy in Eq. (S60) over such patches. Next, we compute the amplitude and the shape of wrinkles by minimizing the coarse-grained elastic energy. In turn, this information is used to calculate the coarse-grained stresses of the wrinkled configuration given by Eq. (S64). Throughout this section we assume that the expansion of the biofilm is much slower than the mechanical relaxation, and hence, the biofilm morphology evolves quasi-statically. To describe the undulation of wrinkles within each coarse-grained patch, we use an ansatz that was previously proposed for herringbone patterns [44, 46] w(x01, x02) = A cos(kx01 − β1 cos β2kx02) ≈ Ah cos(kx01) + β1 sin(kx01) cos(β2kx02)i, (S65) where x01 and x02 denote the coordinates along the two principal axes within the patch, A is the amplitude of undu- lations, k is the wave number of wrinkles in the x01 direction, and β1/k and β2k denote, respectively, the amplitude and the wave number of the in-plane wiggles (see Fig. 4B in the main text). Here, we assumed that β1 (cid:28) 1 and that k, A, β1, β2 are constant over the coarse-grained patch. Following the analysis in Sec. III, we assume that the principal axes are in the radial and circumferential directions. Thus, we choose x01 = rcθ0 and x02 = r0, and write the planar compression F0 22) assuming incompressibility of the biofilm. The slow variation of wrinkle patterns over the biofilm is given by the amplitude A(xc) and by the orientation/shape parameters β1(xc) and β2(xc), which are obtained via the minimization of the coarse-grained areal elastic energy density. Note that the ansatz for the herringbone pattern in the above Eq. (S65) is a perturbative 22, γ0) as a diagonal matrix, where γ0 = 1/(a0 e ≡ diag(a0 11, a0 11a0 expansion around the stripe state (β1 = 0) that would result from the uniaxial compression in the x01 direction. The assumption β1 (cid:28) 1 is thus violated for the zigzag herringbone pattern resulting from the isotropic compression. Below we discuss how to address this limitation to obtain meaningful results for the isotropic compressive stresses as well. 27 The coarse-graining of the elastic energy stored in a compressed biofilm Ψbiofilm given in Eqs. (S60, S61) is done in two steps. First we average the terms that include only the out-of-plane displacements w(x01, x02) by using the ansatz from Eq. (S65). For example, averaging the terms directly related to w in fstrain,lin yields k 1 8 22)2w2 11)2w2 ,1 + (a0 hf (w) strain,lini = h(a0 11)2 − (γ0)2i(kA)2 + h(a0 w (i = 1, 2), and averaging is defined as h. . .i = (∆s0)−1R . . . dx01dx02, where ∆s0 is the projected area ,2(cid:17)+, (γ0)2*(cid:16)w2 ,2i+ − 11)2 − (γ0)2i(kA)2(β1)2 + h(a0 where w,i = ∂x0i of the coarse-grained biofilm patch on to the horizontal plane. Similar calculations can be done for the rest of the terms in the elastic energy density Ψbiofilm 22)2 − (γ0)2i(kA)2(β1β2)2, h(a0 involving only terms with w. ,1 + w2 (S66) 1 2 1 2 1 4 1 8 = k In the second step of the coarse-graining of the elastic energy Ψbiofilm , we discuss how to deal with terms that include the in-plane displacements um. We start by minimizing the energy density in Eqs. (S60, S61) with respect to um for fixed out-of-plane displacements w. This yields k (γ0)2k ∇kwk2 ,1 + (γ0)2(w,1)(w,11 + w,22) = 0, (S67a) (γ0)2k ∇kwk2 ,2 + (γ0)2(w,2)(w,11 + w,22) = 0. (S67b) 3 2 3 2 22)2um1,22 + 11)2um2,11 + h3(γ0)2 + (a0 h3(γ0)2 + (a0 11)2ium1,11 + 3(γ0)2um2,21 + (a0 22)2ium2,22 + 3(γ0)2um1,12 + (a0 The above partial differential equations for the in-plane-displacements um can be solved with the help of Fourier transforms, which yield 22)2 cos(β2kx02) 2(cid:17) kA2 sin(2kx01) cos(2β2kx02) 2i 2(cid:3) + 3(γ0)2β4 22)2 + 3(γ0)2(cid:3)β2 22)2β2 um1 = 4 + β2 1 (−2 + β2 2 ) β1(γ0)2 2(a0 − − 22)2β2 2 )(γ0)2 β2 1 (1 + β2 11)2 + (a0 11)2 + (a0 11)2 + 3(γ0)2(cid:3) kA2 sin(2kx01) + 8(cid:2)(a0 4(cid:16)(a0 11)2 + 3(γ0)2 +(cid:2)(a0 22)2 + 3(γ0)2(cid:3)β2 β1(γ0)2h(8 + β2 2 )(cid:2)4(a0 2h4(a0 2ih4(a0 11)2 + 12(γ0)2 +(cid:2)(a0 1(cid:0) − 1 + 2β2 22)2 + 3(γ0)2(cid:3) kA2 sin(2β2kx02) 8β2(cid:2)(a0 2(cid:2)2(a0 h4(a0 4h(a0 2(cid:1)(γ0)2 β1β2(γ0)2h8(a0 2ih4(a0 11)2 + 3(γ0)2 +(cid:2)(a0 11)2 + 12(γ0)2 +(cid:2)(a0 22)2 + 3(γ0)2(cid:3)β2 22)2β2 11)2 + (a0 β2 1 β2(1 + β2 11)2 + β2 2 )(γ0)2 − β2 + um2 = (S68a) (S68b) 2i kA2 cos(2kx01) cos(β2kx02), and 22)2 − 3(γ0)2(cid:3)i 2i kA2 cos(2kx01) sin(2β2kx02). 22)2 + 3(γ0)2(cid:3)β2 2i kA2 sin(2kx01) sin(β2kx02) The above expressions for um can then be inserted into the elastic energy density Ψbiofilm in Eqs. (S60, S61) and k 28 (S69) (kA)4β4 1 . (S70) For brevity, we report only results in the small deformation limit, i.e. a0 22 → 1. Detailed ex- pressions for the moderate deformations a0 22 can be found in the Mathematica notebook available on Github (https://github.com/f-chenyi/biofilm-mechanics-theory). In a similar way we average all terms in the elastic energy density Ψbiofilm in Eqs. (S60, S61) by using expressions for the in-plane-displacements um in Eq. (S68) and for the out-of-plane displacements w in Eq. (S65). The resulting averaged energy density of the biofilm is given and discussed in the next section below. 11 → 1, and a0 11, a0 Finally, we also compute the elastic energy of deforming the substrate. As discussed in Refs. [44, 45] the elastic k energy of the substrate can be written as 2 + 79β4 22)2(cid:16)hu2 m2,1i(cid:17) + (a0 64(cid:0)4 + β2 2(cid:1)2 2 + 13β6 2 m2,2i(cid:17) m1,2i + hu2 (kA)4β2 1 + 2 + 7β4 7 − 4β2 2 512 (kA)4 + 2 1 3 → Dtr( ∇kum)E = hum1,1 + um2,2i = 0, and k · ∇kuT m1,1i + hu2 m)E = (a0 11)2(cid:16)hu2 32 + 176β2 Dtr( ∇kum · B0 averaged over the patch as described above. For example, (S71) (S72) (S73) Note that the results differ between different substrate models. For the Winkler foundation with elastic constant Ks one obtains hΨsub 1 /2), while for an infinitely thick incompressible elastic substrate with shear i = 1 k k (cid:11) = 1 modulus Gs we find(cid:10)Ψsub 2(cid:1). Here, we treat the agar as an infinitely thick elastic substrate. 2 GskA2(cid:0)1+ β2 p1 + β2 4 KsA2(1 + β2 12 D. Summary and discussion i that includes αα = (a0 2 +23β4 2 )2 2 )/32 are functions of β2 only. Here, we assumed that stresses are compressive (σ0 where σ0 11β4 omitted the constant terms corresponding to the elastic energy stored in the pre-stressed flat configuration. and fb(β2) = (11 + 8β2 2 + 22 ≤ 0). Note that we have Next we comment on the form of the energy density in Eq. (S73). By considering only terms to the order (β1)0, 11, σ0 2 +2β6 4(4+β2 2 , k 1 8 1 2 1 + 1 β2 = − 11 + σ0 11β2 11 → 1, and a0 σ0 12 1 + 22 → 1), the total averaged free energy density hΨtotal 1!(kA)4 } 1 + fa(β2)β2 1q1 + β2 {z αα)2 − (γ0)2 denotes the in-plane pre-stress, and fa(β2) = 64+64β2 2!(kA)2 + {z 2Gb(kH)(cid:16)1 + σ0 22β2 2 )2!(kA)2 } 2(cid:17)(kA)2 } biofilm stretching energy 1 + fb(β2)β4 biofilm bending energy β2 1 (1 + β2 substrate energy {z (kH)2 β2 Gs 1 4 1 2 1 2 1 2 + + hΨtotal GbH k i In the small deformation limit (a0 the energy density of the deformed biofilm and the substrate can be written as [44] we recover the energy density (kA)4 !(kA)2 + 12 i } 22 ≈ 0). [48] The minimum of fσ(k) determines the (S74) 1 8 fσ(k) {z 11 +h Gs σ0 2Gb(kH) (kH)2 + 1 4 = − hΨtotal GbH k i that corresponds to 1D wrinkles for uniaxial compression (σ0 11 < 0, σ0 Ks(k) w(k1, k2)2, 1 2 s0Z dk1dk2 1 ∆ hΨsub k i = where k =(cid:0)k2 1 + k2 2(cid:1)1/2 and we defined the Fourier transform w(k1, k2) = 1 2πZ∆s0 dx01dx02 w(x01, x02) exph − i(k1x01 + k2x02)i. critical stress for the formation of wrinkles, i.e. σc = 4 mink fσ(k) = (3Gs/Gb)2/3, and the corresponding argument kw = H−1(3Gs/Gb)1/3 determines the wavelength of wrinkles λ = 2π/kw. Consequently, when σ0 11 < σc, the coefficient of (kA)2 term in Eq. (S74) is positive, which is minimized for the amplitude A = 0, i.e., a flat biofilm. The 11 surpasses the critical stress bifurcation from the planar to the wrinkled state occurs when the compressive stress σ0 σc, which, in the small deformation limit, corresponds to a phase boundary 0 22/2 = σc/4, where we expressed compressions a0 Going beyond the leading order in β1 and minimizing the Landau-Ginzburg-type free energy in Eq. (S73) with ii. The amplitude of wrinkles grows as kwA =pσ0 ii in terms of strains 0 11 − σc. 11 +0 ii = 1 + 0 respect to the amplitude A and the wave number k leads to 29 , 2 ) 1 (1 + β2 Gb(1 + 1 2 β2 2 β2 k∗w = H−1 3Gs(1 + 1 (k∗wA∗)2 ≈ (cid:16)σ0 2 )2)!1/3 1p1 + β2 1(cid:17) 11 − σc + ∆β2 1 and introduced ∆ = σ0 2(cid:1)/6. The bifurcation from the planar to the wrinkled state occurs when σ0 1 + fb(β2)β4 1 1 + fa(β2)β2 . In Eq. (S75b) we expanded the numerator to the second order term β2 In this case the minimized energy becomes 2 + β4 2 + 2p1 + β2 σc(cid:0)1 + 2β2 (S75a) (S75b) 11/2 + σ0 22β2 11 + ∆β2 2 /2 − 1 > σc. k i hΨtotal GbH (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)A=A∗,k=k∗w (σ0 11 − σc + ∆ β2 1 )2 1 + faβ2 1 + fbβ4 1 . = − (S76) We then solve for the optimal value of β1 by minimizing the free energy density in the above equation, which yields solutions β∗1 = 0 and (β∗1 )2 = (σ0 11 − σc)fa(β2) − 2 ∆ fa(β2) ∆ − 2(σ0 11 − σc)fb(β2) ≈ 8(2σ0 11 − σc) 22 − σ0 3(σ0 11 − σc) β2 2 , (S77) 22 > σ0 11/2 + σc/2. This corresponds to the phase boundary 0 11 > σc), the secondary where the above result was expanded to the lowest order in β2. In the relevant regime (σ0 bifurcation from the stripe pattern (β∗1 = 0) to the zigzag pattern (β∗1 > 0) happens when the compressive stress σ0 22 22 = σc/6, which is is sufficiently large, i.e., when σ0 consistent with previous studies on herringbone patterns [44]. A schematic phase diagram is presented in Fig. S11. In principle, the expression for β∗1 in Eq. (S77) can then be substituted into Eq. (S76) for the minimized energy density, which can then be expressed as a function of β2. As shown in Ref. [44], the optimal value β∗2 that minimizes the resulting energy density scales as 1/2 near the onset of the secondary bifurcation, where = 0 22− σc/6. However, as shown below this would result in anisotropic coarse grained stresses h σki in the wrinkled configuration in Eq. (S64) in response to isotropic compression σ0 22. To overcome this issue, we do not obtain the value of parameter β2 via the energy minimization, but rather we fix the value to β2, such that the coarse grained stresses h σki in the wrinkled configuration remain isotropic in response to isotropic compression. From Eq. (S64) for the stresses σk in the wrinkled configuration, we see that the in-plane displacements um appear only in linear terms that average out to zero in the coarse-grained description. The out-of-plane displacements w appear in quadratic terms ∇kw · ( ∇kw)T and k ∇kwk2I that result in the anisotropic and isotropic stress relaxation, respectively. Their averages over the coarse-grained patches read 11 = σ0 (S78a) (S78b) (S79a) (β1β2)2(kA)2! , (β1β2)2(kA)2iIk. 14 0 + β2 1 β2 2 4 (cid:17), 1 4 1 2 β2 0 12 β2 1 2 2 (kA)2(cid:16)1 + β2 (cid:17) (cid:17) + (kA)2(cid:16)1 + 11 + (kA)2(cid:16)1 + h ∇kw · ( ∇kw)Ti = 1 hk ∇kwk2Iki =h 1 2 hσ11i = −σ0 Therefore, the coarse-grained stresses in the wrinkled configuration become 30 θθ = 1−(F0 e,k )θθ, and the radial compressive strain, measured by 0 Fig. S11: Mechanical characterization of biofilm wrinkling instability. (A) Schematic phase diagram of the coarse-grained model for biofilm wrinkling instability. Axes correspond to the circumferential compressive strain, )rr, in the pre-stressed measured by 0 θθ, the biofilm transitions from the planar state (no state. For the lower portion of the phase diagram, i.e. 0 θθ surpasses the critical stress σc, which, in linear elasticity, corresponds to a pattern) to a wrinkled state when σ0 rr/2 = σc/4 (black solid line). This primary instability results in a radial stripe pattern. The phase boundary 0 rr > σc/6 (linear elasticity phase boundary, secondary transition from radial stripes to radial zigzags occurs when 0 black dotted line). Similarly, for the upper portion of the phase diagram, the modeled biofilm will transition from the planar state to a wrinkled state with concentric rings then to a wrinkled state with distorted rings. The black and magenta arrows illustrate the evolution of the center and the edge of the modeled biofilm, respectively. (B) Magnitude of the compressive stress σθθ at the center (black, top) and edge (magenta, bottom) of the modeled biofilm versus time t, plotted for the chemo-mechanical model with (solid curves) and without (dotted curves) permitting wrinkling. Simulation parameters as in Fig. 4C. rr = 1−(F0 e,k rr < 0 θθ+0 (S79b) + β2 1 β2 2(cid:17). 1 2 β2 (kA)2(cid:16)1 + 1 2 hσ22i = −σ0 22 + 11 = σ0 22), we obtain the condition β2 By requiring that the coarse-grained stresses in the wrinkled state remain equal, i.e. hσ11i = hσ22i, in response to the isotropic compression (σ0 2 − 1). Comparing this condition with Eq. (S77) for β∗1 that minimizes the energy density, we obtain a self-consistent equation 2/(( β2)2 − 1) ≈ 8/3( β2)2 with solution β2 ≈p3/2. Moreover, we observe that by inserting the optimal values of k∗wA∗, β∗1, and β2 into Eq. (S79) for the coarse-grained stresses in the wrinkled state, we find that hσ11i ≡ −σc after the primary bifurcation, and hσ22i ≡ −σc after the secondary bifurcation. Thus, the wrinkling instability prevents the compressive stresses from rising above the critical stress by releasing the compressing mechanical strain via the out-of-plane deformations (see Fig. S11). 1 = 2/(β2 The results presented above can be represented in terms of the dimensionless amplitude of wrinkles A ≡ k∗wA and in terms of the dimensionless shape parameter S2 ≡ (β1 β2)2/6, which is S = 0 for striped wrinkles and S > 0 for zigzag wrinkles ( S = 1 for zigzag wrinkles due to isotropic compression). The energy density in Eq. (S73) can be rewritten in terms of the two parameters A and S. For the optimal wavelength of wrinkles k∗w in Eq. (S75a), the energy density to the leading order becomes (S80) (cid:16)1 + (6b + 3) S2(cid:17) A4, 1 2 22 − σc) S2(cid:17) A2 + 11 − σc)(1 + 3b S2) + 3(σ0 k i 4hΨtotal GbH ≈ −(cid:16)(σ0 where b = ( β2)−2 = 2/3 and σc = (3Gs/Gb)2/3 denotes the normalized critical stress for wrinkling as discussed above. Similarly, we rewrite the stress relaxation due to wrinkles in Eq. (S79) in terms of the two parameters S and A as hσ11i = −σ0 11 + A2h1 + (3b + 3/2) S2i, and hσ22i = −σ0 22 + A2h1/2 + (3b/2 + 3) S2i. The above Eqs. (S80) and (S81) correspond to Eqs. [4] and [5] quoted in the main text, respectively. (S81) Using the coarse-grained model discussed above, the coarse-grained force-balance equation after the wrinkling instability occurs is given by ∇k · h Nki = ξq(k∗wA∗, β∗1 , β2)vk, where q(k∗wA∗, β∗1 , β2) = hk(−w,1,−w,2, 1)ki ≈ 1 + ξ = 4 Without wrinkling With wrinkling σc~ σc~ 40 40 40 t (h) 60 60 60 Center Edge 101 101 101 100 100 100 101 101 101 100 100 100 20 20 20 B Stress σθθ ~ Distorted rings Radial zigzags Radial stripes No pattern rings Concentric Radial compression A Circumferential compression 31 Fig. S12: Formation of wrinkles slows biofilm expansion and affects the subsequent formation of the biofilm wrinkle patterns. (A, B) Kymograph representations of (A) biofilm wrinkling pattern formation ( A ≡ k∗wA is the dimensionless amplitude of wrinkles and S ≡q(β1 β2)2/6 is the dimensionless shape parameter, see text for σ, right) between the chemo-mechanical model with and without wrinkling (denoted by −w), i.e., ∆σ0 θθ + σ0 details), and (B) differences in circumferential pre-stress (denoted by ∆σ0 θθ, left) and in pre-stress anisotropy (denoted by ∆α0 θθ ≡ rr). σ0 θθ − σ0 Axes correspond to time t, and the radial coordinate normalized by the biofilm radius r/rb. Color code of A is the same as in Fig. 4C. The color scale in B left shows (1 + ∆σθθ) on a logarithmic scale for visualization of both small and large values of ∆σθθ. (C) The radial coordinate r of a modeled biofilm with wrinkles, plotted against that of a modeled biofilm when wrinkling is not permitted (denoted by r−w), at the designated time (solid curve). The dashed black line r = r−w is provided as a guide to the eye. Simulation parameters as in Fig. 4C. σ,−w, where the pre-stress anisotropy is defined as α0 θθ,−w and ∆α0 θθ − σ0 σ − α0 σ = (σ0 rr)/(σ0 σ = α0 k (k∗wA∗)2(cid:0)1 + 1 2 (β∗1 )2(1 + ( β2)2)(cid:1) accounts for the increase of the contact area between the wrinkled biofilm and the substrate. The average stress is defined as h Nki = (∆s0)−1R σkdz ds0, where the integral R dz is taken along the z direction in the current deformed configuration and the integral R ds0 is over the reference patch coordinates. Note that dz = dzn/ cos θnz and ds0 = ds cos θnz where zn denotes the coordinate along the normal to the biofilm surface, ds denotes the area of the biofilm patch whose projected area on the x − y plane is ds0, and θnz denotes the angle between the normal to the wrinkled biofilm surface and the z direction. Because dzds0 = dzndsn, we can rewrite h Nki as h Nki = (∆s0)−1R σkdznds = (∆s0)−1R σkdz0ds0, where we used the incompressibility of the biofilm, dznds = dz0ds0, to convert the integration to coordinates dz0 and ds0 in the deformed flat-film configuration. Thus we obtain h Nki = γ0h σki, where the average stress h σki is given by Eq. (S79). In summary, the force-balance equation for the biofilm can be written as ∇k · (γ0h σk)i = ξq(k∗wA∗, β∗1 , β2)vk, 4 (k∗wA∗)2(cid:0)1+ 1 2 (β∗1 )2(1+( β2)2)(cid:1) = 1+ 1 +∆ σk(k∗wA∗, β∗1 , β2) where q = 1+ 1 is given by Eq. (S79) and (S81) with the pre-stress given by σ0 )T −(γ0)2Ik. Before the wrinkling instability = F0 k e,k , and the force-balance Eq. (S82) reduces to Eq. (S15) occurs, the amplitude is A∗ = 0. Thus q = 1, and h σki = σ0 k in Sec. II. In order to simulate the evolution of the wrinkled biofilm, we solve the governing equations (S13) -- (S16) defined in Sec. II, where the force-balance equation is replaced with Eq. (S82). Note that the evolution of the wrinkle pattern is fully coupled to the dynamics of nutrients and biofilm expansion. A2(cid:2)1+3(b+1) S2(cid:3), and the stress h σki = σ0 (F0 e,k (S82) 14 4 According to our model, compared to the case where the wrinkling instability is prevented (force-balance Eq. (S15)), the expansion of the wrinkled biofilm is slowed, the pre-stress anisotropy is reduced, and the magnitude of compressive true stress is reduced while that of the pre-stress is increased (Fig. S11 and S12). Thus, our model suggests that wrinkling due to a growth-induced mechanical instability feeds back and further influences biofilm expansion and pattern formation by modifying the distribution of internal stress. t = 60 h 10 10 0 0 without wrinkling 20 20 Radial coordinate r-w (mm) 20 20 10 10 0 0 with wrinkling r (mm) Radial coordinate C 0.5 0 r / rb 0.5 1 1 60 60 40 40 t (h) 1 1 ~ A 0 0 1 0 1 0.5 ~ S 0 0 Δασ 0 1 0.5 -0.8 0 0.5 log10(1+Δσθθ)~0 0 0.8 1 20 20 B 1 0.5 0.5 0 r / rb 0.5 1 1 60 60 Stripes Zigzags 40 40 t (h) 0 No pattern 0.5 1 20 20 A 32 Fig. S13: Schematic illustration for elastoplastic constitutive models of biofilm materials. Schematics of scalar (left) and tensorial (middle and right) stress-strain diagrams for (A) deformation theory of elasto-plasticity and (B) von Mises theory of elasto-plasticity. For both A and B, the initial yielding surface is illustrated by red in the axis/plane of deviatoric strain (denoted by E or E) and deviatoric stress (denoted by σe or s). Gray dots with designated numbers represent three successive states during a hypothetical loading/unloading process (black curves with arrows). In B, von Mises theory evolves the yielding surface (illustrated by blue dashed circles) and is able to capture hysteresis due to plasticity upon loading and unloading. VII. SUPPLEMENTARY DISCUSSION: BIOFILM MATERIAL PROPERTIES We have focused on the question of how stresses accumulate spatiotemporally in a growing biofilm. For simplicity, we used a hyperelastic model to describe the biofilm material. However, our previous measurements show that the material properties of experimental biofilms are much more complicated, and very closely resemble those of a hydrogel [11]. For example, the biofilm material starts to yield with a dramatic decrease in elastic modulus Gb above a critical strain Y ≈ 10%. The internal stress predicted by our simple elastic model is larger than the yield stress. Thus, we suspect that some yielding of the material likely occurs as the biofilm grows. To explore the effect of yielding on biofilm expansion dynamics, we implement some of the simplest plastic material models to study how the stresses build up before mechanical instability occurs. We adopt the von Mises yield criterion which suggests that yielding of material begins when the second invariant of the deviatoric stress s, i.e. σe = ( 3 2s : s)1/2, reaches a critical value. Incompressible hyperelastic material: We start by rewriting the mechanical stress as the sum of a deviatoric e(cid:17)I be the deviatoric deformation tensor. The dimensionless stress and a hydrostatic stress. Let E = FeFT stress σ can be reorganized as σ = s− pI where s = E denotes the deviatoric stress, and p is the hydrostatic pressure p = γ2 − tr(FeFT Deformation theory of plasticity: The deformation theory of plasticity attempts to develop a one-to-one stress- strain relationship (Fig. S13) [50]. One common power-law deformation theory of plasticity assumes that the total deviatoric deformation can be decomposed into an elastic part and a plastic part, i.e. E = Ee + Ep, where the elastic strain is Ee = s and the plastic strain depends nonlinearly on s, 3 tr(cid:16)FeFT e − 1 e )/3. Ep =(cid:16) σe σY,0(cid:17)n s, (S83) where σY,0 denotes the yield stress and n > 1 is the power-law exponent. When σe < σY,0, the elastic term dominates, and the strain-stress relation recovers the elastic model, i.e. E ≈ Ee = s; when σe > σY,0 the nonlinear term dominates and the material yields. We cannot invert analytically the above equations, i.e. calculating s given the deformation E. However, we know E2 s2 3 1 2 3 1 2 εE εY E1 Deviatoric Strain 𝞼Y Deviatoric Stress s1 E2 s2 3 1 2 3 2 1 εE εY,0 E1 Deviatoric Strain 𝞼Y,0 𝞼Y Deviatoric Stress s1 2 3 1 εY 2 3 1 εY,0 A B 𝞼e 𝞼Y 𝞼e 𝞼Y 𝞼Y,0 33 Fig. S14: Comparison of the modeled biofilm expansion dynamics using an elastic constitutive model versus an elasto-plastic constitutive model. The radial velocity field v (left), the vertical stretch γ (center left), the magnitude of the circumferential stress σθθ (center right), and stress anisotropy ασ = (σθθ−σrr)/(σθθ +σrr) (right), versus radial coordinate r at various times, plotted for the chemo-mechanical model using (A) an elastic constitutive model (dimensionless friction parameter ξ = 4), (B) a deformation elasto-plastic constitutive model (ξ = 4, power law exponent n = 2, dimensionless yielding stress σY = 0.2), (C) a von Mises elasto-plastic constitutive model (ξ = 4, σY = 0.2, dimensionless plastic modulus G0 = 0.1), and (D) an elastic constitutive model but with larger friction (ξ = 30). Note that (B-D) have similar kinematic behaviors, but models using the elasto-plastic constitutive relation lead to stresses with smaller magnitudes, compared to the elastic counterpart. that s and E must be co-linear, i.e. s = cE and hence s : E = cE : E. Therefore, we obtain (S84) (S85) (S86) (S87) =r 2 3 E : E ≡ E. : E = e s σ cE : E 2 E : E cq 3 : E, and hence σe . E : E = s : E +(cid:16) σe σY,0(cid:17)n s E = σe +(cid:16) σe σY,0(cid:17)n 3 2 c22 E, 3 2 e = s : s = c2E : E = σ2 2 3 We can solve the equation above numerically for σe given the deformation (known E). Finally, since Substituting this relation into the expression for E, we derive the co-linear factor is c = 2σe/3E, and the stress s = cE follows. The deformation theory of plasticity essentially describes a material with a nonlinear constitutive relation and a "softened" tangent modulus G0b/Gb ≡ G0 < 1 after yielding. We note that the kinematic behavior of such a material with friction parameter ξ resembles that of a pure elastic material with a larger ξ0 (Fig. S14), but the stress (normalized by the linear elasticity modulus Gb) is significantly smaller. A simple explanation for this effect is that the biofilm enters the yielding regime in the early stage of expansion, and thus the dynamics is governed essentially by the effective friction parameter ξ0 = η(L0/τ0) G0b(H/L0), which is normalized by the effective tangent modulus G0b. Linear isotropic harderning: The deformation theory considers material rules without hysteresis. It appears, however, that more complicated rules are necessary especially for the case of unloading. We now discuss a von Mises- type plasticity model. The stress-strain relation after yielding is essentially given by the evolution of the yield surface, 34 σe(s) − σY(E) = 0, in which σY denotes the current yield stress, initially equal to the material parameter σY,0. Our measurements show that biofilm material starts to yield at about 10% strain in a simple shear experiment [11], from which we can estimate σY,0 = 0.1735 ≈ 0.2. Measuring the stress-strain curve under monotonic loading in the experiments also show that biofilm material is strain hardening,6 in the sense that the yield stress σY increases beyond the yielding point. The von Mises plasticity model also assumes that the material behaves elastically within the yield surface, which leads to hysteresis under cyclic loading (Fig. S13). The evolution of the yield stress σY during plastic flow is described by an additional equation, the hardening law. In the simplest case, the hardening law is isotropic and linear: σY = σY,0 + G0p where plastic modulus G0 < 1 is constant and p is a scalar measure of the plastic strain. The simplest choice of p might seem to be the norm of the plastic strain tensor, i.e. p = (Ep : Ep)1/2, but this variable does not always increase during plastic flow due to the tensorial nature of Ep. Therefore, we use the cumulative plastic strain ¯p, defined by the rate equation, p = ( 3 ¯ 2 Ep : Ep)1/2 to characterize plastic hardening [51]. Consider an infinitesimal (un)loading dE when the current deviatoric stress is s, then the elastic response se = s+dE will occur if σe(se) < σY. If the von Mises stress σe(se) surpasses the current yield stress, two things happen: (1) the yield surface evolves; and (2) the end stress sp falls on the new yield surface with yield stress σY + dσY following the hardening law. Recall the decomposition dE = dEe + dEp, the end stress is corrected by sp = se − dEp. We further assume that the plastic flow has the same direction as the deviatoric stress, i.e. dEp is co-linear with sp, and thus sp is also co-linear with se. Let dEp = (dα)se, then we obtain σe(sp) = (1 − dα)σe(se) hardening law =======⇒ σY + G0d¯p = σY + dα G0 σe(se), and dα = 1 − σY/σe(se) . 1 + G0 (S88) (S89) The stress evolution is hence given by sp = (1 − dα)se, or equivalently ds = [ G0/(1 + G0)]dE. As expected, the magnitude of the compressive stress σθθ is notably smaller than that of an elastic model. Fur- thermore, simulations of such a material model with loading-unloading hysteresis show that the stresses quickly drop when growth is stopped, while the deformation is maintained (data not shown). Thus, plastic deformation may also explain why, in the experiments, the pattern does not seem to change after we stop biofilm growth (for example by putting biofilms in a 4◦C environment). So far we have only discussed how elasto-plasticity of the biofilm material might change the stress state in a non-uniformly growing biofilm before mechanical instability happen. More generally, reorganization and yielding of growing biological materials are commonly observed during morphogenesis, such as in plants [52], fruit flies [53, 54] and brain tissues [55]. Thus, the effects of viscoelasticity [56] and elastoplasticity (Figs. S13 and S14) of biofilms on their morphological development will be an important topic for future studies. Moreover, V. cholerae biofilms are soft (Gb ∼1 kPa) and thin (h ∼100 µm). These features make the interfacial energies relevant in a biofilm system, which affects how mechanical instabilities happen. Note that in conventional abiotic film-substrate systems, surface energy is negligible. For example, for metal films, the surface energy is γ ∼ 1 J/m2 but the shear modulus is G ∼ 50 GPa, and thus the elastocapillary length, i.e. the characteristic film thickness at which surface tension could balance bending, is lec = γ/G ∼ 0.01nm; for plastics like polyethylene, γ ∼ 50 mJ/m2, G ∼ 100 MPa, and lec is about 0.5nm. However, in these systems the film thickness is at least 10 nm, much larger then lec. For V. cholerae biofilms, the interfacial energy between a WT biofilm and water γbw was determined to be around 50 -- 60 mJ/m2 and the biofilm-air interfacial energy is γba ≈30 -- 40 mJ/m2 [11]. Thus, biofilms are "sticky" in the sense that the elastocapillary length (determined by its material properties) is comparable to its natural thickness (γ/Gbh ≈ 0.5 for biofilms). In fact, we previously reported that biofilms can form different types of mechanical instability patterns depending on the interfacial energies [12], and researchers have also started to look at how surface energies affect other type of instabilities [57]. Understanding the effects of interfacial energies on the pattern formation dynamics will be an important future direction. [1] E. K. Rodriguez, A. Hoger, and A. D. McCulloch, "Stress-dependent finite growth in soft elastic tissues," J Biomech, [2] A. Goriely and M. B. Amar, "Differential growth and instability in elastic shells," Phys Rev Lett, vol. 94, no. 19, p. 198103, vol. 27, no. 4, pp. 455 -- 467, 1994. 2005. 6 Not to be confused with the usage "softened tangent modulus". See Fig. S13 for illustrations. 35 [3] J. Dervaux and M. B. Amar, "Morphogenesis of growing soft tissues," Phys Rev Lett, vol. 101, no. 6, p. 068101, 2008. [4] R. de Rooij and E. Kuhl, "Constitutive modeling of brain tissue: current perspectives," Appl Mech Rev, vol. 68, no. 1, p. 010801, 2016. [5] M. Genet, M. Rausch, L. C. Lee, S. Choy, X. Zhao, G. S. Kassab, S. Kozerke, J. M. Guccione, and E. Kuhl, "Heterogeneous growth-induced prestrain in the heart," J Biomech, vol. 48, no. 10, pp. 2080 -- 2089, 2015. [6] M. Uyttewaal, A. Burian, K. Alim, B. Landrein, D. Borowska-Wykręt, A. Dedieu, A. Peaucelle, M. Ludynia, J. Traas, A. Boudaoud, et al., "Mechanical stress acts via katanin to amplify differences in growth rate between adjacent cells in Arabidopsis," Cell, vol. 149, no. 2, pp. 439 -- 451, 2012. [7] T. Matsumoto, M. Kawakami, K. Kuribayashi, T. Takenaka, and T. Tamaki, "Cyclic mechanical stretch stress increases the growth rate and collagen synthesis of nucleus pulposus cells in vitro," Spine, vol. 24, no. 4, pp. 315 -- 319, 1999. [8] R. W. Ogden, Nonlinear Elastic Deformations. Courier Corporation, 1997. [9] J. Magnus and H. Neudecker, Matrix Differential Calculus with Applications in Statistics and Econometrics. Wiley, 1999. [10] L. Landau, E. Lifshitz, and J. Sykes, Theory of Elasticity. Course of theoretical physics, Pergamon Press, 1989. [11] J. Yan, A. Moreau, S. Khodaparast, A. Perazzo, J. Feng, C. Fei, S. Mao, S. Mukherjee, A. Košmrlj, N. S. Wingreen, et al., "Bacterial biofilm material properties enable removal and transfer by capillary peeling," Adv Mater, vol. 30, no. 46, p. 1804153, 2018. [12] J. Yan, C. Fei, S. Mao, A. Moreau, N. S. Wingreen, A. Košmrlj, H. A. Stone, and B. L. Bassler, "Mechanical instability and interfacial energy drive biofilm morphogenesis," eLife, vol. 8, p. e43920, 2019. [13] J. Monod, "The growth of bacterial cultures," Annu Rev Microbiol, vol. 3, no. 1, pp. 371 -- 394, 1949. [14] P. S. Stewart and M. J. Franklin, "Physiological heterogeneity in biofilms," Nat Rev Microbiol, vol. 6, no. 3, pp. 199 -- 210, 2008. 2010. [15] J. W. Costerton, Z. Lewandowski, D. E. Caldwell, D. R. Korber, and H. M. Lappin-Scott, "Microbial biofilms," Annu Rev Microbiol, vol. 49, no. 1, pp. 711 -- 745, 1995. [16] J. N. Reddy, Theory and Analysis of Elastic Plates and Shells. CRC press, 2006. [17] B. Audoly and Y. Pomeau, Elasticity and Geometry: From Hair Curls to the Non-linear Response of Shells. OUP Oxford, [18] S. Budday, C. Raybaud, and E. Kuhl, "A mechanical model predicts morphological abnormalities in the developing human brain," Sci Rep, vol. 4, p. 5644, 2014. [19] A. E. Shyer, T. Tallinen, N. L. Nerurkar, Z. Wei, E. S. Gil, D. L. Kaplan, C. J. Tabin, and L. Mahadevan, "Villification: How the gut gets its villi," Science, vol. 342, no. 6155, pp. 212 -- 218, 2013. [20] T. Tallinen, J. Y. Chung, F. Rousseau, N. Girard, J. Lefèvre, and L. Mahadevan, "On the growth and form of cortical convolutions," Nat Phys, vol. 12, no. 6, p. 588, 2016. [21] S. Walcott and S. X. Sun, "A mechanical model of actin stress fiber formation and substrate elasticity sensing in adherent cells," Proc Natl Acad Sci USA, vol. 107, no. 17, pp. 7757 -- 7762, 2010. [22] P. Sens, "Rigidity sensing by stochastic sliding friction," EPL, vol. 104, no. 3, p. 38003, 2013. [23] E. de Souza Neto, D. Peric, and D. Owen, Computational Methods for Plasticity: Theory and Applications. Wiley, 2011. [24] P. Yu, S. Hu, L. Chen, and Q. Du, "An iterative-perturbation scheme for treating inhomogeneous elasticity in phase-field [25] B. A. Camley, Y. Zhao, B. Li, H. Levine, and W.-J. Rappel, "Periodic migration in a physical model of cells on micropat- models," J Comput Phys, vol. 208, no. 1, pp. 34 -- 50, 2005. terns," Phys Rev Lett, vol. 111, no. 15, p. 158102, 2013. [26] D. Shao, W.-J. Rappel, and H. Levine, "Computational model for cell morphodynamics," Phys Rev Lett, vol. 105, no. 10, p. 108104, 2010. Phys Rev E, vol. 68, no. 3, p. 037702, 2003. vol. 8, p. e42697, 2019. [27] J. Kockelkoren, H. Levine, and W.-J. Rappel, "Computational approach for modeling intra-and extracellular dynamics," [28] S. Srinivasan, C. N. Kaplan, and L. Mahadevan, "A multiphase theory for spreading microbial swarms and films," eLife, [29] V. Berk, J. C. Fong, G. T. Dempsey, O. N. Develioglu, X. Zhuang, J. Liphardt, F. H. Yildiz, and S. Chu, "Molecular architecture and assembly principles of Vibrio cholerae biofilms," Science, vol. 337, no. 6091, pp. 236 -- 239, 2012. [30] W. Hong, Z. Liu, and Z. Suo, "Inhomogeneous swelling of a gel in equilibrium with a solvent and mechanical load," Int J [31] W. Hong, X. Zhao, J. Zhou, and Z. Suo, "A theory of coupled diffusion and large deformation in polymeric gels," J Mech Solids Struct, vol. 46, no. 17, pp. 3282 -- 3289, 2009. Phys Solids, vol. 56, no. 5, pp. 1779 -- 1793, 2008. [32] M. Rubinstein and R. H. Colby, Polymer Physics, vol. 23. Oxford University Press New York, 2003. [33] A. Güner, "Unperturbed dimensions and the theta temperature of dextran in aqueous solutions," J Appl Polym, vol. 72, [34] E. Antoniou and M. Tsianou, "Solution properties of dextran in water and in formamide," Journal of Applied Polymer no. 7, pp. 871 -- 876, 1999. Science, vol. 125, no. 3, pp. 1681 -- 1692, 2012. Phys, vol. 11, no. 11, pp. 512 -- 520, 1943. no. 11, pp. 521 -- 526, 1943. pp. 222 -- 227, 2001. [35] P. J. Flory and J. Rehner Jr, "Statistical mechanics of cross-linked polymer networks I. Rubberlike elasticity," J Chem [36] P. J. Flory and J. Rehner Jr, "Statistical mechanics of cross-linked polymer networks II. Swelling," J Chem Phys, vol. 11, [37] I. W. Sutherland, "The biofilm matrix -- an immobilized but dynamic microbial environment," Trends Microbiol, vol. 9, no. 5, [38] J. Yan, C. D. Nadell, H. A. Stone, N. S. Wingreen, and B. L. Bassler, "Extracellular-matrix-mediated osmotic pressure 36 drives Vibrio cholerae biofilm expansion and cheater exclusion," Nat Commun, vol. 8, no. 1, p. 327, 2017. [39] Y. Tian, N. Pesika, H. Zeng, K. Rosenberg, B. Zhao, P. McGuiggan, K. Autumn, and J. Israelachvili, "Adhesion and friction in gecko toe attachment and detachment," Proc Natl Acad Sci USA, vol. 103, no. 51, pp. 19320 -- 19325, 2006. [40] H. Kweon, S. Yiacoumi, and C. Tsouris, "Friction and adhesion forces of Bacillus thuringiensis spores on planar surfaces in atmospheric systems," Langmuir, vol. 27, no. 24, pp. 14975 -- 14981, 2011. [41] J. C. Fong and F. H. Yildiz, "The rbmBCDEF gene cluster modulates development of rugose colony morphology and biofilm formation in Vibrio cholerae," J Bacteriol, vol. 189, no. 6, pp. 2319 -- 2330, 2007. [42] P.-G. De Gennes, Scaling Concepts in Polymer Physics. Cornell university press, 1979. [43] J. Liu, A. Prindle, J. Humphries, M. Gabalda-Sagarra, M. Asally, D. L. Dong-yeon, S. Ly, J. Garcia-Ojalvo, and G. M. Süel, "Metabolic co-dependence gives rise to collective oscillations within biofilms," Nature, vol. 523, no. 7562, pp. 550 -- 554, 2015. [44] B. Audoly and A. Boudaoud, "Buckling of a stiff film bound to a compliant substrate - Part I: Formulation, linear stability of cylindrical patterns, secondary bifurcations," J Mech Phys Solids, vol. 56, no. 7, pp. 2401 -- 2421, 2008. [45] B. Audoly and A. Boudaoud, "Buckling of a stiff film bound to a compliant substrate - Part II: A global scenario for the formation of herringbone pattern," J Mech Phys Solids, vol. 56, no. 7, pp. 2422 -- 2443, 2008. [46] S. Cai, D. Breid, A. J. Crosby, Z. Suo, and J. W. Hutchinson, "Periodic patterns and energy states of buckled films on compliant substrates," J Mech Phys Solids, vol. 59, no. 5, pp. 1094 -- 1114, 2011. [47] M. Holland, B. Li, X. Feng, and E. Kuhl, "Instabilities of soft films on compliant substrates," J Mech Phys Solids, vol. 98, [48] Z. Huang, W. Hong, and Z. Suo, "Nonlinear analyses of wrinkles in a film bonded to a compliant substrate," J Mech Phys pp. 350 -- 365, 2017. Solids, vol. 53, no. 9, pp. 2101 -- 2118, 2005. [49] E. Lejeune, A. Javili, J. Weickenmeier, E. Kuhl, and C. Linder, "Tri-layer wrinkling as a mechanism for anchoring center initiation in the developing cerebellum," Soft Matter, vol. 12, no. 25, pp. 5613 -- 5620, 2016. [50] V. A. Lubarda, "Deformation theory of plasticity revisited," Proc. Mont. Acad. Sci. Arts, vol. 13, pp. 117 -- 143, 2000. [51] M. Jirasek and Z. P. Bazant, Inelastic Analysis of Structures. John Wiley & Sons, 2002. [52] J. Dumais, S. L. Shaw, C. R. Steele, S. R. Long, and P. M. Ray, "An anisotropic-viscoplastic model of plant cell morpho- genesis by tip growth," Int J Dev Biol, vol. 50, pp. 209 -- 222, 2006. [53] B. He, K. Doubrovinski, O. Polyakov, and E. Wieschaus, "Apical constriction drives tissue-scale hydrodynamic flow to mediate cell elongation," Nature, vol. 508, no. 7496, pp. 392 -- 396, 2014. [54] A. G. Fletcher, M. Osterfield, R. E. Baker, and S. Y. Shvartsman, "Vertex models of epithelial morphogenesis," Biophys J, vol. 106, no. 11, pp. 2291 -- 2304, 2014. [55] O. Foubet and R. Toro, "Mechanical morphogenesis and the development of neocortical organisation," bioRxiv, 2015. [56] D. Matoz-Fernandez, F. A. Davidson, N. R. Stanley-Wall, and R. Sknepnek, "Wrinkle patterns in active viscoelastic thin sheets." preprint, arXiv:1904.08872, 2019. [57] Q. Liu, T. Ouchi, L. Jin, R. Hayward, and Z. Suo, "Elastocapillary crease," Phys Rev Lett, vol. 122, p. 098003, 2019.
1711.01551	1	1711	2017-11-05T09:38:38	Targeted Nanodiamonds for Identification of Subcellular Protein Assemblies in Mammalian Cells	[ "physics.bio-ph" ]	Transmission electron microscopy (TEM) can be used to successfully determine the structures of proteins. However, such studies are typically done ex situ after extraction of the protein from the cellular environment. Here we describe an application for nanodiamonds as targeted intensity contrast labels in biological TEM, using the nuclear pore complex (NPC) as a model macroassembly. We demonstrate that delivery of antibody-conjugated nanodiamonds to live mammalian cells using maltotriose-conjugated polypropylenimine dendrimers results in efficient localization of nanodiamonds to the intended cellular target. We further identify signatures of nanodiamonds under TEM that allow for unambiguous identification of individual nanodiamonds from a resin-embedded, OsO4-stained environment. This is the first demonstration of nanodiamonds as labels for nanoscale TEM-based identification of subcellular protein assemblies. These results, combined with the unique fluorescence properties and biocompatibility of nanodiamonds, represent an important step toward the use of nanodiamonds as markers for correlated optical/electron bioimaging.	physics.bio-ph	physics	Targeted Nanodiamonds for Identification of Subcellular Protein Assemblies in Mammalian Cells Short title: Nanodiamond Labels for Biological EM Michael P. Lakea, Louis-S. Bouchardabcd* aDepartment of Chemistry and Biochemistry, University of California, 607 Charles E. Young Drive South, Los Angeles, CA 90095-1569, USA. bCalifornia NanoSystems Institute, University of California, 570 Westwood Plaza, Los Angeles, CA 90095-7227, USA. cDepartment of Bioengineering, University of California, 420 Westwood Plaza, 5121 Engineering V, Los Angeles, CA 90095-1600, USA dThe Molecular Biology Institute and Jonsson Comprehensive Cancer Center, UCLA. Corresponding author: Louis-S. Bouchard ([email protected]; phone: +1 310 825 1764; fax: +1 310 206 4038) 1 Abstract Transmission electron microscopy (TEM) can be used to successfully determine the structures of proteins. However, such studies are typically done ex situ after extraction of the protein from the cellular environment. Here we describe an application for nanodiamonds as targeted intensity contrast labels in biological TEM, using the nuclear pore complex (NPC) as a model macroassembly. We demonstrate that delivery of antibody-conjugated nanodiamonds to live mammalian cells using maltotriose-conjugated polypropylenimine dendrimers results in efficient localization of nanodiamonds to the intended cellular target. We further identify signatures of nanodiamonds under TEM that allow for unambiguous identification of individual nanodiamonds from a resin-embedded, OsO4-stained environment. This is the first demonstration of nanodiamonds as labels for nanoscale TEM-based identification of subcellular protein assemblies. These results, combined with the unique fluorescence properties and biocompatibility of nanodiamonds, represent an important step toward the use of nanodiamonds as markers for correlated optical/electron bioimaging. 2 Introduction Over the past several decades, transmission electron microscopy (TEM) has allowed imaging of biological samples at the single protein level. With recent advances in electron tomography, it is now possible to resolve protein structures in cell sections with sub-nanometer resolution and field of view greater than 1 µm2 [1]. Currently, attempts to investigate the structures of proteins in situ are hindered by the lack of suitable live cell-compatible labels to identify the protein being observed in TEM. In this study, we show that nanodiamonds (NDs) can be used as targeted intensity contrast labels in biological TEM to image subcellular protein assemblies inside mammalian cells. Nanodiamonds were selected as TEM labels because of their wide range of favorable physical properties for biological applications. Indeed, NDs containing negatively- charged nitrogen vacancy (NV–) defect centers yield strongly fluorescent labels [2] with emission maxima in the near-infrared [3], enabling signal detection in vivo from non-superficial tissues [4], and have recently been used to measure temperatures [5] and magnetic fields [6] in living cells with nanometer precision. Nanodiamonds have been used to effectively deliver drugs [7], siRNAs [8], and proteins [9,10], and are currently being evaluated in preclinical trials for delivery of doxorubicin to brain, mammary, and liver tumors [11,12]. Nanodiamonds are also biologically inert and display minimal or no cytotoxicity [2, 13]. In our demonstration of NDs as targeted intensity contrast labels, we have used the nuclear pore complex (NPC) as a model macro assembly. Specifically, we targeted Nup98, a 98-kilodalton nucleoporin and a component of the central nucleoporin ring structure. Nup98 localizes to both the cytoplasmic and nucleoplasmic solvent accessible faces of the NPC and has a critical role in regulating macromolecular import and mRNA export to and from the nucleus [14]. Nup98 is unique among vertebrate nucleoporins and is known to shuttle in and out of the nucleus, trafficking 3 with mRNA to processing bodies (P-bodies) in the cytoplasm and localizing to distinct sites within the nucleus [15,16]. To date, fluorescence microscopy has been the method of choice to determine the subcellular location of functionalized NDs in cells [17,18]. Unfortunately, this method provides a limited view of the cellular environment immediately surrounding the fluorescent probe. However, TEM images of this cellular environment would be desirable to determine what is being measured by the ND. Existing studies of NDs using TEM have been restricted to untargeted or surface modified NDs that remain restricted to endosomes. Previous work predominantly provides isolated snapshots of NDs in cells without quantification [19], with the exception of a recent quantitative study of 150 nm NDs in endosomes [20]. To our knowledge, our TEM analysis is the first quantification of NDs with average diameter under 100 nm selectively targeted to a specific macromolecular structure and outside of endosomes. By utilizing a TEM compatible label [21], we have been able to verify the targeting of NDs to a specific subcellular structure via direct visualization. To unambiguously quantify the localization of antibody-conjugated NDs in mammalian cells at the nanometer scale, we have measured the location of NDs relative to a visible, uniformly sized macro assembly, the NPC. In this study, we have successfully demonstrated the use of maltotriose-conjugated polypropylenimine (PPI) dendrimers to assist in endosomal escape and delivery of NDs to the cytoplasm of HeLa cells, allowing the NDs to traffic for 12 additional hours to the intended target. The goal of this study was to determine if this delivery and targeting method would result in labeling efficiency sufficient to justify use of the NDs to locate a target of interest when visualization of the target may be ambiguous and to determine if cellular factors or the conjugated antibody would dictate the localization pattern instead. Ultimately, this method will 4 assist in low energy, low magnification identification of the NDs, reducing the electron radiation dose on the target of interest during the location stage of TEM imaging. Materials and Methods Cell Culture The procedure for transfection, preparation of the ND conjugates, and resin embedding is described in Zurbuchen et al. [21] and Mkandawire et al. [22]. Briefly, 2 g (5.7x10-5 mol) of 4th generation PPI dendrimers (SyMO-Chem, Netherlands), 0.92 g (1.82 mmol) and 0.24 ml (1.82 mmol) borane-pyridine complex were mixed in sodium borate buffer at 50° C, stirred continually for 7 days, purified by dialyzing against DI water for 3 days and freeze dried in a lyophilizer. The nanodiamonds were sonicated in a 3:1 mixture of concentrated HNO3:H2SO4 for 24 h to enhance carboxylation of the ND surfaces. For the conjugated NDs, an antibody raised against Nup98 (C-5), a mouse monoclonal IgG1 from Santa Cruz Biotechnology (sc-74578, Antibody Registry ID: AB_2157953), was conjugated to NDs using 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) to covalently attach free amines on the antibody directly to the carboxyl groups on the ND surface. The antibody:ND ratio during conjugation was 6:1. HeLa CCL2 cells (ATCC CCL-2) [23] were grown on polylysine-treated plastic coverslips for 24 h prior to transfection. 24.3 µg Nup98 conjugated and unconjugated NDs mixed with 2.5 µl Nup98 antibodies with an average diameter of 15-20 nm prepared by ball milling were pre-mixed with 100 µg maltotriose-conjugated PPI dendrimers, incubated at room temperature for 20 min. Transfection reagents were added dropwise to cells in a 6-well dish with 2 ml Dulbecco's modified Eagle's medium (DMEM) without serum at a final concentration of 12.2 µg/ml and 50 µg/ml of ND and glycodendrimer, respectively. Cells were 5 incubated for 6 h, the media changed to DMEM with 10% fetal bovine serum and then incubated for an additional 12 h post transfection. Electron Microscopy Sample Preparation To obtain material for electron microscopy, materials were directly fixed in a solution containing 2% (v/v) glutaraldehyde and 2% (v/v) paraformaldehyde in 0.1 M phosphate-buffered saline (PBS) buffer (pH 7.4) for 2 h at room temperature and then incubated at 4°C overnight. Subsequently, 0.5% (w/v) tannic acid was added, and the samples were incubated for 1 h at room temperature. The tissues were then washed five times in 0.1 M PBS buffer and postfixed in a solution of 1% (w/v) OsO4 (in PBS; pH 7.2–7.4). The combined treatment with tannic acid, glutaraldehyde, and paraformaldehyde followed by osmification enhanced the staining of membranes. The samples were washed four times in 0.1 M sodium acetate buffer (pH 5.5) and then block stained in 0.5% (w/v) uranyl acetate (in 0.1 M sodium acetate buffer, pH 5.5) for 12 h at 4°C. Subsequently, the samples were dehydrated in graded ethanol (50%, 75%, 95%, 100%, 100%, and 100%) for 10 min per step, rinsed with propylene oxide, and infiltrated in mixtures of Epon 812 and propylene oxide (1:1 and then 2:1 for 2 h each), followed by infiltration in pure Epon 812 overnight. Embedding was performed in pure Epon 812, and curing was done in an oven at 60°C for 48 h. Sections of 80-90 nm and 200 nm thickness were cut with an ultramicrotome (RMCMTX) using a diamond knife and were deposited on copper 200 mesh Quantifoil grids or single-hole grids coated with formvar and carbon, respectively. Sections were then double-stained in aqueous solutions of 8% (w/v) uranyl acetate for 25 min at 60°C and lead citrate for 3 min at room temperature. 6 Electron Microscopy Images were taken on a T20 iCorr cryo-transmission electron microscope from the FEI Tecnai G2 family. Images were taken over wide cell areas at 200 kV using selected area (SA) magnifications ranging from 3,500× to 29,000×, resulting in resolutions from 21 to 3.8 Å/pixel in 2D projections. Image regions were acquired either manually with defocus ranging from –5 to –70 µm or using automated acquisition with Fourier transform-based autofocus over wide areas to obtain unbiased information from each sample. Images were collected on a 2048×2048 electron camera with 16-bit intensity resolution. Image Analysis To assess ND localization statistically, 25 images from the targeted sample and 25 images from the untargeted control sample were selected to be faithful representations of the larger set of images, but without regard to ND quantity or distribution. Each set of 25 images was manually scored to determine edge-to-edge distances from individual ND to individual NPC for all NDs. We processed the images by applying contrast thresholds, morphology-based gradient filtering, and image arithmetic operations as indicated to enhance identification of NDs during the scoring using the Icy bioimage informatics platform [24]. Zoomed images of ROIs were enlarged 10-fold using a bilinear algorithm post-processing for publication. Results Conjugation of PPI Dendrimers 7 To assess the effectiveness of antibody-conjugated NDs as labels in a cellular landscape, we have evaluated the spatial pattern of NDs twelve (12) hours after transfection into living cells relative to that of a specific, independently identifiable macromolecular assembly, the NPC. Maltotriose conjugated PPI dendrimers were mixed with NDs prior to transfection to aid in delivery and endosomal escape. Maltotriose conjugation was verified using 13C and 1H NMR spectroscopy, and the results were consistent with previously published spectra and descriptions (see Figure. S1, supplementary information) [25,26]. Quantitative comparison of NMR peaks from the 13C spectrum indicates 83% conjugation of PPI dendrimer free amines (-H2C–NH2) with maltotriose (-H2C–NH2–R, R=maltotriose). ND-Nup98 Antibody Conjugates Localize to the NPC To target NDs to Nup98, anti-Nup98 antibodies were conjugated directly to the surfaces of NDs, and the conjugates were then delivered into the cytoplasm of HeLa cells via transfection with 32-branch PPI glycodendrimers to aid in endosomal escape. Twelve (12) hours after transfection, the cells were fixed and processed by standard resin embedding, stained with heavy metals including OsO4 which predominantly binds to membrane regions including the nuclear envelope, and sectioned into ultra-thin slices of 90 nm. Figure 1c shows a higher magnification (29,000×, 3.8 Å/pixel) 2D projection image of a NPC obtained from this preparation revealing the internal structural features including the cytoplasmic ring, the inner ring, the nuclear ring, and the nuclear basket. These features correspond to a recently published structure of the NPC [27] from HeLa cells (Fig 1a) and verify the image scale. A large number of nuclear pores are easily identified in thin sections at magnifications as low as 3400x and appear as clear gaps in the nuclear envelope staining or as visible structures matching the well-defined dimensions of the NPC. Using these 8 features and the ability to detect NDs with high sensitivity, we generated a distribution of the minimum edge-to-edge distances observed from projection images between each ND and the nearest nuclear pore using representative images from the antibody-conjugated sample (targeted) and the unconjugated sample (untargeted) (Fig 2). The experiment was performed from ND conjugation through EM sample preparation 2 times independently and with 3 concentrations of both conjugated and control NDs. Due to lack of saturation at the highest concentration of NDs tested, we selected the highest concentration tested for quantitative analysis. Fig 1. Nup98 and the NPC as a target for NDs. (a) In situ structure of human nuclear pore from isolated intact HeLa nuclei. Reproduced from Ref. [27] with permission. (b) Location of Nup98 9 within the NPC showing that Nup98 can localize in the central channel on both sides of the nuclear pore and outside, where it helps anchor the NPC to the nuclear envelope. Reproduced from Ref. [15] with permission. (c) High magnification (3.8 Å/pixel) TEM image of human nuclear pore in 90 nm epoxy resin slice from HeLa cells showing structural details including the cytoplasmic ring, internal ring, nuclear ring and basket. Scale bar, 20 nm, rotated. (d) Wide area view of the NPC in (c), scale bar 100 nm, no rotation. C and N designate cytoplasmic and nuclear regions. Fig 2. Successful targeting of NDs to the NPC. (a–f) Representative images of anti-Nup98– conjugated NDs (a, b) bound to or (c) located near an NPC, and (d–f) unconjugated NDs. C and N designate cytoplasmic and nuclear regions. Scale bar, 125 nm. Black arrows point to NDs. (g,i) Histograms of edge-to-edge distances from each ND or cluster to the nearest nuclear pore for anti-Nup98–conjugated (g) and unconjugated NDs (i). Central bin counts as a percentage of the total NDs counted are 31.0% and 1.26% for (g) and (i), respectively. (h,j) Rescaled data from (g) and (i). Negative and positive values refer to distances into the nucleus and cytoplasm, respectively. We analyzed twenty-five (25) representative images from both the conjugated and unconjugated TEM samples and identified 158 and 239 NDs or ND clusters, respectively, including approximately 1,000 intracellular NDs for both sets. Minimum distance (bin 0) was 10 defined as ± 7.5 nm to account for antibody length. The center 3 bin edges are: -200, -7.5, 7.5, 200, and all others increment by 200 nm. The blue curve is a normal distribution based on the mean and standard deviation of the targeted NDs not bound to the NPC and the untargeted NDs, respectively. The distances measured are included in the Supplementary Information as Table S2. We selected twenty-five (25) representative images from both conjugated and unconjugated (control) ND samples from over 100 images acquired for each sample and scored them manually to determine the distribution of edge-to-edge distances. An example of the scoring is provided in the supplementary information (Figure S3). Image selection was blind to the position of any image inside of the larger image set. Automatic focus and image acquisition over pristine, random areas was used to collect images without bias, with prescreening only to identify areas that contained cells. To construct a binary statistic, we categorized NDs within 7.5 nm of either side of the pore as bound to account for the size of an antibody and the location of Nup98 at the edge of the NPC, represented by the central bins in Fig 2, panels (g) and (h). Using this criterion to define the bound fraction, 31.0 percent of all intracellular anti-Nup98–conjugated NDs or clusters were bound to a NPC, compared with 1.26 percent for unconjugated ND or clusters. ND counts per cluster had a mean µ = 9.4 and a standard deviation σ = 11 for the targeted NDs and mean µ = 7.5 and standard deviation σ = 6.3 for the untargeted NDs. These values were consistent across several counts. We also analyzed the images with a different binary statistic by defining each NPC as labeled or unlabeled using the same distance criteria (within 7.5 nm), resulting in 32/89 or 36.0% NPCs labeled by NDs in the targeted set and 3/93 or 3.2% NPCs labeled in the untargeted set. Treating the probability of an NPC being bound to a ND as a Poisson statistic, we then performed a test of the null hypothesis that these counts were generated by the same underlying probability of success for unequal sample size as described by Shiue and Bain [28]. 11 Briefly, for two Poisson distributions X ~ POI(λ1=s1γ1) and Y ~ POI(λ2=s2*γ2), where s is the total number of observations and γ is the probability of success, the conditional distribution of Y, given the counts observed are Y = y and (X + Y) = x + y = m, is a binomial distribution with parameters m and p or (Y x + y = m) ~ Bin (m,p) where p = s2/(s1+s2). This conditional distribution provides a universally most powerful unbiased (UMPU) test of the hypothesis H0: γ1=γ2 against the alternative Ha: γ2 > γ1 at the level of 1 – Bin(y-1; m,p). To use this test, we observe y = 32 (targeted), x = 3 (untargeted), x + y = 35 and determined that 1-Bin(31; 35, 89/182) = 1.1x10-7, or 1 – the cumulative probability that y ≤ 32, allowing us to reject the null hypothesis for any alpha ≥ 1.1x10-7. Similar analysis taking the counts of central bin of the above histograms as successes in a Poisson distribution gives y = 49, x = 3, and x + y = 52 yielding 1-Bin(48; 53, 158/397) = 1.5x10-14. This information is summarized in the supplementary information Table S1. Comparing the untargeted data to a normal distribution using a chi-square goodness-of-fit test we determined a p-value of 0.027, which indicates a reasonable deviation from a normal distribution, especially considering that this includes both cytoplasmic and nuclear distances in the same distribution. Assessing the targeted data set by excluding the central bin and assessing all remaining values using the same test, we obtain a p-value of .17, indicating that the distance from nuclear pores of the unbound fraction of targeted diamonds are close to normally distributed. Due to manual identification of nuclear pores it is possible that the closest NPC is not detected, however identification of the nuclear envelope is unambiguous is all images and given the high density of nuclear pores found on the nuclear envelope compared to the distances being measured, an error in identification would only slightly alter the distributions presented. 12 Anti-Nup98–Conjugated NDs as Markers for the Nuclear Pore Anti-Nup98–conjugated NDs bind to the NPC in a manner that reflects the expected localization pattern of Nup98 (Fig 3, a–e). The NDs tend to appear in clusters centered in the pore or off-center, consistent with having been dislodged and left at the edge of the NPC on either side of the membrane, possibly due to interference by endogenous molecular traffic. Considering that the NDs in this study have a size (~20 nm) smaller than the NPC central channel (~60 nm) [27], aggregates of NDs within the pore are not unexpected. Nup98 is known to localize to other cellular structures, and thus the calculated percentage of conjugated NDs localized at the NPC (31.0%) likely represents a lower bound for the targeting efficiency of anti-Nup98–conjugated NDs. Consistent with this, we observed strong localization of NDs to additional distinct structures. Identifying and properly quantifying binding to these other structures is hindered by lack of positive identification, however, and will be the subject of future investigations. Fidelity of the ND conjugates for Nup98 overall appears significantly higher than the percentage reported bound to NPCs, but can only be qualitatively assessed by observation at this time. It is known that Nup98 localizes to multiple subcellular assemblies such as P-bodies and intranuclear bodies however, these structures cannot be unambiguously identified in our TEM images. Most ND localization studies with larger NDs do not observe ND entry in to the nucleus. This claim is not consistent with our observations with ~20 nm NDs for either targeted or untargeted NDs in this study. However, our study was not intended to analyze this characteristic and future systematic studies of this observation coupled with additional measurements should be more conclusive. It has been observed that 4 nm gold particles modified with both and poly(ethylene glycol) [29] and polyethylenimine [30] and 70 nm SiO2 nanoparticles [31] form aggregates in the nucleus, emphasizing the importance of further research on the ability of NDs to enter the nucleus. 13 Fig 3. Nanodiamonds can be detected individually from low magnification (5,000×) images. (a) Original image. (b) Zoom of ROI from (a). (c) Gradient morphometry enhancement of (b). (d) Contrast inverted image of (b). (e) Inverted and gradient images summed with threshold applied to isolate diamond. Red and blue indicate saturated maximum and minimum intensities, respectively, resulting in a binary colored image. Isolating NDs from Cellular Background 14 To further assess the utility of NDs as TEM landmarks, we evaluated the intensity profile of NDs in transmission electron images. We found that the NDs are clearly distinguished from both the surrounding intracellular milieu and heavy metal stained components because of their low transmitted intensity and bright intensity fringes. Figure 3 provides an example of these features in raw and processed images captured under low magnification, demonstrating that they can be used to detect NDs with zero background. The NDs were successfully isolated from the cellular background in heavy metal-stained samples using a simple two-step image processing method. First, a gradient morphometry filter [24] was used to generate a separate image. Then, by inverting the intensity of the original image to turn dark contrast into bright signal and superimposing the two images, a clearly distinguishable object for each ND results. NDs are detected with high sensitivity because they appear as simultaneously bright and dark objects on the scale of tens of nanometers, producing an exceptionally strong spatial gradient, in addition to their intensity contrast. It can be seen from Fig 4 that NDs can be isolated with zero background signal over areas spanning significant fractions of cells that include cytoplasmic, nuclear and extracellular regions. Because the appearance of NDs in TEM images depends on the crystal orientation and the defocus, most (but not all) NDs observed can be isolated with zero background from automatically acquired images with human intervention only to select coverage areas. Based on the results from this acquisition method, alteration of the acquisition process to include a limited series of tilt angles or defocus values and diffraction mode imaging at each location should enable all individual NDs to be isolated with zero background in wide area images. Our previous work has demonstrated that diffraction-mode imaging can identify NDs over a magnification range spanning seven orders of magnitude, and that high resolution TEM can conclusively verify the identity of NDs by observing the atomic lattice and by generating diffraction patterns [21]. 15 Fig 4. Nanodiamond isolation is robust and reproducible. (a) Wide area view of two cells transfected with anti-Nup98–conjugated NDs. (b) Gradient enhanced image of (a). (c) Central ROI from (a). (d) Central ROI from (b). (e) Original image inverted and added to gradient image, then contrasted to show isolation of ND with zero background. (f) Upper ROI from (a). (g) Upper ROI from (b). (h) Original image inverted and added to gradient image with threshold applied to show isolation of ND with zero background. In epoxy resin-embedded samples stained with OsO4, the predominant source of membrane contrast is the osmium, which concentrates in lipid membranes like the nuclear envelope. Because of this, objects near the nuclear membrane are especially challenging to isolate. Figure 5 shows complete isolation of the NDs that are bound to the nuclear pore, with substantial sections of the nuclear membrane in the field of view. Because this data was collected with the most basic single image acquisition method and a simple image processing algorithm, there is substantial opportunity to enhance the specific detection of NDs over wide areas in a biological environment 16 to achieve perfect fidelity for ND recognition. 17 Fig 5. Isolation of NDs bound to NPCs. (a–f) Unprocessed TEM images of anti-Nup98– conjugated NDs bound to NPCs in 90 nm epoxy resin sections from HeLa cells fixed 12 h after transfection with the ND conjugates. NPCs are highlighted in green. (g–i) Gradient morphometry filter applied to (d–f). (j–l) Images processed with a color threshold applied to isolate NDs. All scale bars are 50 nm. Due to their geometry and crystalline volume, NDs produce bright thickness and phase fringes that make them excellent candidates as fiducial markers for computer recognition during tomographic reconstruction and computer-assisted localization. Figure 6 shows that the gradient image alone is sufficient to isolate most individual NDs. From the ND gradient image (center column), the image-based signal-to-noise ratio (SNR) of the ND center to fringe transition, given as the mean intensity of the brightest circle of pixels of the ring divided by the standard deviation of the local background are 14.0 and 11.3 (15,555/1,044 and 11,791/1,041) with the brightest pixel in each having an SNR of 18.3 and 14.4, for panels (a) and (b), respectively. Because the fringes derive mainly from the shape of the nanocrystal [32], the ring-shaped fringe indicates that the particles are quasi-spherical. The development of consistently shaped NDs from processes other than ball milling should result in reproducible signals such as the two shown in Fig. 6, significantly enhancing the detectability of NDs in cells using TEM. 18 Fig 6. Gradient enhancement alone allows detection of quasi-spherical NDs. Cropped sections from (a) central ROI and (b) upper ROI of Fig. 4. From left to right are the original images (left), gradient morphometry enhanced images (center), and contrast thresholds applied to the gradient images (right). Discussion Fluorescent labeling of proteins has been highly successful for sub-micron visualization of protein localization. Nonetheless, the resolution of optical measurements is subject to Abbe's diffraction limit [33] and super-resolution techniques do not allow for visualization of the cellular environment surrounding fluorescent labels at the nanometer scale. Fluorescence microscopy studies indicate that the cellular uptake mechanism for unconjugated NDs is endocytosis and that NDs are subject to several cellular trafficking mechanisms [19]. 19 Although fluorescence microscopy can determine the localization of particles to the perinuclear region, more advanced optical techniques and carefully controlled experiments are required in order to differentiate targeting to more specific cellular regions, such as localization to the nuclear v. cytoplasmic sides of the nuclear envelope [14]. Moreover, the diffraction limit of light (~200 nm) imposes a significant obstacle to the mapping of 20 nm nanoparticles particles relative to a 120 nm macro assembly such as the NPC. Light microscopy precludes discrimination among particles that are bound to the intended target, that are near the target, that have displaced the target protein from its assembly, or particles that are bound in aggregates or nanoscale structures preventing them from directly contacting the intended target. While sub-diffraction methods such as STED microscopy can visualize fluorescent molecules such as NV- NDs, these methods are limited by substantial sample restrictions, the requirement for high laser powers that can bleach fluorophores and damage intact cell samples, inability to distinguish large NDs from clusters because they do not detect the diamond material and the requirement to co-stain any target, severely limiting the interpretation of the context surrounding the target and label. In practice, most sub-diffraction imaging systems are restricted to 40-80 nm resolution in cells [34], especially because STED resolution depends strongly on the power of the depletion beam. Furthermore, even ~20 nm resolution is insufficient for structural identification of individual macromolecular assemblies such as the nuclear pore or to separate individual NDs when clustered, precluding accurate quantification [35]. In fact, this methodology was specifically inspired as a way to complement STED and other fluorescent imaging techniques, by providing much of the information inaccessible through these methods. Nanoscale localization information about the Nup98 nucleoporin of the NPC by immuno- gold staining of cells post fixation has been studied [36]. However, the ability of antibodies to 20 successfully localize NDs to subcellular regions with nanoscale precision and outside of endosomes in live cells has not been demonstrated prior to this study. Given the resolution limit of light microscopy and the emerging use of NDs as nanoscale optical sensors in live cell thermometry and magnetometry, as well as in drug and gene delivery, knowledge of the precise location of NDs relative to a target of interest is increasingly important. In previous work, we demonstrated that NDs possess the physical and biological properties necessary to act as landmarks in a cellular environment for TEM using resin-embedded ultra-thin sections from HeLa cells [21]. Whether the localization of antibody-conjugated NDs to specific protein assemblies is sufficiently robust to enable live cell labeling of protein assemblies, accounting for transfection and trafficking throughout cells, has not been previously determined and is the subject of this work. In general, quantification of the subcellular location of nanoparticles is difficult due to the large number of sections and images required to unambiguously identify nanoparticles, usually at higher magnifications than the present study, and the difficultly of normalizing to the variable geometry or volumes of the sub cellular regions being compared. [37, 38] By utilizing a recognizable, homogeneously sized macromolecular structure, the NPC, and the binary criteria of being inside or outside of the nucleus, our analysis avoids these complications. We show here that NDs provide a robust, identifiable landmark that can be used to direct expensive and time-consuming tomography studies to specific regions of interest (ROIs), providing a high probability of detecting the target of interest, even if the identity of the protein assemblies and their constituents cannot be verified until after image processing and reconstruction. Using this method, low magnification images with wide fields of view can provide quantification of NDs or identification of ROI's around NDs with automated image collection. 21 Low SNR is an inherent limitation in TEM of cells; increasing beam exposure enhances SNR at the cost of increasing damage to the sample. Targeting of biologically inert contrast agents such as NDs with high specificity to subcellular structures can extend the applications for TEM imaging of biological structures by providing a multi-functional label capable of correlating fluorescently detected sensor signals and images with structural information obtainable exclusively by TEM. Nanodiamonds offer several distinct advantages over other nanoparticles used for probing biological systems using TEM. Quantum dots (QDs) are amenable to surface functionalization but lack the versatility of binding, especially directly to proteins, offered by a heavily carboxylated surface. They are also generally cytotoxic [39, 40] and are prone to intermittent fluorescence [41]. Moreover, QD fluorescence is quenched by osmium tetroxide fixation [42] limiting the practicality of QDs for multimodal imaging. Gold nanoparticles (AuNPs) lack the biosensing capabilities of NDs [5], do not readily undergo endosomal escape, accumulating in endosomes even after accounting for diverse strategies to aid in cytoplasmic delivery [43,44], and are not easily distinguished by appearance in electron micrographs from OsO4-stained cytosolic lipid droplets [45]. Unlike NDs, AuNPs are not bioinert and produce cytotoxic effects at concentrations as low as 20 nM [46]. AuNPs may also induce protein aggregation at physiological pH due to the high affinity of gold for free thiols on proteins, which are abundantly present in the reducing environment of the cytoplasm [47]. While capping and surface modifications can reduce these effects, their removal in vivo and resulting exposure of the bare gold surface makes modified AuNPs unattractive for long-term biological labeling or clinical applications [48]. Finally, metal oxide nanoparticles are useful as contrast agents in TEM but exhibit high cytotoxicity [49,50] 22 and lack the versatility of NDs, especially the sensing capabilities and the availability of surfaces for direct covalent conjugation of proteins. Based on our results, targeted NDs can identify specific sub-volumes or cellular ROIs with sufficient efficiency to target high-resolution imaging or electron tomography. To improve detection, acquisition of a minimal defocus series including positive defocus at each location will enable detection of variations in ND fringes with the defocus, further capitalizing on the material differences between crystalline and amorphous carbon structures, which ideally show no contrast at positive defocus. Furthermore, this signature can be analyzed using image-based, high-content recognition algorithms similar to those currently used in fluorescence microscopy to identify subcellular geometric patterns in drug screening assays [51]. Automating the recognition of fiducial or correlation markers over whole cell areas provides a major advantage for in situ imaging approaching the structural level, fiducial marker identification for reconstruction of electron tomograms and single particle reconstructions of large numbers of repetitive images to refine structures. This method has direct applications in traditional TEM as well as in focused ion beam thinned cryo sections, where the imaging necessary to locate structures of interest can significantly degrade the sample. We are currently optimizing methods to extend this technique to correlated light (live cell) and electron microscopy without the need for specialized microscopes or stages. Beyond research, this method has applications for improving the analysis of both ND and monoclonal antibody-based therapeutics. TEM-based ND imaging enables the localization of therapeutic antibodies bound to NDs in tissues of humans or animals to be determined unambiguously at the nanometer scale. To date, an increasing array of monoclonal antibodies have been approved by the Food and Drug Administration (FDA) to treat cancer, autoimmune 23 diseases, allergic asthma, viruses, blood disorders, organ transplant rejection, osteoporosis, and several other disorders [52]. PPI dendrimers also have promise in therapeutic applications. In vivo studies with rats have demonstrated that 25% conjugation with maltotriose reduces cytotoxicity to a minimal level and 100% conjugation eliminates detectable toxicity at all doses tested [25]. When tested on panels of human cells for pharmaceutical applications, densely maltotriose-conjugated dendrimers were the least cytotoxic form of PPI glycodendrimers, compared to unconjugated, partially conjugated and maltose-conjugated PPI dendrimers [53]. Given their low cytotoxicity profile, maltotriose-conjugated dendrimers hold promise as a component of therapeutics for use in humans [54]. The third part of our targeting strategy, the NDs, are also generating increasing interest as therapeutic agents. In vivo studies in mice have shown that intraperitoneal injection of NDs did not cause detectable cytotoxicity in the central or peripheral nervous system or impair neural function based on gross animal behavior and hippocampal novel object recognition (NORT) tests which can report damage to the hippocampus [55]. A recent study concluded that 20 and 100 nm carboxylated NDs showed no cytotoxicity or genetoxicity measured in real time in six human cell lines derived from liver, kidney, intestine and lung tissue, with ND doses up to 250 µg/ml and recommended that carboxylated NDs be considered for use as a negative control in nanoparticle toxicity studies [13]. The added advantage of detectability by TEM at the nanoscale compared to other candidate medicinal nanoparticles or antibodies alone could improve and expedite clinical trials for NDs as vehicles for antibodies, proteins, nucleic acids or drugs, potentially lowering costs and reducing ambiguity about the subcellular fate of these particles in preclinical studies. 24 Due to the long-term chemical stability of NDs, this imaging method could also help assess tissue penetration and persistence of therapeutic NDs in biopsies or post mortem tissue immediately following or months after ND administration. Nanoscale imaging of NDs identified as bound to or near a target of interest or accumulating inside cellular structures where they are isolated from the target of interest can correlate therapeutic efficacy measurements with ND location by direct visualization. Determination of the mechanisms of action and drug resistance for monoclonal antibody therapies are increasingly important for improving clinical outcomes and studies on trastuzumab emtansine show that acquired resistance may be due to reduced binding of the antibody to cancer cells, inefficient internalization or intracellular trafficking, heightened recycling of the antibody- target complex or impaired lysosomal degradation [56]. The ability to successfully target and identify the subcellular location of ND-antibody-target complexes by TEM could contribute significantly to further studies of these mechanisms. Conclusions We have demonstrated a method to prepare, deliver and image ND-antibody conjugates for labeling of cellular protein assemblies. We chose the Nup98 nucleoporin, a component of the NPC, as the target for our study because of the recognizable structure of the NPC, its well defined location and the significance of its physiological functions for sensing and drug delivery applications. Using TEM, we assessed the percentage of NDs bound to the Nup98 nucleoporin at the NPC on a single diamond–single complex basis. In addition to localization at the NPC, we observed clustering of diamonds at other locations, distinct from the NPC, where Nup98 is known to localize. Our results demonstrate that NDs can be identified individually over the area of entire 25 cells at low magnification with near zero background signal from first pass automated image acquisition. Using the unique signature of NDs observed under TEM in biological samples, we have measured the distance of individual NDs to individual NPCs from images taken over wide areas and with automated acquisition, demonstrating the minimal image requirements necessary to identify NDs. Given the established efficacy of NV–-containing NDs for fluorescence microscopy applications, our results highlight the value of NDs for correlated imaging. Widespread use of correlated fluorescence and electron imaging has been hindered by practical limitations surrounding the markers and sample preparations. Here we have described a TEM-based method that enables high precision localization of particles and the ability to recognize ROIs with a novel TEM marker. This method is compatible with live cell optical microscopy using an established non-bleaching fluorophore (NV– ND), allowing for practical correlated imaging that can be performed on separate light and electron microscopes, thereby facilitating extension of correlated imaging beyond specialized instruments to equipment readily available at most research institutions. Our technique has the ability to complement and outperform the common method of fluorescence-based localization. We have recently extended this technique into plunge-frozen, whole, unstained cells to provide a versatile and unambiguous standard that can provide localization of particles at the level of proteins and protein assemblies without requiring costly embedding and sectioning. We further investigated the appearance of NDs observed by TEM, and noted that the maximum and minimum transmitted intensities from NDs are each more extreme than the signal variation from the resin-embedded cellular background or heavy metal staining. We also observed geometrically consistent, ring-shaped intensity gradients likely due to a combination of thickness- 26 and phase-based intensity fringes, capable of providing well defined geometric features for computer-based recognition. These features will likely prove valuable for automated acquisition and tomographic reconstruction. References 1. Mahamid J, Pfeffer S, Schaffer M, Villa E, Danev R, Cuellar LK, et al. Visualizing the molecular sociology at the HeLa cell nuclear periphery. Science 2011;351: 969–972. doi: 10.1126/science.aad8857 2. Yu SJ, Kang MW, Chang HC, Chen KM, Yu YC. Bright fluorescent nanodiamonds: no photobleaching and low cytotoxicity. J. Am. Chem. Soc. 2005;127: 17604–17605. doi: 10.1021/ja0567081 3. Jelezko F, Tietz C, Gruber A, Popa I, Nizovtsev A, Kilin S, et al. Spectroscopy of single N-V centers in diamond. Single Mol. 2001;2: 255–260. doi: 10.1002/1438-5171(200112)2:4<255::aid- simo255>3.0.co;2-d 4. Tuchin VV. Tissue Optics: Light Scattering Methods and Instruments for Medical Diagnosis. 3rd ed. Bellingham: Society of Photo-Optical Instrumentation Engineers; 2015. 5. Kucsko G, Maurer PC, Yao NY, Kubo M, Noh HJ, Lo PK et al. Nanometre-scale thermometry in a living cell. Nature 2013;500: 54–58. doi:10.1038/nature12373 6. Le Sage D, Arai K, Glenn DR, DeVience SJ, Pham LM, Rahn-Lee L et al. Optical magnetic imaging of living cells. Nature 2013;496: 486–489. doi:10.1038/nature12072 7. El-Say, KM. Nanodiamond as a drug delivery system: applications and prospective. J. Appl. Pharm. Sci. 2011;1: 29–39. 27 8. Chen M, Zhang XQ, Man, HB, Lam R, Chow EK, Ho, D. Nanodiamond vectors functionalized with polyethylenimine for siRNA delivery. J. Phys. Chem. Lett. 2010;1: 3167– 3171. doi: 10.1021/jz1013278 9. Moore, L, Gatica, M, Kim, H, Osawa, E, Ho, D. Multi-protein delivery by nanodiamonds promotes bone formation. J. Dent. Res. 2013, 92, 976–981. DOI: 10.1177/0022034513504952 10. Shimkunas RA, Robinson E, Lam R, Lu S, Xu X, Zhang XQ et al. Nanodiamond-insulin complexes as pH-dependent protein delivery vehicles. Biomaterials 2009;30: 5720–5728. doi: 10.1016/j.biomaterials.2009.07.004 11. Xi G, Robinson E, Mania-Farnell B, Vanin EF, Shim KW, Takao T et al. Convection- enhanced delivery of nanodiamond drug delivery platforms for intracranial tumor treatment. Nanomedicine 2014;10: 381–391. doi: 10.1016/j.nano.2013.07.013 12. Chow EK, Zhang XQ, Chen M, Lam R, Robinson E, Huang H et al. Nanodiamond therapeutic delivery agents mediate enhanced chemoresistant tumor treatment. Sci. Transl. Med. 2011;3: 73ra21. doi: 10.1126/scitranslmed.3001713 13. Paget V, Sergent JA, Grall R, Altmeyer-Morel S, Girard HA, Petit T, Gesset C et al. Carboxylated nanodiamonds are neither cytotoxic nor genotoxic on liver, kidney, intestine and lung human cell lines. Nanotoxicology. 2014;8:46-56. 14. Chatel G, Desai SH, Mattheyses AL, Powers MA, Fahrenkrog, B. Domain topology of nucleoporin Nup98 within the nuclear pore complex. J. Struct. Biol. 2012;177: 81–89. doi: 10.1016/j.jsb.2011.11.004 28 15. Franks TM, Hetzer MW. The role of Nup98 in transcription regulation in healthy and diseased cells. Trends Cell Biol. 2013;23: 112–117. doi: 10.1016/j.tcb.2012.10.013 16. Souquere S, Weil D, Pierron, G. Comparative ultrastructure of CRM1-Nucleolar bodies (CNoBs), Intranucleolar bodies (INBs) and hybrid PML/p62 bodies uncovers new facets of nuclear body dynamic and diversity. Nucleus 2015;6: 326–338. doi: 10.1080/19491034.2015.1082695 17. Faklaris O, Joshi V, Irinopoulou T, Tauc P, Sennour M, Girard H et al. Photoluminescent diamond nanoparticles for cell labeling: study of the uptake mechanism in mammalian cells. ACS Nano 2009;3: 3955–3962. doi: 10.1021/nn901014j 18. Mkandawire M, Pohl A, Gubarevich T, Lapina V, Appelhans D, Rodel G et al. Selective targeting of green fluorescent nanodiamond conjugates to mitochondria in HeLa cells. J. Biophotonics 2009;2: 596–606. doi 10.1002/jbio.200910002 19. Faklaris O, Joshi V, Irinopoulou T, Tauc P, Sennour M, Girard H et al. Photoluminescent diamond nanoparticles for cell labeling: study of the uptake mechanism in mammalian cells. ACS Nano 2009;3: 3955–3962. doi: 10.1021/nn901014j 20. Nagarajan S, Pioche-Durieu C, Tizei LHG, Fang CY, Bertrand JR, Cam EL et al. Simultaneous cathodoluminescence and electron microscopy cytometry of cellular vesicles labeled with nanodiamonds. Nanoscale. 2016;8:11588-11594. 21. Zurbuchen MA, Lake MP, Kohan SA, Leung B, Bouchard, L-S. Nanodiamond landmarks for subcellular multimodal optical and electron imaging. Sci. Rep. 2013;3: 2668. doi:10.1038/srep02668 29 22. Mkandawire M, Pohl A, Gubarevich T, Lapina V, Appelhans D, Rodel G et al. Selective targeting of green fluorescent nanodiamond conjugates to mitochondria in HeLa cells. J. Biophotonics 2009;2: 596–606. doi 10.1002/jbio.200910002 23. Scherer WF, Syverton JT, Gey GO. Studies on the propagation in vitro of poliomyelitis viruses. IV. Viral multiplication in a stable strain of human malignant epithelial cells (strain HeLa) derived from an epidermoid carcinoma of the cervix. J Exp Med. 1953;97(5):695-710. 24. De Chaumont F, Dallongeville S, Chenouard N, Hervé N, Pop S, Provoost T et al. Icy: an open bioimage informatics platform for extended reproducible research. Nat. Methods 2012;9: 690–696. doi: 10.1038/nmeth.2075 25. Ziemba B, Janaszewska A, Ciepluch K, Krotewicz M, Fogel WA, Appelhans D et al. In vivo toxicity of poly(propyleneimine) dendrimers. J. Biomed. Mater. Res. A 2011;99: 261–268. doi: 10.1002/jbm.a.33196 26. Klajnert B, Appelhans D, Komber H, Morgner N, Schwarz S, Richter S et al. The influence of densely organized maltose shells on the biological properties of poly(propylene imine) dendrimers: new effects dependent on hydrogen bonding. Chemistry 2008;14: 7030–7041. doi: 10.1002/chem.200800342 27. von Appen A, Kosinski J, Sparks L, Ori A, DiGuilio AL, Vollmer B. In situ structural analysis of the human nuclear pore complex. Nature 2015;526: 140–143. doi:10.1038/nature15381 28. Shiue WK, Bain LJ. Experiment Size and Power Comparisons for Two-Sample Poisson Tests. J. Roy. Statist. Soc. Ser. C. 1982;31(2):130-134. 30 29. Gu YJ, Cheng J, Lin CC, Lam YW, Cheng SH, Wong WT. Nuclear penetration of surface functionalized gold nanoparticles. Toxicol Appl Pharmacol. 2009;237:196-204. 30. Thomas M, Klibanov AM. Conjugation to gold nanoparticles enhances polyethylenimine's transfer of plasmid DNA into mammalian cells. Proc. Natl. Acad. Sci. U.S.A. 2003;100(16):9138- 9143. 31. Chen M, von Mikecz A. Formation of nucleoplasmic protein aggregates impairs nuclear function in response to SiO2 nanoparticles. Exp Cell Res. 2005;305(1):51-62. 32. Hirsch PB, Howie A, Nicholson RB, Pashley DW, Whelan ML. Electron Microscopy of Thin Crystals. 2nd ed. New York: Krieger; 1977. 33. Abbe E. On the estimation of aperture in the microscope. J. R. Microsc. Soc. 1881;1: 388– 423. doi: 10.1111/j.1365-2818.1881.tb05909.x 34. Tzeng YK, Faklaris O, Chang BM, Kuo Y, Hsu JH, Chang HC. Superresolution Imaging of Albumin-Conjugated Fluorescent Nanodiamonds in Cells by Stimulated Emission Depletion. Angew. Chem. Int. Ed. 2011;50:2262-2265. 35. Rothen-Rutishauser B, Kuhn DA, Ali Z, Gasser M, Amin F, Parak WJ et al. Quantification of gold nanoparticle cell uptake under controlled biological conditions and adequate resolution. Nanomedicine. 2014;9(5):607-621. 36. Griffis ER, Xu S, Powers MA. Nup98 localizes to both nuclear and cytoplasmic sides of the nuclear pore and binds to two distinct nucleoporin subcomplexes. Mol. Biol. Cell 2003;14: 600– 610. doi: 10.1091/mbc.e02-09-0582 31 37. Mayhew TM, Mühlfeld C, Vanhecke D, Ochs M. A review of recent methods for efficiently quantifying immunogold and other nanoparticles using TEM sections through cells, tissues and organs. Ann. Anat. 2009;191:153-170. 38. Mayhew TM, Mapping the distributions and quantifying the labelling intensities of cell compartments by immunoelectron microscopy: progress towards a coherent set of methods. J. Anat. 2011;219:647-660. 39. Derfus AM, Chan WCW, Bhatia SN. Probing the cytotoxicity of semiconductor quantum dots. Nano Lett. 2004;4: 11–18. doi: 10.1021/nl0347334 40. Oh E, Liu R, Nel A, Gemill KB, Bilal M, Cohen Y et al. Meta-analysis of cellular toxicity for cadmium-containing quantum dots. Nat Nanotechnol. 2016;11:479-486. 41. Frantsuzov P, Kuno M, Jankó B, Marcus RA. Universal emission intermittency in quantum dots, nanorods and nanowires. Nat. Phys. 2008;4: 519–522. doi: 10.1038/nphys1001. 42. Deerinck, TJ. The application of fluorescent quantum dots to confocal, multiphoton, and electron microscopic imaging. Toxicol. Pathol. 2008;36: 112–116. doi: 10.1177/0192623307310950. 43. Guo S, Huang L. Nanoparticles escaping RES and endosome: challenges for siRNA delivery for cancer therapy J. Nanomater. 2011;2011: 1– 12. doi: 10.1155/2011/742895. 44. Lèvy R, Shaheen U, Cesbron Y, Sèe V. Gold nanoparticles delivery in mammalian live cells: a critical review. Nano Rev. 2010; doi: 10.3402/nano.v1i0.4889. 32 45. Thiberge S, Nechushtan A, Sprinzak D, Gileadi O, Behar,V, Zik O et al. Scanning electron microscopy of cells and tissues under fully hydrated conditions. Proc. Natl. Acad. Sci. U.S.A. 2004;101: 3346–3351. doi: 10.1073/pnas.0400088101. 46. Soenen SJ, Manshian B, Montenegro JM, Amin F, Meermann B, Thiron T et al. Cytotoxic effects of gold nanoparticles: a multiparametric study. ACS Nano 2012;6: 5767– 5783. doi: 10.1021/nn301714n. 47. Zhang D, Neumann O, Wang H, Yuwono VM, Barhoumi A, Perham M et al. Gold nanoparticles can induce the formation of protein-based aggregates at physiological pH. Nano Lett. 2009;9: 666–671. doi: 10.1021/nl803054h. 48. Kreyling WG, Abdelmonem AM, Ali Z, Alves F, Geiser M, Haberl N et al. In vivo integrity of polymer-coated gold nanoparticles. Nat. Nanotechnol. 2015;10: 619–623. doi: 10.1038/nnano.2015.111. 49. Ivask A, Titma T, Visnapuu M, Vija H, Kakinen A, Sihtmae M et al. Toxicity of 11 metal oxide nanoparticles to three mammalian cell types in vitro. Curr. Top. Med. Chem. 2015;15: 1914– 1929. doi: 10.2174/1568026615666150506150109. 50. Wingard CJ, Walters DM, Cathey BL, Hilderbrand SC, Katwa P, Lin S et al. Mast cells contribute to altered vascular reactivity and ischemia-reperfusion injury following cerium oxide nanoparticle instillation. Nanotoxicology. 2011;5(4):531-545. 51. Buchser W, Collins M, Garyantes T, Guha R, Haney S, Lemmon V et al. Assay Development Guidelines for Image-Based High Content Screening, High Content Analysis and High Content Imaging. In: Sittampalam GS, Gal-Edd N, Arkin M, Auld D, Austin C, Bejcek B et al., editors. 33 Assay Guidance Manual. Bethesda: Eli Lilly & Company and the National Center for Advancing Translational Sciences; 2014. Available from: https://www.ncbi.nlm.nih.gov/books/NBK53196/. 52. Rodgers KR, Chou RC. Therapeutic monoclonal antibodies and derivatives: Historical perspectives and future directions. Biotechnol. Adv. 2016;34:1149-1158. 53. Ziemba B, Franiak-Pietryga I, Pion M, Appelhans D, Munoz-Fernández MÁ, Voit B et al. Toxicity and proapoptotic activity of poly(propylene imine) glycodendrimers in vitro: considering their contrary potential as biocompatible entity and drug molecule in cancer. Int. J. Pharm. 2014; 461: 391–402. doi: 10.1016/j.ijpharm.2013.12.011. 54. Franiak-Pietryga I, Ziółkowska E, Ziemba B, Appelhans D, Voit B, Szewczyk M et al. The influence of maltotriose-modified poly(propylene imine) dendrimers on the chronic lymphocytic leukemia cells in vitro: dense shell G4 PPI. Mol. Pharm. 2013;10: 2490–2501. doi: 10.1021/mp400142p. 55. Huang YA, Kao CW, Liu KK, Huang HS, Chiang MH, Soo CR et al. The effect of fluorescent nanodiamonds on neuronal survival and morphogenesis. Sci Rep. 2014; doi:10.1038/srep06919. 56. Barok M, Joensuu H, Isola J. Trastuzumab emtansine: mechanisms of action and drug resistance. Breast Cancer Res. 2014;16:209-220. 34 35 Figure S1. NMR Measurement of PPI Dendrimer Conjugation. 13C (top) and 1H (bottom) spectrum of maltotriose-conjugated PPI dendrimers recorded on Bruker AVX-500 and DRX-500 NMR systems, respectively. Quantitative comparison of NMR peaks from the 13C spectrum indicates 83% conjugation of PPI dendrimer free amines (-H2C–NH2) with maltotriose (-H2C– NH2–R, R=maltotriose). Figure S2. Impact of Binning Choice on Distribution Presentation. (a) Histogram of simulated counts to confirm that the relative frequency/bin size reproduces the normal distribution for the untargeted ND data. Data was simulated by generating a 10,000,000 element random data set from the normal distribution fit to the untargeted ND data set (µ = 0.168 µm, σ = .770), then binned identically. Bin edges are -2.4, -2.2, -2, -1.8, -1.6, -1.4, -1.2, -1, -0.8, -0.6, -0.4, -0.2, -0.0075, 0.0075, 0.2, 0.4, 0.6, 0.8, 1, 1.2, 1.4, 1.6, 1.8, 2, 2.2, 2.4, 2.6, 2.8, 3. Counts were then divided by 10,000,000 and this relative frequency was divided by the bin width. (b) Histogram of simulated counts to confirm that the relative frequency/bin size reproduces the normal distribution for the unbound targeted ND data. Data was simulated by generating a 10,000,000 element random data 36 set from the normal distribution fit to the unbound targeted ND data set (µ = -.101 µm, σ = .796). This data set was the targeted ND distances with the bound fraction removed consisting of 108. All 10,000,000 elements were binned identically to the entire targeted ND data set. Bin edges are -2.4, -2.2, -2, -1.8, -1.6, -1.4, -1.2, -1, -0.8, -0.6, -0.4, -0.2, -0.0075, 0.0075, 0.2, 0.4, 0.6, 0.8, 1, 1.2, 1.4, 1.6, 1.8, 2, 2.2, 2.4, 2.6, 2.8, 3. Counts were then divided by 10,000,000 and the central bin (bound fraction) was discarded. Figure S3. Scored Targeted ND Image. Example of image with scoring from targeted ND image set. Only intracellular NDs were included. 37 Table S1. Poisson Statistics for Labeling of NPCs Table S2. NPC-ND Distances Measured from TEM Images. 38
1503.06507	1	1503	2015-03-23T01:38:57	Pitch Perfect: How Fruit Flies Control their Body Pitch Angle	[ "physics.bio-ph", "q-bio.QM" ]	Flapping insect flight is a complex and beautiful phenomenon that relies on fast, active control mechanisms to counter aerodynamic instability. To directly investigate how freely-flying D. melanogaster control their body pitch angle against such instability, we perturb them using impulsive mechanical torques and film their corrective maneuvers with high-speed video. Combining experimental observations and numerical simulation, we find that flies correct for pitch deflections of up to 40 degrees in 29 +/- 8 ms by bilaterally modulating their wings' front-most stroke angle in a manner well-described by a linear proportional-integral (PI) controller. Flies initiate this corrective process after only 10 +/- 2 ms, indicating that pitch stabilization involves a fast reflex response. Remarkably, flies can also correct for very large-amplitude pitch perturbations--greater than 150 degrees--providing a regime in which to probe the limits of the linear-response framework. Together with previous studies regarding yaw and roll control, our results on pitch show that flies' stabilization of each of these body angles is consistent with PI control.	physics.bio-ph	physics	Pitch Perfect: How Fruit Flies Control their Body Pitch Angle Samuel C. Whitehead1∗, Tsevi Beatus1∗, Luca Canale2, and Itai Cohen1 1Department of Physics, Cornell University, Ithaca, New York 14853, USA; 2D´epartement de M´ecanique, ´ Ecole Polytechnique 911128, Palaiseau, France. ∗ Equal contributors (Dated: October 5, 2018) Flapping insect flight is a complex and beautiful phenomenon that relies on fast, active con- trol mechanisms to counter aerodynamic instability. To directly investigate how freely-flying D. melanogaster control their body pitch angle against such instability, we perturb them using im- pulsive mechanical torques and film their corrective maneuvers with high-speed video. Combining experimental observations and numerical simulation, we find that flies correct for pitch deflections of up to 40◦ in 29 ± 8 ms by bilaterally modulating their wings' front-most stroke angle in a manner well-described by a linear proportional-integral (PI) controller. Flies initiate this corrective process after only 10 ± 2 ms, indicating that pitch stabilization involves a fast reflex response. Remarkably, flies can also correct for very large-amplitude pitch perturbations -- greater than 150◦ -- providing a regime in which to probe the limits of the linear-response framework. Together with previous stud- ies regarding yaw and roll control, our results on pitch show that flies' stabilization of each of these body angles is consistent with PI control. I. Introduction From walking humans to flying insects, many fascinat- ing forms of bio-locomotion are contingent upon robust stabilization control. Implementing this control is par- ticularly difficult in the case of small, flapping-wing in- sects, since flapping flight is inherently subject to rapidly- divergent aerodynamic instabilities [1 -- 13]. As such, fly- ing insects have evolved stabilization techniques relying on reflexes that are among the fastest in the animal king- dom [14] and robust to the complex environment that insects must navigate [15 -- 19]. In particular, pitching instability is a prominent obsta- cle for flight control in flapping insects. Analytical and numerical modeling suggest that, for many two-winged insects (e.g. flies), periodic flapping couples with longi- tudinal body motion to produce rapidly-growing oscilla- tions of the body pitch angle [4, 11, 13, 20]. This oscilla- tory instability can be understood as the result of differ- ential drag on the wings due to longitudinal body motion [4, 11]. For example, if a fly pitches down while hovering, its re-directed lift propels its body forward, causing an increased drag on the wings during the forward stroke relative to the backward stroke. Because dipteran wings are attached to their bodies above the center of mass, this drag asymmetry generates a torque that pitches the fly up. Rather than acting as a restoring torque, the drag -- together with the body inertia -- pitches the fly up, beyond its initial pitch orientation. The fly then begins to move backwards, and oscillation ensues in the oppo- site direction. This mechanism results in an undulating instability of the body pitch angle, which doubles over a timescale of ∼ 9 wing-beats [13]. Mitigating the effects of this instability requires flies to actively adjust their wing motion on time scales faster than the growth of these oscillations. ture on insect flight control, a sizable portion of which addresses the pitch degree of freedom. Experimental studies subjecting tethered insects to both mechanical pitching perturbations and visual pitching stimuli [1, 21 -- 25] have elucidated stereotyped kinematic responses for pitch correction, including manipulation of wing stroke angle, stroke plane orientation, wing-beat frequency, and body configuration. However, tethered insects do not constitute a closed-loop feedback system, since changes to their wing kinematics do not affect their body orienta- tions. Moreover, in the case of tethered flies, it has been shown that the wing kinematics are qualitatively different than those in free-flight [26, 27]. Thus, free-flight stud- ies are necessary for a comprehensive understanding of pitch control. Significant analysis has been performed on freely-flying insects executing voluntary maneuvers [28 -- 30] or responding to visual stimuli [31 -- 33], but the gen- eral challenge of systematically inducing mechanical per- turbations on untethered insects has traditionally been a barrier to the study of stabilization reflexes. Some no- table exceptions to this include methods of mechanical perturbation using air-flow vortices [15, 17, 18] or gusts of wind [19]. However, such fluid-impulse methods are difficult to tune, and are thus not ideal for of inducing the fast, precise mechanical perturbations that are re- quired for a quantitative understanding of pitch control. To achieve the necessary speed and precision for mea- surements of body pitch control, we use a perturbation scheme that has previously been applied to analyzing control of the yaw [34] and roll [14] degrees of freedom. We mechanically perturb free-flying D. melanogaster by gluing small magnetic pins to their dorsal thoracic sur- faces and applying short bursts (5-8 ms) of a vertical magnetic field that pitches their bodies up or down. As the flies correct their orientation, we measure their body and wing kinematics using high-speed video (Figure 1a). Our work builds upon an already rich corpus of litera- We recorded perturbation events with amplitudes typ- arXiv:1503.06507v1 [physics.bio-ph] 23 Mar 2015 ically ranging 5-40◦ for both pitching up and pitching down. For these perturbations, flies recover 90% their pitch orientation within ∼30 ms. Moreover, we find that the corrective process is initiated ∼2 wing-beats (∼ 10 ms) after the onset of the impulsive torque; such a short latency time indicates that this corrective process is a re- flexive behavior largely governed by input from the hal- teres, the flies' rate-gyro-like mechanical sensing organs [22, 23, 35]. To generate corrective pitching torques, flies bilater- ally modulate their wings' front-most stroke angle, i.e. they flap their wings more or less in the front to pitch up or down, respectively. This corrective mechanism is in general agreement with previous findings on active body pitch stabilization in fruit flies [20, 21, 23, 36]. We show that flies' modulation of front stroke angle is well- modeled by a linear, continuous, proportional-integral (PI) controller with a time delay, ∆T = 6 ± 1.7 ms (mean ± standard deviation). Our results indicate that pitch control in fruit flies is an extremely fast and ro- bust process, which can be accurately modeled by a sim- ple controller for a wide range of perturbation ampli- tudes. Moreover, we find that flies are capable of cor- recting for pitch deflections of 150◦ or more, a perturba- tion regime in which the linear controller theory begins to break down. Together with previous results on how flies control yaw [34] and roll [14], the analysis of pitch control presented here addresses a missing piece in our understanding of how flies control each of their body an- gles individually. II. Methods A. Fly Preparation We perform each experiment using common fruit flies (Drosophila melanogaster, females) from an out-bred lab- oratory stock. Individual flies are anesthetized at 0-4◦C, at which point we carefully glue 1.5-2 mm long, 0.15 mm diameter ferromagnetic pins to their notum (dorsal tho- racic surface), oriented so that the pin lies in the flies' sagittal plane. The pin is shown in Figure 1a (false- colored blue) and Figure 2 (images). Control experiments with untreated flies show that the addition of the pin does not qualitatively alter flies' flight kinematics. When at- tached, the pins add ∼20% to the mass and pitch mo- ment of inertia of the fly, which falls within the range of their natural body mass variations. Moreover, the pin contributes negligibly to off-diagonal components of the flies' inertia tensors, and therefore does not introduce any coupling between the rotational degrees of freedom of the body. The primary effect of the added pin mass is a dorsal shift of ∼0.2 mm in the flies' center of mass, which must be accounted for during calculations of aero- dynamic torque. 2 FIG. 1: Pitch perturbation and correction. (a) snapshots and a 3D model reconstruction from a representative pitch up event at t = -25.5, 13, 56.5 ms. For the full video of this event, see Movie S1. The middle snapshot (t = 13 ms) corresponds to the maximum pitch up deflection. Ferromagnetic pin is false-colored blue, the fly's long body axis is given by the green arrows, and the red arrow indicates the direction of the perturbation. (b) and (c) definitions for the body and wing Euler angle coordinates, respectively. ϕb, θb, and ρb indicate the body Euler angles, while φw, θw, and ηw indicate wing Euler angles; the stroke plane is shown in gray (c). Also shown are the lab (b) and body (c) frames of reference. (d-f) time series of body Euler angles for 18 perturbation events, with the highlighted curves corresponding to the event from (a). The yellow bar in (d-f) gives the timing of the magnetic pulse (0-5.8 ms). Body angles in (d-f) are spline-smoothed from raw data, with body yaw in (e) shifted by its value at t = 0. B. Videography and Mechanical Perturbation Once 15-30 flies have been prepared as above, we re- lease them into a transparent cubic filming chamber of side length 13 cm. On the top and bottom of the chamber are attached two horizontally-oriented Helmholtz coils, which produce a vertical magnetic field. The central re- gion of the chamber is filmed by three orthogonal high- speed cameras (Phantom V7.1) at 8000 frames per sec- 50 0 -50 0 20 Time0[ms] 40 Angle,00000[deg] Body0Roll0 20 0 -20 Angle,00000[deg] Body0Yaw0 f) Magnetic0 Pulse 80 60 40 20 Angle,00000[deg] Body0Pitch0 e) d) a) b) xb roll000 xlab c) zb zlab yaw000 ylab pitch000 xb yb Stroke0Angle ond. The cameras are calibrated using a direct linear transformation scheme detailed in [37, 38]. When flies enter the filming volume, an optical trigger simultane- ously signals the cameras to record and supplies a 5-8 ms current pulse to the Helmholtz coils. We varied both the strength and duration of the magnetic pulse produced by the coils across experiments. Maximal field strengths reached ∼10−2 Tesla, and most experiments were per- formed with a pulse that lasted 5.8 ms. The magnetic pulse exerts a torque on the ferromagnetic pin, pitching the fly either up or down. Of the movies collected using the above method, we se- lected 18 to analyze in full; 16 additional movies were par- tially analyzed to collect more data on pitch correction time and to observe correction for very large-amplitude perturbations (∼ 150◦). We chose movies to fully ana- lyze based on the criteria that i) the time window dur- ing which the fly is in the field of view for all three cameras is sufficiently long to observe pre- and post- perturbation kinematics, ii) the perturbation primarily affects the fly's pitch orientation, iii) the fly is not per- forming any maneuver other than correction, and iv) we sample a wide range of perturbation magnitudes for both pitching up and pitching down across our data set. To glean kinematic data from the raw footage, we use a custom-developed image analysis algorithm detailed in [14, 30]. This 3D hull reconstruction algorithm provides a kinematic description of 12 degrees of freedom for the fly (body orientation and center of mass position, as well as three Euler angles for each wing). III. Results A. Body and Wing Kinematics During Pitch Correction Representative kinematics for perturbation events are shown in Figures 1 and 2. Figure 1b shows definitions for the body Euler angle coordinates -- yaw (ϕb), roll (ρb), and pitch (θb). Figures 1d-f show time series of these Euler angles before and after the application of a 5.8 ms magnetic pulse (yellow strip) for 18 perturbation events. Before the perturbation, flies typically maintain a pitch angle of roughly 50◦. Perturbations deflect the pitch an- gle by as much as 40◦ either up or down. While the flies' yaw and roll angles are sometimes altered by the pertur- bation, pitch is the most consistently and significantly affected degree of freedom immediately following the ap- plication of the pulse (at t ≈ 6 ms). Highlighted curves in Figures 1d-f show an event in which the fly was pitched up by 25◦, attaining its maximal angular deflection at 15 ms after the onset of the perturbation. By 29 ms it has corrected for 90% of the pitch deflection. The maximum pitch velocity due to the perturbation was 2400 ◦/s. For the full movie of this perturbation event, see Movie S1. The wing kinematics for two representative perturba- tions, one in which the fly is pitched up by 25◦ (the 3 same event highlighted in Figures 1d-f) and another in which the fly is pitched down by 23◦, are shown in Fig- ure 2. Wing Euler angles -- stroke (φw), pitch (ηw), and deviation (θw) -- are defined in Figure 1c. In general, the wing kinematics we observe following the perturbation are left/right symmetric. Hence, these kinematics can be attributed to pitch correction, since both yaw [34] and roll [14] correction require left/right asymmetric wing motion. For the pitch up event, ∼10 ms, or 2 wing- beats, after the onset of the perturbation, the minima of the wing stroke angles shift upward for both the left and right wing, as indicated by the orange arrows in Figure 2a. During a given wing-beat cycle, the minimum of the stroke angle for each wing corresponds to its front-most position. Since pitch correction is left/right symmetric, we refer to the average of the front-most positions for the left and right wings as the front stroke angle, φfront . By 15 ms, the fly in Figures 2a-d has increased its φfront from its pre-perturbation value by ∼25◦. Physically, this means that the fly is significantly reducing the amplitude of its ventral stroke, i.e. flapping less forward. The du- ration of this increase in φfront is 3 wing-beats. We do not observe any shifts in the fly's back-most stroke angle, φback w , during the correction maneuvers. Conversely, for the pitch down event in Figures 2e-h, ∼10 ms after the perturbation onset the pitched-down fly begins to decrease its φfront . This corresponds to the fly increasing the amplitude of its ventral stroke, i.e. flap- ping further forward. Again, there appears to be little to no change in the back stroke angle. Put together, these results are consistent with previous kinematic mea- surements [11, 21, 23, 36] and suggest that flies modu- late their front stroke angle to produce corrective pitch torques, increasing φfront to pitch themselves down, and decreasing it to pitch themselves up. w w w w w B. Aerodynamic Forces and Torques Intuitively, the relationship between front stroke angle and pitching torque can be understood as follows. To within a good approximation, the net aerodynamic force generated by a flapping wing is directed perpendicular to the wing's surface [39, 40], so that portions of the wing stroke during which the wing is in the front half of the stroke plane (φw ≤ 90◦) generate pitch up torques, while portions in the back half (φw ≥ 90◦) generate pitch down torques. During non-maneuvering flight, these torques cancel over a wing stroke. By biasing a wing stroke so that it spends a smaller fraction of the stroke period in the forward position, a smaller pitch up torque is gen- erated during that cycle, such that the net pitch torque will be directed downward. Conversely, by increasing the front stroke angle, and thus increasing the portion of the stroke spent in the front position, flies can generate a net pitch-up torque over the course of a full stroke. This can be observed in Figures 2 (orange arrows) and 3, in which active adjustments to front stroke angle result in 4 FIG. 2: Wing kinematics and aerodynamic torques for two representative perturbation events. (a-d) correspond to a pitch up perturbation with maximum amplitude 25◦; (e-h) correspond to a pitch down perturbation with maximum amplitude 23◦. Plotted wing kinematics as a function of time include: stroke angle (a),(e); wing pitch angle (b),(f); and deviation angle (c),(g). Instantaneous aerodynamic torques about the flies' pitch axis are given in (d),(f). Orange arrows in (a) and (e) highlight corrective front strokes, with corresponding arrows in (d) and (h) highlighting the changes in pitch torque resulting from the corrective kinematics. Images above (a) and (d) show side views of the flies (raw data) at different points during the movie, illustrating the changes in body pitch that accompany a perturbation. White and gray bars indicate forward and back strokes respectively; yellow bar corresponds to the perturbation duration (0 - 5.8 ms). net corrective pitching torques over the course of individ- ual wing-beats. Modulating the balance of two opposing forces or torques is used by flies in other maneuvers such as forward [41] and sideways [30] flight, as well as yaw [34] and roll [14] maneuvers, and appears in other bio- locomoting systems, such as knifefish [42]. To quantify the effect of changing the front stroke an- gle, we calculate the pitching torque generated by the wings during the maneuvers shown in Figure 2 using the full 3D fly kinematics. To calculate the aerodynamic force generated by the wings, we used a quasi-steady aerodynamic force model that was previously calibrated on a mechanical, scaled-up fly model [39, 40]. This model gives the lift (FL) and drag (FD) forces generated by the wings as: FL = 1 FD = 1 2CL(α) 2CD(α) 2 ρ0SU 2 2 ρ0SU 2 t r2 t r2 (1) (2) CL and CD are the wing's lift and drag coefficients re- spectively, and are given as functions of angle of attack (α) by [40]; S is the wing area, r2 2 the non-dimensional second moment of wing area (given as 0.313 by [43]), ρ0 the density of air, and Ut the wingtip velocity. Drag is directed anti-parallel to the wing tip velocity, and lift is perpendicular to both drag and the wing span vec- tor. The total aerodynamic force is the vector sum of the lift and drag forces. While this is a simple method for calculating forces on flapping wings, we find that it Pitch Up Perturbation of 25° Pitch Down Perturbation of 23° -2 -10 0 10 20 Timer[ms] 30 40 pitch down -10 0 10 20 Timer[ms] 30 40 e) back front f) g) h) pitch up 170 130 90 50 10 180 140 100 60 20 40 20 0 02 Angle,rrrrrr[deg] WingrStroke Angle,rrrrrr[deg] WingrPitch Angle,rrrrrr[deg] WingrDeviation rrrrr[10-9rN⋅m] PitchrTorquer a) b) c) d) 5 FIG. 4: The magnitude of body pitch angle change resulting from different combinations of wing Euler angles for both a pitch up and pitch down event. Individual points correspond to unique combinations of pre- and post-perturbation wing angle kinematics, which are used with our quasi-steady aero- dynamic model to calculate pitching torques, which in turn are used to estimate pitch angle deflection over a wing-stroke. Wing kinematics φw (stroke), θw (deviation), and ηw (wing pitch) are defined in Figure 1c. Points above the axis corre- spond to kinematics taken from the pitch up event in Figures 2 and 3; points below the axis correspond to kinematics taken from the pitch down event in Figures 2 and 3. ing motion, mitigating the overshoot in body pitch angle caused by the initial correction response. This allows for faster correction times, since the initial corrective maneu- ver can generate larger torques, and thus more quickly return the fly to pitch angles near its original orientation. In movies that allowed us to track the fly for long times after the perturbation, we observe that the front stroke angle and the net aerodynamic torque often oscillate with decaying amplitude and a period of ∼3-4 wing-beats. Importantly, passive damping of pitch motion con- tributes negligibly to the correction maneuvers we ob- serve, since the characteristic passive decay time for pitch velocity is & 150 ms, much longer than the entire correc- tion maneuver [44]. Taken together, our results indicate that pitch correction for flies is an active process involv- ing modulation of phifront . w C. The Importance of Stroke Angle Relative to Other Degrees of Freedom To assess the effect of front stroke angle modulation on body pitch correction, we calculate the changes to body pitch angle generated by changes in wing kinemat- ics. We isolate the corrective effect of each wing kine- matic variable by first identifying wing-strokes that cor- respond to both non-maneuvering (no net torque) and corrective flight. We then separate out the kinematic variables for each type of flight, and calculate the changes to body pitch angle resulting from different combinations of corrective and non-corrective kinematics, shown by the color codes in Figure 4. For example, a point with color combination blue-blue-red corresponds to a wing- beat with wing pitch and deviation angles taken from the corrective maneuver and the stroke angle taken from FIG. 3: Mechanism for generating corrective pitch torques. Data in (a),(b) corresponds to the pitch up event shown in Figures 1d-f and 2a-d, while (c),(d) correspond to the pitch down event shown in Figures 2e-h. (a),(c) wing- stroke-averaged pitch torque (black line, spline interpolated) and negative front stroke angle deviation (orange dots) vs time. Front stroke angle deviation is defined by ∆φfront (t) = φfront to illustrate correlation with torque. (b),(d) body pitch angle (red line) vs time. As in previous plots, the yellow strip in (a)-(d) cor- responds to the duration of the magnetic pulse. (0). We flip the sign of ∆φfront w (t) − φfront w w w quantitatively captures the relevant force production for both pre- and post-perturbation wing kinematics. We tested the effect of adding rotational forces to the aero- dynamic model [40], which should give the next largest contribution to force, but found negligible changes to our calculation results. From these lift and drag forces, we calculate aerody- namic torques exerted by the wings on the body, shown in Figures 2d,h. The moment arm for the torque is given by the vector from the fly's center of mass to the wing's center of pressure, assumed to be in the chord center, 70% along the length of the wing's span. In Figure 2d, during the active correction, the fly's wings generate a net downward pitching torque to oppose the perturba- tion. The main effect we observe is that, during a given wing cycle the wings generate less upward torque, result- ing in a net downward bias for the torque on the body. This net pitch down torque is highlighted by the orange arrows in Figure 2d. Similarly, the pitched down fly in Figure 2h generates net upward pitching torque during active correction, also highlighted by orange arrows. The corrective torques for these two events are well-correlated with the measured modulations of front stroke angle (Fig- ure 3a,c), a trend that we observe across all perturbation events. Interestingly, after the flies generate a corrective torque for 2-3 wing-beats, we also observe a few wing-beats in which they generate net torque in the opposite direction (Figure 2d,h and 3a,c). As with the initial corrective torque, this subsequent counter-torque arises from mod- ulations of the front stroke angle, evident in Figures 3a,c. The counter-torque acts to brake the corrective pitch- Time-AveragedvBodyvv Torquevvvvvvvv[10-9vN⋅m] 036 -3 0 50 Timev[ms] 100 c) 20 10 0 03 -3 -6 -10 d) 10 0 -10 -20 -30 -50 0 Timev[ms] 50 10 0 -10 -20 20 10 0 -10 vvvvvvvvvvvvv[deg] vvStrokevAnglevv v-ΔvWingvFrontv a) vvvvvv[deg] ΔvBodyvPitchvAnglev b) CorrectiveaaΔaBodyaPitchaAngle ForaMixedaWingaKinematicsa =acorrection =apre-correction 0 1 2 3 4 5 6 non-maneuvering flight. Our calculation uses the aerody- namic model detailed above to determine the torques pro- duced by a given set of wing kinematics, assuming rigid wings attached to a stationary body at a point above the body center of mass. From the calculated torques, we determine the net change in body pitch angle over the course of each wing-stroke using numerical integration. We perform this analysis for all 8 possible combina- tions of wing kinematics for data from two different per- turbation events, the pitch up and pitch down events in Figures 2 and 3. Points corresponding to combinations of wing kinematics from the pitch up and pitch down events are shown above and below the axis, respectively. In both cases, the sign for ∆θb is chosen so that positive indicates a corrective rotation. The grouping of points in Figure 4 indicates that body pitch correction is most closely as- sociated with changes to wing stroke angle. The blue- blue-red point, corresponding to corrective stroke angle but non-maneuvering wing pitch and rotation, achieves at least 60% of the correction to body pitch angle, con- sistent with [33]. Moreover, the only combinations that give more than 30% of the total correction have correc- tive stroke angle (points of the form x-x-red). Changes to wing pitch or deviation can contribute ∼ 40% to the corrective process, but cannot alone account for correc- tive body kinematics. These results motivate a minimal model for body pitch stabilization that considers only variations in front stroke angle to drive pitch correction. D. The Corrective Effect of Stroke Angle Over a Range of Perturbations w To further flesh out the relationship between front stroke angle and corrective torque, we plot the maximum measured corrective pitch acceleration generated by the fly in each maneuver as a function of the correspond- ing change in front stroke angle measured at that time, ∆φfront (Figure 5a). The maximum pitch acceleration was measured at the extremum of θb, using a quadratic polynomial fit. The plot demonstrates a strong correla- tion between changes in front stroke angle and correc- tive acceleration (linear R2 = 0.87). Consistent with the two maneuvers in previous figures, Figure 5a shows that, across our data set, flies increase ∆φfront (flap less for- ward) to pitch themselves down, and decrease ∆φfront (flap further forward) to pitch themselves up. w w w The correlation between ∆φfront and corrective pitch acceleration in Figure 5a is also predicted by a calcula- tion based on the quasi-steady aerodynamic force model in Equations 1,2 (Figure 5a, gray line). To calculate the aerodynamic pitch torques, we use a simplified wing kine- matic model similar to that in [20] in which only the front stroke angle is varied (see Supplementary). We average the computed torques over a wing-beat, and divide by the moment of inertia to obtain pitch acceleration. The calculation relies only on the wing kinematics and fly morphology [43], and has no fitted parameters. More- 6 over, we verified that the resultant acceleration is insen- sitive to body pitch rotations, suggesting that passive damping contributes negligibly to correction [44]. The results of our calculation, shown in Figure 5a, quantita- tively reproduce the measured pitch acceleration, further corroborating a model for pitch control that includes only modulation of φfront . w To rule out an alternative corrective mechanism, based on modulation of back stroke angle, we plot corrective pitch acceleration as a function of φback w , shown in Figure 5b. We find no discernible correlation between these two variables. Calculating aerodynamic forces predicts that changes to the back stroke angle could produce corrective pitching torques in the same way that changes to front stroke angle do; the fact that we do not observe this in the data hints that morphological constraints favor modulation of the front stroke angle. E. Correction Timescales w We also analyze the pitch correction timescales across our data set. Figure 5c shows a histogram of latency times for each corrective maneuver, defined as the time between the onset of the perturbation and the first mea- > 4◦). surable change in the front stroke angle (∆φfront The mean latency time is 9.9 ± 2.1 ms corresponding to ∼ 2 wing-beats (mean ± standard deviation, n = 18). Figure 5d plots the total correction time for each maneu- ver as a function of the maximum body pitch deflection in each perturbation event. We define the correction time as the time between the onset of the perturbation and the fly's correcting 90% of the pitch deflection. The median correction time is 29 ± 8 ms (mean ± standard devia- tion, n = 32). Finally, we find that the correction time is weakly correlated with the perturbation amplitude (lin- ear R2 = 0.093), which is consistent with a linear control model. F. Control-Theory Model We use a control-theoretic framework to describe flies' method for pitch stabilization. In particular, we model actuated changes to the front stroke angle as the output of a proportional-integral (PI) controller with time delay ∆T , for which the input is body pitch velocity (block diagram in Figure 6a). The response ∆φfront is given by: w ∆φfront w (t) = Kp θb(t − ∆T ) + Ki∆θb(t − ∆T ) (3) w Equation 3 states that adjustment of the front stroke angle (∆φfront ) at a given time t is given by a linear combination of the body pitch angle deviation from non- maneuvering orientation (∆θb) and body pitch velocity ( θb) at an earlier time t − ∆T . The parameters Kp and Ki are the proportionality constants that determine the relative weights of body pitch angle and pitch velocity. 7 FIG. 6: Control theory model. (a) A block diagram for the PI controller model. The effect of external torque (τext) is sensed by the halteres as angular velocity. The measured body pitch velocity ( θb) is subject to a time delay (∆T ), after which the signal is split into two branches. One branch is multiplied by Ki and integrated to yield pitch displacement, while the other branch is multiplied by Kp. These signals are recombined as an output, ∆φfront , that adjusts the front stroke angle of the wings. This adjustment to front stroke angle results in a corrective wing torque (τw), which is in turn sensed by the fly. (b) Measured front stroke angle as a function of time for the pitch up event in Figures 1, 2, and 3 (orange dots) compared with the output of the fitted PI controller model (blue line). The relative contributions from the P and I terms are shown in the gray solid line and the brown dashed line, respectively. Shaded blue region corresponds to the confidence interval of the three control parameters. w w tem. An interplay between the haltere and visual sys- tems, as in [45], is necessary for comprehensive pitch stability. The PI model presented here can accurately account for the fly's fast reflex response, which stabilizes it against rapid pitch perturbations. Using measured values for ∆φfront , ∆θb, and θb, we fit for the parameters Kp, Ki, and ∆T . The three param- eters are fit for each movie individually, using one data point per wing-stroke. The fit is performed by evaluat- ing Equation 3 on a dense 3D grid in parameter space and finding the global minimum for the sum of squared residuals between the control model and real data. The results of a controller fit for the pitch up event in Fig- ures 1, 2, and 3 are shown in Figure 6b. The orange dots give the measured values of the front stroke angle, while the blue line shows the output of the fitted con- troller model. We find excellent quantitative agreement between the controller fit and our measured data with an RMS error of 1.9◦, which is on the order of the mea- FIG. 5: Statistics from many perturbation events. (a-b) Max- imum corrective pitch acceleration generated by the fly as a function of change in front stroke angle and back stroke an- gle. The gray line in (a) is the calculated pitch acceleration for simplified wing kinematics. Note that the gray line has no fit parameters, and is based purely on morphological parameters [43], wing kinematics [20], and the quasi-steady aerodynamic model. (c) A histogram of latency times across our data set, with latency time defined as the time between the onset of the magnetic perturbation and the beginning of a measur- able corrective wing response (± 4◦ change in front stroke angle). The orange background in corresponds to the mean delay time ± standard deviation obtained from our controller model fits. (d) The time for the pitch correction -- defined as the time it takes for the fly to recover 90% of its original pitch orientation -- plotted as a function of the maximum pitch de- flection for each perturbation event. Solid and dashed lines give mean ± standard deviation. The lack of discernible cor- relation in d) is a hallmark of linear control. The same controller could be termed a PD controller, if the input to the controller were the body pitch angle. Be- cause the fly halteres are known to measure body angular velocities [22, 23, 35], we choose the PI nomenclature. We exclude controller models that depend on angular accel- eration (PID) based on previous studies that have shown flies' corrective pitch response to be insensitive to angu- lar acceleration [23]. Further analysis of the data in [23] confirms this model framework (see Supplementary). Importantly, a PI controller model with only angular velocity as an input cannot account for pitch stabiliza- tion on long timescales, due to integration errors affect- ing measurement of the absolute pitch angle. Controlling pitch on long timescales requires direct measurement of the pitch angle, as could be achieved by the visual sys- 5 40 w5 w10 ΔmBackmStrokemAnglefm 0 back mmmmmmmmmmmmm[deg] ΔΦwm 20 10 30 Max8mBodymPitchm DeflectionfmΔθbfmaxmm[deg] 036 w3 w6 b) mθbmm[105mdeg/s2] BodymPitchmAcceleration Data Theory a) 036 w3 w6 θbmm[105mdeg/s2] BodymPitchmAcceleration 50 40 30 20 0 d) CorrectionmTimem[ms] w20 w10 0 10 20 30 ΔmFrontmStrokemAnglef front mmmmmmmm[deg] ΔΦwm 8 6 LatencymTimem[ms] 10 12 0246 4 c) Counts -10 0 10 20 TimeP[ms] 30 40 ++ Halteres - correctiveP torque Wings a) + b) ΔPFrontPStrokePAngle Data PIPCtrl PPterm IPterm 20 10 0 -10 ΔΦwP PPPPPPPP[deg] front w w w surement uncertainty for ∆φfront . The controller model not only captures the sharp rise in ∆φfront in response to the perturbation, but also the subsequent decrease in ∆φfront corresponding to the braking counter-torque that slows the fly's downward pitching motion (Section III B). The proportional (P) and integral (I) contributions to the controller model are shown in solid gray and dashed brown curves, respectively. The fast rise time of the re- sponse can be attributed to the proportional term. We apply the same fitting process to nine movies. We find the values of fitted control parameters (Table I) to be Ki = 0.3 ± 0.15, Kp = 7 ± 2.1 ms, and ∆T = 6 ± 1.7 ms (mean ± standard deviation), with an average RMS fit- ting error of 3.0◦. The mean value of ∆T corresponds to roughly 1 wing-beat, providing a lower bound for mea- sured latency times (see Discussion). Figure 5c shows the region corresponding to mean ∆T ± 1σ (highlighted in orange) compared with measured latency times. Con- fidence intervals (CI) in Table I are calculated based on a χ2 test for the fitting residuals in control parameter space. The confidence interval size is large relative to the fitted control parameters (> 50% in some cases). The large confidence intervals, combined with the accuracy of the fit, indicate that the model is robust to deviations in the controller parameters. 8 for the simulated controllers, we fit our experimental data to each model, and to mimic the perturbation conditions in our experiments we impose a 5 ms external mechanical torque on the simulated flies (yellow strip), with magni- tude comparable to our real system. For details on the simulation, see Supplementary. We find that flies with no control or I control are sub- ject to large, rapid oscillations of body pitch angle (Fig- ure 7b), while flies with P control are subject to slightly smaller, long-timescale oscillations (Figure 7a). In all three of these cases, the simulated fly fails to remain aloft and rapidly loses altitude. Among the four candi- date models, PI control is the only one that is consistent with the fast, robust pitch control that we observe exper- imentally (Figure 7a). Simulated flies implementing PI control correct their orientation over timescales similar to those in our experimental data. In contrast to the other three controller models, the simulated flies using PI con- trol maintain pitch stability over long times and remain aloft. The general features of each control scheme show little sensitivity to the values of the control parameters, in agreement with the large confidence intervals that we find for fitted control parameters in Section III F. Mov. ∆θb Ki ± CI Kp ± CI ∆T ± CI RMS Err. Num. [deg] [deg] 1 2 3 4 5 6 7 8 9 [none] [ms] 25 0.5 ± 0.24 6 ± 1.9 16 0.1 ± 0.25 8 ± 3.4 25 0.5 ± 0.19 4 ± 1.7 0.5 ± 0.23 12 ± 7 6 0.2 ± 0.61 7 ± 7.3 7 15 0.3 ± 0.24 8 ± 3.2 -24 0.3 ± 0.29 6 ± 2.4 -23 0.2 ± 0.31 6 ± 2.0 -21 0.2 ± 0.26 8 ± 2.8 [ms] 4 ± 2.1 8 ± 2.5 4 ± 1.7 4 ± 3.6 7 ± 5.8 6 ± 2.2 7 ± 2.1 9 ± 1.5 7 ± 3.6 1.9 2.9 3.2 1.7 2.5 2.7 3.7 2.9 4.5 TABLE I: Fit results for PI controller model with confidence intervals (CI) for each parameter. G. Numerical Simulation To corroborate our experimental evidence for the PI controller, we perform a dynamical simulation of a me- chanically perturbed fruit fly. The simulation solves the equations of motion for the pitch, longitudinal, and ver- tical degrees of freedom, assuming the fly's geometry, simplified wing kinematics, and the quasi-steady aero- dynamic force model detailed above. The body pitch angle over time for simulated flies implementing different control strategies is shown in Figure 7. The four simu- lated control schemes shown are: i) proportional-integral (blue), ii) proportional (green), iii) integral control (or- ange), and iv) no control (red). To determine parameters FIG. 7: Numerical simulation results for different controller models. (a) Time series of body pitch angle for simulated flies implementing proportional-integral (PI, blue) and pro- portional (P, green) control. (b) Time series of body pitch angle for simulated flies implementing integral (I, orange) and no (None, red) control. Time axes for both (a) and (b) are the same, but the range of the pitch angle axis differs significantly between the two. Gray region in (a) and (b) corresponds to 45 ± 2◦, where 45◦ is the reference body pitch angle for each controller. Hence, curves returning to and remaining within this region indicate successful control. The yellow strip in- dicates the duration of the mechanical perturbation (0 - 5 ms). -100 -200 0 50 100 150 TimeI[ms] 200 250 300 350 a) Perturbation θbI[deg] b) BodyIPitchIAngleI PII PI II None 70 60 50 40 30 20 200 100 0 θbI[deg] BodyIPitchIAngleI H. Extreme Perturbations IV. Discussion 9 In addition to the 18 perturbation events analyzed in full (Figure 1, 5), we examine two large-amplitude pertur- bation events. Snapshots and time courses of body pitch angle are shown in Figure 8 for a pitch up and a pitch down event, both with maximum pitch deflection greater than 130◦(Movie S2). Remarkably, both flies performed successful correction maneuvers, although they were not in-frame long enough to observe them returning to their original orientation. The correction time for both large- amplitude events (> 50 ms) is longer than the correction times shown in Figure 5, which can be attributed to the fact that the controlled quantity ∆φfront is biologically front stroke angle is limited, for instance, constrained: by the angle at which the body or the other wing ob- structs a wing's forward motion. If we input the body pitch kinematics for the events in Figure 8 into our PI controller model, the model predicts changes to φfront in excess of 100◦ -- a value that is physiologically impossible in the forward direction and not observed in the back- ward direction. Assuming the flies' corrective response is ≤ 30◦, a physiologically reasonable bounded by ∆φfront estimate, our numerical simulation predicts a response time of ∼70 ms for a perturbation ∆θb = −150◦, in ex- cellent agreement with the experimental data. w w w FIG. 8: Large perturbations. Overlaid snapshots from raw data and time series of body pitch angle for (a) pitch up perturbation and (b) pitch down perturbation (Movie S2). The pitched up fly reaches a maximum pitch deflection of 130◦ at ∼20 ms after the onset of the perturbation. The pitched down fly reaches a maximum pitch deflection of -155◦ after ∼30 ms. The loss of altitude during the correction is evident in both cases and shown to scale. Yellow strips indicate the 5.8 ms magnetic pulse; red arrows indicate the direction of each perturbation. A. Front Stroke Angle as the Controlled Quantity We show that front stroke angle modulation is the pri- mary mechanism for body pitch control in fruit flies, con- sistent with previous experiments [11, 21, 23] and is in the same spirit as other proposed mechanisms that include modulation of the mid-stroke angle [20]. Our compu- tational results (Figures 5 and 7) show that a minimal model, which only incorporates changes to front stroke angle, and uses control parameters extracted from fits to our measurements, is the simplest linear, continuous model capable of stabilizing the body pitch angle on time scales similar to those observed in the experiments. Kinematic variables other than φfront w may also con- tribute to pitch correction. Previous studies have asso- ciated changes to stroke plane deviation [46], wing-beat frequency [23], and body posture [36] with pitch correc- tion. In particular, we observe transient alterations in both wing pitch and deviation angle during corrective maneuvers. Figure 4 suggests that, when combined with modulation of front stroke angle, changes to wing pitch and deviation angle can account for up to 40% of body pitch correction, consistent with [33]. The detailed role of these kinematic variables in pitch control and whether they are actively or passively actuated remains unknown. B. Discrete vs. Continuous Control Models The periodic motion of wing flapping introduces inher- ent discreteness to insect flight. For processes occurring on timescales comparable to a wing-stroke period -- like the perturbations and maneuvers we reported here -- we expect discrete effects to be more pronounced. In partic- ular, modulations of front stroke angle can, by definition, occur only once per wing-beat. Because perturbations can be induced at any time during the wing-beat, but the actuated kinematics are discretely constrained, la- tency times for correction depend on the phase of the perturbation relative to the wing-stroke. Latency times will be bounded from below by the flies' neural response time, but could potentially be as much as one wing-beat longer as a result of the phase of the perturbation within the wing-beat. Measured latency times can also depend on discrete sensing. The temporal sampling resolution with which flies can measure mechanical perturbations is likely deter- mined by motion of their halteres, the rate-gyro sensory organs used in fast perturbation response [23]. Dipteran halteres beat at the wing frequency and use Coriolis forces to measure body angular velocities [22, 47]. The largest sensitivity to mechanical perturbations is likely to occur at times during the fly's mid-stroke, when Coriolis forces on the halteres are the largest [22]. Sensing at dis- crete times introduces a second relevant phase for correc- tion latency time: the phase of the perturbation relative a) b) t = -15 ms t = -1ms t = 60 ms t = 65 ms 50 0 -50 -100 60 0 20 40 Time [ms] 0 20 40 Time [ms] 200 150 100 50 θb [deg] Body Pitch Angle to sensing. Similar to discrete actuation, discrete sensing would lead to latency times longer than the neural re- sponse time. Moreover, even during the fly's mid-stroke, its halteres only have finite sensitivity. It is likely that there exists some threshold for angular velocities that are large enough to elicit a control response [48]. The continuous PI controller model does not account for the effects of discrete actuation, discrete sampling, or sensing threshold. Hence, the measured latency time should constitute an upper bound for the delay time that we obtain from the controller model. Indeed, the time delays from our controller model (∆T = 6 ± 1.7 ms) is roughly one wing-beat shorter than the measured latency times (9.9 ± 2.1 ms). Despite the inherent discreteness of the fly control sys- tems, our continuous PI controller model quantitatively captures the behavior of flies in response to pitch pertur- bations (Figure 6). This quantitative agreement leads to an interesting open question: under what conditions does it become necessary to use a discrete controller model to describe flight stabilization? To answer this question would require precise perturbation timing, in order to probe the short timescales at which discretization be- comes relevant. Such an analysis could provide signifi- cant insight into the timing and thresholding of fruit fly reflexes. C. Physiological Basis for Pitch Control Both the mechanism and timing for the pitch correc- tion indicate a likely candidate muscle for control actu- ation: the first basalare muscle (b1), as suggested by previous studies [20, 49]. Among dipteran flight control muscles, b1 is unique in that it is active during every wing-stroke [50 -- 52], which would allow for the wing-beat timescale pitch control that we observe. Moreover, b1 activity is strongly correlated with modulations of ven- tral stroke amplitude, i.e. changes in φfront [53, 54]. In blowflies, b1 activity is also correlated with changes in de- viation angle during the ventral stroke [55], which could explain the slight shifts in deviation angle that we observe during correction (Figure 2b,f). Our results indirectly support the hypothesis that the b1 muscle is responsible for pitch control through the regulation of φfront . Testing flies with disabled or altered b1 muscles could provide an avenue for confirming the role of b1 in the pitch control process. w w 10 control is a generic feature of pitch stabilization in in- sects. Beyond insects, what we refer to as PI control has also been observed in fast obstacle avoidance in pigeons [56]. Future research on the ubiquity of PI control could have fascinating implications for the evolution of flight stabilization mechanisms. Extending beyond pitch stabilization, our results, to- gether with previous studies on yaw [34] and roll [14] control in fruit flies, show that the strategies flies use to control each of their body Euler angles can be modeled as PI controllers. However, the overarching structure in which these three individual controllers are embedded is still unknown. Given the non-commutativity of rotations in 3D, the relationship between controllers that measure different angular coordinates is likely to be non-trivial. For example, in response to perturbations that simulta- neously affect both the roll and yaw degrees of freedom, previous studies observed preferential correction for roll over yaw [14]. Because roll is known to be aerodynam- ically unstable, while yaw is passively stable [7, 9, 13], the results from [14] suggest that control for different body angles may be imposed hierarchically, with pref- erence given to correcting degrees of freedom that are most unstable. However, in our data set we observe per- turbation events in which pitch is corrected prior to roll, despite roll being the more unstable degree of freedom, implying that the response to complex perturbations has amplitude dependent features. Taken together, these re- sults hint at a complex and intriguing control architec- ture used by flies to stabilize their orientation. An understanding of the relationship between control of different Euler angles could have profound implications for how the fly encodes information about its body orien- tation. In the case of vision, organism-specific demands have spurred the development of novel, specialized neural structures in both mammals [57, 58] and insects [59, 60]. Pioneering work on information processing from halteres has suggested similar morphology/function relationships for the gyroscopic rate sensing in insects [61]. Connect- ing such analyses with the resultant control structure ob- served in free-flight behavioral experiments could provide a window into the most basic ways in which flies sense and interpret the world. Acknowledgments We thank Grace (Li) Chi and Andy Clark for providing flies; Ty Hedrick for advice on camera calibration; Andy Ruina and the Cohen group for useful conversations. D. Linear Control of Body Orientation Funding In addition to the results on fruit flies reported here, PI control has also been identified in pitch control for hawkmoths [31, 32]. The anatomical similarities found across species suggest that pitch instability is an obstacle faced by many flapping insects [4, 11]; a natural question raised by these collective findings is whether or not PI This work was supported in part by an NSF DMR award (no. 1056662) and in part by an ARO award (no. 61651-EG). S.C.W. was supported by the NDSEG Fel- lowship. T.B. was supported by the Cross Disciplinary Post-Doctoral Fellowship of the Human Frontier Science Program. [1] G. K. Taylor and A. L. Thomas, Journal of Experimental Biology 206, 2803 (2003). [2] M. Sun and Y. Xiong, The Journal of experimental biol- ogy 208, 447 (2005). [3] G. K. Taylor and R. 11 [30] L. Ristroph, G. J. Berman, A. J. Bergou, Z. J. Wang, and I. Cohen, Journal of Experimental Biology 212, 1324 (2009). [31] B. Cheng, X. Deng, and T. L. Hedrick, The Journal of Zbikowski, Journal of The Royal Experimental Biology 214, 4092 (2011). Society Interface 2, 197 (2005). [4] M. Sun, J. Wang, and Y. Xiong, Acta Mechanica Sinica 23, 231 (2007). [5] H. Liu, T. Nakata, N. Gao, M. Maeda, H. Aono, and W. Shyy, Acta Mechanica Sinica 26, 863 (2010). [6] I. Faruque and J. S. Humbert, Journal of theoretical bi- ology 264, 538 (2010). [7] Y. Zhang and M. Sun, Acta Mechanica Sinica 26, 175 (2010). [32] S. P. Windsor, R. J. Bomphrey, and G. K. Taylor, Journal of The Royal Society Interface 11, 20130921 (2014). [33] F. T. Muijres, M. J. Elzinga, J. M. Melis, and M. H. Dickinson, Science 344, 172 (2014). [34] L. Ristroph, A. J. Bergou, G. Ristroph, K. Coumes, G. J. Berman, J. Guckenheimer, Z. J. Wang, and I. Cohen, Proceedings of the National Academy of Sciences 107, 4820 (2010). [35] J. W. S. Pringle, Philosophical Transactions of the Royal [8] Y.-L. Zhang and M. Sun, Acta Mechanica Sinica 27, 823 Society B: Biological Sciences 233, 347 (1948). (2011). [9] N. Gao, H. Aono, and H. Liu, Journal of Theoretical Biology 270, 98 (2011). [10] N. O. Prez-Arancibia, K. Y. Ma, K. C. Galloway, J. D. Greenberg, and R. J. Wood, Bioinspiration & Biomimet- ics 6, 036009 (2011). [11] L. Ristroph, G. Ristroph, S. Morozova, A. J. Bergou, S. Chang, J. Guckenheimer, Z. J. Wang, and I. Cohen, Journal of The Royal Society Interface 10 (2013). [12] N. Xu and M. Sun, Journal of theoretical biology 319, 102 (2013). [13] M. Sun, Rev. Mod. Phys. 86, 615 (2014). [14] T. Beatus, J. M. Guckenheimer, and I. Cohen, Journal of The Royal Society Interface 12 (2015), ISSN 1742-5689. [15] S. A. Combes and R. Dudley, Proceedings of the National Academy of Sciences 106, 9105 (2009). [16] A. K. Dickerson, P. G. Shankles, N. M. Madhavan, and D. L. Hu, Proceedings of the National Academy of Sci- ences 109, 9822 (2012). [17] S. Ravi, J. D. Crall, A. Fisher, and S. A. Combes, The Journal of Experimental Biology 216, 4299 (2013). [18] V. M. Ortega-Jimenez, J. S. Greeter, R. Mittal, and T. L. Hedrick, The Journal of experimental biology 216, 4567 (2013). [19] J. Vance, I. Faruque, and J. Humbert, Bioinspiration & biomimetics 8, 016004 (2013). [20] S. Chang and Z. J. Wang, Proceedings of the National Academy of Sciences 111, 11246 (2014). [21] J. Zanker, Philosophical Transactions of the Royal Soci- ety of London. B, Biological Sciences 327, 43 (1990). [22] G. Nalbach, Neuroscience 61, 149 (1994). [23] M. H. Dickinson, Philosophical Transactions of the Royal Society of London.Series B: Biological Sciences 354, 903 (1999). [24] A. Sherman and M. H. Dickinson, Journal of experimen- tal biology 207, 133 (2004). [25] A. Sherman and M. H. Dickinson, Journal of Experimen- tal Biology 206, 295 (2003). [26] S. N. Fry, R. Sayaman, and M. H. Dickinson, Journal of Experimental Biology 208, 2303 (2005). [27] J. A. Bender and M. H. Dickinson, The Journal of ex- perimental biology 209, 3170 (2006). [36] G. K. Taylor, Biological Reviews 76, 449 (2001). [37] M. A. Lourakis and A. Argyros, ACM Trans. Math. Soft- ware 36, 1 (2009). [38] D. H. Theriault, N. W. Fuller, B. E. Jackson, E. Bluhm, D. Evangelista, Z. Wu, M. Betke, and T. L. Hedrick, The Journal of Experimental Biology 217, 1843 (2014). [39] M. H. Dickinson, F.-O. Lehmann, and S. P. Sane, Science 284, 1954 (1999). [40] S. P. Sane and M. H. Dickinson, Journal of Experimental Biology 204, 2607 (2001). [41] L. Ristroph, A. J. Bergou, J. Guckenheimer, Z. J. Wang, and I. Cohen, Phys. Rev. Lett. 106, 178103 (2011). [42] S. Sefati, I. D. Neveln, E. Roth, T. R. Mitchell, J. B. Snyder, M. A. MacIver, E. S. Fortune, and N. J. Cowan, Proceedings of the National Academy of Sciences 110, 18798 (2013). [43] B. Cheng, S. Fry, Q. Huang, W. Dickson, M. Dickinson, and X. Deng, in Robotics and Automation, 2009. ICRA '09. IEEE International Conference on (2009), pp. 1889 -- 1896, ISSN 1050-4729. [44] B. Cheng and X. Deng, pp. 39 -- 44 (2010). [45] S. J. Huston and H. G. Krapp, The Journal of Neuro- science 29, 13097 (2009). [46] J. M. Zanker, Physiological Entomology 13, 351 (1988), ISSN 1365-3032. [47] J. Pringle, The Journal of physiology 108, 226 (1949). [48] J. Fox and T. Daniel, Journal of Comparative Physiology A 194, 887 (2008), ISSN 0340-7594. [49] A. Fayyazuddin and M. H. Dickinson, Journal of Neuro- physiology 82, 1916 (1999), ISSN 0022-3077. [50] G. Heide, BIONA-report 2, 35 (1983). [51] G. Heide and K. G. Gotz, The Journal of experimental biology 199, 1711 (1996). [52] J. A. Miyan and A. W. Ewing, Philosophical Transac- tions of the Royal Society B: Biological Sciences 311, 271 (1985). [53] M. H. Dickinson and M. S. Tu, Comparative Biochem- istry and Physiology Part A: Physiology 116, 223 (1997). [54] S. M. Walker, D. A. Schwyn, R. Mokso, M. Wicklein, T. Mller, M. Doube, M. Stampanoni, H. G. Krapp, and G. K. Taylor, PLoS Biol 12, e1001823 (2014). [55] C. N. Balint and M. H. Dickinson, Journal of experimen- [28] A. R. Ennos, Journal of Experimental Biology 142, 49 tal biology 207, 3813 (2004). (1989). [56] H.-T. Lin, I. G. Ros, and A. A. Biewener, Journal of The [29] S. N. Fry, R. Sayaman, and M. H. Dickinson, Science Royal Society Interface 11, 20140239 (2014). 300, 495 (2003). [57] T. Hafting, M. Fyhn, S. Molden, M.-B. Moser, and E. I. Moser, Nature 436, 801 (2005). [58] M. M. Yartsev, M. P. Witter, and N. Ulanovsky, Nature 479, 103 (2011). [59] T. A. Ofstad, C. S. Zuker, and M. B. Reiser, Nature 474, 204 (2011). [60] J. D. Seelig and V. Jayaraman, Nature 503, 262 (2013). [61] J. L. Fox, A. L. Fairhall, and T. L. Daniel, Proceedings of the National Academy of Sciences 107, 3840 (2010). 12 Pitch Perfect: How Fruit Flies Control their Body Pitch Angle Supplementary Information Samuel C. Whitehead1∗, Tsevi Beatus1∗, Luca Canale2, and Itai Cohen1 1Department of Physics, Cornell University, Ithaca, New York 14853, USA; 2D´epartement de M´ecanique, ´ Ecole Polytechnique 911128, Palaiseau, France. ∗ Equal contributors (Dated: October 5, 2018) I. Movies Movie 1:- A fruit fly undergoing a typical pitch up perturbation and correction maneuver, corresponding to the data presented in Figures 1-4 and 6 in the main text. The side panels of the 3D box correspond to raw movies from the three orthogonal high-speed cameras. The green line, and its white projection onto the floor of the 3D box, show the center-of-mass trajectory of the fly, with red regions of the line representing the duration of the magnetic pulse. The timing for the event, in milliseconds, is given in the bottom left corner. If not attached, the video can be found at https://youtu.be/2FGj7HUCL7E. Movie 2:- Raw footage of three camera views of a fly undergoing a large-amplitude pitch down perturbation, corresponding to the data presented in Figure 8. The center panel corresponds to the overhead view, while the left and right panels correspond to side views. The timing for the event, in milliseconds, is given in the bottom left corner. The magnetic perturbation is applied from 0-5.8 ms. If not attached, the video can be found at https: //youtu.be/EUIE9jRcusg. II. Simplified Wing Kinematics For both the calculation of aerodynamic torques in Figure 5a and the dynamical simulation in Figure 7, we use an analytic form for simplified wing kinematics taken from [1]. These kinematics closely resemble the motion of real fly wings, but are simple enough to write down concisely. φw(t) = φ0 + φm asin(Ksin(ωt)) asin(K) ηw(t) = η0 + ηm tanh(Csin(ωt + δη)) tanh(C) θw(t) = θ0 + θmcos(2ωt + δθ) (S1) (S2) (S3) The wing Euler angles are defined in Figure 1c in the main text. The terms in Equations S1, S2, and S3 are defined as follows: • φ0, η0, θ0 are angle offsets. • φm, ηm, θm are amplitudes. • K, C are waveform parameters. K tunes the stroke angle from pure sine wave to triangle wave, while C tunes the wing pitch angle from sinusoid to square wave. • ω is the wing-beat frequency. • δη, δθ are phase offsets. For the calculation in Figure 5a we use ω = 250 × (2π) rad/s; ηm = 53◦; η0 = 90◦ ; δη = 72.4◦ ; C = 2.4; θm = 25◦; θ0 = 0◦; δθ = 90◦; K = .7; φm ∈ [60◦, 87.5◦]; φ0 ∈ [87.5◦, 115◦]. For the simulation in Figure 6, we use similar parameters to the above, but set θm = θ0 = 0◦ for simplicity and vary φm and φ0 according to our controller model. Note that in the main text we refer to front and and φback back stroke angles (φfront w ), which are related to φm and φ0 by the linear relations: w φfront w = φ0 − φm φback w = φ0 + φm (S4) (S5) III. Numerical Simulation Using the simplified wing kinematics above and the quasi-steady aerodynamic model detailed in the main text (Equations 1 and 2), our numerical simulation solves the Newton-Euler equations for vertical, forward, and pitch rotational motion. Control is implemented by ad- justing the front stroke angle of the prescribed wing kine- matics according to Equation 3 in the main text. The Ki and Kp parameters were determined by the fit to the ex- perimental data. The inputs for the controller in each wing beat -- the body pitch angle and pitch velocity -- were taken as their mean values during the previous wing beat. This scheme represents a time-delay of one wing-beat while avoiding the effects of the inherent small-scale pitch oscillations. Before we apply the perturbation, we let the simulated fly stabilize to a steady-state body pitch angle of 45◦. The perturbation is then applied, with magni- tude roughly corresponding to the accelerations observed in the experiments. In simulation runs that tested con- troller models other than PI, we let the system stabilize at θb = 45◦ using a PI controller and only then we ap- plied the perturbation and simultaneously changed the controller type. arXiv:1503.06507v1 [physics.bio-ph] 23 Mar 2015 2 FIG. S1: Data extracted from [2] for the left wing amplitude in response to sinusoidal pitch perturbation. On the top, the red circles represent the response to the pitch angle; on the bottom panel, the blue circles represent the response to the pitch velocity in the same measurement. The black solid ellipses show the predicted response of a PI control model. IV. Analysis of Data from Previous Studies Following [3], we further test the PI control model by using it to predict the pitch response of tethered fruit flies previously published by Dickinson [2]. In [2], flies were tethered to a gimbal apparatus that oscillated about different rotational axes. The left wing stroke amplitude of the flies, Φleft, was measured using photodetectors that recorded the wings' shadows. As noted by [2], flies do not adjust their back stroke angle during pitch correction, so stroke amplitude is a good proxy for φfront , the quantity we measure in free flight experiments. w In one of the measurements reported in [2], pitch per- turbations were imposed so that: θb(t) = Asin(ωt) and θb(t) = Aωcos(ωt), with A = 25◦, period T = 0.63 s, and maximum pitch velocity 250◦s−1. The left wing stroke amplitude was plotted against both pitch angle and pitch velocity (Figure 3a,c in [2]). Using standard image pro- cessing techniques, we extract the data from these plots. The pitch oscillations in the tethered experiments have period 630 ms, which is much longer than the observed pitch correction latency times from our experiments (≈ 10 ms), so we consider the controller time delay negligi- ble. We then write the form for our controller, now in terms of left wing-stroke amplitude, as: Φleft(t) = Kp θb(t) + Ki∆θb(t) + Φmean (S6) where the left wing stroke amplitude, Φleft(t), and mean stroke amplitude, Φmean, are related to the controlled quantity from the main text, ∆φfront (t), by the linear relation: ∆φfront (t) = Φleft(t) − Φmean. Note that con- sidering only the left wing does not reduce generality of this analysis, since pitch correction is left/right symmet- ric. We manually fit for for the control parameters from w w FIG. S2: Data extracted from [2] for the left wing amplitude in response to sinusoidal pitch perturbation is plotted in the 3D space whose axes are (θb, θb, Φleft). The data points are colored according to their original color in Figure S1. The black solid line shows the predicted response of the same PI control model shown in Figure S1. the data. The fitted parameters obtained are Ki = 0.3 and Kp = 8 ms, comparable to the parameters from the main text. The predictions of the PI controller fit are shown in Figures S1, S2, and S3. The output of the PI controller -200 -100 100 Pitch Velocity [deg s-1] 0 200 165 160 155 150 -20 0 -10 10 Pitch Angle [deg] 20 165 160 155 150 Left wing amplitude [deg] 20 0 -200 -20 Pitch [deg] 200 0 Pitch Velocity [deg s-1] 165 160 155 150 Left wing Amp. [deg] 3 is plotted as a function of both pitch angle and pitch velocity in Figure S1, yielding an ellipse in both cases (R2 = 0.761 and 0.842 respectively). The linear model for Φleft as a function of θb gives R2 = 0.556. To show the full dependence of the PI controller on both angle and velocity, we also plotted the PI controller prediction θb, Φleft), shown in in the 3D space whose axes are (θb, Figure S2. The PI controller predicts an inclined ellipse in this space, the projections of which onto the horizontal axes yield the plots in Figure S1. The inclination of the ellipse shows that corrective response depends on both pitch angle and pitch velocity, i.e. the inclination of the measured data in [2] is consistent with a PI controller. FIG. S3: Data extracted from [2] for the left wing ampli- tude in response to sinusoidal pitch perturbation is plotted as a function of pitch acceleration. The data corresponds to the same measurement (same fly and same perturbation) as shown in Figures S1 and S2. The black solid line shows the predicted response of the same PI control model as in Fig- ures S1 and S2. Note that the PI controller model includes no information about the pitch acceleration, suggesting that pitch acceleration is unimportant in determining the flies' cor- rective wing kinematics, and thus excluding a PID controller model. Additionally, we show the predicted output of the PI controller plotted as a function of pitch acceleration in Figure S3. Consistent with [2], Figure S3 shows that the fly's corrective response can be quantitatively captured without including information about the pitch accelera- tion. Figure S3 suggests that pitch acceleration is unim- portant in determining the flies' corrective wing kinemat- ics, and thus excludes a PID controller model. [1] S. Chang and Z. J. Wang, Proceedings of the National (1999). Academy of Sciences 111, 11246 (2014). [2] M. H. Dickinson, Philosophical Transactions of the Royal Society of London.Series B: Biological Sciences 354, 903 [3] T. Beatus, J. M. Guckenheimer, and I. Cohen, Journal of The Royal Society Interface 12 (2015), ISSN 1742-5689. RawRdata PIRmodel 165 160 155 LeftRwingRAmp.R[deg] 150 -3000 -2000 -1000 1000 PitchRAccelerationR[degRs-2] 0 2000 3000
1907.07179	1	1907	2019-07-12T12:59:49	Quantum Intelligence on Protein Folding Pathways	[ "physics.bio-ph", "cond-mat.soft", "quant-ph" ]	We study the protein folding problem on the base of the quantum approach we proposed recently by considering the model of protein chain with nine amino-acid residues. We introduced the concept of distance space and its projections on a $XY$-plane, and two characteristic quantities, one is called compactness of protein structure and another is called probability ratio involving shortest path. Our results not only confirmed the fast quantum folding time but also unveiled the existence of quantum intelligence hidden behind in choosing protein folding pathways.	physics.bio-ph	physics	Quantum Intelligence on Protein Folding Pathways Wen-Wen Mao,1 Li-Hua Lu,1, ∗ Yong-Yun Ji,2 and You-Quan Li1, 3, † 1Zhejiang Province Key Laboratory of Quantum Technology & Device and Department of Physics, Zhejiang University, Hangzhou 310027, P.R. China. 2Department of Physics, Wenzhou University, Wenzhou 325035, P.R. China. 3Collaborative Innovation Center of Advanced Microstructure, Nanjing University, Nanjing 210008, R.R. China. (Received July 17, 2019) We study the protein folding problem on the base of the quantum approach we proposed recently by consider- ing the model of protein chain with nine amino-acid residues. We introduced the concept of distance space and its projections on a XY -plane, and two characteristic quantities, one is called compactness of protein structure and another is called probability ratio involving shortest path. Our results not only confirmed the fast quantum folding time but also unveiled the existence of quantum intelligence hidden behind in choosing protein folding pathways. Protein folding problem has been an important topic in in- terdisciplinary field involving molecular biology, computer science, polymer physics as well as theoretical physics etc.. Levinthal noted early in 1967 that a much larger folding time is inevitable if proteins are folded by sequentially sam- pling of all possible conformations. It was widely assumed that a random conformational search does not occur in the folding process, for which various hypotheses with the help of a series of meta-stable intermediate states have been of- ten proposed. There have been substantial theoretical mod- els which are useful for understanding the essentials of the complex self-assembly reaction of protein folding with dif- ferent simplifying assumptions, such as Ising-like model [1, 2], foldon-dependent protein folding model [3], diffusion- collision model [4, 5], and nucleation-condensation mecha- nism [6, 7]. However till now, this often brings in certain diffi- culties in connecting analytical theory to experimental results because some hypotheses can not be easily put into a practi- cal experimental measurement since they often rely on vari- ous hypotheses[8 -- 12]. As the approach of computing simula- tions introduced less hypotheses in comparison to those the- oretical models, the atomistic simulations [13 -- 15] have been also used to investigate the protein folding along with nowa- days' advances in computer science. All-atom computational method [16], including physics-based and knowledge-based approaches, have provided useful insights on protein folding and design by building high-accuracy atomistic models of pro- teins. However, all those models are computationally costly for high-throughput folding studies and still need certain arti- ficial hypotheses. Recently, we introduced a quantum strategy [17] to formu- late protein folding as a quantum walk on a definite graph, which provides us a general framework without making hy- potheses, where we merely studied the model with six amino- acids as toy model. We know the shortest peptide chain in nature contains more than twenty amino-acid residues. To- ward a genuine understanding, it is obligatory to study more complicate case with more residues. Here we consider the ∗ email: [email protected] † email: [email protected] model with nine amino-acid residues and obtain that the fold- ing time via our quantum approach is much shorter than the one obtained via classical random walk. As the number of amino acid residues increases, the protein structure set be- comes more complicate, whereas, this drives us to introduce the projection method and find some new features on the pro- tein folding pathways in addition to the fast protein folding time. MODELING In the course-grained model the protein is considered as a chain of non-own intersecting unit [18 -- 21], usually refer- ring an amino acid residue of a given length on the two- dimensional square lattice. We indicated recently [17] that for a protein with n amino-acid residues, there will be totally Nn distinct lattice conformations that distinguish various pro- tein intermediate structures, which provide us a point set with Nn objects. We call such a point set as structure set and de- }. The entire structure set note it by Sn = {s1, s2,··· , sNn includes various structures that may be an unfolded, partially folded or completely folded. To distinguish their difference, we introduce a concept of compactness of a structure, (cid:2)(rk − ¯r)2(cid:3) 1 2 C = 1 n n(cid:88) k=1 (1) where the geometric center of the protein is located at ¯r = (r1 + r2 + ··· + rn)/n, and rk refers to coordinates of the k-th residues. Clearly, the compactness Ca represents the av- erage distance from each residues to its geometric center of a given structure sa. We consider in this paper the case of n = 9 and obtain N9 = 388 through DSA-Depth Search Algorithm. The 388 distinct structures that are unrelated by rotational, reflection or reverse-labeling symmetries [20] were plotted in Appendix I. The straight-line structure is labeled lastly as s388 for convenience since it is the farthest end of the depth-first al- gorithm search. We plot it together with the three most com- pact structures s186,s193 and s236 in Fig. 1 as an illustration. For the structure set S9, the compactness of each structure is calculated and shown in the Appendix Table 2. The largest 9 1 0 2 l u J 2 1 ] h p - o i b . s c i s y h p [ 1 v 9 7 1 7 0 . 7 0 9 1 : v i X r a 2 The largest distance in D9 stretches between the completely unfolded straight-line structure s388 and the most compact structures s186 or s193, namely, d388,186 = d388,193 = 17. While the distance between the structure s388 and the other most compact structure s236 is no more the farthest, it is just d388,236 = 14. It will be very helpful to make a projection of the space D9 into a XY -plane. If mapping the s388 to the origin point (0, 0) and either s186 or s193 to the farthest point at (17, 0), we will have a set of points in the XY -plane where each point corresponds to one or several objects in the space D9. Their concrete locations in the XY -plane are determined by the the distance away from (0, 0) and that away from (17, 0), re- spectively. The former represents the distance away from the straight-line structure s388 and the later represents that away from either s186 or s193 in the distance space D9. The color scatter diagram shown in Fig. 2 exhibits the aforementioned projection of D9 on XY -plane, in which the color of each dot measures the degeneracy that is defined as the number of distinct structures mapping to the same point in the XY -plane. The degeneracy of each dots in the above figure and the concrete structures contained in the original im- ages are all listed in Appendix III. Those points lain on the X-axis in between (0, 0) and (17, 0) are the images of the objects referring to the shortest path leaving away from the initial structure s388 and approaching to a most compact struc- ture. Those objects on the shortest path together with the one mapped to (17, 0) constitute a subset SI ⊂ S9. If let SII de- note the set constituted by the other objects, then the structure set S9 can be partitioned into two subsets SI and SII namely, S9 = SI ⊕ SII. The density matrix ρ(t) describing quantum dynamics is of 388 × 388 that is too large to manifest some useful informa- tion. The above projection picture helps us to properly re- duced the large density matrix to a 2× 2 one to extract certain useful information, namely, where ρI,I = (cid:80) a,b ρab and I,II with a ∈ SI and b ∈ SII. Here the large density ρII,I = ρ∗ matrix stands for ρ(t) = Ψ(t)(cid:105)(cid:104) Ψ(t) . (cid:18) ρI,I ρI,II a ρaa, ρII,II = (cid:80) (cid:19) b ρbb, ρI,II = (cid:80) ρII,I ρII,II ρ = (2) B. On the time evolution a = 1, 2,··· , 388}, namely, H = −(cid:80) Letting sa (cid:105) denote the state of a protein structure in the shape of the a-th lattice conformation, we will have a quantum Hamiltonian in a 388-dimensional Hilbert space H = { sa (cid:105) a,b Jab sa (cid:105)(cid:104) sb where Jab refers to the connections between different objects in the structure set S9, i.e., Jab is not zero if the a-th protein structure can be transited into the b-th one by a one-step fold- ing while vanishes otherwise. With these physics picture one can also investigate quantum walk [23 -- 25] on the aforemen- tioned graph, which gives a quantum mechanical understand- ing of the protein folding, i.e., the process by which proteins FIG. 1. Structure Examples There are 388 distinct structures for the amino-acid chain with 9 residues, thus the corresponding struc- ture set S9 contains 388 objects. Here we plot the straight-line struc- ture and the three most compact structures, together with two exam- ples related to them via a one-step folding. For n = 9 the connection graph G9 includes 388 sites which well defines the kinetic term of a quantum Hamiltonian. value C = 2.222 refers to the completely unfolded straight- line structure s388, while the smallest value C = 1.073 refers to the most compact structure s186, s193 and s236. On the basis of the lattice model, we introduced previously the concept of one-step folding to describe the protein fold- ing process [17]. The one-step folding is defined by one dis- placement of an amino acid in one of the lattice sites: two protein structures are connected via one-step folding if their chain conformations differ in one site only. For example, in Fig. 1, the structures s388 and s386 can be obtained by a one- step folding from s387, and so did s193 from s129. This makes us to establish certain connections between distinct objects in the structure set S9 so that we have a connection graph G9 which is described by a 388×388 adjacency matrix Mat(Jab) that characterizes a classical random walk [22] on the connec- tion graph. A. Distance space and its projections If the aforementioned graph Gn is completely connected, a distance between any two structures, saying sa and sb, is well defined. Then we will have a distance space Dn := {sa, dab} in which the magnitude of dab equals to the number of steps of the shortest path connecting sa and sb in the graph Gn. There has been four line-crossings already when the 22-vertex graph G6 is plotted on a plane [17]. The G9 contains 388 ver- tices and appears very complicate if it is plotted on a plane. 388387386129193186236Jab/deg(b) where deg(b) =(cid:80) vertex-b in the graph. 3 c Jcb represents the degree of RESULTS It is widely believed that the native structure of a protein possesses the lowest free energy [26]. This can be interpreted by the hydrophobic force that drives the protein to fold into a compact structure with as many hydrophobic residues inside as possible [19]. How fast does a protein initially in a straight- line structure folds into the most compact structure and what is the main behavior during the folding process will be important issues to study. C. Folding time and folding pathway We let Pa,b(t) denote the probability of a state being the basis state b(cid:105) at time t if starting from the state a(cid:105) at initial time t = 0. Quantum mechanically, Pa,b(t) = ψ(a) (t)2 as long as we solved the Schrodinger equation [3] in terms of the initial condition Ψ(0)(cid:105) = sa (cid:105). Here the superscripts are introduced to distinguish the solution from different initial conditions. As we known, the first-passage probability Fa,b(t) from a state sa (cid:105) to another state sb (cid:105) after t time obeys the known convolution relation [27 -- 31] b Pa,b(t) = 0 t Fa,b(t(cid:48))Pb,b(t − t(cid:48))dt(cid:48) (5) b (t)2 arises from the solution of Eq. [3] where Pb,b(t) = ψ(b) from another initial condition Ψ(0)(cid:105) = sb (cid:105). In the classical case, Pa,b and Pb,b refer to the pb(t) solved from equation [4], respectively, with initial conditions pc(0) = δac and pc(0) = δbc. As protein folding is the process that proteins achieve their native structure, the folding time is the case that the start- ing state is chosen as s388 (cid:105) and the target states are the most compact states sc (cid:105). For example, they are s186 (cid:105), s193 (cid:105) for n = 9 as aforementioned. With the help of the first-passage probability solved from [5], we can calculate the folding time by the formula ´ ´ τ0 0 tFa,c(t)dt τ0 0 Fa,c(t)dt τfd = (6) where τ0 represents the time period when the first-passage probability vanishes Fa,b(τ0) = 0 [17]. Here a = 388 and c = 186 or 193. We know in a previous work that the quan- tum folding is faster than the classical folding with about four to six times even for the simplest model of 4 residues and it is faster than the classical folding with almost ten to hundred times or more for n = 6 [17]. Here we calculate the folding time for n = 9 that is given in Table 1. We can see that the quantum folding time τ q fd is much shorter than the classical folding time τ c fd, and their difference becomes more signifi- cant in the of n = 9 in comparison to n = 6. FIG. 2. Projection of the distance space D9. The color of the dot measures the degeneracy which is defined by the number of struc- tures that concise on the same projection point. The points lain on X-axis correspond to those structures on the shortest path. Here s388 was mapped to the origin point (0, 0), and a, s186 as the farthest point, b, s193 as the farthest point. achieve their native structure. As the time evolution is gov- erned by the Schrodinger equation i d dt Ψ(t)(cid:105) = H Ψ(t)(cid:105) in which Ψ(t)(cid:105) = (cid:80)388 (3) a=1 ψa(t) sa (cid:105), and the expan- sion coefficients ψa(t) can be solved numerically at least. In our numerical calculation, we set and J to be unity and take the time step as ∆t = 0.02. For the initial condition, Ψ(0)(cid:105) = s388 (cid:105), we solve the first-order differential equa- tion [3] by means of Runge-Kutta method and obtain the mag- nitude of ψa(t) at any later time, t = j∗∆t with j = 1, 2,··· . In order to compare with the protein folding problem in the classical literature, i.e., a random conformation search pro- cess, let us revisit the classical random walk. The continuous- time classical random walk [24] on a graph Gn is described by the time evolution of the probability distribution p(t) that obeys the equation of motion (cid:88) b d dt pa(t) = Kab pb(t) (4) where Kab = Tab − δab with Tab being the probability- transition matrix. In the conventional classical random walk, the probability-transition matrix is determined by the ad- jacency matrix of an undirected graph, namely, Tab = ab4 FIG. 4. Time evolution of average distance and average compact- ness. a, The average distance leaving away from the initial struc- ture s388. b, The average distance approaching to the most compact structure s186 while leaving from the structure s388. c, The average distance approaching to the most compact structure s193 while leav- ing from the structure s388. d, The time dependence of the overall average of compactness. Our calculation of those two average quantities are given in Fig. 4. We can see that the mean distance of leaving away from the initial straight-line structure s388 increases much more rapidly in the quantum folding process than in classi- cal case ( see in Fig. 4 a). Correspondingly, the mean distance approaching to the most compact structures s186 and s193 de- creases more rapidly (see in Fig. 4 b,c) in the quantum fold- ing process than in classical one. Additionally, the average of compactness also decreases more rapidly in the quantum folding process than in class case, (see Fig. 4 d), thus the pro- tein classically shrinks slower and it quantum mechanically shrinks faster. for the classical result to compare. Here pI = (cid:80) and pII =(cid:80) As we know, quantum mechanically, the off-diagonal ele- ments of a density matrix reflects certain quantum coherent properties, but our discussion till now have not yet involve it directly. The partition of entire structure set into subset by considering the concept of shortest path help us to have a 2 × 2 density matrix. It it then worthwhile to calculate von Neumann entropy [32] EN(ρ) = −tr(ρ log2 ρ). We also cal- culated the Shannon entropy [33] ES = −pI log pI−pII log pII pa pb. The aforementioned 2 × 2 density ma- trix ρ can also define a quantity called the degree of coher- ence [34, 35] η(t) = 2trρ(t)2 − 1, which is related to vari- ous quantum coherence phenomena, such as Rabi osicillation, self-trapping etc.. After making some calculus, one can found that the degree of coherence and the Von Neumann entropy reach the extreme value at the same time, precisely, η(t) takes minimal value when EN(t) takes its maximum. b∈SII a∈SI The time dependence of both Von Neumann and Shannon entropies are plotted in Fig. 5. We can see that both entropies FIG. 3. The probability ratio γ. The time evolution of γ in the folding process from s388 to s186 and s193 for a, quantum walk and for b, classical random walk. In order to explore whether there are any intelligence hid- den behind in choosing the protein-folding pathway, we com- pare the total probability on the shortest path and the other path by a probability ratio γ to demonstrate how they changes relatively after it leaves the initial state, namely, ΓII(t) − Pa,a(t) ΓI(t) (7) γ(t) = where Γσ(t) = (cid:80) b∈Sσ Pa,b(t) with σ = I, II. Our re- sult is plotted in Fig. 3, we can see that the magnitude of γ changes from 0 to 1 monotonously during a short period of time for both quantum and classical cases. This implies that the probability on the shortest path is more favorable af- ter leaving the initial state for it is on the denominator of γ. In order to rule out the ambiguity on the time scales about classical and quantum literature, we evaluate a mean ratio, dt and show them in Table I. The ¯γ = γ(t)dt/ ´ ´ γ=1 0 γ=1 0 TABLE I. The probability ratio and the folding time τ c structure fd 0.341002 0.551697 8.549224 2153.938 0.353285 0.542997 6.440695 3305.09 s186 s193 τ q fd ¯γq ¯γc mean ratio in quantum case ¯γq is smaller than that in classical case ¯γc. This implies that the probability distribution is more aggregated on the shortest path in quantum case. D. More on folding behavior Although the random walk reflects a stochastic process, it is characterized by the probability distribution defined on all the distinct structures that can always give us intuitive infor- mation if we calculate some weighted average of the entire system. We can attain the time dependence of mean distance (t)2. We can also observe the time evolution of the mean com- In classical case, the mean dis- b (t) away from any structure sa by calculating(cid:80) pactness (cid:80) tance and mean compactness is evaluated by(cid:80) and(cid:80) b dabψ(a) b Caψ(a) b dabp(a) b (t) respectively. b Cap(a) (t)2. b b ababcd5 SUMMARY In the above, we studied the protein folding problem on the base of the quantum approach we proposed recently [17] by considering the model of protein chain with nine amino-acid residues. We show that the protein folding can be modelled as a quantum walk on the graph S 9 of 388 vertices. As such a graph appears complicate if it were plotted on a plane, we introduced the concept of distance space D9 and its projec- tions on a XY -plane by choosing two farthest structures (one is the completely unfolded straight-line structure and the other is a most compact structure) as reference-base points. Accord- ing to our scheme [17], we obtained the protein folding time by making use of the first-passage probability. In order to attain more understandings on the folding behavior, we intro- duced two characteristic quantities, one is called compactness of protein structure another is called probability ratio involv- ing shortest path. The introduction of the concept of shortest path help us to reduce the 388 × 388 large density matrix to a 2 by 2 density matrix. Then we can conveniently evaluate the Von Neumann entropy etc.. We also calculated the Shan- non entropy on the base of classical random walk approach to compare. Our results not only confirmed the fast quantum folding time [17] but also unveiled the existence of a quan- tum intelligence hidden behind in the choosing protein folding pathways. This work is supported by National Key R & D Program of China, Grant No. 2017YFA0304304, and partially by the Fundamental Research Funds for the Central University. FIG. 5. Time-dependent evolution of entropy a, The quantum and classic time evolution of the entropy from s388 to s186, b, The quantum and classic time evolution of the entropy from s388 to s193. change from 0 monotonously to a maximal value and then varies. The maximal Von Newmann entropy means the quan- tum state is in a completely mixed state (the degree of coher- ence vanishes), while a maximal Shannon entropy implies the maximal information uncertainty. In our present problem, the population (or probability summation in classical case) on the shortest path is dominated during the time before the entropy reaches the maximal value. Clearly, such a period in quantum case is longer than in classical case. This illustrates from an- other angle of view that quantum folding process posses more intelligence in protein folding pathway. [1] Munoz, V., & Eaton, W. A., A simple model for calculating the kinetics of protein folding from three-dimensional structures, PNAS 96, 11311 (1999). [2] Henry, E. R., & Eqton, W. A., Combinatorial modeling of pro- tein folding kinetics: Free energy profiles and rates, Chem Phys 307, 163 (2004). [3] Englander, S. W., & Mayne, L., The case for defined protein folding pathways, PNAS, 114, 8253 (2017). [4] Karplus, M., & Weaver, D. L., Protein-folding dynamics, Na- ture 260, 404 (1976). [5] Sali, A., Shakhnovich, E., & Karplus, M., How does a protein folds, Nature 369, 248 (1994). [6] Guo, Z. Y., & Thirumalai, D., Kinetics of protein-folding - nuleartion mechanicsm, time scales and pathways, Biopolymers 36, 83 (1995). [7] Fersht, A. R., Transition-state structure as a unifying basis in protein-folding mechanism: contact order, chain topology, sta- bility, and the extended nucleus mechanism, PNAS 97, 1525 (2000). [8] Oliveberg, M., & Wolynes, P. G., The experimental survey of potein-folding energy landscapes, Q. Rev. Biophys. 38, 245 (2005). [9] Shakhnovich, E., Protein folding thermodynamics and dynam- ics: where physics, chemistry, and biology meet, Chem. Rev. 106, 1559 (2006). [10] Wolynes, P. G., Eaton, W. A., & Fersht, A. R., Chemical physics of protein folding, PNAS 109, 17770 (2012). [11] Dill, K. A., & MacCallum, J. L., The protein-folding problem, 50 years on, Science 338, 1042 (2012). [12] Thirumalai, D., Liu, Z. X., O'Brien, E. P., & Reddy, G., Protein folding: from theory to practice, Curr. Opin. Struct. Biol. 23, 22 (2013). [13] Piana, S. Lindorff-Larsen, K., & Shaw, D. E., Protein folding kinetics and thermodynamics from atomistic simulation, PNAS 109, 17845 (2012). [14] Henry, E. R., Best, R. B., & Eaton, W. A., Comparing a simple theoretical model for protein folding with all-atom molecular dynamics simulation, PNAS 110, 17880 (2013). [15] Snow, C. D., Nguyen, H., Pande, V., & Gruebele, M., Absolute comparison of simulated and expereimental protein-folding dy- namics, Nature 420, 102 (2002). [16] Mirny, L., Shakhnovich, E., Protein folding theory: From lat- tice to all-atom models, Annual Review of Biophysics and Biomolecular Structure, 30 361 (2001). [17] Lu, L. H, Li, Y. Q., Quantum approach to fast protein-folding time, https://arxiv.org/abs/1906.09184. [18] Taketomi, H., Ueda, Y., & Go, N., Studies on protein fold- ing, unfolding and fluctuations by computer simulation, Int. J. Prept. Protein Res. 7, 445 (1975). [19] Dill, K. A., Theory for the folding and stability of globular pro- teins, Biochemistry 24, 1501 (1985). [20] Li, H., Helling, R, Tang, C., & Wingreen, N. S., Emergence of preferred structures in a simple model of protein folding, Sci- ence 273, 666 (1996). [21] Li, Y. Q., Ji, Y. Y., Mao, J. W., & Tang, X. W., Media effects on the selection of sequences folding into stable proteins in a absimple model, Phys. Rev. E 72, 021904 (2004). [22] van Kampen, N. G., Stochastic Processes in Physics and Chem- istry (revised edition) (North-Holland, Amsterdam, 1997). [23] Aharonov, Y., Davidovich, L., & Zagury, N., Quantum random walks, Phys. Rev. A 48, 1687 (1993). [24] Farhi, E. & Gutmann, S., Quantum computation and decision trees, Phys. Rev. A 58, 915 (1998). [25] Manouchehri, K., & Wang, J. B., Physical Implementation of Quantum Walks (Springer-Verlag, Berlin, 2014). [26] Anfinsen, C. B., Principles that govern the folding of protein chains, Science 181, 223 (1973). [27] Montroll, E. W., Random walks on lattices III: calculation of first-passage times with application to exciton trapping on pho- tosynthetix units, J. Math. Phys. 10, 753 (1969). [28] Redner, S., A guide to first-passage processes (Cambridge Univ. Press, Cabridge, 2001). [29] Noh, J. D. & Rieger, H., Random walk on complex newworks, 6 Phys. Rev. Lett. 92, 118701 (2004). [30] Condamin, S., Benichou, O., Tejedor, V., Voituriez, R., & Klafter, J., First-passage times in complex scale-invariant me- dia, Nature 450, 77 (2007). [31] Guerin, T., Levernier, N., Benichou, O., & Voituriez, R., Mean first-passage times of non-Markovian random walkers in con- finement, Nature 534, 356 (2016). [32] Nielsen, M. A., & Chuang, I. L., Quantum computation and quantum information, (Cambridge Univ. Press, Cambridge, 2000). [33] Pathria, R. K & Beale D. Paul., Statistical Mechanics (Third Edition) (Butterworth-Heinemann Elsevier, New York, 2011). [34] Leggett, A. J., Bose-Einstein condensation in the alkali gases: Some fundamental concepts, Rev. Mod. Phys 73, 307 (2001). [35] Lu, L. H., & Li, Y. Q., Dynamics for partially coherent Bose- Einstein condensates in double wells, Phys. Rev. A 80, 033619 (2009).
1802.04965	1	1802	2018-02-14T05:56:26	Structural-elastic determination of the mechanical lifetime of biomolecules	[ "physics.bio-ph" ]	The lifetime of protein domains and ligand-receptor complexes under force is crucial for mechanosensitive functions, while many aspects of how force affects the lifetime still remain poorly understood. Here, we report a new analytical expression of the force-dependent molecular lifetime to understand transitions overcoming a single barrier. Unlike previous models derived in the framework of Kramers theory that requires a presumed one-dimensional free energy landscape, our model is derived based on the structural-elastic properties of molecules which is not restricted by the shape and dimensionality of the underlying free energy landscape. Importantly, the parameters of this model provide direct information of the structural-elastic features of the molecules between the transition and the native states. We demonstrate the applications of this model by applying it to explain complex force-dependent lifetime data for several molecules reported in recent experiments, and predict the structural-elastic properties of the transition states of these molecules.	physics.bio-ph	physics	Structural-elastic determination of the mechanical lifetime of biomolecules Shiwen Guo,1, ∗ Qingnan Tang,2, ∗ Mingxi Yao,1 Shimin Le,2 Hu Chen,3 and Jie Yan1, 2, 4, † 1Mechanobiology Institute, National University of Singapore, Singapore 117411 2Department of Physics, National University of Singapore, Singapore 117542 3Department of Physics, Xiamen University, Xiamen, China 361005 4Centre for Bioimaging Sciences, National University of Singapore, Singapore 117546 (Dated: February 15, 2018) Abstract The lifetime of protein domains and ligand-receptor complexes under force is crucial for mechanosensitive functions, while many aspects of how force affects the lifetime still remain poorly understood. Here, we report a new analytical expression of the force-dependent molecular lifetime to understand transitions overcoming a single barrier. Unlike previous models derived in the frame- work of Kramers theory that requires a presumed one-dimensional free energy landscape, our model is derived based on the structural-elastic properties of molecules which is not restricted by the shape and dimensionality of the underlying free energy landscape. Importantly, the parameters of this model provide direct information of the structural-elastic features of the molecules between the transition and the native states. We demonstrate the applications of this model by applying it to explain complex force-dependent lifetime data for several molecules reported in recent experiments, and predict the structural-elastic properties of the transition states of these molecules. PACS numbers: 87.80.Nj, 82.37.Rs, 87.15.A- 8 1 0 2 b e F 4 1 ] h p - o i b . s c i s y h p [ 1 v 5 6 9 4 0 . 2 0 8 1 : v i X r a 1 I. INTRODUCTION It has been known that single cells can sense mechanical properties of their micro- environment, and transduce the mechanical cues into biochemical reactions that eventu- ally affect cell shape, migration, survival, and differentiation [1]. This mechanotransduction requires transmission of force through a number of mechanical linkages, each of which is often composed of multiple linearly arranged force-bearing proteins that are non-covalently linked to one another. Under force, the domains in each protein in the linkage may undergo mechanical unfolding. In addition, two neighbouring proteins in the linkage can dissoci- ate. Therefore, the force-dependent lifetime of the protein domains and protein-protein complexes is a key factor that affects the mechanotransduction on a particular mechanical linkage. Determining the force-dependent rate (the reciprocal of lifetime) of rupturing a ligand-receptor complex or unfolding a protein domain has been a focus of experimental measurements [2–7] and theoretical modelling [8–18]. Previous single-molecule force spec- troscopy measurements have revealed complex kinetics for a variety of molecules [2–6], yet the mechanisms still remain elusive. An extensively applied phenomenological expression of k(F ) was proposed by Bell et al.[8]: k(F ) = k0eβF δ* , where β = (kBT )-1, k0 is the rate in the absence of force and δ∗ is the constant transition distance. This model assumes that the force applied to the molecule results in change of the energy barrier by the amount of −F δ, while the physical basis of this assumption is weak. The limitation of Bell's model has been revealed in many recent experiments that reported complex deviations from its predictions [2–6]. In order to explain such deviations, several analytical expressions of k(F ) were derived based on extending the Brownian dynamics theory from Kramers [19] for force-dependent dissociation of bonds [9–12]. The Kramers theory was originally proposed to study kinetics of particle escaping from an energy well through diffusion on a presumed one-dimensional free energy landscape. The theory showed that for sufficiently high barrier, the escaping rate exponentially decreases with the height of the barrier, which proves Arrhenius law for the one dimensional case. In order to derive k(F ) in the framework of Kramers theory, the force- dependent unfolding/dissociation kinetics was paralleled to the kinetics of particle escaping from the energy well when the particle is subject to force. Specifically, a fixed zero force free energy landscape U0(x) is assumed, which can be described as a function of the molecular 2 extension x along the pulling direction. Under force, the free energy landscape becomes U (x) = U0(x) − F x. Assuming a sufficiently high energy barrier such that the energy well and the barrier are well separated and for the cases where U0(x) can be approximated by a cusp or a linear-cubic function, an analytical expression of the force-dependent transition rate k(F ) was derived by Dudko et al. [11], which has been extensively applied to explain experimental data. F(cid:82) δ∗(F (cid:48))dF (cid:48) β In general, the applications of the expression of k(F ) derived based on the framework of Kramers theory are limited by two factors, namely 1) the transition pathway that is one dimensional with the molecular extension as good reaction coordinate, and 2) the shape of the presumed free energy surface U0(x). A more recent publication [20] shows that k(F ) 0 can be re-expressed as k(F ) = k0e in the framework of Kramers theory, where δ∗(F ) is the average extension difference of the molecule between the transition state and the native state. This expression does not have an explicit dependence on a presumed free-energy landscape U0(x). However, in order to actually apply this formula, a presumed one-dimensional free energy landscape is still needed to calculate δ∗(F ). As a result, its application is still limited by these two factors mentioned above. Due to these limitations, although k(F ) derived in the framework of Kramers theory can explain mild deviations from Bell's model [10–12], they typically predict monotonic k(F ) and fail to explain more complex experimentally observed kinetics, such as the non-monotonic k(F ) reported in several recent experiments [2, 3, 5]. Previously, non-monotonic k(F ) were typically explained by high-dimension phenomeno- logical models involving multiple competitive pathways or force-dependent selection of mul- tiple native conformations that have access to different pathways [13–17]. For example, the transition rate described by two competitive transition pathways, k(F ) = k1(F ) + k2(F ) , each following Bell's model, can explain non-monotonic k(F ) with one of the transition distances being negative [13]. On the other hand, models based on force-dependent selec- tion of multiple native conformations that have access to different pathways are much more complex and lack of analytical simplicity for general cases [14]. Simplification of such mod- els must require additional assumptions on the force-dependence of the selection of native conformations [14–16]. A limitation of all these models is that the model parameters do not provide insights into the structural and physical properties of the molecules in the native 3 and transition states. We recently reported that k(F ) of mechanical unfolding of titin I27 immunoglobulin It (Ig) domain exhibits an unexpected "catch-to-slip" behaviour at low force range [2]. switches from a decreasing function (i.e., "catch-bond" behaviour) at forces below 22 pN to an increasing function (i.e., "slip-bond" behaviour) at forces greater than 22 pN. The transition state of the titin I27 domain is known to involve a peeled A-A(cid:48) peptide containing 13 residues [21–24]. Taking the advantage of the known structures of I27 in its native and the transition states, we analysed the effects of the structural-elastic properties of I27 on its force-dependent unfolding kinetics by applying Arrhenius law. We demonstrated that the entropic elasticity of titin I27 in the two states is responsible for the observed "catch- to-slip" behaviour of k(F ). Besides suggesting the structural-elastic property of a molecule as a critical factor affecting the force-dependent transition rate, the result also points to a possibility of deriving k(F ) based on the structural-elastic properties of molecules in the framework of Arrhenius law. As the derivation of k(F ) based on Arrhenius law does not depend on any presumed free energy scape, it is not limited by the dimensionality of the system and the choice of the transition coordinate. Therefore, it is promising to be applied to a broader scope of experimental cases. The result described in our previous work [2] is obtained based on the prior knowledge of the structural-elastic properties of I27 in the native and transition states. Unfortunately, such prior knowledge is unavailable for most of other molecules. In order to interpret the force-dependent transition rate based on the structural-elastic properties of molecules for generic cases, it is necessary to derive an expression of k(F ) that contains parameters related to the potential structural-elastic properties of the molecule based on Arrhenius law. If this can be achieved, fitting the experimental data using the derived k(F ) not only can be applied to explain experimental data but also can provide important insights into the structural- elastic properties of the molecule based on the best-fitting values of the model parameters. To our knowledge, k(F ) with such capability has not been derived before. 4 II. RESULTS A. Deriving k(F ) based on the structural-elastic properties of molecules In this work, we derived an analytical expression of k(F ) on the basis of the structural- elastic features of molecules by applying Arrhenius law. In our derivation, the native state is modelled as a deformable folded structure with a relaxed length b0, which is the linear distance between the two force-attaching points on the relaxed folded structure, as depicted in Fig. 1A. Here a deformable folded structure refers to that the structure has a certain Yong's modulus, and it can be slightly deformed along the force direction without causing local structural changes. One example is the B-form DNA, which can be extended beyond its relaxed contour length without breaking any Watson-Crick basepairs at forces in 20-40 pN [25, 26]. FIG. 1. The entropic elastic free energies of the native and the transition states. The native state is sketched as a folded structure, with a length of b0 and stiffness of γ0. The transition state is modelled as a structure consisting of a folded core with a length of b and stiffness of γ* as well as a flexible polymer with a contour length of L. The force-dependent entropic free energies of the states are indicated. kBT ) − kBT F b0 )(1 + F γ0 ), where b0(coth( F b0 A deformable folded structure has a very simple analytical force-extension curve [27], xb0,γ0(F ) = b0(coth( F b0 F b0 ) is the solution of the force-extension curve of an inextensible rod with a length b0, and the factor (1 + F γ0 ) takes into account the force-dependent elongation of the rod. Here γ0 describes the stiffness of the folded structure along the force direction. The ratio b0/γ0 describes the deformability of the folded structure. γ is in the order of 102 − 103 pN for typical protein domains and nucleic acids structures (SI: SI-II, Tab. S1). kBT ) − kBT The transition state is assumed to consist of a deformable folded structure and a semi- flexible polymer that is a peptide for protein or a single-stranded DNA (ssDNA) for DNA 5 Φ0(F)=−F!0xb0,γ0(F′)dF′bbL0FFFFNative StateTransition State ABΦ(F)=−F!0(xb∗,γ∗(F′)+xL∗(F′))dF′BNative StateTransition State Denatured StateEnergyExtensionΔGΔG(F)0δ(F)k(F)=k0e−β∆Φ∗(F)∆Φ∗(F)=−Fδ∗ ∆Φ∗(F)=∆G∗(F)−∆G0structure or a single-stranded RNA (ssRNA) for RNA structure. Its force-extension curve is a sum of the contributions from the folded core and the flexible polymer: x(F ) = xb,γ(F )+ xL∗(F ). Here b and γ* are the length and stiffness of the folded core, respectively. L∗ = n∗lr is the contour length of the flexible polymer that contains n∗ residues, where lr is the contour length per residue. xL∗(F ) can be described by the worm-like chain (WLC) polymer model that contains two parameters, the bending persistence length A and the contour length L∗ = n∗lr. In previous single-molecule manipulation experiments, values of A ∼ 0.8 nm and lr ∼ 0.38 nm for peptide chain have been experimentally determined [28]. Based on the WLC model, xL∗(F ) can be analytically approximated by the Marko-Siggia formula [29]: kBT = x L∗ + F A f (x(cid:48))dx(cid:48) Force F introduces an entropic conformational free energy Φ(F ) to a molecule in a partic- ular structural state, in addition to other chemical interactions that maintain the molecule in the structural state. Φ(F ) can be calculated based on the force-extension curve of the molecule as: Φ(F ) = − F x(F ), where f (x) is the equilibrium force-extension curve of the molecule, and x(F ) is the equilibrium extension of the molecule at the applied force. It is straightforward to show that this energy can be rewritten into a simpler form: x(f(cid:48))df(cid:48) [30, 31] (SI: SIII), where x(f ) is the inverse function of f (x). If the Φ(F ) = − molecule has different force-extension curves between the native and the transition states, force applied to the molecule will result in a force-dependent transition distance: x(F )(cid:82)0 F(cid:82)0 1 4. 4(1−x/L∗)2 − 1 δ∗(F ) = xb,γ∗(F ) − xb0,γ0(F ) + xL∗(F ), (1) and cause a change in the transition free energy barrier that can be computed as: ∆Φ∗(F ) = − F(cid:82)0 δ(F (cid:48))dF (cid:48). ∆Φ∗(F ) can be rewritten as a linear combination of three terms: ∆Φ∗(F ) = Φb∗,γ∗(F ) − Φb0,γ0(F ) + ΦL∗(F ), (2) F(cid:82)0 xi(F (cid:48))dF (cid:48) denotes the contributions from the folded native state (b0,γ0), where Φi(F ) = − the folded core of the transition state (b∗,γ∗), and the flexible polymer in the transition state (L∗), respectively. The three force-dependent entropic conformational free energy terms scaled by β−1 = kBT have the following analytical solutions: 6 βΦb,γ(F ) = − ln sinh(βF b) βF b + Li2(e−2βF b)−ξ(2) 2 − 1], 2βγb γ [ln(1 − e−2βF b) + βF b − F 2AL∗ − xL∗ (F )+L∗ 4A(L∗−xL∗ (F )) − F xL∗ (F ) L∗ (F ) L∗2 kBT 4A . βΦL∗(F ) = x2 + (3)  ∞(cid:80)k=1 zk Here, Li2(z) = k2 is the second order polylogarithm function (also known as Jonquire's function), and ξ(2) ∼ 1.645 is the Riemann-Zeta function evaluated at z = 2. k(F ) is then determined by applying the Arrhenius law k(F ) = k0e−β∆Φ∗(F ): k(F ) = k0e −β(Φb∗,γ∗ (F )−Φb0,γ0 (F )+ΦL∗ (F )). (4) At forces (cid:29) kBT /b0, (cid:29) kBT /b∗ and (cid:29) kBT /A, k(F ) has a simple asymptotic expression: k(F ) = k0eβ(σF +αF 2/2−ηF 1/2), (5) ξ(2) 2 ( kBT γ∗b∗ − kBT b0 b∗ e kBT γ∗ − kBT γ0 ), α = b∗ − b0) − ( kBT γ0 , and η = L∗(cid:113) kBT which contains a kinetics parameter k0 = k0 γ0b0 ), and three model parameters σ = L∗ + (b∗ γ∗ − b0 γ0 and A . Typical values of kBT γ∗ are in the range of 10−3 nm - 10−2 nm (SI: SI-II, Tab. S1); therefore, σ ∼ L∗ + (b∗ − b0). An alternative derivation of Eq. 5 is provided in Supporting Information (SIV : "Alternative derivation of Eq. 5"). Here we emphasize that, since Eq. 5 is an large-force asymptotic formula, k0 should not be interpreted as the zero-force transition rate. The zero-force rate k0 predicted by the model should be based on Eq. S5, which is related to k0 by the following equation: k0 = k0 − ξ(2) 2 ( b∗ b0 e kBT γ∗b∗ − kBT γ0b0 ). (6) For experiments that record transition force distribution p(F ) under a time-varying force constraint, it is straightforward to apply Eq. 5 to fit such data by a simple transformation [11]: p(F ) = k(F )/ F exp− 7 k(F (cid:48)) F F(cid:90)0 dF (cid:48) , (7) where F is the time derivative of F (t). p(F ) in Eq. 7 is a density function, therefore the transition force histogram obtained from experiments should be reconstructed as (cid:104)number of counts per bin(cid:105) / (cid:104)the total number of counts(cid:105) / (cid:104)bin size(cid:105). Clearly, in the three model parameters of Eq. 5, σ is the contour length difference and α describes the deformability difference between the folded core of the transition state and the native state. η only depends on the contour length of the flexible polymer in the transition state. Therefore, the best-fitting values of these parameters can provide important insights regarding how the different structural-elastic properties between the transition state and the native state affect k(F ). It is even possible to use this model to obtain further insights into the structural-elastic properties of the transition state based on the best-fitting values of σ, α and η. The native state structure is often known and therefore b0 is determined. In addition, γ0 can be estimated with reasonable accuracy using all-atom molecular dynamics (MD) simulations (SI: SI-II). Hence, for molecules with a known native state structure, the structural-elastic parameters of the transition state can be solved from σ, α and η. B. Applications in interpreting experimental data We firstly applied Eq. 5 to fit k(F ) observed for titin I27 domain, and tested whether the fitting parameters can provide insights into how the structural-elastic properties of the molecule play a role in determining the transition kinetics. The titin I27 domain has a known transition state structure, which allows us to examine the quality of the prediction of the transition state properties based on the best-fitting parameters. As described earlier, the experimental data of I27 exhibits a "catch-to-slip" switching behaviour, where k(F ) switches from a decreasing function to an increasing function when force exceeds a certain threshold value at around 22 pN (Fig. 2A, black squares) [2]. At forces larger than ∼ 60 pN, the force-dependent unfolding rate converges to a Bell-like behaviour (Fig. 2A). The best-fitting parameters according to Eq. 5 without any restriction are determined as: k0 = 0.026± 0.014 s-1, with 95% confidence bounds of (−0.020, 0.073) s-1; σ = 1.099 ± 0.243 nm, with 95% confidence bounds of (0.510, 1.689) nm; α = 0.002 ± 0.003 nm/pN, with 95% confidence bounds of (−0.004, 0.007) nm/pN; and η = 10.519 ± 1.542 nm·pN1/2, with 95% confidence bounds of (6.356, 14.682) nm·pN1/2. Here, the errors indicate standard deviations obtained with bootstrap analysis (SI: SV, Tab. S2) and the 95% confidence bounds are determined by 8 fitting of all the data points (Fig. 2A, black squares). We also tested the robustness of the convergence of the fitting by repeating the fitting procedure with 10 different well-separated initial sets of values, and found that the best-fitting parameters converged to the same set regardless of the initial values (SI: SVI, Tab. S5). Based on the structure of I27 and steered MD simulations, b0 ∼ 4.32 nm and γ0 ∼ 1900 pN were estimated (SI: SI-II, Figs. S2 and S6). From the best-fitting parameters, L* = 4.6± 0.7 nm, b* = 0.8 ± 0.4 nm and γ* = 194 ± 41 pN were solved for the transition state. The value of L* corresponds to a peptide of 12 ± 2 residues, which is in good agreement with the previously known result that the transition state of I27 involves a peeled A-A(cid:48) peptide chain of 13 residues (SI: Fig. S2) [2, 21–24]. This result shows that our model indeed can provide information of the structural-elastic properties of the transition state. The zero-force transition rate predicted by the model is estimated to be k0 ∼ 5 × 10−3 s−1 according to Eq. 6. This value is consistent with that recently reported in [2] but differs from the value extrapolated based on Bell's model in earlier studies [32] (see discussions in the discussion section). Based on the best-fitting parameters, one can predict the I27 unfolding force probability density function p(F ) using Eq. 7 at any loading rate. Figure 2B shows predicted p(F ) at several loading rates from 0.01 pN/s to 10 pN/s. We next compare the predicted p(F ) of I27 with experiments. Previous AFM experiments suggest that the native state of I27 transits to an intermediate state with the A strand detached from the B strand at forces >100 pN, and unfolding transition starts from this intermediate state at forces above 100 pN [33]. Since the k(F ) data in Fig. 2A were measured at forces below 100 pN, we chose to conduct experiment with a loading rate of 0.08 pN/s at which the unfolding forces are mainly below 100 pN for the comparison. Figure 2C shows the unfolding force density function constructed from 210 unfolding forces of I27 from 7 independent molecular tethers (vertical bars with a bin size of 5 pN) and the predicted p(F ) according to Eq. 7 using the best-fitting values of the parameters (k0 = 0.026 s-1, σ = 1.099 nm, α = 0.002 nm/pN and η = 10.519 nm·pN1/2) described in the preceding section. The comparison shows good agreement between the predicted and experimental results. We next investigated the force-dependent rupturing rate of the monomeric PSGL-1/ P- selectin complex, which also demonstrates a "catch-to-slip" switching behaviour (Fig. 3A, black squares) [3]. In addition, the k(F ) profile does not approach a Bell-like shape in the 9 FIG. 2. Application of Eq. 5 to interpret experimental data of titin I27. (A) The k(F ) data for titin I27 domain unfolding [2] are indicated with black squares and fitted with Eq. 5 (black line). The goodness-of-fit was evaluated by a R-Square of ∼ 0.997 and a Root mean squared error (RMSE) of ∼ 0.162. The best-fitting model parameters and the structural-elastic parameters determined based on the native state structure, steered MD simulation, or solved from the best- fitting parameters are indicated in the panel. (B) The panel shows the predicted I27 unfolding force distribution p(F ) using Eq. 7 based on the best-fitting parameters for k(F ), with different loading rates of 0.01 pN/s (solid line), 0.1 pN/s (short dash line), 1 pN/s (short dot line) and 10 pN/s (dash line). (C) Comparison between the predicted p(F ) of I27 (solid black curve) and the experimental data (grey bars) shows good agreement at a loading rate of 0.08 pN/s. slip bond region when force is further increased. Therefore, this protein complex represents a more complicated situation compared with I27. The best-fitting parameters without any restriction are determined as k0 = 51.786 ± 27.083 s-1, with 95% confidence bounds of (12.499, 91.072) s-1; σ = 0.723± 0.162 nm, with 95% confidence bounds of (0.468, 0.978) nm; 10 bLb0L∗∼4.6nm∼12a.a.,b∗∼0.8nm,γ∗∼194pNb0∼4.32nm,γ0∼1900pNσ∼1.099nmα∼0.002nm/pNη∼10.519nm·pN1/2ABCα = −0.005 ± 0.001 nm/pN, with 95% confidence bounds of (−0.008,−0.002) nm/pN; and η = 5.760 ± 1.275 nm·pN1/2, with 95% confidence bounds of (4.019, 7.501) nm·pN1/2. The errors and the robustness of the parameter convergence are generated/tested similar to the case of I27 (SI: SV-VI, Tabs. S3 and S6). b0 ∼ 7.28 nm was determined based on the structure of the PSGL-1/ P-selectin complex (SI: Fig. S3). As P-selectin occupies most of the volume of the complex, its stiffness should be the determining factor for the deformability of the folded structure/core for both the native state and the transition state (i.e., γ0 ∼ γ∗). From these values, L∗ = 2.5 ± 0.6 nm, b∗ = 5.5± 0.4 nm, and γ0 = γ∗ = 364± 48 pN were solved. These results predict a partially peeled peptide/sugar polymer in the transition state, which suggests that detachment of the sugar molecule covalently linked to the PSGL-1 from P-selectin is a necessary step that has to take place before rupturing (SI: Fig. S3). The zero-force transition rate predicted by the model is estimated to be k0 ∼ 39.1 s−1 according to Eq. 6. The predicted p(F ) using Eq. 7 at several loading rates from 20 pN/s to 200 pN/s are shown in Fig. 3B. To the best of our knowledge, loading rate-dependent p(F ) for the rup- turing of monomeric PSGL-1/P-selectin complex has not been experimentally measured in the force range similar to the k(F ) data; therefore, the predicted p(F ) in Fig. 3B will be awaiting for future experimental tests. We also applied the theory to understand the unfolding of the src SH3 domain under a special stretching geometry that causes a significant deviation from Bell's model (Fig. 4A, black squares) [4] On the logarithm scale, it exhibits a convex profile increasing with force, term in the exponential of Eq. 5 with a positive α is which strongly suggests that the αF 2 2 the cause of the observed k(F ). Unconstrained fitting resulted in a negative value of b∗, which is physically impossible. We found that η < 4.3 is needed to ensure a positive b∗. Good quality of fitting was obtained for any values of η < 4.3 (SI: SVII), suggesting that the length of peptide produced in the transition state is not responsible for the observed k(F ) data. The value of α ∼ 0.042 − 0.048 nm/pN is insensitive to changes in η (SI: Tab. S8), strongly suggesting the deformability of the folded core in the transition state as the key factor of the observed k(F ) . In order to further obtain more accurate structural-elastic properties of the transition state of src SH3 domain, additional information of the peptide length in the transition state is needed. Previous study estimated a small transition distance ∼ 0.45 nm in the force 11 FIG. 3. Application of Eq. 5 to interpret experimental data of monomeric PSGL-1/ P-selectin. (A) The k(F ) data obtained for rupturing of monomeric PSGL-1/ P-selectin complex (Fig. 4b in Ref. [3]) are indicated with black squares and fitted with Eq. 5 (black line). The goodness-of-fit was evaluated by a R-Square of ∼ 0.991 and a Root mean squared error (RMSE) of ∼ 0.032. The best-fitting model parameters and the structural-elastic parameters determined based on the native state structure, steered MD simulation, or solved from the best-fitting parameters are indicated in the panel. (B) The panel shows the predicted sPSGL-1 / sP-selectin rupturing force distribution p(F ) using Eq. 7 based on the best-fitting parameters for k(F ), with different loading rates of 20 pN/s (solid line), 50 pN/s (short dash line), 100 pN/s (short dot line) and 200 pN/s (dash line). range of 15-25 pN [4], suggesting insignificant fraction of peptide in the transition state (SI: SVII). Consistently, our steered MD simulation shows a negligible production of peptide under force during transition (SI: Fig. S4). Based on these information, we estimated b∗ and γ∗ by approximating η ∼ 0. The resulting best-fitting parameters are determined as 12 b0Structural-mechanicaldeterminationofthelifetimeofbiomoleculesunderforceShiwenGuo,1QingnanTang,2MingxiYao,1HuijuanYou,3ShiminLe,2HuChen,4andJieYan1,2,5,∗1MechanobiologyInstitute,NationalUniversityofSingapore,Singapore1174112DepartmentofPhysics,NationalUniversityofSingapore,Singapore1175423SchoolofPharmacy,HuazhongUniversityofScienceandTechnology,Wuhan,China4300224DepartmentofPhysics,XiamenUniversity,Xiamen,China3610055CentreforBioimagingSciences,NationalUniversityofSingapore,Singapore117546(Dated:May28,2017)Severalrecentexperimentshavesuggestedthatthestructural-elasticpropertiesofthenativeandthetransitionstatesofbiomoleculesareakeydeterminantoftheirmechanicalstability.How-ever,mostofthecurrenttheoreticalmodelswerederivedbasedonconformationdiffusionofthemoleculealongaphenomenologicalenergysurface,lackingadirectrelationtothestructural-elasticparametersofthemolecules.Here,basedontheArrheniuslawandtakingintoconsiderationofthestructural-elasticfeaturesofthemolecules,wederivedasimpleanalyticalexpressionfortheforce-dependentlifetimeofthenativestateofthemolecules.Weshowthatthismodelcanexplainavarietyofcomplexforce-dependenttransitionkineticsobservedinexperiments.Thisworkhighlightsthatstructural-elasticpropertiesasakeydeterminantofthelifetimeofbiomoleculesunderforce,whichhasbeenlargelyignoredpreviously.Thenewtheoreticalframeworkprovidedinthispaperwillenableustoexplainawidescopeofexperimentsfromanovelstructural-elasticperspective.PACSnumbers:87.80.Nj,87.15.hm,82.37.RsTTTb0∼1.90nm,γ0∼3000pN,L∗∼0nmσ∼−0.441nmα∼0.049nm/pNη=0nm·pN1/2b∗∼1.6nmγ∗∼32pNb0∼4.32nm,γ0∼1800pNσ∼1.099nmα∼0.002nm/pNη∼10.519nm·pN1/2L∗∼4.6nm∼12a.a.,b∗∼0.8nm,γ∗∼188pNb0∼7.28nm,γ0∼γ∗σ∼0.723nmα∼−0.005nm/pNη∼5.760nm·pN1/2L∗∼2.5nm,b∗∼5.5nm,γ0=γ∗∼364pNδ1orδ2mustbenegative!k0=1s-1,b0=5nm,b∗=7nmΦ0(F)=−F!0xb0,γ0(F′)dF′Φ(F)=−F!0(xb∗,γ∗(F′)+xL∗(F′))dF′τ(F)=1/k(F)k(F)=1/τ(F)b0=4.32nm,b∗∼1.6nm,L∗=0nmk0∼0.0295/s,σ∼−0.44nm,α∼0.05nm/pNη=5.98nm.pN1/2σ∼0.723nmα∼−0.005nm/pNη∼5.760nm·pN1/2b0∼7.28nm,γ0∼γ∗L∗∼2.5nm,b∗∼5.5nm,γ0=γ∗∼364pNABk0 = 0.030±0.043 s-1, with 95% confidence bounds of (−0.033, 0.092) s-1; σ = −0.441±0.249 nm, with 95% confidence bounds of (−1.083, 0.202) nm; and α = 0.049 ± 0.009 nm/pN, with 95% confidence bounds of (0.027, 0.071) nm/pN. The errors and the robustness of the parameter convergence are generated/tested similar to the case of I27 (SI: SV-VI, Tabs. S4 and S7). The structural-elastic parameters of the native state were determined to be b0 ∼ 1.90 nm and γ0 ∼ 2900 pN based on the structure and steered MD simulations (SI: SI-II, Figs. S4 and S7). Finally, based on the best-fitting values, b∗ = 1.6 ± 0.2 nm and γ∗ = 32 ± 9 pN were solved. The estimated value of γ∗ is reasonably in agreement with the value estimated based on steered MD simulations for the transition state of src SH3 (SI: Fig. S7). The predicted p(F ) for src SH3 using Eq. 7 at several loading rates from 0.1 pN/s to 10 pN/s are shown in Fig. 4B. The unfolding force histogram of SH3 was measured at a loading rate of 8 pN/s [4], which was converted to probability density function. The comparison between the experimental data and p(F ) predicted by Eq. 7 using the best-fitting parameters reported in this study shows very good agreement (Fig. 4C). As shown in the previous paragraphs, the five structural-elastic parameters (b0, γ0, b, γ, L) for I27, monomeric PSGL-1/ P-selectin and src SH3 are determined based on the best- fitting model parameters (σ, α, η), the molecular structures and steered MD simulations. With these structural-elastic parameters, the force-dependent transition distance δ(F ) and the change of the free energy barrier ∆Φ(F ) can be computed using Eq. 1 and Eq. 2 (Fig. 5). The results reveal that the three molecules have markedly different profiles of δ(F ) and ∆Φ(F ). For all the three molecules, the complex shapes of δ(F ) over 1-100 pN force range deviate from Bell's model that assumes a force-independent transition distance. These complex profiles of δ(F ) result in complex force-dependent changes of free energy barrier (∆Φ(F )), which in turn affects the force-dependence of the transition rate in a very complex manner. For I27 and PSGL-1/ P-selectin, the transition distances can become negative over a broad force range up to ∼ 20 pN, which results in a "catch-bond" behaviour at forces below 20 pN. Remarkably, the force-dependent transition distance drops dramatically when force increases from 0 pN to a few pN. These behaviours of the force-dependent transition distance are a result from the highly flexible nature of the peptide chain with a contour length L∗ and a persistence length A ∼ 0.8 nm [28] produced in the transition state. According to the WLC polymer model [29], the force-extension curve of a peptide polymer at low force regime 13 FIG. 4. Application of Eq. 5 to interpret experimental data of src SH3. (A) The k(F ) data obtained for src SH3 (Fig. 3A in Ref. [4]) are indicated with black squares and fitted with Eq. 5 (black line). The goodness-of-fit was evaluated by a R-Square of ∼ 0.992 and a Root mean squared error (RMSE) of ∼ 0.224. The best-fitting model parameters and the structural-elastic parameters determined based on the native state structure, steered MD simulation, or solved from the best-fitting parameters are indicated in the panel. (B) This panel shows the predicted src SH3 unfolding force density function p(F ) using Eq. 7 based on the best-fitting parameters for k(F ), at different loading rates of 0.1 pN/s (solid line), 0.5 pN/s (short dash line), 5 pN/s (short dot line) and 10 pN/s (dash line). (C) The predicted p(F ) of src SH3 (solid black curve) agrees with the previously published experimental data (Fig. 2B in Ref. [4]) (grey bars) at a loading rate of 8 pN/s. kBT (F < kBT /A ∼ 5 pN) can be approximated by a Hookean spring with a spring constant of AL∗ , which are ∼ 1.7 pN/nm for I27 and ∼ 3.1 pN/nm for PSGL-1/ P-selectin based on the respective best-fitting values of L∗. 3 2 14 b0b0∼1.90nm,γ0∼2900pN,L∗∼0nmbσ∼−0.441nmα∼0.049nm/pNη=0nm·pN1/2b∗∼1.6nmγ∗∼32pNABCFIG. 5. Force-dependent transition distance and change of free energy barrier. The force-dependent transition distance δ(F ) (solid line) calculated by Eq. 1 and the force-dependent change of the free energy barrier ∆Φ(F ) (dash dot line) calculated by Eq. 2 for I27 (A), monomeric PSGL-1/ P-selectin (B) and src SH3 (C) are shown. δ(F ) and ∆Φ(F ) are calculated based on the values of the five structural-elastic parameters (b0, γ0, b, γ, L) determined based on the best- fitting parameters (σ, α, η), the molecular structures and steered MD simulations for the respective molecules described in the Results section. III. DISCUSSION In summary, we have derived a novel analytical expression of k(F ) for single-barrier tran- sitions. Most importantly, the parameters are functions of the structural-elastic parameters of the molecules; therefore, their values directly inform us about the structural-elastic prop- erties of the molecule. We have shown that it is possible to determine the structural-elastic parameters of the molecule in both the native and the transition states by combining this 15 ABCmodel with the steered MD simulations. In our previous publication [2], based on the prior knowledge of the crystal structure of the native state of I27 (PDB ID:1TIT) and the structure of its transition state suggested from MD simulations [21–23], we calculated ∆Φ(F ) without any model parameters, with an assumption that both the native state and the folded portion of the transition state are non-deformable. Applying the Arrhenius law, we showed that this parameter-free ∆Φ(F ) correctly describes the shape of the experimentally measured k(F ) up to 100 pN. The only free parameter in that calculation is the attempting rate k0, which only affects the value of k(F ) (i.e., this parameter is unrelated to the force-dependence of k(F )). The work described in this paper differs from that earlier work in that: 1) it does not require prior knowledge of the structural-elastic properties of the molecule, 2) the best-fitting parameters (σ, α and η) reflect differences in the structural-elastic properties of the molecule between the transition and native states, and 3) with additional knowledge on the structural-elastic properties of the native state that can often be obtained from crystal structure and MD simulations, these best-fitting parameters can provide important information about the nature of the transition state. In most of experiments, k(F ) is measured over certain force range. Fitting to the data based on any kinetics model, it is attempting to extrapolate the fitted k(F ) to forces beyond the experimentally measured range. However, this is dangerous if the force extrapolated to is far away from the experimentally measured range. This is because the nature of the transition may vary with the force, while most of the models [9–12] , including ours, are derived based on assuming a unique initial folded state and a single transition barrier. Such assumption may only be valid in limited force range. For example, previous AFM experiments and MD simulations [22, 33] suggest that at forces below 100 pN, the initial folded state of I27 has all the seven β-strands folded in the native structure. However, at forces > 100 pN, the initial folded state transits to an intermediate state with the A strand detached from the B strand [22, 33]. Therefore, k(F ) fitted based on experimental data at forces below 100 pN should not be extrapolated to forces above 100 pN and vice versa. As an example, the zero-force transition rate of I27 was estimated to be ∼ 0.0005 s−1 by extrapolating experimental data obtained at forces above 100 pN according to Bell's model [32], which is about 10 times slower than that estimated by our model. However, that extrapolation did not take into consideration of the difference in the initial states between 16 forces below and above 100 pN. In addition, Bell's model cannot describe the recently reported "catch-bond" behaviour of I27 at forces below 22 pN [2], which further contributes to the discrepancy. The simple expression of Eq. 5 is derived based on large force asymptotic expansion (F (cid:29) kBT /b0, F (cid:29) kBT /b∗ and F (cid:29) kBT /A). The typical sizes of protein domain and the folded core in the transition state are in the order of a few nanometers; therefore, kBT /b0 and kBT /b∗ are close to 1 pN. If in the transition state a protein peptide or a ssDNA/ssRNA polymer is produced, due to their very small bending persistence of A ∼ 1 nm [28, 34], kBT /A ∼ 5 pN becomes the predominating factor that imposes a restriction to the lower boundary of force range to apply Eq. 5. In actual application, the applicable forces do not have to be much greater than 5 pN, since the force-extension curve of a flexible polymer with A ∼ 1 nm calculated based on the asymptotic large force expansion differs from the one according to the full Marko-Siggia formula [29] by less than 10% at forces above 3 pN (SI: Fig. S9). Therefore, Eq. 5 can be applied to forces > 3 pN. Consistently, we have shown that Eq. 5 can fit three different experimental data in this force range. Since Eq. 5 is not applicable at forces (cid:28) 3 pN, k0 should not be interpreted as the transition rate at zero force. Under cases where the five physical parameters (b0, γ0, b, γ* and L) can be solved from the best-fitting parameters (σ, α and η), extrapolation to lower forces is possible using the complete solution of Eq. S5. A better quantity that is more indicative of zero force transition rate is k0 in Eq. S5 that is derived without using asymptotic large force expansion, which can be computed based on the best fitting value of k0 according to Eq 6. However, caution should still be taken for such extrapolation since at very low forces the WLC model of the flexible protein peptide or ssDNA/ssRNA may no longer be valid due to potential formation of secondary structures on these polymers. The effects of the elastic properties of molecules on the force-dependent transition rate have been discussed in several previous works [35, 36]. In particular, in a pioneering work published by Dembo et al.[35], by treating the native and the transition states as molecular springs with different mechanical stiffness and lengths, the authors were able to predict the existence of catch, slip and ideal bonds. However, that model is too simple to explain complex k(F ) such as the "catch-to-slip" behaviour. In addition, treating the native and the transition states as molecular springs makes it impossible to relate the force dependence of transition rate to the actual structural parameters of the molecules in the native and 17 In another work by Cossio et al. transition states. For instance, it cannot predict whether there is a peptide produced in the transition state. [12], the authors discussed a free energy landscape that has a force-dependent transition distance, based on which k(F ) was derived by applying the Kramers kinetics theory. A phenomenological form of the force- dependent transition distance is proposed to describe the kinetic ductility that results in a monotonically decreased transition distance as a function of force, which could only describe transition with "slip" kinetics. Different from these previous studies, our derivation is based on the structural-elastic properties of molecules in the transition state and the native state. Therefore, its force dependence can be much richer. Depending on the structural-elastic properties of the molecules, the resulting force-dependent transition distance can be an increasing, decreasing or non-monotonic function of force. F(cid:82) δ∗(F (cid:48))dF (cid:48) β The analytical expressions of k(F ) (Eq. S5 and Eq. 5) are derived by applying Arrhenius law based on the structural-elastic parameters of molecules. The resulting relation between 0 the rate and the force-dependent transition distance, k(F ) = k0e , is identical to that obtained in the framework of the Kramers theory [20]. However, they differ from In our theory δ∗(F ) is calculated based on the structural- each other in a key aspect: elastic parameters of molecules; therefore it does not involve describing the system using any transition coordinate and it does not depend on the dimensionality of the system. In contrast, in the framework of Kramers theory, δ∗(F ) has to be calculated based on a presumed one-dimentional free energy landscape that must be expressed by the extension as the transition coordinate. As a result, δ∗(F ) depends on the structural-elastic parameters of the molecules in our theory, while it relies on the parameters associated with shapes of the presumed one-dimension free energy landscape in the framework of the Kramers theory [20]. Owing to this difference, our theory can be applied to a broader scope of experimental cases and the best-fitting parameters can provide important insights into the structural-elastic properties of the molecules in the native and the transition states. The three molecules selected to test the application of k(F ) derived in this work have markedly different profiles. The fact that the expression of k(F ) is able to perfectly fit the experimental data for all the three molecules reveals an exquisite interplay between the structural-elastic properties of molecules and the force-dependent transition rate. 18 IV. METHODS Titin I27 domain unfolding experiments – A vertical magnetic tweezers setup [37] was used for conducting in vitro titin I27 domain stretching experiments. The sample pro- tein (8I27) was designed with eight repeats of titin I27 domains spaced with flexible linkers (GGGSG) between each domain; The 8I27 was labeled with biotin-avi-tag at the N-terminus and spy-tag at the C-terminus. The expression plasmid for the sample protein was synthe- sised by geneArt. In a flow channel, the C-terminus of the protein was attached to the spycatcher-coated bottom surface through specific spy-spycather interaction, while the N- terminus was attached to a streptavidin-coated paramagnetic bead (2.8 µm in diameter, Dynabeads M-270) through specific biotin-streptavidin interaction. During experiments, the force on a single protein tether was linearly increased from ∼ 1 pN up to ∼ 120 pN with a loading rate of ∼ 0.08 pN/s, to allow the unfolding of each I27 domain; after unfolding of the domains, the force was decreased to ∼ 1 pN for ∼ 60 sec to allow refolding of the domains before next force-increase scans. Each I27 unfolding events and its corresponding unfolding force were detected by a home-written step-finding algorithm. All experiments were performed in buffered solution containing 1× PBS, 1% BSA, 1 mM DTT, at 22 ± 1 ◦C. Additional information of the magnetic tweezers setup, force calibration, step-finding algorithm, protein sequences, protein expression, and flow channel preparation can be found in previous publications [2, 6, 37]. MD simulations – The all-atom molecular dynamics (MD) simulations used to estimate the value of γ of the folded structure are introduced in the Supporting Information (SI: SI-II). Data extraction – The data of k(F ) for monomeric PSGL-1/P-selectin complex and src SH3 domain, and the histogram of unfolding force for src SH3 were obtained by digitizing previously published experimental data ( Fig. 4b in [3] for PSGL-1/P-selectin data, and Fig. 3A and Fig. 2B in [4] for src SH3 data). The values of k(F ) and the histogram of unfolding force were extracted using ImageJ with the Figure Calibration plugin developed by Frederic V. Hessman from Institut für Astrophysik Göttingen. 19 V. AUTHOR CONTRIBUTIONS J.Y., S.G. and H.C. developed the theory. Q.T. performed steered MD simulations. S.G. and M.Y. performed the calculation and data fitting. Q.T. and S.L. performed the titin I27 domain unfolding experiments. J.Y. and S.G. wrote the paper. J.Y. conceived and supervised the study. VI. ACKNOWLEDGEMENT The authors thank Jacques Prost (Institut Curie) for many stimulating discussions. This work was supported by the National Research Foundation (NRF), Prime Minister's Of- fice, Singapore under its NRF Investigatorship Programme (NRF Investigatorship Award No. NRF-NRFI2016-03 to JY) and through the Mechanobiology Institute (to JY), Hu- man Frontier Science Program (RGP00001/2016 to JY), and the National Nature Science Foundation of China (11474237 to HC). ∗ The first two authors contributed equally to this work † [email protected] [1] T. Iskratsch, H. Wolfenson, and M. P. Sheetz, Nature Reviews Molecular Cell Biology 15, 825 (2014). [2] G. Yuan, S. Le, M. Yao, H. Qian, X. Zhou, J. Yan, and H. Chen, Angewandte Chemie 129, 5582 (2017). [3] B. T. Marshall, M. Long, J. W. Piper, T. Yago, R. P. McEver, and C. Zhu, Nature 423, 190 (2003). [4] B. Jagannathan, P. J. Elms, C. Bustamante, and S. Marqusee, Proc. Natl. Acad. Sci. U.S.A. 109, 17820 (2012). [5] S. Rakshit, Y. Zhang, K. Manibog, O. Shafraz, and S. Sivasankar, Proceedings of the National Academy of Sciences 109, 18815 (2012). [6] H. Chen, G. Yuan, R. S. Winardhi, M. Yao, I. Popa, J. M. Fernandez, and J. Yan, J. Am. Chem. Soc. 137, 3540 (2015). 20 [7] D. T. Edwards, J. K. Faulk, A. W. Sanders, M. S. Bull, R. Walder, M.-A. LeBlanc, M. C. Sousa, and T. T. Perkins, Nano letters 15, 7091 (2015). [8] G. Bell, Science 200, 618 (1978). [9] G. Hummer and A. Szabo, Biophysical journal 85, 5 (2003). [10] E. Evans and K. Ritchie, Biophysical journal 72, 1541 (1997). [11] O. K. Dudko, G. Hummer, and A. Szabo, Physical review letters 96, 108101 (2006). [12] P. Cossio, G. Hummer, and A. Szabo, Biophysical journal 111, 832 (2016). [13] Y. V. Pereverzev, O. V. Prezhdo, M. Forero, E. V. Sokurenko, and W. E. Thomas, Biophysical journal 89, 1446 (2005). [14] V. Barsegov and D. Thirumalai, Proceedings of the National Academy of Sciences of the United States of America 102, 1835 (2005). [15] E. Evans, A. Leung, V. Heinrich, and C. Zhu, Proceedings of the National Academy of Sciences of the United States of America 101, 11281 (2004). [16] C. A. Pierse and O. K. Dudko, Physical Review Letters 118, 088101 (2017). [17] D. Bartolo, I. Derényi, and A. Ajdari, Physical Review E 65, 051910 (2002). [18] A. A. Rebane, L. Ma, and Y. Zhang, Biophysical journal 110, 441 (2016). [19] H. A. Kramers, Physica 7, 284 (1940). [20] O. K. Dudko, G. Hummer, and A. Szabo, Proceedings of the National Academy of Sciences 105, 15755 (2008). [21] H. Lu, B. Isralewitz, A. Krammer, V. Vogel, and K. Schulten, Biophys. J. 75, 662 (1998). [22] H. Lu and K. Schulten, Chem. Phys. 247, 141 (1999). [23] R. B. Best, S. B. Fowler, J. L. Herrera, A. Steward, E. Paci, and J. Clarke, J. Mol. Biol. 330, 867 (2003). [24] P. M. Williams, S. B. Fowler, R. B. Best, J. L. Toca-Herrera, K. A. Scott, A. Steward, and J. Clarke, Nature 422, 446 (2003). [25] C. Bouchiat, M. Wang, J.-F. Allemand, T. Strick, S. Block, and V. Croquette, Biophysical journal 76, 409 (1999). [26] P. Cong, L. Dai, H. Chen, J. R. van der Maarel, P. S. Doyle, and J. Yan, Biophysical journal 109, 2338 (2015). [27] S. B. Smith, Y. Cui, and C. Bustamente, Science 271, 795 (1996). [28] R. S. Winardhi, Q. Tang, J. Chen, M. Yao, and J. Yan, Biophysical Journal 111, 2349 (2016). 21 [29] J. Marko and E. Siggia, Macromolecules 28, 8759 (1995). [30] S. Cocco, J. Yan, J.-F. Léger, D. Chatenay, and J. F. Marko, Physical Review E 70, 011910 (2004). [31] I. Rouzina and V. A. Bloomfield, Biophysical journal 80, 882 (2001). [32] M. Carrion-Vazquez, A. F. Oberhauser, S. B. Fowler, P. E. Marszalek, S. E. Broedel, J. Clarke, and J. M. Fernandez, Proceedings of the National Academy of Sciences 96, 3694 (1999). [33] P. E. Marszalek, H. Lu, H. Li, M. Carrion-Vazquez, A. F. Oberhauser, K. Schulten, and J. M. Fernandez, Nature 402, 100 (1999). [34] A. Bosco, J. Camunas-Soler, and F. Ritort, Nucleic acids research 42, 2064 (2013). [35] M. Dembo, D. Torney, K. Saxman, and D. Hammer, Proceedings of the Royal Society of London B: Biological Sciences 234, 55 (1988). [36] J. Valle-Orero, E. C. Eckels, G. Stirnemann, I. Popa, R. Berkovich, and J. M. Fernandez, Biochemical and biophysical research communications 460, 434 (2015). [37] H. Chen, H. Fu, X. Zhu, P. Cong, F. Nakamura, and J. Yan, Biophys. J. 100, 517 (2011). [38] M. J. Abraham, T. Murtola, R. Schulz, S. Páll, J. C. Smith, B. Hess, and E. Lindahl, SoftwareX 1, 19 (2015). [39] I. Ivani, P. D. Dans, A. Noy, A. Pérez, I. Faustino, A. Hospital, J. Walther, P. Andrio, R. Goñi, A. Balaceanu, et al., Nature methods 13, 55 (2016). [40] R. B. Best, X. Zhu, J. Shim, P. E. Lopes, J. Mittal, M. Feig, and A. D. MacKerell Jr, Journal of chemical theory and computation 8, 3257 (2012). [41] X.-J. Lu and W. K. Olson, Nucleic acids research 31, 5108 (2003). [42] S. Improta, A. S. Politou, and A. Pastore, Structure 4, 323 (1996). [43] H. Yu, M. K. Rosen, and S. L. Schreiber, FEBS letters 324, 87 (1993). [44] W. S. Somers, J. Tang, G. D. Shaw, and R. T. Camphausen, Cell 103, 467 (2000). [45] W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L. Klein, The Journal of chemical physics 79, 926 (1983). [46] C. Bustamante, S. B. Smith, J. Liphardt, and D. Smith, Current opinion in structural biology 10, 279 (2000). [47] H. Fu, H. Chen, X. Zhang, Y. Qu, J. F. Marko, and J. Yan, Nucleic Acids Res. 39, 3473 (2011). 22 [48] X. Zhang, H. Chen, S. Le, I. Rouzina, P. S. Doyle, and J. Yan, Proc. Natl. Acad. Sci. U.S.A. 110, 3865 (2013). 23 Appendix: Supplemental Information The supporting information describes: 1) the steered molecular dynamics simulation method (SI); 2)the extraction of the structural and elastic parameters of the native and the transition states of molecules investigated in this paper (SII); 3) the conformational free energy of a molecule under external force (SIII); 4) an alternative derivation of Eq. 5 (SIV); 5) bootstrap analysis to determine fitting errors (SV); 6) the robustness of convergence of the best-fitting parameters (SVI); and 7) fitting of Eq. 5 to experimental data of src SH3 domain over different presupposed peptide length in the transition state (SVII). SI. Molecular dynamics simulations All-atom molecular dynamics (MD) simulations were performed in Gromacs 5.1.1 [38] with Parmbsc1 force field [39] for DNA and with CHARMM36 force field [40] for proteins. Molecular structures of DNA is built by x3DNA software [41], and the structures of titin I27 domain (PDB: 1tit) [42], src SH3 domain (PDB: 1srl) [43], monomeric PSGL-1/P-selectin complex (PDB: 1g1s) [44] are public data from protein data bank. All the simulations used explicit water TIP3P [45] with 150 mM NaCl to mimic physiological condition. Simulation boxes were heated to 300 K and then kept at constant temperature and pressure for 200 ps to relax. During steered molecular dynamics simulations, a constant force was applied to the force-bearing residues, therefore the end-to-end distance (extension) fluctuation of the molecules could be analysed. Standard deviation and mean value of extension were calculated from the last 20 ns of simulation. The transition state of src SH3 is determined by steered MD simulations. A sequence of harmonic traps with same stiffness of 1000 pN/nm and different center separation of 2.1-2.6 nm were applied to the same force-bearing residues as in experiment [4]. With this stretching setup, the force slowly build up between the stretching residues, and the structure has enough time to relax to equilibrium. The force on stretching residues were recorded and concatenated (Fig. S1). Structural transition is indicated by the force drop occured at a trap separation of 2.4 nm. The structure after force drop was regarded as a transition state of the protein domain during unfolding. 24 SII. Structural and elastic properties of the native and the transition states of molecules The contour length of the folded structure in native state or the folded core in transition state of the molecules were estimated based on structures of the molecules (Fig. S2-S4). Molecular dynamics simulations were used to determine the stretching stiffness of typical folded structures. Denoting b(0) and b(F ) the folded structure lengths in the absence or presence of force, and assuming Hookean stretching elasticity, we have: b(F ) = b(0) + b(0) γ F, (S1) where γ is the stretching stiffness and b(0) γ describes the stretching deformability of the folded structure. Therefore, for a folded structure, γ could be calculated from the linear dependence of b(F ) on F. We calibrated this method for double-stranded DNA (dsDNA), whose stretching stiffness is in the range of 1000 − 3000 pN, as measured from single-molecule stretching experiments [29, 46–48]. The estimated γ of dsDNA (150 mM NaCl) from our MD simulation is around 1500 pN (Fig. S5), which is consistent with experimentally measured values. Using this approach, we estimated γ for the native state of titin I27 (Fig. S6), as well as the native and transition states of src SH3 (Fig. S7). TABLE S1. γ (pN) for different structures dsDNA I27 SH3 native SH3 transition 1500 1900 2900 86 SIII. The conformational free energy of a molecule stretched by an external constant force In general, an external constant force F applied to a molecule in a given state introduces a conformational free energy to the state by: Φ(x, F ) = x(cid:90)0 f (x(cid:48))dx(cid:48) − F x, 25 (S2) where x is the extension of the molecule in this state, and F is the applied force. The external force contributes to a potential energy of −F x. At equilibrium, x is no longer independent from F , since it depends on F through the force-extension curve x(F ). Therefore, this − F x(F ), which can be energy becomes dependent only on force: Φ(x(F ), F ) = rewritten to a simpler form by Legendre transformation [30, 31]: f (x(cid:48))dx(cid:48) x(F )(cid:82)0 Φ(F ) = x(F (cid:48))dF (cid:48). F(cid:90)0 This can be easily seen by the relation (Fig: S8): f (x(cid:48))dx(cid:48) + xeq(cid:90)0 F(cid:90)0 x(F (cid:48))dF (cid:48) = F xeq. (S3) (S4) SIV. Alternative derivation of Eq. 5 Based on the force-dependent free energies of the molecule in both native and transition states that are shown in Eq. 3, k(F ) is determined by applying the Arrhenius law k(F ) = k0e−β∆Φ∗(F ): k(F ) = k0e −β(Φb∗,γ∗ (F )−Φb0,γ0 (F )+ΦL∗ (F )). (S5) In the main text, the large force expression Eq. 5 can be derived based on direct asymptotic expansion from Eq. 4. Here we provide an alternative derivation based on large-force expansion of force-extension curves of folded structure and flexible polymer. At large forces (F (cid:29) kBT /b and F (cid:29) kBT /A), the force-extension curves of the extensible folded structure and the flexible polymer have very simple asymptotic expressions:  xb,γ(F ) ≈ b(1 − kBT xL(F ) ≈ L(1 −(cid:113) kBT F b )(1 + F 4AF ). γ ), (S6) These expressions are derived based on large force expansion (F (cid:29) kBT /b and F (cid:29) kBT /A). The typical sizes of protein domain and the folded core in the transition state are in the order of a few nm; therefore, kBT /b are close to 1 pN. If in the transition state a protein peptide or a ssDNA/ssRNA polymer is produced, due to their very small bending 26 persistence length of A ∼ 1 nm, kBT /A ∼ 5 pN becomes the predominating factor that imposes a restriction to the lower boundary of force range. In actual applications, however, the applicable forces do not have to be much greater than 5 pN, since the force-extension curve of a flexible polymer with A = 0.8 nm and L = 5 nm calculated based on the asymptotic large force expansion differs from the one according to the full Marko-Siggia formula by less than 10% (Fig. S9) at forces above 3 pN. Based on these large-force asymptotic expressions of the force-extension curves, it is straightforward to show that the force-dependent change in the free energy barrier is: Here σ = L∗ + (b∗ − b0) − ( kBT ∆Φ∗(F ) ≈ −(cid:0)σF + αF 2/2 − ηF 1/2(cid:1) . γ∗ − kBT γ0 ), α = b∗ γ∗ − b0 γ0 , and η = L∗(cid:113) kBT (S7) kBT γ0 and kBT Eq. 5 is obtained by applying the Arrhenius law: γ∗ are in the range of 10−3 nm - 10−2 nm (SI: Sec II); therefore, σ ∼ L∗ + (b∗ A . Typical values of − b0). k(F ) = k0eβ(σF +αF 2/2−ηF 1/2). (S8) SV. Bootstrap analysis to determine fitting errors In order to test the robustness of fitting of Eq. 5 to experimental data, for each molecule studied in our work, we performed 1000 times of fitting with 80% data points that are randomly chosen from the experimentally measured k(F ) data for every fitting. We found that all the 1000 sets of the best-fitting parameters are in the reasonable range around the best-fitting parameters that are determined using the whole experimental data. Table S2, Table S3 and Table S4 have shown the averages and the standard deviations of the best- fitting parameters (k0, σ, α, and η) for 1000 times of fitting with the randomly chosen data points, which occupy 80% of the whole experimental data in each fitting. The structural- elastic parameters in the transition state determined based on the native state structure, steered MD simulation, or solved from the best-fitting parameters are also indicated in the tables. 27 TABLE S2. Parameters for I27 by fitting Eq. 5 to 1000 sets of 80% data points Best-fitting parameters Structural-elastic parameters k0 (s−1) σ (nm) α (nm/pN) η (nm·pN1/2) L∗ (nm) b∗ (nm) 0.026 0.9 0.4 Standard deviation 0.014 10.023 1.542 0.003 0.003 1.014 0.243 Average 4.4 0.7 γ∗ (pN) 179 41 TABLE S3. Parameters for sPSGL-1 / sP-selectin by fitting Eq. 5 to 1000 sets of 80% data points Structural-elastic parameters k0 (s−1) σ (nm) α (nm/pN) η (nm·pN1/2) L∗ (nm) b∗ (nm) γ0=γ∗ (pN) 58.498 Standard deviation 27.083 Best-fitting parameters -0.005 0.001 5.752 1.275 0.727 0.162 Average 379 48 2.5 0.6 5.5 0.4 TABLE S4. Parameters for src SH3 by fitting Eq. 5 to 1000 sets of 80% data points Structural-elastic parameters Best-fitting parameters k0 (s−1) σ (nm) α (nm/pN) b∗ (nm) 0.045 Standard deviation 0.043 -0.475 0.249 0.050 0.009 Average 1.4 0.2 γ∗ (pN) 29 9 SVI. Robustness of convergence of the best-fitting parameters We have tested whether the best-fitting values of these parameters in Eq. 5 (σ, α and η) are uniquely determined for a given shape of k(F ), by starting from many well-separated different initial values for the fitting of k(F ). We used lsqcurvefit function in Matlab to solve the nonlinear curve-fitting (data-fitting) problems in least-squares sense. By starting with different initial points for the fitting, lsqcurvefit may find a local solution that is not particularly close to the global best-fitting parameter values. So if another set of solutions exists that can fit equally well the data, one of the well-separated initial values may lead to a new set of solutions due to the existence of possible local minimums. However, for each of the three molecules tested in the study, we have found that the parameters always converge to the same set regardless of the initial values (Table S5 for I27, Table S6 for sPSGL-1/ sP-selectin, Table S7 for src SH3), which means the best-fitting parameters can be uniquely determined when applying Eq. 5 to fit experimental data of k(F ), at least for all the three cases studied in this work. 28 TABLE S5. Best-fitting parameters for I27 with different initial values Best-fitting values Initial values Case k0 (s−1) σ (nm) α (nm/pN) η (nm·pN1/2) k0 (s−1) σ (nm) α (nm/pN) η (nm·pN1/2) 1 2 3 4 5 6 7 8 9 10 10.519 10.519 10.519 10.519 10.519 10.519 10.519 10.519 10.519 10.519 0.026 0.026 0.026 0.026 0.026 0.026 0.026 0.026 0.026 0.026 1.099 1.099 1.099 1.099 1.099 1.099 1.099 1.099 1.099 1.099 0.002 0.002 0.002 0.002 0.002 0.002 0.002 0.002 0.002 0.002 0.01 0.01 0.01 0.01 0.01 0.01 0.01 1 -10 -10 -10 0 0 10 0 0 0 10 -10 -10 -10 0 0 0 10 0 0 10 0 5 10 5 10 5 5 10 10 10 0.00001 1 TABLE S6. Best-fitting parameters for sPSGL-1/ sP-selectin with different initial values Initial values Best-fitting values Case k0 (s−1) σ (nm) α (nm/pN) η (nm·pN1/2) k0 (s−1) σ (nm) α (nm/pN) η (nm·pN1/2) 1 2 3 4 5 6 7 8 9 10 51.786 51.786 51.786 51.786 51.786 51.786 51.786 51.786 51.786 51.786 -0.005 -0.005 -0.005 -0.005 -0.005 -0.005 -0.005 -0.005 -0.005 -0.005 0.723 0.723 0.723 0.723 0.723 0.723 0.723 0.723 0.723 0.723 5.760 5.760 5.760 5.760 5.760 5.760 5.760 5.760 5.760 5.760 1 1 1 1 1 1 1 1 10 100 0 0 0 0 0 10 -5 -10 0 0 0 0 0 5 10 10 0 0 0 0 0 5 10 0 0 10 5 5 5 5 TABLE S7. Best-fitting parameters for src SH3 with different initial values Initial values Best-fitting values Case k0 (s−1) σ (nm) α (nm/pN) k0 (s−1) σ (nm) α (nm/pN) 1 2 3 4 5 6 7 8 9 10 -0.441 -0.441 -0.441 -0.441 -0.441 -0.441 -0.441 -0.441 -0.441 -0.441 0.049 0.049 0.049 0.049 0.049 0.049 0.049 0.049 0.049 0.049 0.030 0.030 0.030 0.030 0.030 0.030 0.030 0.030 0.030 0.030 0 0 0 10 10 -10 -100 10 10 100 0 10 100 -10 -100 10 10 10 10 100 0.1 0.1 0.1 0.1 0.1 0.1 0.1 10 100 100 29 SVII. Fitting of Eq. 5 to experimental data of src SH3 In the fitting of Eq.5 to the k(F ) data of src SH3, η < 4.3 nm·pN1/2 is needed to ensure a positive b∗. By restricting the number of residues of the flexible peptide in the transition state of src SH3, good quality of fitting can be obtained (Fig. S10), which suggests that the peptide length is not a key factor for the k(F ) profile. At each peptide length, the best-fitting parameters predict α > 0 and γ∗ (cid:28) γ0, indicating that a much softer folded core in the transition state than that of the native state is the predominant factor of k(F ) (Table S8). Previous study estimated a small transition distance ∼ 0.45 nm in the force range of 15 − 25 pN [4], suggesting insignificant fraction of peptide in the transition state. Otherwise, considering 0.22 − 0.28 nm per residue of typical peptide in 15 − 25 pN force range [28], one would expect a significantly larger transition distance if a long peptide (> 3 a.a) is produced in the transition state. TABLE S8. Fitting parameters for src SH3 n∗ L∗ (nm) η (nm·pN1/2) σ (nm) α (nm/pN) γ∗ (pN) 25 1 20 2 16 3 4 10 4 5 -0.317 -0.196 -0.066 0.049 0.179 0.048 0.046 0.044 0.043 0.042 0.38 0.76 1.14 1.52 1.90 0.86 1.7 2.6 3.4 4.3 n∗ is the number of residues assumed for the peptide length in the transition state of src SH3. L∗ is the contour length of the flexible polymer, which is determined based on L∗ = n∗lr and lr ∼ 0.38 nm for peptide chain. The value of η is restricted by η = L∗(cid:113) kBT in the fitting of Eq. 2 to the experimental data of src SH3 for each peptide length. σ and α are the besting fitting values. Based on the structure of the native state and using steered MD simulation, the structural-elastic parameters of the native state are determined to be b0 ∼ 1.90 nm and γ0 ∼ 2900 pN (SII). From these parameters, the value of γ∗ was solved for each presupposed peptide length. The goodness-of-fit is evaluated by R-square ∼ 0.992 A for all the fittings. 30 Supplementary figures FIG. S1. Force applied on src SH3 domain during steered MD simulation. Forces from a sequence of simulation (20 ns each) with increasing harmonic trap separation from 2.1nm-2.6nm were concatenated. At beginning, stepwised increase in force was observed as trap separation increased. As the trap separation increased to 2.4nm, the force firstly increased and suddenly dropped off, indicating a structural transition in the protein domain. The force drop was followed by a much weaker dependence of force on trap separation as the separation continued to increase, indicating a very different structure produced under stretching. Thus the structure after force drop was characterised as a transition state. FIG. S2. The structure of titin I27 domain in native state and transition state. (A) The native state of I27 is a folded structure with b0 = 4.32 nm. (B) The transition state of I27 is composed of a peptide of 13 residues [21–24] under force and a folded core with a relaxed length of b∗ = 0.64 nm. 31 020406080100120Time(ns)-200-1000100200300400Force(pN)b0 ~ 4.32 nmbL* ~ 4.94 nmFIG. S3. The structure of monomeric PSGL-1/ P-selectin complex in native state. The protein complex has a folded structure length of b0 = 7.28 nm. It contains a SLex sugar chain (red) covalently linked to PSGL-1 (cyan) that binds to P-selectin (green). FIG. S4. The structure of src SH3 in native state and transition state under force. (A) In the native state, the distance between force-bearing residues is 1.90 nm, thus it is regarded as a folded structure with a relaxed length of b0 = 1.90 nm. (B) A snapshot of the transition state produced by sequential stretching by harmonic traps (SI: Sec I). The hydrogen bonds in N-terminal remains(as shown in yellow dashed lines, key residues involved are shown in line representation), while the C-terminal peptide peels off, which is not under force, so the released peptide under force is negligible. 32 b0 ~ 7.28 nmb0 ~ 1.90 nmb*FIG. S5. Elasticity of dsDNA molecule. Extension of 50 bp dsDNA was measured at constant forces (squares in the figure show the average value of extension and vertical error bars indicate the standard deviation). The simulations run for 50 ns before which no structural transition occurs, so the extensions were a pure elastic response. The fitting parameter of the slope s is determined to be (with 95% confidence bounds): s = 0.011 (0.011, 0.011) nm/pN. Based on the well-known B-form DNA contour length ∼ 0.34 nm per basepair, the value of b is determined as b ∼ 0.34 × 50 = 17 nm. As a result, the elasticity of dsDNA molecule is estimated to be γ ∼ 1500 pN. FIG. S6. Elasticity of titin I27 domain in native state. Extension of titin I27 domain in the native state was measured at constant forces (squares in the figure show the average value of extension and vertical error bars indicate the standard deviation). The simulations run for 50 ns before which no structural transition occurs, so the extensions were a pure elastic response. The fitting parameter of the slope s is determined to be (with 95% confidence bounds): s = 0.0023 (0.0006, 0.0040) nm/pN. According to the value of b ∼ 4.3 nm, which is obtained based on the structure of I27 in native state, the elasticity of folded titin I27 domain is estimated to be γ ∼ 1900 pN. 33 406080100120140Force (pN)1616.51717.5Extension (nm)y=0.011x+160100200300Force (pN)44.24.44.64.85Extension (nm)y=0.0023x+4.2FIG. S7. Elasticity of src SH3 domain in native state and transition state. Extension of src SH3 domain in the native and transition states were measured at constant forces (squares in the figure show the average value of extension and vertical error bars indicate the standard deviation). The simulations run for 50 ns before which no structural transition occurs, so the extensions were a pure elastic response. For the native state, the fitting parameter of the slope s is determined to be (with 95% confidence bounds): s = 0.00065 (0.00057, 0.00073) nm/pN. According to the value of b ∼ 1.90 nm, which is obtained based on the structure of src SH3 in native state, the elasticity of folded src SH3 is estimated to be γ ∼ 2900 pN. Similarly, the slope s for transition state is determined to be s = 0.022 (0.007, 0.037) nm/pN. Since the folded core of the transition state maintains the overall structure as in the native state, the value of b is also estimated to be b ∼ 1.90 nm for the transition state, from which the elasticity of the transition state of src SH3 is calculated to be γ ∼ 86 pN. FIG. S8. The conformational free energy of a molecule under external force. At equilib- rium, the conformational free energy of a molecule under force Φ(x(F ), F ) = equals Φ(F ) = x(F (cid:48))dF (cid:48). F(cid:82)0 f (x(cid:48))dx(cid:48) − F x(F ) x(F )(cid:82)0 34 0100200300Force (pN)11.522.53Extension (nm)y=0.00065x+1.5y=0.022x+1.1transition statefolded state𝑥𝑓(𝑥)𝑥𝑒𝑞𝐹∫𝑥𝑓)𝑑𝑓)+,∫𝑓𝑥)𝑑𝑥)-./,𝑓𝑥𝑒𝑞= FFIG. S9. Force-extension curves of flexible polymer. The force-extension curve of a flexible polymer with A = 0.8 nm and L = 5 nm calculated based on the asymptotic large force expansion (dash line) differs from the one from the full Marko-Siggia formula (solid line) by less than 10%. FIG. S10. Fitting of Eq. 5 to experimental data of src SH3. By restricting the number of residues of the flexible peptide in the transition state of src SH3, good quality of fitting can be obtained. The figure shows the fitting curves of experimental data for src SH3 domain when the number of peptide residue is presupposed to be 2, 3 and 5. The goodness-of-fit is evaluated by R-square ∼ 0.992 for all the fittings. 35
1703.04392	3	1703	2018-06-11T19:01:32	Enhancing human color vision by breaking binocular redundancy	[ "physics.bio-ph", "cs.CV", "physics.optics" ]	To see color, the human visual system combines the response of three types of cone cells in the retina--a compressive process that discards a significant amount of spectral information. Here, we present an approach to enhance human color vision by breaking its inherent binocular redundancy, providing different spectral content to each eye. We fabricated a set of optical filters that "splits" the response of the short-wavelength cone between the two eyes in individuals with typical trichromatic vision, simulating the presence of approximately four distinct cone types ("tetrachromacy"). Such an increase in the number of effective cone types can reduce the prevalence of metamers--pairs of distinct spectra that resolve to the same tristimulus values. This technique may result in an enhancement of spectral perception, with applications ranging from camouflage detection and anti-counterfeiting to new types of artwork and data visualization.	physics.bio-ph	physics	Enhancing human color vision by breaking binocular redundancy Bradley S. Gundlach1, Michel Frising1,2, Alireza Shahsafi1, Gregory Vershbow3, Chenghao Wan1,4, Jad Salman1, Bas Rokers5,6, Laurent Lessard1, Mikhail A. Kats1,4,6* 1Department of Electrical and Computer Engineering, University of Wisconsin-Madison, Madison, WI 2Department of Mechanical and Process Engineering, ETH Zurich, Zurich, Switzerland 3Department of Art, University of Wisconsin-Madison, Madison, WI 4Department of Materials Science and Engineering, University of Wisconsin-Madison, Madison, WI 5Department of Psychology, University of Wisconsin-Madison, Madison, WI 6McPherson Eye Research Institute, University of Wisconsin-Madison, Madison, WI ABSTRACT To see color, the human visual system combines the response of three types of cone cells in the retina-a compressive process that discards a significant amount of spectral information. Here, we present an approach to enhance human color vision by breaking its inherent binocular redundancy, providing different spectral content to each eye. We fabricated a set of optical filters that "splits" the response of the short-wavelength cone between the two eyes in individuals with typical trichromatic vision, simulating the presence of approximately four distinct cone types ("tetrachromacy"). Such an increase in the number of effective cone types can reduce the prevalence of metamers-pairs of distinct spectra that resolve to the same tristimulus values. This technique may result in an enhancement of spectral perception, with applications ranging from camouflage detection and anti-counterfeiting to new types of artwork and data visualization. 1 Introduction In the typical human eye, the three cone types-labeled "S" for short wavelengths, "M" for medium, and "L" for long-are sensitive primarily to light with wavelengths in the 390 - 530 nm, 400 - 670 nm, and 400 - 700 nm bands, respectively1–4. When excited by light, the signal from the cones is relayed though retinal ganglion cells, to the optic nerve, and then the brain, where it is further processed to produce a color sensation5,6. This process can be understood as a type of lossy compression from an N-dimensional spectrum, where N is the number of wavelength bins necessary to sufficiently approximate a continuous spectrum, into a color, which is a three-dimensional object (Fig. 1). A manifestation of this N-to-three compression is metamerism, a phenomenon in which different spectra resolve to the same tristimulus values (i.e., they appear as the same color, neglecting possible contextual effects)2. The number of cone types and the widths and separations of their spectral sensitivities govern the degree to which metamerism is a limitation of the visual system; for example, a hypothetical increase in the number of distinct cone types should result in a decrease in the prevalence of metamers. Cast in a signal processing perspective, an increase in the sampling rate of a spectrum (i.e., the number of distinct cone types) improves the ability to detect sharp features in the spectrum7. Several studies have reported that a small percentage of humans, primarily women, express a mutated L cone in addition to the standard one, resulting in a total of four cone types, which may in principle enable vision with four color dimensions (tetrachromacy)8–10. Reports suggest that a few of these individuals can utilize this fourth photopigment type, and thus "perceive significantly more chromatic appearances" compared to typical, healthy humans with three cone types (trichromats)11,12. More broadly, it is reasonable to infer that an additional cone type would enhance spectral perception, provided subsequent neural processing can capitalize on its presence. 2 Here, we aim to simulate tetrachromatic (and possibly higher-dimensional) color vision in typical trichromatic humans by increasing the number of effective cone types in the visual system comprising the two eyes and a passive optical device. The term "effective" is used here to differentiate between true tetrachromatic vision, which would be defined by four distinct retinal photopigments that generate distinct neural responses. Our approach breaks the binocular redundancy of the two eyes, where the visual fields of each eye are overlapping, providing different spectral content to each eye via a wearable passive multispectral device comprising two optical transmission filters (Fig. 2). Figure 1: Compression of spectral information (a) A sample spectrum generated by a cathode ray tube (CRT) monitor displaying a purple color. The filled rectangle represents a single spectral bin, if the continuous spectrum is divided into N bins. (b) (i) Normalized spectral sensitivity of the cone types for a typical trichromatic observer (𝑀 = 3). (ii) Normalized spectral sensitivity of the effective cone types for a typical trichromatic observer enhanced using our device (𝑀 =෥ 4). (c) A representation of the perceived color of the spectrum in (a). The case with 𝑀 =෥ 4 cannot be displayed, but extra spectral information would be present compared to the 𝑀 = 3 case. 3 A number of existing vision-assistive devices or techniques operate by breaking binocular redundancy, though usually in the spatial rather than spectral domain. Examples include hemianopia (partial blindness in the left or right visual field) treatment using spectacles with a monocular sector prism that selectively relocates the visual field in one eye, leaving the other eye unaffected, and thus conferring an additional 20 of visual-field sensitivity for binocular vision13,14, and the treatment of presbyopia by correcting one eye for near vision and the other for distance vision15. Figure 2: Wearable passive multispectral device comprising two distinct transmission filters (a) Simplified schematic of an optical filter comprising several thin-film layers. (b) Measured transmission spectra of fabricated Filters 1 (black) and 2 (blue). (c) Magnified portion of the transmission spectrum from (b), including angle dependence with angle of incidence (AOI) from 0 to 15°. (d) Colors of an example metameric pair (1, 2) and D65 broadband white light (3), as perceived by a typical trichromat. The traces in each box are the underlying unfiltered spectra (arbitrary units), taken from our experiments, as described in Fig. 3. (e, f) Rendered monocular colors of the spectra in (d) after passing through Filter 1 and Filter 2, respectively. Note that e(1) and e(2), f(1) and f(2) are substantially distinct, while e(3) and f(3) are similar in color due to the white-balance constraint enforced in our design. 4 We break binocular redundancy spectrally by using filters that selectively attenuate different wavelength bands to yield effective cone sensitivities (i.e., the products of the cone sensitivities and the filter transmission spectra) that are different between the two eyes. This approach increases the number of effective cone types while preserving most spatial information. In this vein, the use of two simple band- pass filters was previously demonstrated to increase the dimensionality of color vision in dichromatic individuals (i.e., those with two functioning cone types)16. Conversely, the goal in the present work is to enhance the dimensionality of a trichromat's visual system to beyond that of a typical human. Such an approach was briefly suggested by Cornsweet in 197017, but to our knowledge no specific design has been proposed or realized. We note that the use of even a single filter positioned in front of both eyes can help distinguish certain metamers17, with the caveat that previously distinguishable spectra can become metamers when viewed through the filter; that is, a similar number of metamers (usually more) are created as are destroyed. In contrast, the use of two filters might be used to decrease the overall number of possible metamers. For this work, if at least one of the two filters can be used to differentiate a pair of spectra, we consider the pair to no longer be metamers. Results and Discussion Filter Design and Construction The filter pair was designed using a standard psychophysical model to determine the perceived (monocular) colors corresponding to particular spectra2,18. The perceived colors were calculated using the International Commission on Illumination (CIE) 1931 2° standard-observer matching functions, and monocular color differences (e.g., between colors 1 and 2) were calculated in the CIELAB color space using a standard color-difference metric (see Methods for further details)2,19: 5 ∆𝐸12 = √(𝐿2 − 𝐿1)2 + (𝑎2 − 𝑎1)2 + (𝑏2 − 𝑏1)2 (𝟏) The filters were designed to enhance the ability of a typical trichromatic viewer to discriminate spectra while limiting adverse effects. For simplicity, we focused on a design that splits the response of the S cone, thus transforming the trichromatic visual system into one that simulates tetrachromatic vision. The S cone was chosen because its responsivity has relatively little overlap with those of the M and L cones (Fig. 1b(i)), so it can be attenuated while minimizing the impact on the effective responsivity (i.e., the product of cone responsivity and the filter transmission response) of the other two cone types. To provide approximate parity between eyes, we partitioned the S cone responsivity such that each eye retains approximately half of the original response spectrum (Fig. 1b(ii)). Our secondary design goal was to ensure that the transmission of broadband white light (defined using CIE illuminant D65)20 through the two filters results in similar tristimulus values. This constraint was put in place to minimize potential baseline disparities (e.g., when viewing broadband white objects) between the eyes when the device is used in daylight. Though a particular implementation of this type of filter-based device generally depends on the illuminant chosen, the design presented here should work well for most illuminants along the Planckian locus21. The final device presented here comprises a 450 nm long-pass filter (Filter 1) and a 450 – 500 nm, 630 – 680 nm double-band-stop filter (Filter 2) (Fig. 2b-c); the filter designs were optimized by varying their stopband/passband positions and transmittances using simulated annealing to minimize the CIE ΔE color difference of D65 white light passing through Filters 1 and 2 (See Methods for further details). The 450 nm transition between Filters 1 and 2 is at the peak sensitivity of the S cone, and partitions it in half. However, due to the non-zero sensitivity of the M and L cones in the 450 – 500 nm region [Fig. 1b(i)], the M and L cones are also (unintentionally) attenuated by Filter 2. A second stopband at 630 – 680 nm was introduced to attenuate the effective responsivity of the M and L cones to broadband white light to preserve 6 color balance. Though the filter designs were optimized for these constraints, we note that the design presented here is a proof of concept, and is not a unique or globally optimal solution. To reduce cost and manufacturing time, an off-the-shelf component (450LP RapidEdge, Omega Optical), was used for Filter 1. The optimized transmission function of Filter 2 was realized using conventional thin-film technology22, with alternating layers of silicon oxide (SiO2 , n = 1.46) and tantalum oxide (Ta2O5, n = 2.15) (Fig. 2a-b), deposited on an NBK7 glass substrate (see Methods). The two filters were then characterized by angle-dependent transmission spectroscopy, demonstrating that the transmission spectra are robust to incidence angles up to 5 away from the normal (Fig. 2c). Following fabrication, the filters were constructed into a pair of glasses. Experiments To test the performance of this design, we constructed a setup that generates metameric spectra using a liquid crystal display (LCD, True HD-IPS display on LG G3 smartphone) and a cathode ray tube (CRT, Dell E770P) monitor (Fig. 3a-b). The displays use different emission mechanisms, and thus produce distinct spectra when displaying the same color (See Supplementary Information for further analysis)23,24. Blocks of color generated by the displays were presented side by side using a 50/50 beam splitter, and the colors were individually adjusted until no perceivable color difference was present. The emission spectra of each monitor were recorded using a free-space spectrometer, allowing for chromaticity and color- difference calculations to be made given a standard observer. A threshold value of 2.3 for the CIE ΔE "just noticeable difference" was taken to define perceptually indistinguishable spectra (i.e., ΔE < 2.3)25. See Methods for further details of the experimental setup. One representative example from this dual-display setup, using a pair of metamers that appear purple, is shown in Fig. 3b-d. Without the use of either filter, the two different spectra appeared as identical patches of color. However, when observed through either of the filters, the two can be differentiated. 7 Subjectively, we observed that, by looking at a particular patch through both filters simultaneously (i.e., Filter 1 over the left eye, Filter 2 over the right), a color percept is observed that is different from the color perceived through either filter individually or with no filter. This new percept may be a manifestation of the approximately four effective cone types created by the pair of filters. We note that a related study involving dichromats demonstrated an increase in color dimensionality using band-pass filters, which the authors suggest the effect could be related to binocular lustre16. Our proposed capability to distinguish metamers by breaking binocular redundancy may be affected by binocular lustre and/or rivalry, which might be advantageous provided it can be used as a cue to a difference in hue. For example, a recent report has demonstrated that compression artifacts of pixels in virtual reality images can be easily detected due to lustre26. Note that lustre and rivalry are both dynamic phenomena, even for static stimuli27–29; thus the use of lustre/rivalry may result in a tradeoff between temporal and spectral resolution. Though our current experiments do not directly investigate the effect of lustre or rivalry, or probe to what degree the neuronal processing system of a trichromat can take advantage of the extra spectral information resulting from binocular filtering, this can be explored in future work. We note that substantial differences in luminance between the two eyes, such as for spectra that transmit chiefly through only one filter, might lead to the Pulfrich effect30; however, our design was optimized to minimize differences in appearance of Illuminant D65, and the binocular luminance disparity required for the Pulfrich effect to occur is unlikely for most commonly occurring (i.e., smooth/broad) spectra. 8 Figure 3: Splitting metamers by breaking binocular redundancy (a) Schematic of our metamer generation setup. Images from two monitors, an LCD and a CRT display, are combined using a 50/50 beam splitter (i), to be viewed at location (ii). The two monitors use different emission mechanisms, and thus generate different spectra for the same color. (b) Photograph taken at position (ii) in the schematic. (c) Measured emission spectra from the LCD (solid) and CRT (dashed) monitors while displaying the same purple color. Spectra are shown with arbitrary units. (d) Rendered colors of the spectra in (c) viewed through no filter, Filter 1, and Filter 2, respectively, showing that the metamers can be distinguished using either filter. See the supplementary information for discussion of the difference between rendered colors in (d) and those in the photograph in (b). Calculation of metamer reduction Broadly stated, the number of cone types and their frequency-dependent responsivities determines the extent to which metamerism is a limitation to the visual system. Our method increases the number of effective cone types, which should decrease the number of potential metamers, provided the subsequent neuronal processing can adapt appropriately (which seems to occur in the case of spatial multiplexing used for vision-assistive devices13,14). In general, quantitatively determining the decrease in the metamer frequency is difficult because the set of possible metamers is not bounded. Nevertheless, various metrics can be applied to roughly estimate this quantity. For this work, we developed two separate metrics that describe this decrease in metamer frequency given the following conditions. Condition 1: Without the use 9 of filters, a metamer pair is defined by two spectra with a color difference ΔE < 2.325. Condition 2: With the use of binocular filters, such as those in Fig. 2b, a metamer pair is defined by two spectra with a monocular color difference ΔE < 2.3 in each eye. That is, a pair of spectra is a metamer if and only if it is a metamer in each eye individually. We do not consider the possibility of other perceptual effects such as binocular rivalry or dichoptic color mixing. Our first metric uses a Monte Carlo simulation to probe the effect binocular filters have on the perception of pairs of spectra, given the conditions above (Fig. 4). To start, a pair of reflectance spectra is generated by stochastically sampling intensity values from a uniform distribution at regularly spaced intervals within the visible wavelength range. Particularly, samples were taken uniformly at points: 𝜆1, 𝜆2…𝜆𝑁𝑠−1, 𝜆𝑁𝑠, where 𝑁𝑠 is the total number of sampling points. The sharpness of the reflectance spectra was adjusted by changing 𝑁𝑠, with larger numbers leading to sharper features, and were interpolated at 10 nm intervals using a cubic spline to create smooth spectra. We assumed illuminant D65, and then filtered the reflected spectra by the filter transmission responses given in Fig. 2b. ΔE color differences were calculated between the pairs of spectra for the unfiltered case, and through Filter 1 and Filter 2. The method was performed for various number of iterations (𝑁𝑖), which varied from 1,000 to 20,000,000, and the number of unfiltered (𝑀𝑢) and filtered (𝑀𝑓) metamers were recorded for each trial. We then defined a metric that represents the decrease in metamer frequency upon filtering: 𝑃𝑚 = 𝑀𝑢 𝑀𝑓 For example, 𝑃𝑚= 2 represents a two-fold decrease in the number of metamers using the two filters. The results from this simulation, for several sampling values (𝑁𝑠) and iteration numbers (𝑁𝑖), are given in Fig. S5 of the Supplementary Information. Given the simulation conditions, the filters in this work result in up to a ~15× decrease in the number of metamers for randomly generated spectra; this effect appears to be greatest for moderately sharp spectral features (𝑁𝑠 = 15), and drops off for very broad or very sharp 10 spectra. We note that, though the given metric seems to converge for larger iteration numbers (See Methods and Supplementary Information), these measures are only meaningful to within a factor of ~2 due to the stochastic nature of this calculation. Figure 4: Graphic representation of the Monte-Carlo metamer-reduction calculation, where four pairs of randomly generated spectra are selected as an illustration. For the actual calculation, many pairs are generated. The corresponding rendered colors when viewed under illuminant D65 are shown for unfiltered, filter 1, and filter 2 cases, respectively. If a metamer is present in the unfiltered case, 𝑀𝑢 is incremented by 1. If metamers are present when viewed through both Filter 1 and Filter 2, 𝑀𝑓 is incremented by 1. Note that if a metamer exists when viewing through Filter 1 but not Filter 2, or vice versa, the spectra are considered distinguishable and 𝑀𝑓 is not incremented. 11 As further verification of the apparent decrease in metamer frequency, we also developed a more- abstract mathematical method (See Supplementary Information for a complete description of this calculation, abridged here for clarity). Rather than comparing stochastically generated spectra, as above, this method aims to calculate the overall number of spectra that map to perceptually indistinguishable tristimulus values (ΔE < 2.3). For a given reference point in LAB space, [Lo, ao, bo], the number of metamers (with respect to the reference point) was determined by counting the spectra, 𝐼(𝜆), that map to LAB coordinates within a sphere of radius 2.3 around the reference point. We determined the number of metamers by calculating the volume of spectra, represented by an ellipsoid in 𝑁-dimensional space, where 𝑁 is the number of discrete wavelength bins that define a spectrum. However, calculating the exact volume of high dimensional ellipsoids in this case is difficult; instead, we calculate the volume of the max-inscribed ellipsoid subject to box constraints, which represents an upper-bound of the true value and is more computationally efficient31 (see Supplementary Information for more details). The volume of this ellipsoid represents the number of metamers, for a given reference point, for the unfiltered case (𝑉𝑢). For the filtered case, the union of two ellipsoids, corresponding to each filter individually, represents the number of metamers (𝑉𝑓); this is equivalent to Condition 2 above, where we assume that monocular metamerism must be present in both eyes simultaneously to yield indistinguishable color percepts in the filtered case. Thus, the overall decrease in metamer frequency is given by: 𝐹𝑚 = 𝑉𝑢 𝑉𝑓 Where 𝐹𝑚= 2, as an example, represents a two-fold decrease in metamer frequency. This process was repeated for 500 LAB reference points from randomly generated spectra to adequately sample the color space. The number of wavelength samples (𝑁𝑆) was also varied to again explore the effect of spectral sharpness; as in the Monte-Carlo simulation, the decrease in metamer frequency occurs around 12 – 16 bins. Using this metric, we estimate a decrease in metamer frequency by one-to-two orders of magnitude 12 when using our passive multispectral device (see Table S1 in the Supplementary Information). By the same metric, a single-filter system designed to improve vision in color-vision-deficient individuals32 seems to provide no decrease in the frequency of apparent metamers. Conclusion By breaking the inherent chromatic redundancy in binocular vision, our method provides the user with more spectral information than he/she would otherwise receive. In the present design, the S cone is partitioned using a pair of filters that results in photoreceptor responses consistent with a visual system that utilizes approximately four cone types (i.e., simulated tetrachromacy). The current demonstration splits the S cone, which is more selectively sensitive to blue-colored objects, and might find direct applications such as differentiating structural color versus natural pigments (See Supplementary Information)33. However, it is possible to use similar methods to design filters that more strongly affect metamers that appears as green and red, that are more prevalent in nature34. While the possibility of natural tetrachromacy in a fraction of the population has received both academic and popular interest10–12, the technology demonstrated here has the potential to simulate tetrachromatic vision in anyone with typical, healthy trichromatic vision. The extent to which observers can (or can learn) to take advantage of the additional spectral information is yet to be determined. Given two eyes and three types of cones, it should be possible to increase the number of effective cones up to six using our approach, and potentially even more with spatial or temporal multiplexing. It may also be possible to generate personalized designs to improve color discrimination for individuals with color- vision deficiencies. This technology can be integrated in a simple pair of eyeglasses or sunglasses, and could have immediate applications in camouflage detection, quality control, anti-counterfeiting, and more. More broadly, the ability to see many more colors has intriguing opportunities for design and artwork, and for data representation with extra color channels. 13 Methods Color calculations and CIE color differences. The International Commission on Illumination (CIE) standard was used for color calculations, represented by the equation2,18: 𝜆2 Θ = ∫ 𝜃(𝜆)𝑇(𝜆)𝐼(𝜆)𝑑𝜆 𝜆1 𝜆2 ⁄ ∫ 𝑦(𝜆)𝐼(𝜆)𝑑𝜆 𝜆1 where Θ = [X; Y; Z] are the XYZ tristimulus values, 𝜃(𝜆) = [𝑥(𝜆), 𝑦(𝜆), 𝑧(𝜆)] are the 1931 CIE 2° standard observer matching functions, 𝑇(𝜆) is the transmission spectrum of the filter, and 𝐼(𝜆) is the spectral irradiance of light passing through the filter. The XYZ tristimulus values can be transformed to a different color space (e.g., RGB); here, we use the CIELAB color space because it is more perceptually uniform and allows for straightforward calculations of perceived color differences. The XYZ to LAB transformation is given by19: ) − 16, 𝑎1 = 500 [𝑓 ( 𝑋 𝑋𝑛 ) − 𝑓 ( 𝑌 𝑌𝑛 )] , 𝑏1 = 200 [𝑓 ( 𝑌 𝑌𝑛 ) − 𝑓 ( 𝑍 𝑍𝑛 )], 𝑌 𝑌𝑛 𝐿1 = 116𝑓 ( where 𝑓(𝑡) = 𝑡1/3 , 2 ) ( 6 29 𝑡 + 4 29 , 𝑡 > ( 6 29 3 ) 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 { 1 3 and 𝑋𝑛, 𝑌𝑛, 𝑍𝑛 are the tristimulus values of the reference white point. Here, white light is defined by the CIE D65 standard illuminant, which roughly corresponds to average mid-day solar illuminance. The white point of D65 is (95.047, 100.000, 108.883)20. Filter Design An iterative optimization approach was used to design the transmission of Filter 1 and Filter 2, where the filter responses became more complex as our intuition grew between iterations. This approach was used to meet the primary design goal, splitting the spectral response of the S cone between eyes, while 14 also enforcing other optimization conditions such as a perceptual color balance for D65 white light between eyes. The final filter designs comprise a 450nm longpass filter (Filter 1) and a 450-500nm, 630-680nm double bandstop filter (Filter 2). The 450nm transition region, where Filter 1 cuts on and the first bandstop of Filter 2 cuts off, occurs roughly at the peak sensitivity of the S cone. Therefore, Filter 1 transmits the long wavelength half of the S cone, while the first stopband of Filter 2 transmits the short wavelength half of the S cone. However, due to the nonzero sensitivity of the M and L cones between 450-500nm, their sensitivities are also inadvertently attenuated, impacting the D65 color balance between eyes. The second bandstop of Filter 2, between 630-680nm, attenuates the long wavelength tails of the M and L cone sensitivities, restoring color balance between eyes. A 450nm longpass filter (Omega Optical, 450LP RapidEdge) was chosen as Filter 1. Filter 2 was optimized using constrained optimization by linear approximation (COBYLA) to minimize the merit function: Δ𝐸 𝑊𝑖𝑑𝑡ℎ𝑏𝑎𝑛𝑑𝑠𝑡𝑜𝑝 ⁄ , where Δ𝐸 is the color difference between Filter 1 and Filter 2 when transmitting D65 white light and Widthbandstop is the spectral width of the short-wavelength band-stop region of Filter 2. This merit function ensures satisfactory D65 color balance between eyes while also maximizing the difference between filters, enhancing their ability to distinguish spectra. For Filter 2, the transmittance of the pass and stop-bands were constrained between 5- 15% and 80-95%, respectively, and the longer wavelength stopband was constrained between 600-700nm to prevent attenuation of the M and L cones at their peak sensitivities (~550nm, ~580nm respectively). This procedure yielded an optimized response for Filter 2 with stopbands at 450 - 500 nm and 630 - 680 nm, and stopband/passband transmittance of 10% and 90%, respectively; the color difference for illuminant D65 between Filter 1 and Filter 2 is Δ𝐸 = 5.21, with chromaticities of (0.348,0.406) and (0.35,0.415), respectively. 15 Thin-film optimization and construction The required film thicknesses were determined by conventional thin-film optimization methods, including gradual evolution35 and needle optimization36, to implement the target transmission function. The final stack was constrained to be less than 75 total layers, and each layer between 10 and 500 nm thick. The filter was optimized such that the transmission would not change significantly for incident angles up to 5 away from the normal. The films were deposited using ion-assisted sputtering onto an NBK7 glass substrate at a thin-film foundry (Iridian Spectral Technologies, Ontario, Canada). See Supplementary Information for more information about the thin-film design. Metamer Generation An LCD (True HD-IPS on LG G3 smartphone) and CRT (Dell E770P) display were used to generate metameric pairs. The monitors were placed at a 90° angle from one another, and a large 50/50 beam splitter (Edmund Optics) was placed at 45° between the displays such that images from the two displays could be projected directly next to each other with no border. To find a metamer, a block of color was displayed on the CRT display, and a user-controlled 3-axis joystick was used to adjust the LCD image until no perceivable color difference was detected by the observer; the 3-axis joystick controlled colors in the HSV color space. The entire experimental setup was enclosed in a wooden box, painted black on the inside, and square apertures were placed on each display to ensure the images were displayed with black backgrounds to mitigate possible contextual perception effects28. Spectra from each monitor were acquired using a free-space spectrometer (Ocean Optics FLAME VIS-NIR with cosine corrector), normal to and adjacent to each display screen. Spectra for the white point of each monitor are shown in the Supplementary Information. 16 Monte Carlo Simulation Reflectance spectra were calculated by generating random values at a defined number of sampling numbers (𝑁𝑠) within the visible wavelengths (400 – 720 nm) using Matlab's rand function; 𝑁𝑠 was varied between 4-35 points to define the sharpness of spectral features, and the spectra were interpolated using a cubic spline. CIE 1931 2° matching functions were used to calculate tristumlus values for illuminant D65 reflected from the objects. Illuminant D65 was used as the white-point for conversion to the CIELAB space. A threshold of Δ𝐸 < 2.3 was used to define indistinguishable tristimulus values25. The simulation was performed for several number of iterations (𝑁𝑖), from 1,000-20,000,000, to determine if the defined metric 𝑃𝑚 converged for a given 𝑁𝑠. For 𝑁𝑖 greater than 1,000,000 values for 𝑃𝑚 converged to values within ~20% of each other within a given 𝑁𝑠. More-abstract calculation of metamer frequency The volume approximation ratios were computed using CVX, a package for specifying and solving convex programs37. Reference points [𝐿0, 𝑎0, 𝑏0] were chosen by uniformly sampling discretized spectra and mapping them to LAB tristimulus values. The computation was performed for various wavelength binning values (𝑁𝑏), between 7-18, to vary the broadness/sharpness of spectra, and 500 reference points were computed for each binning value. A detailed summary of this calculation, including a formalized mathematical treatment, can be found in the Supplementary information. 17 Data Availability Statement All data generated or analyzed during this study are included in this published article (and its Supplementary Information files). Acknowledgments. B.G. is supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE-1256259. This work was partially supported by startup funds from UW Madison, and partially by the AFOSR (FA9550-18-1-0146). We also thank Middleton Spectral Vision for access to their hyperspectral imaging systems. Competing Interests. UW-Madison has filed a patent based on the technology presented in this manuscript, Pub. No: US2018/0041737 A1. B.G. and M.K. are listed as co-inventors on this patent. Author Contributions. M.K. and B.G. developed the concept. B.G. performed the design and optimization, and carried out the majority of the experiments. B.R. assisted with brainstorming and experimental design. M.F., A.S., C.W., J.S., and G.V. assisted with the experiments. B.R., M.F., and L.L. assisted with analysis and interpretation. L.L. performed the "more abstract" metamer reduction calculation. B.G. and M.K. wrote the manuscript, with input from all authors. M.K. supervised the research. 18 References 1. Schnapf, J. L., Kraft, T. W. & Baylor, D. A. Spectral sensitivity of human cone photoreceptors. Nature 325, 439–441 (1987). 2. Wyszecki, G. & Stiles, W. S. Color Science: Concepts and Methods, Quantitative Data 3. and Formulae. (Wiley-Interscience, 2000). Stockman, A. & Sharpe, L. T. The spectral sensitivities of the middle- and long- wavelength-sensitive cones derived from measurements in observers of known genotype. Vision Res. 40, 1711–1737 (2000). 4. Mollon, J. D. Color vision: Opsins and options. Proc. Natl. Acad. Sci. 96, 4743–4745 5. 6. 7. 8. 9. (1999). Backhaus, W., Kliegl, R. & Werner, J. S. Color Vision: Perspectives from Different Disciplines. (Walter de Gruyter, 1998). Hurvich, L. & Jameson, D. An Opponent-Process Theory of Color Vision. Psychol. Rev. 64, 384–404 (1957). Barlow, H. B. What causes trichromacy? A theoretical analysis using comb-filtered spectra. Vision Res. 22, 635–643 (1982). Neitz, J., Neitz, M. & Jacobs, G. H. More than three different cone pigments among people with normal color vision. Vision Res. 33, 117–122 (1993). Neitz, M., Neitz, J. & Grishok, A. Polymorphism in the number of genes encoding long- wavelength-sensitive cone pigments among males with normal color vision. Vision Res. 35, 2395–2407 (1995). 10. Nagy, A. L., MacLeod, D. I. A., Heyneman, N. E. & Eisner, A. Four cone pigments in 11. 12. 13. 14. women heterozygous for color deficiency. JOSA 71, 719–722 (1981). Jameson, K. A., Highnote, S. M. & Wasserman, L. M. Richer color experience in observers with multiple photopigment opsin genes. Psychon. Bull. Rev. 8, 244–261 (2001). Jordan, G. & Mollon, J. D. A study of women heterozygous for colour deficiencies. Vision Res. 33, 1495–1508 (1993). Peli, E. Treating with Spectacle Lenses: A Novel Idea!? Optom. Vis. Sci. 79, 569–580 (2002). Peli, E. Vision Multiplexing: an Engineering Approach to Vision Rehabilitation. Optom. Vis. Sci. 78, 304–315 (2001). 15. Chernyak, D. Binocular optical treatment for presbyopia. (2005). 16. Knoblauch, K. & McMahon, M. J. Discrimination of binocular color mixtures in dichromacy: evaluation of the Maxwell-Cornsweet conjecture. J. Opt. Soc. Am. A. Opt. Image Sci. Vis. 12, 2219–2229 (1995). 17. Cornsweet, T. N. Visual Perception. (Academic Press, 1970). 18. Colorimetry - Part 1: CIE standard colorimetric observers. (2007). 19. Colorimetry - Part 4: CIE 1976 Lab* Colour space. (2008). 20. Colorimetry - Part 2: CIE standard illuminants. (2007). 21. S., M. C. Correlated color temperature as an explicit function of chromaticity coordinates. Color Res. Appl. 17, 142–144 (1992). 22. MacLeod, A. Thin-Film Optical Filters, Fourth Edition. (2010). 23. Berns, R. S., Motta, R. J. & Gorzynski, M. E. CRT colorimetry part I: Theory and practice. Color Res. Appl. 18, 299–314 (1993). 19 24. Sharma, G. LCDs versus CRTs-color-calibration and gamut considerations. Proc. IEEE 90, 605–622 (2002). 25. Mahy, M., Eycken, L. & Oosterlinck, A. Evaluation of Uniform Color Spaces Developed after the Adoption of CIELAB and CIELUV. Color Res. Appl. 19, 105–121 (1994). 26. Bal, C., Jain, A. K. & Nguyen, T. Q. Detection and removal of binocular luster in compressed 3D images. in 2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 1345–1348 (2011). doi:10.1109/ICASSP.2011.5946661 27. Blake, R. A Primer on Binocular Rivalry, Including Current Controversies. Brain Mind 2, 5–38 (2001). 28. Brascamp, J. W., Klink, P. C. & Levelt, W. J. M. The 'laws' of binocular rivalry: 50 years of Levelt's propositions. Vision Res. 109, 20–37 (2015). 29. Levelt, W. J. M. On Binocular Rivalry. (Leiden University, 1965). 30. Lit, A. The Magnitude of the Pulfrich Stereophenomenon as a Function of Binocular Differences of Intensity at Various Levels of Illumination. Am. J. Psychol. 62, 159–181 (1949). 31. Yang, E. K. & Tolle, J. W. A class of methods for solving large, convex quadratic 32. programs subject to box constraints. Math. Program. 51, 223–228 (1991). Schmeder, A. W. & McPherson, D. M. Multi-band color vision filters and method by lp- optimization. (2014). 33. Vukusic, P., Sambles, J. R., Lawrence, C. R. & Wootton, R. J. Quantified interference and diffraction in single Morpho butterfly scales. Proc. R. Soc. London B Biol. Sci. 266, 1403– 1411 (1999). 34. Bulina, M. E. et al. New class of blue animal pigments based on Frizzled and Kringle protein domains. J. Biol. Chem. 279, 43367–43370 (2004). 35. Tikhonravov, A. V, Trubetskov, M. K., Amotchkina, T. V & Kokarev, M. A. Key role of the coating total optical thickness in solving design problems. in 5250, 5210–5250 (2004). 36. Tikhonravov, A. V, Trubetskov, M. K. & DeBell, G. W. Optical coating design approaches based on the needle optimization technique. Appl. Opt. 46, 704–710 (2007). 37. Boyd, S. & Grant, M. CVX: Matlab software for disciplined convex programming. 20 Supplementary Information Enhancing human color vision by breaking binocular redundancy Bradley S. Gundlach1, Michel Frising1,2, Alireza Shahsafi1, Gregory Vershbow3, Chenghao Wan1,4, Jad Salman1, Bas Rokers5,6, Laurent Lessard1, Mikhail A. Kats1,4,6* 1Department of Electrical and Computer Engineering, University of Wisconsin-Madison, Madison, WI 2Department of Mechanical and Process Engineering, ETH Zurich, Zurich, Switzerland 3Department of Art, University of Wisconsin-Madison, Madison, WI 4Department of Materials Science and Engineering, University of Wisconsin-Madison, Madison, WI 5Department of Psychology, University of Wisconsin-Madison, Madison, WI 6McPherson Eye Research Institute, University of Wisconsin-Madison, Madison, WI Narrative explanation of the filter design process The filters in this work were designed using a white-balance condition that ensures that broadband "white light" that passes through the two filters is perceived similarly, to prevent significant color clashing between the two eyes under typical viewing conditions. Simultaneously, the filters are designed to be sufficiently distinct, which results in each eye receiving different spectral information. We used a design approach where each revision increased in complexity, building intuition at each design stage. For the first revision, a brick-wall longpass filter and a band-stop filter were used to split the blue cone response without significantly affecting the other cone types (Fig. S1). The filters were constrained such that the longpass filter cut-on wavelength was equal to the band-stop filter cut-off wavelength, which ensured that at least one eye was sensitive to every region of the visible wavelengths (i.e., no wavelength was attenuated by both filters). The band-stop filter cut-on wavelength was chosen to be 450 nm to minimize the effect of the filters on the M and L cone responses; thus, we were left to decide the longpass cut-on and band-stop cut-off wavelengths (which are enforced to be equal), and which must be below 450nm. With these constraints, the position of the cut-on/cut-off wavelength was optimized in order to minimize the CIE Δ𝐸 color difference between D65 white light passing through each filter. Figure S1(a) shows the result of this optimization, with a minimum Δ𝐸 color difference of 0.596 with the cut-on/cut-off wavelength at 437 nm. A Δ𝐸 color difference less than 2.3 typically means that the colors are indistinguishable. The resulting optimized filters are: a longpass filter (Filter 1) with a 437 nm transition wavelength, and a band-stop filter (Filter 2) with a stopband of 437 - 450 nm. Figure S1(c) shows the rendered color of D65 white light through each filter, and demonstrates the excellent white-balance between filters, because the color samples are 1 indistinguishable. However, the band-stop width of the resulting Filter 2 is only 13 nm, quite low in comparison to the ~300 nm range of the visible wavelengths. In order to effectively differentiate spectral features in everyday scenes, which are typically quite broad (>10 - 20 nm), the band-stop width must be larger. (a) (b) (c) Filter 1 ΔE = 0.596 Filter 2 Figure S1: (a) Plot showing the CIE ΔE color difference between filters 1 and 2 when transmitting CIE D65 white light versus cut-on/cut-off wavelength. (b) Optimized transmission response for filters 1 and 2 after the first design revision, with a band-stop region between 437 – 450 nm. (c) Rendered colors for CIE D65 white light transmitted through each filter from (b), with an optimized ΔE color difference of 0.596. Because this initial design produced a filter set with good white-balance, the same general approach was used in the following design revision. A similar optimization was performed as above, but the bandstop width of Filter 2 was constrained to be larger than 25 nm; this width was chosen such that it was larger than many typical spectral features found in nature, which could therefore be resolved with these filters. Instead of minimizing just the Δ𝐸 color difference between the filters, a modified merit function (MF) was used that also accounted for the filter width: 𝑀𝐹 = Δ𝐸 𝑊𝑖𝑑𝑡ℎ𝑏𝑎𝑛𝑑𝑠𝑡𝑜𝑝 ⁄ . A brute-force optimization was performed, which varied the cut-on and cut-off wavelengths of the two filters (again with the longpass cut- on and bandstop cut-off wavelengths being equal). The transmission spectra of the optimized filters are given in Figure S2: a longpass filter (Filter 1) with a 450 nm cut-on wavelength, and a bandstop filter (Filter 2) with a stopband of 450 – 500 nm. The resulting filter set has a Δ𝐸 color difference of 14.15, significantly above the 2.3 just noticeable difference threshold. Therefore, it is clear that for the simple design using a longpass and single-bandstop filter, there is a tradeoff between white-balance and width of the band stop region. 2 (a) (b) Filter 1 ΔE = 14.15 Filter 2 Figure S2: (a) Optimized transmission response for filters 1 and 2 after the second design revision, with a band-stop region of 450 - 500 nm. (b) Rendered colors for CIE D65 white light transmitted through each filter from (b), with an optimized ΔE color difference of 14.15 By considering the cone responses (Fig. 1c), it is clear why this filter design results in worse white balance. Beyond 450 nm, the longpass filter allows all wavelengths of visible light through; the M and L cones have very little response below 450 nm. However, the band-stop filter attenuates wavelengths between 450 and 500 nm, where the M and L cones have significant sensitivity. Therefore, the two filters affect the M and L cones very differently, which results in poor color balance. In particular, the band-stop filter (Filter 2) introduces a slight red tint compared to the longpass filter (Filter 1), which is also clear from the rendered colors in figure S2. Therefore, in order to maintain the high band-stop width of Filter 2 while improving the white balance of the filter pair, a second stopband was introduced to Filter 2 in order to soften the "red response" of the previous design. In this design, the shorter wavelength band-stop region of Filter 2 splits the S cone, whereas the longer wavelength band-stop region improves the white balance of the filter pair. In the final design revision that was implemented and described in the main text, the response of the longpass filter (Filter 1) was not optimized any further; in fact, the transmission response of a commercially available 450 nm longpass filter was used to define Filter 1. This was done to decrease manufacturing costs, such that only Filter 2 required custom design and manufacturing. The short- wavelength band-stop region of the Filter 2 was also used from before (allowed to change only slightly during optimization), and a preliminary long wavelength band-stop region was added between 600 and 700 nm, to be optimized further. An error function was also implemented to smooth the transition regions; in the previous design revisions, the transition regions had a sharp vertical slope, which is difficult to achieve in practice. The smoothness of the filters can be adjusted by changing the proportionality constant (a) of the error function: 𝑦 = erf⁡(𝑎𝑥). 3 Unlike the previous revisions, Filter 2 was optimized using a more rigorous method compared to the brute force method used above. A constrained optimization by linear approximation (COBYLA) method was implemented in a stochastic basin-hopping algorithm to optimize the position of the cut-on/cut-off wavelengths, transmission of the pass and stop-bands, and slope of the transition regions. The transition regions of the filter's short-wavelength band-stop region was constrained within +/- 10 nm of their previous values; this was done to maintain the overall shape of the previous design while allowing some room for color balance optimization. The transmittance was constrained between 5 and 15% in the stopbands, and between 80 and 95% in the passbands, to allow for high throughput and relative ease of manufacturing. The error function proportionality constant was constrained between 0.25 and 1. The long-wavelength stopband region was constrained between 600 – 700 nm, but the band-stop width was not constrained; this was done to prevent attenuation of the M and L cones at their peak sensitivities (~550nm, ~580 nm respectively), while also preventing needless optimization beyond the visible wavelengths (> 700 nm). With these constraints in place, Filter 2 was optimized in order to minimize the modified merit function: Δ𝐸 𝑊𝑖𝑑𝑡ℎ𝑏𝑎𝑛𝑑𝑠𝑡𝑜𝑝 ⁄ , where Δ𝐸 is the color difference between Filter 1 and Filter 2 when transmitting D65 white light and Widthbandstop is the spectral width of the short-wavelength band-stop region of Filter 2. This procedure yielded an optimized response for Filter 2 with stopbands at 450 - 500 nm and 630 - 680 nm, and stopband/passband transmittance of 10% and 90%, respectively (Figure S3(a)). The rendered color of transmitted D65 white light through the filters is given in Figure S3(b), with a Δ𝐸 color difference of 5.21. (a) (b) Filter 1 ΔE = 5.21 Filter 2 (i) (ii) (ii) (ii) Figure S3: (a) Optimized transmission response for filters 1 and 2 after the third design revision, with band-stop regions between 450 – 500 nm (i) and 630 – 680 nm (ii). Region (i) splits the short- wavelength cone, region (ii) improves the white-balance of the filter pair (b) Rendered colors for CIE D65 white light transmitted through each filter from (a), with a ΔE color difference of 5.21. 4 Methodology for counting metamers in "more abstract" calculation: In this section, we describe the technique by which we estimated the reduction of the occurrence of metamerism using our vision enhancement device. Each spectrum 𝐼(𝜆) is mapped to LAB tristimulus values [𝐿, 𝑎, 𝑏] via the CIE matching functions described earlier in the supplementary. Counting the number of metamers for a particular LAB reference point [𝐿0, 𝑎0, 𝑏0] amounts to counting the number of different spectra 𝐼(𝜆) that map to tristimulus values that are within a sphere in LAB space of radius ΔE of the reference point. The number of metameric spectra is infinite, so we instead compute a surrogate quantity. Roughly, we discretize each spectrum by wavelength so each spectrum can be abstracted as a point in a finite-dimensional space. We then count metamers by computing the volume that they occupy in this space. The details of the computation are described below. 1. Represent spectra by using 𝑁𝑆 equally spaced samples in wavelength. For example, 𝐼(𝜆) is represented as a vector [𝐼1, 𝐼2, … , 𝐼𝑁𝑆], which corresponds to the intensities at the wavelengths [𝜆1, 𝜆2, … , 𝜆𝑁𝑆]. 2. The map [𝐼1, … , 𝐼𝑁𝑆] → [𝐿, 𝑎, 𝑏] from the discretized spectrum to LAB tristimulus values is smooth and nonlinear. Since the map is only being evaluated in a local neighborhood of the reference point, the map is well approximated by its first order Taylor expansion. This allows us to replace the nonlinear map with an affine function 𝑔(𝐼1, … , 𝐼𝑁𝑆) = [𝐿, 𝑎, 𝑏]. 3. Let 𝑆0 = {⁡[𝐿, 𝑎, 𝑏]⁡⁡(𝐿 − 𝐿0)2 + (𝑎 − 𝑎0)2 + (𝑏 − 𝑏0)2 ≤ Δ𝐸2⁡} be the set of tristimulus values indistinguishable from the reference point. The set of metameric spectra is given by the image of 𝑆0 under the inverse map 𝑔−1. 4. Since the inverse map 𝑔−1 is affine and the set 𝑆0 is a sphere in LAB space, the image 𝑔−1(𝑆0) is an ellipsoid in the discretized spectrum space [S1]. Note that this ellipsoid is degenerate; it will be infinite in the directions corresponding to the kernel of 𝑔. 5. We assume that we are counting "reflection metamers" or "transmission metamers" under a certain illuminant. That is, we are excluding metamers generated by active emissive sources for the sake of this calculation, since including unbounded emissive sources complicates this calculation further. Under this assumption, the allowed intensities are not infinite, since every admissible spectrum has intensities bounded by the corresponding intensities of the illuminant. More formally, 𝐷65⁡for⁡𝑘 = 1, … , 𝑁𝑆}, where define the set of admissible spectra as 𝐶0 = {⁡[𝐼1, … , 𝐼𝑁𝑆]⁡⁡0 ≤ 𝐼𝑘 ≤ 𝐼𝑘 𝐷65 is the intensity at 𝜆𝑘 of the D65 illuminant. 𝐼𝑘 6. The volume of metameric spectra is therefore the volume of the set 𝑔−1(𝑆0) ∩ 𝐶0. 5 If the spectrum is filtered through a filter 𝑇(𝜆), the map 𝑔 must be replaced by a map 𝑔𝑇 that accounts for −1(𝑆0) ∩ 𝐶0. If the spectrum is filtered through 𝑇1(𝜆) for one eye and 𝑇2(𝜆) for the other eye, spectra are only counted if the filter 𝑇. The derivation is otherwise identical; metameric spectra are given by the set 𝑔𝑇 they are metameric for both eyes. This results in the set 𝑔𝑇1 −1(𝑆0) ∩ 𝑔𝑇2 −1(𝑆0) ∩ 𝐶0. We can compare configurations (e.g. natural human vision versus vision augmented by our device, or vision augmented by two different filters sets) by comparing the volumes of their respective metameric spectra. For example, to compare the unfiltered case (natural human vision) to the case of vision modified by our two-filter passive multispectral device, we would compute the ratio: 𝜌 = Vol(𝑔−1(𝑆0) ∩ 𝐶0) −1(𝑆0) ∩ 𝑔𝑇2 −1(𝑆0) ∩ 𝐶0) Vol(𝑔𝑇1 A ratio of 𝜌 = 20 would signify that metameric spectra are 20 times less abundant when the two-filter passive multispectral device is used as compared to the unfiltered case. Specifically, if spectra are sampled from a uniform distribution on intensities, a spectrum is 20 times less likely to be metameric. Computing the ratio 𝜌 is challenging because the volumes involved have irregular shapes; they are intersections of degenerate (high-dimensional) ellipsoids with box constraints. In order to approximate the ratio 𝜌, we approximate each volume by the volume of its max-volume inscribed ellipsoid. An illustration of a max-volume inscribed ellipsoid is shown below (Fig. S4). Figure S4: The shaded region is the intersection of ellipsoids and box constraints. The area is approximated by the max-area inscribed ellipsoid (dotted curve). 6 It turns out the max-volume inscribed ellipsoid can be efficiently computed using semidefinite programming techniques. See for example [S2]. We approximated the ratio 𝜌 by using the inner ellipsoid approximation for both the numerator and denominator. Volume approximation ratios were computed using CVX, a package for specifying and solving convex programs [S3]. We computed approximate volume ratios for the two-filter case as well as the one-filter case, using the transmission spectrum of a filter sold by EnChroma for alleviating some of the adverse effects of red-green color vision deficiency (EnChroma filter [S4]). In each case, we tried several different discretization points 𝑁𝑆 and we repeated the computation for 500 different reference points [𝐿0, 𝑎0, 𝑏0]. The reference points were selected by choosing discretized spectra at random, with reflectance values sampled from a uniform distribution between 0 and 1 (i.e., 𝑁𝑆 randomly sampled values per spectrum), and mapping them to LAB tristimulus values. A summary of the results is shown in Table S1. Using two filters results in a dramatic decrease in metamers, roughly consistent over the range of tested discretizations. In contrast, using a single filter (EnChroma in this case) has little effect on the number of metamers. Discretization points (𝑁𝑆) 7 9 12 14 16 18 Two filters, mean 22.977 115.823 120.580 160.634 178.020 111.243 Two filters, median 17.251 80.690 57.4080 85.885 77.153 41.369 EnChroma, mean 0.945 0.887 1.090 0.944 1.125 1.086 EnChroma, median 0.945 0.870 1.082 0.944 0.924 1.017 Table S1 – Results of the metamer ratio approximation. Mean and median 𝜌 values are computed over 500 randomly generated spectra for each discretization. Using two filters reduces the frequency of metamers by a factor of about 50 on average, while using a single EnChroma filter has a negligible effect on the frequency of metamers 7 Monte Carlo Metamer Calculation The Monte Carlo simulation, as described in the main text, was performed for several values of spectral sharpness 𝑁𝑠, with higher numbers signifying sharper features, and number of iterations 𝑁𝑖 (Fig. S5). As discussed in the main text, 𝑃𝑚 is largest for moderately sharp spectral features (𝑁𝑠 = 15). The metamer reduction metric seems to converge to within ~20% of neighboring values when 𝑁𝑖 reaches ~1,000,000. Note there are missing values for low number of iterations because the sample size was not large enough to generate metameric spectra. Figure S5: Convergence of the metamer reduction metric (𝑃𝑚) as a function of 𝑁𝑠 and number of iterations, 𝑁𝑖 8 Thin-film filter design The filter response design goal (Fig. S3(a)) used in this work is realized by conventional thin-film design methods. A commercial thin-film design software (Optilayer) was used to optimize a two-material thin-film stack to adequately meet the design goal. Tantalum oxide (Ta2O5) was chosen as the high index (n = 2.15) material and silicon dioxide (SiO2) was chosen as the low index (n = 1.46) material, as they are both easily deposited. The substrate was NBK7, a common optical glass. The final stack was constrained to be less than 75 total layers to keep costs down, and each layer between 10 – 500 nm to prevent stress cracks in thick films. Using these constraints in tandem with the provided filter design goal, the thin-film stack was optimized for incident angles between 0 - 10. A representative stack design produced by Optilayer for the design of Filter 2 is given in Fig. S5. The actual design for the device experimentally demonstrated in the main text was slightly modified from that of Fig. S6 by a thin-film foundry (Iridian Spectral Technologies, Ontario, Canada), though they did not share the precise thicknesses of the films with us due to their standard disclosure policy. Nevertheless, the specifics of the design are not critical as long as it implements the desired transmission spectrum (Fig. 2(b)). Figure S6: Thin-film filter stack design for filter 2 (Fig. S3(a)), using Ta2O5 (n = 2.15, blue) and SiO2 (n = 1.46, red) dielectric layers. Comparison of CRT and LCD monitors: 9 Figure S7: (a) Measured emission spectra of the LCD and CRT displays used in this work, displaying a white color (RGB = 255, 255, 255). (b, c) Measured emission spectra of each individual color channel (R, G, B) for the LCD and CRT display, respectively. For blue curves, the displayed color was RGB = (0, 0, 255), red curves RGB = (255, 0, 0), and green curves RGB = (0, 255, 0) The displays used in this work to generate metameric spectra were a True HD-IPS liquid crystal display (LCD) on a LG G3 smartphone and a conventional cathode ray tube (CRT) monitor (Dell E770P). These displays use very different mechanisms to generate colors, which results in significantly different spectra when displaying the same color (i.e. metamers). The LCD display uses a backlight, typically a white LED, which is transmitted through color filter arrays (red, green and blue color filters) to produce its color response. Therefore, the emitted spectrum is the product of the LCD backlight and color filter transmission response. The CRT monitor uses an electron gun, and relies on a phosphorescent screen to control its spectrum in the visible wavelength range. Because the two display types use significantly different methods to generate colors, the two emission technologies have different spectral features for the individual red, green and blue color channels. The distinct features of the two displays are demonstrated in Figure S7(a), which shows the measured emitted spectrum of white light from each display (RGB = [255, 255, 255]). Figures S7(b, c) show the spectrum of each pure color channel (red, green and blue) for the LCD and CRT display, respectively. Color accuracy of photographs: In this work, we used digitally generated color samples from spectroscopic data to demonstrate the splitting of a metameric pair using an LCD and CRT monitor (Fig. 3). This method was used because, due to the difference in spectral response between a camera sensor and the human eye, it is difficult to obtain a precisely color-accurate photograph. This difficulty is shown in Figure S8, which shows the original photograph of the experimental setup in Fig. 3(c), an edited photograph that better approximates the actual color, and a digitally rendered color sample showing the "actual color". The colors are rendered using CIE matching functions, as described above; the [X,Y,Z] values calculated using the matching functions and the measured spectrum can then be converted to the sRGB color space, which is the working color space 10 of most computers. Though the sRGB values will be the same across all display devices, the actual displayed color depends on the calibration of the monitor used. Therefore, the rendered color is Fig. S8(c) only represents the perceived color seen in the experiment when the monitor used to view the image has a perfect color calibration. It is clear that Fig. S8(a) is significantly different than the generated color sample in Fig. S8(c), even though the camera used in S8(a) and the spectrometer used to acquire the sample that generated the color in S8(c) sampled the same light. Fig. S8(b) is the figure used in the main text (Fig. 3(b)), and was edited to more closely represent the color in S8(c) to prevent confusion. The rendered color samples using spectroscopic data represent the perceived colors of both monitors much more accurately. (C) Figure S8: (a) Photograph of the setup shown in Fig. 3(c) of the main text, taken using a Sony α7R II camera. (b) The edited photograph that appears in Fig. 3(b) of the main text, which was modified to appear close in color to the rendered colors. This manipulation was performed to prevent confusion in the main text. (c) The actual color displayed during the experiment, rendered using spectra acquired using a grating spectrometer and cosine corrector. Hyperspectral images To demonstrate the utility of our wearable passive multispectral device in a more natural setting, we acquired a hyperspectral image of a complex scene, and applied the filters digitally (Fig. S9). The scene included a variety of blue and violet objects, including patches of color made using paints and pastels, plants, and a Morpho butterfly featuring a structural blue color [S5]. The image was obtained using a Middleton Spectral Vision MSV-500 High Sensitivity VNIR hyperspectral camera. 11 The enlarged images in Fig. S9(d) show the butterfly wing next to six similarly colored samples made using oil pastels, with no filters applied. In Fig. S9€ and Fig. S9(f), filters 2 and 1, respectively, are applied to these enlarged images. The numbers in each panel represent the CIE ΔE color difference between the oil pastel color and the butterfly wing, averaged over a small area to reduce pixel noise. Using filter 2 (Fig. S9(e)), the appearance of the butterfly wing becomes more dissimilar to the pastel samples compared to no filter (i.e., the butterfly "blue" becomes easier to distinguish from the pastel "blue" using the filter). This again demonstrates the effect of partitioning the S cone to provide more spectral information. The improvement is absent for filter 1 (Fig. S9(f)), demonstrating the need for both filters in the design. We note that each filter creates a new set of metamers that may have been distinguishable before; by using two filters, the set of overlapping newly created metamers becomes significantly smaller. Therefore, as long as at least one filter creates an increase in contrast, more spectral information can be communicated to the visual system while decreasing the overall number of possible metamers. Although a slight yellow/green tint is applied to both filtered images, Fig. S9(b) and Fig. S9(c) have similar "white-balance" due to the white-balance condition enforced during the design process. Fig. S9: (a) RGB rendering of a hyperspectral image with natural and artificially colored objects, with no filter applied. (b) The render with filter 2 applied, and (c) the render with filter 1 applied. (d) Magnified view of butterfly wing and paint samples (outlined in green in (a)). (e) and (f) are the same samples as in (d), with filters 2 and 1 applied, respectively. The numbers inside each paint sample are the ΔE color difference, rounded to the nearest integer, between the paint sample and the butterfly wing with each respective filter applied. 12 Supplementary References S1. Giuseppe C, Ghaoui LE (2014) Optimization Models (Cambridge University Press). S2. Boyd S, Vandenberghe L (2004) Convex optimization (Cambridge University Press). S3. Grant M, Boyd S CVX: Matlab software for disciplined convex programming, version 2.1, http://cvxr.com/cvx, October 2016. S4. Schmeder AW, McPherson DM (2014) Multi-band color vision filters and method by lp-optimization. S5. P. Vukusic, J. R. Sambles, C. R. Lawrence, R. J. Wootton, Quantified interference and diffraction in single Morpho butterfly scales. Proc. R. Soc. Lond. B Biol. Sci. 266, 1403–1411 (1999). 13
0902.3891	2	0902	2010-06-04T10:35:34	Wave-train induced unpinning of weakly anchored vortices in excitable media	[ "physics.bio-ph", "nlin.PS", "q-bio.TO" ]	A free vortex in excitable media can be displaced and removed by a wave-train. However, simple physical arguments suggest that vortices anchored to large inexcitable obstacles cannot be removed similarly. We show that unpinning of vortices attached to obstacles smaller than the core radius of the free vortex is possible through pacing. The wave-train frequency necessary for unpinning increases with the obstacle size and we present a geometric explanation of this dependence. Our model-independent results suggest that decreasing excitability of the medium can facilitate pacing-induced removal of vortices in cardiac tissue.	physics.bio-ph	physics	Wave-train Induced Unpinning of Weakly Anchored Vortices in Excitable Media Alain Pumir1,2, Sitabhra Sinha3, S. Sridhar3, M´ed´eric Argentina2, Marcel Horning4, Simonetta Filippi5, Christian Cherubini5, Stefan Luther6 and Valentin Krinsky6,7 1Laboratoire de Physique, ENS de Lyon and CNRS, 46 All´ee d'Italie, 69007, Lyon, France. 2Lab J. A. Dieudonn´e, Universit´e de Nice and CNRS, Parc Valrose, F-06108 Nice Cedex, France. 3The Institute of Mathematical Sciences, CIT Campus, Taramani, Chennai 600113, India. 4 Dept. of Physics, Graduate School of Science, Kyoto University, Kyoto 606-8502, Japan. 5Nonlinear Physics & Math Modeling Lab, University Campus Bio-Medico, I-00128, Rome, Italy. 6MPI for Dynamics and Self-Organization, Am Fassberg 17, D-37077, Gottingen, Germany. and 7Institut Non Lin´eaire de Nice and CNRS, F-06560, Valbonne, France. (Dated: August 11, 2018) A free vortex in excitable media can be displaced and removed by a wave-train. However, simple physical arguments suggest that vortices anchored to large inexcitable obstacles cannot be removed similarly. We show that unpinning of vortices attached to obstacles smaller than the core radius of the free vortex is possible through pacing. The wave-train frequency necessary for unpinning increases with the obstacle size and we present a geometric explanation of this dependence. Our model-independent results suggest that decreasing excitability of the medium can facilitate pacing- induced removal of vortices in cardiac tissue. PACS numbers: 87.19.Hh,05.45.-a,87.19.lp,05.45.Gg Rotating spiral waves of propagating excitation char- acterize the disruption of ordered behavior in excitable media describing a broad class of physical, chemical and biological systems [1]. In the heart, spiral waves of elec- trical activity have been associated with life-threatening arrhythmias [2 -- 4], i.e., breakdown of the normal rhyth- mic pumping action of the heart. Controlling such spa- tial patterns with low-amplitude external perturbation is a problem of fundamental interest [5 -- 9] with significant implications for the clinical treatment of cardiac arrhyth- mias [10]. In a homogeneous active medium, a spiral wave can be controlled by a wave train, induced by periodic stimula- tion from a local source (pacing)[4]. If the frequency of stimulation is higher than that of the spiral wave, the wave train induces the spiral to drift. In a finite medium, the vortex is eventually driven to the boundary and thereby eliminated from the system [11 -- 13]. Inho- mogeneities in the medium, such as inexcitable obsta- cles, can anchor the spiral wave preventing its removal by a stimulated wave-train [14]. This mechanism is anal- ogous to pinning of vortices in disordered superconduc- tors [15, 16]. In the heart, obstacles such as blood vessels or scar tissue, can play the role of pinning centres [17], leading to anatomical reentry, the sustained periodic ex- citation of the region around the obstacle. In the immediate neighborhood of the obstacle, pinned vortices are qualitatively equivalent to waves circulating in a one-dimensional ring. They can be removed by ex- ternal stimulation provided the electrode is located on the reentrant circuit, i.e., the closed path of the vortex around the obstacle, and the stimulus is delivered within a narrow time interval [18]. However, for the more gen- eral situation of pacing waves generated far away from the reentrant circuit, a classical result due to Wiener and Rosenblueth (WR) states that, all waves circulat- ing around such obstacles are created or annihilated in pairs (see Ref. [19], in particular, pp.216-224). This im- plies that it is impossible to unpin the spiral wave by a stimulated wave train. In this paper, we demonstrate that the WR mecha- nism for the failure of pacing in unpinning spiral waves is valid only when the radius of the free spiral core (i.e., the closed trajectory of the spiral tip defined as a phase singularity [20]) is small compared to the size of the ob- stacle. We elucidate the transition between the case of a free vortex and one attached to a large obstacle by sys- tematically decreasing the core radius of the free spiral, RF S, relative to the obstacle size, Robst, by increasing the excitability of the medium. Our main result is that an anchored rotating wave can be removed by a stimulated wave train provided RF S > Robst. To illustrate our arguments, we use the simple model of excitable media introduced in Ref. [21], described by an excitatory (u) and a recovery (v) variable: ∂tu = 1 ǫ u(1 − u)[u − (v + b a )] + ∇2u, ∂tv = (u − v), (1) where, a and b are parameters describing the kinetics. The relative time-scale ǫ between the local dynamics of u and v is set to 0.02. We discretize the system on a square spatial grid of size L × L, with a lattice spacing of ∆x = 0.25 and time step of ∆t = 0.01. For our sim- ulations L = 200. We solve Eq. 1 using forward Euler scheme with a standard nine-point stencil for the Lapla- cian. No flux boundary conditions are implemented at the edges of the simulation domain. An obstacle is im- plemented by introducing a circular region of radius Robst in the center of simulation domain inside which diffusion 1 S 0 ( a ) 2 1a 1b S 0 ( b ) 2 1b S 0 1a ( c ) 3 2b 1b 2a S 1 S 0 ( d ) FIG. 1: (a) Wave S0, pinned to an obstacle (shaded), ro- tates counterclockwise; wave 1 is the first pacing wave. (b) Wave 1 hits the obstacle, and separates into a wave rotating counterclockwise (1a) and a wave rotating clockwise (1b). (c) Waves S0 and 1b collide and merge leaving only one rotating wave 1a denoted S1 hereafter. (d) The wave resulting from the merging of S0 and 1b leaves the system. The interaction between the following pacing wave, 2 and S1, is similar to that shown in (a-c). Thus, the pinned vortex persists. Numerical simulation of the Barkley model with parameters: a = 0.9, b = 0.17; the pacing period is Tp = 6.7 and the radius of the obstacle is R = 6.5. is absent. Pacing is delivered by setting the value of u to up = 0.9 in a region of 6×3 points at the center of the up- per boundary of the simulation domain. The maximum pacing frequency is limited by the refractory period, Tref , the duration for which stimulation of an excited region does not induce a response. When the obstacle size is large relative to the core ra- dius of the free spiral, RF S, the failure of a wave train in unpinning the vortex is illustrated in Fig. 1. Initially, the spiral wave S0 rotates counterclockwise around the obstacle. During the interaction with pacing waves, the number of waves attached to the obstacle can change due to two possible processes (see Ref. [19], p.216 and 220). First, when the pacing wave reaches the obstacle, it splits into two oppositely rotating waves: one clockwise and the other counterclockwise. Second, collision between two ro- tating waves, as seen in Fig. 1(c), results in the annihila- tion of a pair of counterclockwise and clockwise waves. In both cases, the number of waves rotating counterclock- wise is always larger than the number rotating clockwise by 1. Thus, in addition to conservation of total topolog- ical charge (i.e., sum of the individual chiralities, +1 or −1) for all spiral waves in a medium [20, 22], topolog- ical charge around the obstacle also appears to be con- served. However, in the limiting case of infinitesimally small obstacle corresponding to a free vortex, a stimu- lated wave-train with frequency higher than that of the spiral wave will always succeed in displacing the latter, eventually removing it from a finite medium. Thus, there is a transition from failure to successful pacing as Robst is reduced relative to RF S. The primary fact responsible for this transition is that the spiral wave is no longer in physical contact with an obstacle of size smaller than RF S [17], contrary to the fundamental assumption of Ref. [19]. Fig. 2 shows an explicit example of successful detachment of a pinned wave from the obstacle boundary, where the core radius 2 2a 1a 3 2b 1b S 0 ( c ) 3 2 1b S 0 ( b ) 3 2a 2b S 1 ( e ) ( f ) 2 1a 1b 1a S 0 ( a ) 2a 3 2b 1a 1c ( d ) 1b S 0 FIG. 2: Lowering excitability results in successful detachment of pinned vortex by pacing. S0 is a rotating wave whose core (dashed line) is larger than the pinning center (shaded). (a-c) are topologically as in Fig. 1. (d) A wavelet 1c is produced after collision of waves S0 and 1b, in contrast with Fig. 1(d). (e) The wavelet 1c collides with 1a and the resulting wave S1 is displaced away from the obstacle. (f) Subsequent pacing induces drift of the spiral wave S1 to the boundary, eventu- ally removing it from the medium. The parameters are as in Fig. 1, except for a = 0.895 and b = 0.1725, resulting in increasing the vortex core size. of a free spiral in the medium is made larger than Robst by diminishing the excitability of the system. The possibility of unpinning the wave in Fig. 2 can be traced to the following fact: the collision between S0 and the pacing wave-branch 1b occurs a small distance away from the obstacle boundary and does not result in complete annihilation of both waves. A small fragment 1c survives in the spatial interval between the collision point and the obstacle [Fig. 2(d)]. If the tip of S0 is close to the obstacle, the fragment 1c is small, and rapidly shrinks and disappears. However, if the gap between the reentrant wave tip and the obstacle is large at the colli- sion point, such that the size of 1c is larger than a critical value ln, the fragment can survive. As 1c propagates fur- ther away, it collides with the pacing wave 1a and forms a new broken wave S1 that is completely detached from the obstacle. Interaction with successive pacing waves progressively pushes the vortex further away from the obstacle, and eventually from a finite medium. The dif- ference between the number of spirals rotating counter- clockwise and clockwise around the obstacle changes from 1 initially (Fig. 2, a), to 0 in Fig. 2(e), contrary to what happens for a larger obstacle (Fig. 1). The absence of topological charge conservation for waves rotating around a smaller obstacle underlines the breakdown of the fun- damental assumption behind the WR argument for why pacing cannot detach pinned waves. The unpinned wave is subsequently driven outside the system boundaries by b 0.3 0.2 0.1 0 ( a ) NW SE R R R obst obst obst = 0 = 1.25 = 6.5 SW BI 0.4 0.6 0.8 a 1 1.2 i s u d a R 10 8 6 4 2 0 0 ( b ) RFS Rmax obst 0.03 d 0.06 FIG. 3: (a) Parameter space of the Barkley model. Unpin- ning is possible in the shaded portion of the SW region, which exhibits persistent spiral waves. The thick line indicates the boundary with the SE region, where spirals cannot form. The domain where unpinning is possible shrinks with increasing size of the pinning center, the three dashed lines correspond- ing to Robst = 0, i.e., no obstacle (square), Robst = 1.25 (plus) and Robst = 6.5 (diamond). (b) Radius RF S of the free spi- ral and the maximum obstacle radius Rmax obst from which wave trains can unpin vortices, as a function of the distance d from the SE-SW boundary, along the dot-dashed line indicated in (a). Note that RF S > Rmax obst , and both increase with decreas- [In (a), NW (BI) indicates the parameters for which ing d. steady waves are absent (the medium is bistable).] pacing (Fig. 2, f), thus eventually also reducing the total topological charge of the finite medium to 0. The relative size of the obstacle, compared to the free spiral core, is the key parameter that decides whether a pinned reentrant wave can be removed or not. Indeed, the radius of the free spiral core in the successful case, RF S = 9.05 (Fig. 2) is significantly larger than in the unsuccessful one, RF S = 5.80 (Fig. 1). It is further con- firmed by a detailed numerical study of the interaction between a pacing wave train and a pinned spiral over the (a, b) parameter space of the Barkley model. As shown in Fig. 3(a), the rotating wave anchored to the obsta- cle can be removed by pacing only in the neighborhood of the sub-excitable (SE) region (using the terminology of Ref. [23]), where RF S diverges [Fig. 3(b)]. This is explained by noting that in the SE regime, the tangen- tial velocity of a broken wavefront is negative, thus caus- ing the front to shrink and not form a spiral. As we approach the regime where spiral waves are persistent (SW), the tangential velocity of the wave break gradu- ally increases to zero and becomes positive on crossing the SE-SW boundary, so that the broken wave front can now evolve into a spiral. As RF S increases with decreas- ing tangential velocity of the wave front, the spiral core becomes large close to the SE region resulting in success- ful pacing-induced termination of pinned reentry. We observe that there is a maximum radius of the ob- stacle (Rmax obst ) close to RF S above which pacing is unsuc- cessful in detaching the anchored spiral wave [Fig. 3(b)]. 3 ( b ) C 1c S θ l T FS Rmax obst ( a ) 1 1a T ref 2 6 Radius of obstacle, R obst 4 25 20 15 10 5 0 x a m T p , i d o i r e P g n c a P m u m x a M i 1 S 1b C S 1a 1a 1c ( c ) ( d ) ( e ) 1c 1a ( f ) p (a) The maximum pacing period T max FIG. 4: at which unpinning is possible as a function of the obstacle radius Robst. For the parameters a = 1.1323, b = 0.2459 that we have used, the maximum radius of obstacle from which depinning can occur is Rmax obst = 4. TF S is period of a free spiral wave and Tref is the refractory period. The dashed line indicates the prediction from Eq. 2. (b) The wavelet formation mechanism leading to the detachment of the pinned vortex (schematic). (c-f) Numerical simulation of the Barkley model. S collides with wave 1 at point C at an angle θ. The part 1b of the pacing wave merges with S, moving out of the system. The remaining part of the pacing wave collides with the obstacle (shaded) separating into 1a and a small wavelet 1c. When the length l of wavelet 1c is larger than the critical nucleation length, 1c survives and collides with S. This results in unpinning of S. p Fig. 4(a) shows that the pacing period for successful unpinning from the obstacle is bounded by the refrac- tory period (Tref ) and a maximum value T max that is independent of Robst for small obstacles. As we ap- proach Rmax obst , the upper bound sharply decreases, be- coming equal to the refractory time at Rmax obst , which indi- cates that pacing will be unsuccessful in unpinning waves attached to obstacles of radii larger than Rmax obst . Thus, the results shown in Figs. 3(b) and 4(a) demonstrate our earlier assertion that pacing induced removal of an- chored waves will be possible only when the obstacle is smaller than the core radius of the free spiral wave in the medium. Our numerical results indicate that the maximum pac- ing period necessary for detaching a pinned spiral wave is a decreasing function of the obstacle size [Fig. 4(a)]. This can be explained semi-quantitatively by the following ge- ometric argument, valid when the size of the obstacle is small compared to the core size of the spiral, and sup- ported by the simulations shown in Fig. 4(c-f). The tip of the spiral S moves along its circular trajectory, shown by the broken line in Fig. 4(b), and interacts with the pacing wave coming from the top, represented by a solid line. The part 1b of the pacing wave collides with S at the point C characterized by an angle θ that the spiral tip makes with the symmetry axis (i.e., the line joining the centers of the obstacle and spiral core); the resulting wave eventually leaves the system [Fig. 4(d)]. The re- maining section of the pacing wave splits into two waves, 1a and 1c, propagating along either side of the obstacle. The wave tip moves approximately in a straight line from C, so that the length of the wave 1c at the symmetry axis is l = RF S(1 + cos θ) − 2Robst. When the fragment 1c is larger than the nucleation size ln, it expands into a wavefront that reconnects with wave 1a. This results in a displacement of the wave 1a away from the obstacle, leading to unpinning (as in Fig. 2). For l < ln, 1c shrinks and eventually disappears, resulting in unsuccessful pac- ing. Thus, the condition for detachment is l ≥ ln. The length l is a decreasing function of the angle θ, which in turn, is a decreasing function of the pacing period, Tp, as explained below. The relation between Tp and θ can be established by estimating the time interval for two successive collisions of the spiral with the pacing waves. From the point of collision C, the pacing wave reaches the obstacle after time T1 = (RF S sin θ − Robst)/v, and the symmetry axis after time T2 = T1 + (RobstTF S/4RF S). From the symmetry axis, the new reentrant wave S moves by an angle (θ + π) to arrive at C at time T3 = T2 + [TF S(θ + π)/2π], where it collides with the next pacing wave. Noting that T3 = Tp allows us to implicitly express Tp as a function of θ, and thereby, l. The maximum pacing period leading to detachment is obtained when l = ln, as: T max p = RF S v (sin θc − fR) + fRTF S 4 + TF S(θc + π) 2π , (2) where, θc = arccos(2fR − 1 + [ln/RF S]) and fR = Robst/RF S. When Robst > Rmax obst = RF S − (ln/2), T max has complex values, indicating that for larger obstacles the fragment is too small to survive. The nucleation length ln can thus be estimated from Rmax obst , which al- lows us, in turn, to determine the dependence of T max as a function of Robst from Eq. 2. Fig. 4(a) shows this to be in fair agreement with our numerical simulations. p p We stress that the arguments used here are model inde- pendent, and are based only on the property that waves in excitable media annihilate on collision. We verified nu- merically that wave-train induced unpinning is observed also in a more detailed and realistic description of car- diac tissue, the Luo-Rudy model [24] [see the movie on line], under conditions of reduced excitability. Meander- ing, which occurs in the Barkley model at low a, b values (Fig. 3, a), does not affect the physical effect discussed here. Note that the proposed unpinning mechanism is for the case of an obstacle smaller than the vortex core. It is possible under certain circumstances to unpin waves from obstacles larger than the core because of other ef- fects such as the presence of slow conduction regions [25] and nonlinear wave propagation (alternans) [26]. 4 Our results thus predict that in cardiac tissue, the re- moval of spiral waves pinned to a small obstacle by high- frequency wave trains is facilitated by decreasing the ex- citability of the medium. This is consistent with previ- ous experimental results on cardiac preparations using Na-channel blockers [17] and our prediction could be di- rectly tested in a similar experimental setup [12, 17]. In conclusion, we have shown that for a pinned vortex interacting with a pacing wave train, unpinning is pos- sible when the size of the obstacle is smaller than that of the spiral core. The minimum wave train frequency necessary for unpinning in the presence of an inexcitable obstacle is higher than that for inducing drift in a free vortex towards the boundaries, and it increases with the size of the pinning center. Our results suggest that low- ering the excitability of the medium makes it easier to unpin vortices by pacing. This research was initiated at the Kavli Institute for Theoretical Physics, and was supported in part by IFCPAR (Project 3404-4) and IMSc Complex Systems Project (XI Plan). [1] M. C. Cross and P. C. Hohenberg, Rev. Mod. Phys. 65, (1993). [2] V. I. Krinsky, Vestnik A.N. SSSR 1-12 (1980). [3] J. M. Davidenko et al., Nature (London) 355, 349 (1992). [4] Cardiac Electrophysiology: From Cell to Bedside, edited by D. P. Zipes and J. Jalife (Saunders, Philadelphia, 2004). [5] S. Sinha, A. Pande, and R. Pandit, Phys. Rev. Lett. 86, 3678 (2001); [6] S. Takagi et al., Phys. Rev. Lett. 93, 058101 (2004). [7] H. Zhang et al., Phys. Rev. Lett. 94, 188301 (2005). [8] S. Sinha and S. Sridhar, in Handbook of Chaos Control (2nd edition, Wiley-VCH, Weinheim, 2008), p. 703; S. Sridhar and S. Sinha, EPL 81, 50002 (2008). [9] A. Isomura et al., Phys. Rev. E 78, 066216 (2008); M. Horning et al., Phys. Rev. E 79, 026218 (2009). [10] F. Fenton et al., Circulation 120, 467 (2009). [11] V. Krinsky and K. Agladze, Physica D 8, 50 (1983). [12] K. Agladze et al., Am. J. Physiol. Heart Circ. Physiol. 293, H503 (2007). [13] G. Gottwald et al., Chaos 11, 487 (2001). [14] A. M. Pertsov et al., Physica D 14, 117 (1984); M. Vin- son et al, Physica D 72, 119 (1994); Z. A. Jimenez et al, Phys. Rev. Lett. 102, 244101 (2009); B. T. Ginn et al, Phys. Rev. Lett. 93, 158301 (2004). [15] G. Blatter et al., Rev. Mod. Phys. 66, 1125 (1994). [16] D. Pazo et al., Phys. Rev. Lett. 93, 168303 (2004). [17] Z. Y. Lim et al., Circulation 114, 2113 (2006). [18] L. Glass and M. E. Josephson, Phys. Rev. Lett. 75, 2059 (1995). [19] N. Wiener and A. Rosenblueth, Arch. Inst. Cardiol. Mex- ico, 16, 205 (1946). [20] A. T. Winfree, When Time Breaks Down, Princeton Univ Press, Princeton NJ (1987). [21] D. Barkley et al., Phys. Rev. A 42, 2489 (1990). [22] L. Glass, Science 198, 321 (1977). [23] S. Alonso et al., Science 299, 1722 (2003). [24] C. H. Luo and Y. Rudy, Circ. Res. 68, 1501 (1991). [25] S. Sinha et al., Chaos 12, 893 (2002); S. Sinha and D.J. Christini, Phys. Rev. E 66, 061903 (2002). [26] J. Breuer and S. Sinha, Pramana 64, 553 (2005). 5
1808.02750	3	1808	2018-08-13T17:12:18	Variability of collective dynamics in random tree networks of strongly-coupled stochastic excitable elements	[ "physics.bio-ph" ]	We study the collective dynamics of strongly diffusively coupled excitable elements on small random tree networks. Stochastic external inputs are applied to the leaves causing large spiking events. Those events propagate along the tree branches and, eventually, exciting the root node. Using Hodgkin-Huxley type nodal elements, such a setup serves as a model for sensory neurons with branched myelinated distal terminals. We focus on the influence of the variability of tree structures on the spike train statistics of the root node. We present a statistical description of random tree network and show how the structural variability translates into the collective network dynamics. In particular, we show that in the physiologically relevant case of strong coupling the variability of collective response is determined by the joint probability distribution of the total number of leaves and nodes. We further present analytical results for the strong coupling limit in which the entire tree network can be represented by an effective single element.	physics.bio-ph	physics	Variability of collective dynamics in random tree networks of strongly-coupled stochastic excitable elements Ali Khaledi-Nasab,1, 2, ∗ Justus A. Kromer,3, † Lutz Schimansky-Geier,1, 4, 5, ‡ and Alexander B. Neiman1, 2, § 1Department of Physics and Astronomy, Ohio University, Athens, Ohio 45701, USA 2Neuroscience Program, Ohio University, Athens, Ohio 45701, USA 3Stanford University, Department of Neurosurgery, Stanford, CA, 94305, USA 4Department of Physics, Humboldt-Universitat zu Berlin, Newtonstrasse 15, 12489 Berlin, Germany 5Bernstein Center for Computational Neuroscience, Berlin, Germany (Dated: August 14, 2018) We study the collective dynamics of strongly diffusively coupled excitable elements on small random tree networks. Stochastic external inputs are applied to the leaves causing large spiking events. Those events propagate along the tree branches and, eventually, exciting the root node. Using Hodgkin-Huxley type nodal elements, such a setup serves as a model for sensory neurons with branched myelinated distal terminals. We focus on the influence of the variability of tree structures on the spike train statistics of the root node. We present a statistical description of random tree network and show how the structural variability translates into the collective network dynamics. In particular, we show that in the physiologically relevant case of strong coupling the variability of collective response is determined by the joint probability distribution of the total number of leaves and nodes. We further present analytical results for the strong coupling limit in which the entire tree network can be represented by an effective single element. PACS numbers: 87.19.ll, 87.19.lb, 87.19.lc, 05.45.Xt, 05.10.Gg 8 1 0 2 g u A 3 1 ] h p - o i b . s c i s y h p [ 3 v 0 5 7 2 0 . 8 0 8 1 : v i X r a ∗ [email protected] † [email protected] ‡ [email protected] § [email protected] I. INTRODUCTION 2 The study of the dynamical properties of complex networks of nonlinear elements [1, 2] is an important trend in nonlinear science [3, 4]. In particular, networks of coupled stochastic excitable elements are commonly used as model systems for a wide range of natural phenomena, such as pattern formation in chemical reactions [5, 6], the dynamics of gene regulatory networks [7 -- 10], the electrical activity of single [11 -- 13] neurons and large neuronal populations [14]. Network topology strongly influences the collective dynamics of coupled excitable elements [3, 15]. For instance, coherent collective oscillations can emerge for certain coupling strengths or particular choices of network connectivity [16, 17]. Furthermore, the emergent correlated activity of large recurrent neural networks can be linked to their connectivity [18]. The sensitivity of complex networks to external signals and the dynamic range of the network's collective response can be maximized for the so-called critical networks [19], i.e being on the verge of a phase transition. This criticality can be achieved either by tuning the coupling strength and signal propagation parameters in the network [12, 13] or by tuning its topology [20]. The majority of works in this area are devoted to large networks with well-defined statistical properties such as degree distributions and spectra of the adjacency or Laplace matrices [3, 4, 9]. Due to the large number of nodes and interconnections one can average over the network structure. Then, the emergent collective dynamics can be related the statistical properties of the network's architecture [15, 18, 21]. The situation is different when the collective dynamics of small random networks is studied. Although the network topology can be specified in terms of statistical properties, such as degree distributions, etc. , individual network realizations may differ significantly. In consequence, a detailed analysis of the relation between the collective dynamics and statistical properties of the network topology require studies of ensembles of networks realizations. FIG. 1. Examples of connectivity of nodes of Ranvier in myelinated branching terminals of muscle spindle afferents (a -- c) from [22, 23] and of a touch receptor afferent (d) from [24]. Red semicircles represent heminodes (leaf nodes) which receive external inputs. Internal nodes of Ranvier are marked by blue circles; the green circle marks the root node. In the present paper we focus on a class of small networks: small random trees of strongly-coupled stochastic excitable elements. Such networks serve as models for certain types of peripheral sensory neurons whose morphol- ogy includes tree-like branched myelinated distal terminals. Examples of such sensory neurons include cutaneous mechanoreceptors [24, 25], pain receptors [26], mechanoreceptors in lungs (pulmonary afferents) [27, 28], some elec- troreceptors [29], and muscle spindles [30, 31]. Terminal branches of such neurons are wrapped by myeline which is interrupted at the nodes of Ranvier, located at branching points. Figure 1 exemplifies such networks for terminal branching of muscle spindle afferent neurons and for a touch receptor afferent. Starting from a primary node (green circle), branching continues for a few generations (2 -- 5), terminating at leaves, called heminodes, at which myeline ends. Heminodes receive sensory inputs from thinner neurite processes. In response, the sensory neuron generates a sequence of action potentials. The number of nodes and heminodes, their connectivity and the size of terminals vary among individual neurons. Due to the high density of voltage-gated Na+ ion channels at the heminodes, an action potential (AP) can be triggered at any heminode. Therefore, such neurons obey multiple stimulus encoding zones [30]. Furthermore, despite the inevitable randomness of input signals to the individual spatially-separated terminal endings, these sensory neurons often exhibit pacemaker-like activity, characterized by noisy periodic spiking [22]. Several dynamic mechanisms were proposed in order to explain AP generation, the periodicity of firing, and the observed nonlinear responses of these neurons. Those mechanisms include random mixing [32], nonlinear competition between multiple pacemakers, associated with heminodes [22], and additional mechanical coupling between sensory receptors [33]. In an alternative approach, the low resistance of myelinated segments, interconnecting the individual nodes of Ranvier, leads to strong coupling of their activity. In consequence, the stochastic firing of heminodes and nodes is synchronized and the whole branched terminal can be viewed as a single effective excitable system, which produces the corresponding firing (a) (b) (c) (d)3 statistics [34]. This proposal was supported by modeling studies using star [35] and regular tree [36] networks of stochastic excitable elements. In particular, in [36], a strong-coupling theory was developed, allowing prediction of the firing rate and spike train variability of strongly coupled excitable elements. Reconstructions of myelinated terminals of sensory neurons revealed that their tree structures varies among neurons [22 -- 24], see Fig. 1. This gives rise to a description of those terminals using random tree networks as models. Within this paradigm the specific coupling structure in a single myelinated terminal is just one possible realization of a random branching process, which generates random tree networks with certain statistical properties. Those properties can, for instance, be specified by providing a branching probability mass function, which, back in the experimental setup, would characterize myelinated terminal of a certain kind of neuron. As terminals of individual neurons may differ significantly, this raises the question of how this structural variability affects the statistics of neuronal firing [24]. The present paper is organized as follows. In Sec.II A we describe a model of Hodgkin-Huxley type excitable elements coupled on random tree network. The deterministic dynamics and measures of spike train variability for a tree network are described in Sec. II B,C. In Section II D we introduce a statistical description of random tree networks. The latter is then applied for particular examples of random binary trees in Sec.III A. Section III B and AppendixC are devoted to the strong coupling theory, which is applied to three examples of random trees in Sec.III C. We end with our conclusions in Sec. IV. II. MODEL AND METHODS In the present paper, we study the collective dynamics of excitable elements located at the branching points of random tree networks. Elements are interconnected by passive branches. This setup is illustrated in Fig. 1. A. Hodgkin-Huxley type model We assume that all nodes and passive links are identical, except for the leaf nodes, representing heminodes, which receive external inputs. Given a tree with N nodes, the dynamics of the nodes' membrane potentials are approximated by a discrete cable model [37] in which the membrane potential of the nth node is governed by N(cid:88) j=1 C Vn = −Iion + κ An,j(Vj − Vn) + Jn(t). (1) Here the index n = 1, 2, ...,N marks the respective node. In particular, n = 1 refers to the root node. In Eq. (1) C is the nodal membrane capacitance and the term Iion represents the nodal ionic currents. In the following, we use a Hodgkin-Huxley type (HH) model for the ionic currents of nodes of Ranvier, which is a simplified version of the model used in [38]. Ionic currents are represented by Na+ and leak currents, Iion = INa + IL, [35, 36]. The Na+ current is INa = gNam3h(V − VNa), where gNa = 1100 mS/cm2 is the maximal value of the sodium conductance and VNa = 50 mV is the Na+ reversal potential. The gating activation and inactivation variables obey the dynamics m = αm(V )(1 − m) − βm(V )m, h = αh(V )(1 − h) − βh(V )h, (2) with the following rate functions: αm(V ) = 1.314(V + 20.4)/[1 − exp[−(V + 20.4)/10.3]], βm(V ) = −0.0608(V + 25.7)/[1 − exp[(V + 25.7)/11]], αh(V ) = −0.068(V + 114)/[1 − exp[(V + 114)/11]], βh(V ) = 2.52/[1 + exp[−(V + 31.8)/13.4]]. In Eq. (1) the coupling between nodes is described by κ(cid:80)N The leak current is given by IL = gL(V − VL) with gL = 20 mS/cm2, VL = −80 mV, and the nodal capacitance is set to C = 2 µF/cm2. j=1 An,j(Vj − Vn), where A is the adjacency matrix of the undirected rooted tree graph. It is a N × N symmetric matrix with elements Ai,j = 1 for connected nodes i and j, and Ai,j = 0 for unconnected nodes, see Appendix A for more details. In the following the coupling strength, κ, is used as a control parameter. However, the physiologically-relevant values of κ can be estimated from the sizes of a node, the myelinated links, and the axoplasmic resistivity [37], giving the range of ≈ 125 -- 1500 mS/cm2 [35, 36]. The external currents Jn are applied to the leaves only and consist of a constant and a noisy part, i.e Jn(t) = δn,l[J + √2S ξl(t)], (3) 4 FIG. 2. Sufficiently-strong input current to leaf nodes fires up the root node. (a): A sample tree with N = 17 nodes. The H = 8 leaf nodes are marked by red semicircles. The root node is marked by the green circle. Star symbols point at "recording" sites of voltage traces shown in Fig. 3b. The shown tree structure reproduces a reconstruction of an experimentally-observed muscle spindle afferent neuron presented in Ref. [23]. (b): Threshold current, Jth, for the onset of repetitive firing of the root node, as a function of the coupling strength, κ, for the tree shown in panel (a). The dashed horizontal line marks the theoretical estimate of the threshold current in the strong coupling limit, J∞ = (N /H)JAH = 61.75 µA/cm2, see in Sec. III B. where l denotes indices of leaf nodes; δn,l is the Kronecker delta. The zero-mean Gaussian white noise ξl(t) with intensity S is uncorrelated for different leaves, i.e. (cid:104)ξi(t)(cid:105) = 0, (cid:104)ξi(t)ξj(t + τ )(cid:105) = δi,j δ(τ ). Thus, leaf nodes receive random uncorrelated inputs. Other sources of noise, e.g. due to fluctuations of nodal ion channel conductances are neglected. In contrast to regular tree networks where inputs are administered only to nodes in the last generation of a tree [36], leaves can occur at any generation in random tree networks, see Fig. 1. Eqs.(1 -- 3) were integrated numerically using the explicit Euler - Maruyama method with time step of 0.1 µs for 60 -- 600 seconds long simulation runs. B. Deterministic dynamics We first discuss repetitive action potential generation at the root node for deterministic input currents, S = 0 in Eq. (3). Qualitatively different dynamical operation modes of the root node are separated by a threshold value, Jth, of the constant current, J, applied to leaf nodes. While APs evoked at the leaf nodes do not fire up the root node for low currents, values of J exceeding Jth result into sustained periodic firing of the root node. This is reminiscent of the dynamic behavior of a single isolated node. The latter is at resting state in the absence of external input, J = 0. A sufficiently high constant current results in an Andronov-Hopf bifurcation of the equilibrium state rendering an isolated node to fire a periodic sequence of action potentials (APs). For a single node this Andronov-Hopf bifurcation occurs at JAH ≈ 29.06 µA/cm2 [36]. As in Ref. [36], we numerically calculated the threshold current Jth, which is the minimum constant current applied to the leaf nodes, which for a given coupling strength, results in the repetitive sustained generation of full-size APs (a voltage spike of at least 60 mV magnitude) at the root node. As for regular trees [36], the threshold current depends on the coupling strength, Jth(κ), as exemplified in Fig. 2b. As the coupling strength increases, more input current is required to sustain firing of the root node. In consequence, the threshold current increases with κ and, finally, saturates at the limiting value J∞ := limκ→∞ Jth(κ) for strong coupling. The strong coupling regime spans the range of physiologically realistic values, κ > 100 mS/cm2, for branched myelinated terminals of sensory neurons [35, 36]. For a given value of κ the root node shows a sustained sequence of APs, if values of the input current are above Jth(κ), shown in Fig. 2b, and no APs if the value of Jn is below that curve. These two regimes are referred to as oscillatory and excitable, respectively, in the following. Note that for weak coupling the dynamics of the tree can be quite complex, e.g. not every AP generated at leaf nodes may propagate all the way to the root node. Such regimes will be studied elsewhere. For strong enough coupling and sufficient input current, all nodes in the tree are synchronized and fire periodic sequences of APs. As in the case of regular trees, the limiting value of the threshold current for strong coupling, J∞, matches the threshold value of the current of an effective single node with parameters re-scaled by the ratio of the number of inputs (leaf nodes) and the total number of nodes as, J∞ = (N /H)JAH, see dashed line in Fig. 2b. This value is derived in Sec. III B and Appendix C. (a)(b)C. Spike train statistics 5 In the presence of noise, the AP generation becomes stochastic. We are particularly interested in statistical prop- erties of spike trains generated at the root node. We extracted a sequence of spike times of the root node, {ti}, from 60 − 600 s long simulation runs. The sequence of interspike intervals (ISIs), τi = ti+1 − ti, is characterized by the firing rate, r, and by the coefficient of variation (CV), Cτ , r = (τ )−1 , σ2 τ = τ 2 − (τ )2, Cτ = rστ , (4) where τ and στ are the mean and the SD of the sequence of ISIs, respectively. The bar stands for the averaging over all ISIs in the spike train of the root node. The CV quantifies the ISI variability. FIG. 3. Stochastic dynamics of excitable elements coupled on the tree network shown in Fig. 2a. (a): Time traces of the membrane potentials of two leaf nodes and the root marked by stars in Fig. 2a, for the indicated coupling strengths. (b,c): Firing rate, r, and coefficient of variation, Cτ for spike trains generated at the root node as a function of coupling strength. Dashed lines in panels (b) and (c) refer to theoretical estimates of the firing rate and CV in the strong coupling limit. The parameters of input currents to leaves are: J = 50 µA/cm2, S = 500 (µA/cm2)2ms. Figure 3 exemplifies the stochastic firing dynamics for the tree shown in Fig. 2a. The constant current was chosen such that the tree is in the excitable regime for κ > 20 mS/cm2. As in regular trees [36], the firing rate, r(κ), depends non-monotonously on the coupling strength. For weak coupling, spikes generated at the leaves often fail to propagate to the root, leading to its sparse firing, Fig. 3a. This results in low values of the firing rate and large values of the CV. Furthermore, the root and leaf nodes fire asynchronously. Increasing the coupling strength leads to stronger interaction between nodes. In consequence, synchronous coherent firing with large firing rates and small CVs emerges, see middle panel of Fig. 3a. As the coupling increases, more input current is required to sustain the network firing, see Fig. 2b. Consequently, for constant input current the firing rate decrease again. Competition of these two tendencies results in a maximum of the root node firing rate as a function of the coupling strength. For strong coupling, nodal firing is perfectly synchronized, see rightmost traces in Fig. 3a, and the firing rate and the CV saturate at their limiting values, indicated by dashed lines in Fig. 3b,c. In Sec. III B we show that in the case of strong coupling, a tree can be well represented by its root node which receives effective input with rescaled constant current and noise intensity. D. Statistical description of random trees networks An ensemble of random trees can be constructed as a set of realizations of a stochastic branching process [39]. In this paradigm, the tree shown in Fig. 2a is just one possible realization of a random tree network. Each network realization causes certain statistical properties of its root node's firing characteristics, such as those depicted in Fig. 3. (a)(b)(c)In each realization, branching starts at the root and continues up to a prescribed maximal number of generations, G, or until all branches end in leaf nodes. Each node in a certain generation g is located at the same path distance, g, from the root. The latter defines the generation g = 0. Furthermore, each node in the g-th generation, g < G, is a parent to a random number of offspring in the (g + 1)-st generation. However, each child node has only one parent. The k-th realization of a tree network consists of several generations each containing a random number of nodes, Dg,k, which we model by using the Galton-Watson process [40], Dg+1,k = dg,i,k, g = 0, 1, ..., G − 1, D0,k = 1. (5) (cid:88) i∈gen.g 6 Here g indicates the generation and k = 1, 2, . . . indicates a particular realization of a tree network. The sum runs over all nodes in generation g. The number of offspring, dg,i,k, of a certain parent node i in generation g is an independent random variable generated from the probability mass function (PMF), pg(d). In the following sections, we will consider several examples of branching PMFs. We consider basic properties of a single tree network realization, first. The total number of nodes Nk of a tree network realization k is a random variable and obtained by summing Dg,k over all generations, Nk = Dg,k = Dg,k, (6) G(cid:88) g=0 Gk(cid:88) g=0 (cid:88) i∈gen.g G(cid:88) g=1 Hk = where Gk ≤ G is the height of the tree and is given by the actual number of generations in the current tree network realization. Thus, there might be no nodes in the outer generations for particular realizations of the Galton-Watson process, Eq. (5). The number of leaf nodes in a particular generation g is also a random variable given by hg,k = δdg,i,k,0. (7) By construction, branching terminates at the maximum generation G and all nodes in that generation are leaf nodes. In general, however, leaf nodes can be found in any but the 0-th generation. Thus, a node i in generation 0 < g ≤ G becomes a leaf node with probability pg(0), i.e. if dg,i,k = 0. The total number of leaf nodes, Hk, is a random variable, obtained by summing hg,k over all generations, i.e. hg,k. (8) if pg(0) = 0 for 0 ≤ g < G, all leaf nodes are in the peripheral generation, In the special case of no extinction, i.e. g = Gk = G and Hk = DG,k. Next, given the branching PMF, pg(d), and the maximum allowed number of generations, G, some statistical properties of random trees formed by the Galton-Watson process, such as the mean number of total nodes and leaves, their variances, etc. can be determined by the standard method of probability generating functions [1, 39, 40]. In the present paper, we are mainly interested in the influences of structural variability, as arising from different tree network realizations for the same branching PMF, on the firing statistics of the root node. We characterize that statistics for a single tree network realization k by calculating the firing rate, rk(J, S, κ), and the CV, Ck(J, S, κ) according to Eqs. (4). Note that these quantities depend not only on the network realization with the particular coupling structure, but also on input current, i.e. J and S, and the coupling strength κ. The variability of a certain quantity Qk, of the k-th tree realization, can be assessed by calculating its ensemble average mean and standard deviation (SD): (cid:104)Q(cid:105) = lim K→∞ 1 K K(cid:88) k=1 Qk, σ2 Q = lim K→∞ 1 K Q2 k − (cid:104)Q(cid:105)2. K(cid:88) k=1 This yields the ensemble averaged firing rate (cid:104)r(J, S, κ)(cid:105) and its normalized standard deviation: Cr(J, S, κ) = σr(J, S, κ) (cid:104)r(J, S, κ)(cid:105) . The latter provides a measure of variability for the root nodes' firing rates resulting from tree network realizations from the same branching PMF with respective input currents and coupling strengths. (9) (10) 7 {T} (cid:80) By defining sets of identical or isomorphic trees in an ensemble, averaging can be performed using the probability distribution of sets of identical trees. Denoting the set of parameters, which uniquely defines trees by {Tk}, the k-th realization of the quantity Q is denoted as Qk := Q({Tk}). Its average can be formally written as, (cid:104)Q(cid:105) = Q({T}) P ({T}), where P ({T}) is the PMF of non-identical trees and the summation run over all possible values of the parameters set, {T}. Various levels of coarse-graining methods can be used to simplify the ensemble averaging. In the present paper we restrict to identical nodes and interconnections except for the external input, which is only applied to leaf nodes. As a first way of coarse-graining, we consider trees with the same number of nodes and leaf nodes in each generation as identical. We get (2G − 1)-tuple {Tk} = (D1,k, ...,DG,k, h1,k, ..., hG−1,k) = ({Dk},{hk}). We note that trees with identical tuples ({Dk},{hk}) may still possess different connectivities. The ensemble of trees is then characterized by the joint (2G − 1)-dimensional PMF of number of the nodes and leaves in all generations, P2G−1(D1, ...,DG; h1, ..., hG−1) ≡ P2G−1({D},{h}). The tree-ensemble averages, Eqs. (9), are then approximated by (cid:104)Q(J, S, κ)(cid:105) ≈ Q(J, S, κ,{D},{h})P2G−1({D},{h}), σ2 Q(J, S, κ) ≈ Q2(J, S, κ,{D},{h})P2G−1({D},{h}) − (cid:104)Q(J, S, κ)(cid:105)2, (11) (cid:88) (cid:88) {D},{h} {D},{h} where the summations run over all possible values of D1, ...,DG and h1, ..., hG−1. As a second way of coarse-graining, we consider trees with the same total number of nodes and leaf nodes as identical. As we will show in Sec. III B, this simplification yields sufficient results in the strong-coupling limit κ → ∞. We parameterize a particular realization of tree network by the tuples (Hk,Nk) and then carry out averaging similar to Eq. (11), but with 2-dimensional PMF of the total number of leaves and nodes, P2(H,N ). Throughout the paper we focus on small trees, 2 < G ≤ 4, with branching supported on a bounded interval. This is consistent with the topology of branched myelinated terminals of sensory neurons [22 -- 25]. For such trees the number of configurations with distinct tuples ({D},{h}) in the same ensemble, is rather small. In particular, for binary trees this enables us to list all non-identical trees in the ensemble and calculate the corresponding joint PMF. Furthermore, dynamical measures, such as the firing rate and CV, can be calculated numerically for the complete small set of trees. Thus, the structure-induced variability of these measures can be calculated according to Eq. (11). III. RESULTS We use small binary trees for illustration. However, our approach is applicable to any random tree network, generated by the Galton-Watson process, as we demonstrate at the end of this section. A. Statistics of binary random trees In binary trees each node has at most two offspring. Here we consider two types of binary trees: full and non-full binary trees. The so-called full binary tree, is a tree in which every internal node has two offspring and leaves have none. In contrast, in non-full binary trees, which we term as general binary trees, the number of offspring of any internal node can be either one or two. 1. Full binary trees To avoid a large number of short trees in the ensemble, we allow extinction only after 2-nd generation. In conse- quence, the smallest tree possesses two generations, each with branching two. In our particular example branching after the 2-nd generation is characterize by the PMF: δd,2, pg(d) = 0 < g ≤ 2 p0δd,0 + (1 − p0)δd,2, 2 < g < G, δd,0, g = G. (12) The resulting ensemble is parameterized by two quantities: the probability of zero branching, p0, and the maximum number of generations, G. For such trees the numbers of nodes in the first two generations are fixed, D1 = 2, D2 = 4, while the numbers of nodes in higher generations are random variables ranging from 0 to 2g, for g > 2. The limit p0 → 0 corresponds to a tree with N = 2G+1 − 1 nodes and H = 2G leaf nodes, located in shell G. In the opposite limit, p0 → 1, trees are extinct after the 2-nd generation, resulting in a tree with the total number of nodes and leaf nodes N = 7 and H = 4, respectively. 8 FIG. 4. Tree networks for all 25 possible configurations of 2-tuples of (D3,D4) resulting for the branching PMF given by Eq. (12) for G = 4. Leaves are marked red, the root node is marked green and internal nodes are depicted as blue circles. Trees are arranged in columns with the same total number of leaves and nodes, (H,N ), shown on the bottom. The ratio of the total number of nodes to leaves, N /H, increases from left to right, from 1.75 to 1.9375 respectively. In the following, we consider a particular ensemble of full binary trees with at most G = 4 generations. While the numbers of nodes in 1-st and 2-nd generation are fixed, the number of nodes in 3rd and 4th generation are random integers. The latter take even values in the intervals 0 ≤ D3 ≤ 8 and 0 ≤ D4 ≤ 16. Furthermore, the numbers of leaf nodes in the 2nd and 3rd generation are determined by the number of nodes in those shells as, h2 = 4 − D3/2 and h3 = D3 − D4/2 [39]. In consequence, trees with the same numbers of nodes in the third and fours generation can be considered as identical. This leads to 25 unique 2-tuples of (D3,D4), or equivalently to 25 distinct trees in the ensemble, shown in Fig. 4. The joint PMF, P2(D3,D4), describing this ensemble, is given by (see Appendix B) (cid:18) 4 (cid:19)(cid:18)2n3 (cid:19) n3 P2(D3,D4) = D3 = 2n3, D4 = 2n4, n3 = 0, 1, .., 4, n4 = 0, 1, ..., 8. (1 − p0)n3+n4, p4+n3−n4 0 n4 (13) This two-dimensional joint PMF is shown in Fig. 5a. FIG. 5. Node statistics of full binary trees with the branching PMF given by Eq. (12) with p0 = 0.5 and G = 4. (a): The joint probability mass function of numbers of nodes in 3rd and 4th generation, P2(D3,D4). All 25 connectivity states are shown by filled circles. Probabilities for realizing the respective trees are indicated by circle diameter and color. (b): The PMF of respective combinations of total number of leaf nodes and nodes given by Eq. (14). Of particular interest is the statistics of the total number of nodes and leaf nodes. As we show in Sec. III B, both are used to derive approximations for certain measures of spike train statistics of the root node in the strong coupling limit. The number of distinct tuples, (Hk,Nk), i.e. the number of trees with identical total numbers of leaves and nodes, is smaller than the number of trees with identical (D3,D4) tuples, as trees with different numbers of nodes (4,7) (5,9) (6,11) (7,13) (8,15) (9,17) (10,19) (11,21) (12,23) (13,25) (14,27) (15,29) (16,31)02468D30481216D4p0=0.5(a)481216H7152331N(b)0.020.050.080.110.149 in certain generations may possess the same total number nodes and leaves. This is illustrated for the example of full binary tree ensemble in Fig. 4. As illustrated in the figure for G = 4, the number of distinct tuples (Hk,Nk) is 13 and thereby smaller than the total of 25 distinct trees in the ensemble. Furthermore, for the considered example of full binary trees, the number of leaves is exclusively determined by the number of nodes as H = 4 + (N − 7)/2. Therefore, the statistics of the total number of nodes and leaves is characterized by the one-dimensional PMF of the total number of nodes, P1(N ) (see Appendix B), (cid:34) 4(cid:88) 2n3(cid:88) n3=0 n4=0 (cid:18) 4 (cid:19)(cid:18)2n3 (cid:19) n3 n4 (cid:35) P1(N ) = δn3+n4,m p4+n3−n4 0 (1 − p0)m, m = (N − 7)/2. (14) The PMF P1(N ) is depicted in Fig. 5b. 2. General binary trees In general binary trees, each node has either zero, one, or two offspring. In our particular example, general binary trees are generated from the following branching PMF:  1 2 p0δd,0 + (δd,1 + δd,2) , 1 − p0 2 δd,0, pg(d) = 0 ≤ g < 1, (δd,1 + δd,2) , 1 ≤ g < G, g = G. (15) The root node may have either one or two offspring with equal probability. Other nodes may have either zero offspring, with probability p0, or either one or two offspring, each with probability (1− p0)/2. As in the case of full binary trees, an ensemble of general binary trees is parametrized by the probability of zero branching, p0, and by the maximal number of generations, G. In full binary trees, the number of nodes in each generation is an even number. In contrast, for general binary trees with the branching PMF (15) both odd and even number of nodes allowed. Compared to full binary trees, this leads to a larger number of trees with distinct tuples ({D},{h}). In order to reduce computational costs, we consider trees with the branching PMF (15) and set G = 3. The resulting network ensemble possesses 50 trees with distinct sequences of numbers of nodes and leaves, {D1,D2,D3, h1, h2}. It is characterized by a 5-dimensional joint PMF, P5(D1,D2,D3; h1, h2), which we calculated numerically for various values of the zero-branching probability p0. Figure 6a shows a small sample of possible tree realizations. The limit p0 = 0 corresponds to non-extinct binary trees, i.e. all leaf nodes are located in the 3rd generation. The opposite limit, p0 = 1, results in only two possible configurations with a total number of either 2 or 3 nodes (one or two leaf nodes, respectively). In contrast to full binary trees, where the total number of nodes uniquely determines the total number of leaf nodes, general binary trees allow for several distinct configurations with the same total number of leaf nodes, but different total numbers of nodes. For the considered case of G = 3 and p0 (cid:54)= 0, the PMF given in Eq. (15) leads to 28 distinct tuples (H,N ), whose PMF function is illustrated in Fig. 6b. It shows the multiplicity of possible total numbers of nodes for the same total number of leaf nodes. B. Strong coupling approximation In the physiologically-relevant case of strong coupling, the stochastic firing of all nodes is synchronized. In the synchronized state, the dynamics of individual tree network realizations, Eqs. (1 -- 3), can be approximated by that of an effective single node (see Appendix C): (16) Here Vk(t) is the effective membrane potential and Jeff,k(t) is the effective stochastic input current. The index k refers to the actual tree realization. The latter encodes the tree structure. The nodal ionic current, Iion[Vk, mk, hk], and the gating variables, mk and hk, are given by the same equations as for the network model in Sec. II A. Equations for Jeff,k(t) for regular trees were derived in [36]. In Appendix C, we extend their approach to random trees and obtain, C Vk = −Iion[Vk, mk, hk] + Jeff,k(t). (cid:112) 2Seff,k ξ(t) = R∞,k J + 2SS∞,k ξ(t), . (17) (cid:112) Jeff,k(t) = Jeff,k + R∞,k = Hk Nk , S∞,k = Hk N 2 k 10 FIG. 6. Realizations and statistics for general binary trees with the branching PMF (15) and maximal allowed number of generation, G = 3. (a): A sample of 12 different realizations of general binary trees. (b): The PMF of the total number of nodes and leaves, P2(H,N ); 28 possible trees with distinct total numbers of nodes and leaves are shown by filled circles, whose diameter and color represent probability values. Here J and S are the constant current and the noise intensity, respectively. Both specify the strengths of the noisy input current applied to the leaves in the actual tree realization; ξ(t) is Gaussian white noise. In the following, we use Eqs. (16) and (17) as an approximation for the dynamics of the root node, i.e. vk(t) ≈ V1,k(t), in the strong coupling limit. Thus, we replace the ensemble of trees by an ensemble of their effective root nodes. The respective total numbers of nodes and leaves, (Hk,Nk), in the individual tree realization determine the effective current, RkJ, and noise intensity, SSk, in Eq. (17). 1. Spike train statistics of tree realizations The strong-coupling theory allows for several important predictions. In the strong coupling limit, the dynamics of tree nodes is determined by the total numbers of leaves and nodes, {Hk,Nk}, only. In consequence, it does not depend on the particular configurations with unique sequences of numbers of nodes in shells, {D1,k, ...,DG,k}, on particular locations of leaves within shells, {h1,k, ..., hG,k}, as well as on nodal connectivity. Thus a set of distinct trees with identical total numbers of nodes and leaves, would show identical dynamics in the strong-coupling limit. In the deterministic case, S = 0, inputs to leaf nodes larger than the threshold current of a tree realization result in a sustained repetitive firing of the root node. From Eq. (17), we find for the threshold current J∞,k = R−1 where JAH is the threshold current of a single isolated node. ∞,kJAH, In the stochastic case, the firing statistics of the whole ensemble of trees can be predicted by evaluating the effective currents and noise intensities for all possible tuples (Hk,Nk). For a single effective node, the firing rate,(cid:101)r(Jeff, Seff), and the CV, (cid:101)Cτ (Jeff, Seff), solely depend on these effective parameters. For a certain tree realization, k, in the strong- coupling limit this yields, r(J, S, κ,{Dk},{hk}) ≈(cid:101)r(Jeff,k, Seff,k), Cτ (J, S, κ,{Dk},{hk}) ≈ (cid:101)Cτ (Jeff,k, Seff,k), (18) (19) where the tilde symbol indicates the firing rate and the CV of the effective node. This relation is illustrated in Fig. 7. The figure shows a heat map of (cid:101)r(Jeff, Seff), where symbols mark combinations of effective parameters that can actually be realized in the presented tree network ensembles. 2. Spike train statistics of tree ensembles As follows from the previous subsection, the ensemble-averaged dynamics and its variability can be predicted from the dynamics of a single isolated node, see Eq. (16 -- 19), and statistics of the total numbers of nodes and leaves. This (a)(b)11 FIG. 7. Heat map of the firing rate of a single HH node vs input current and noise intensity, (cid:101)r(Jeff, Seff). Symbols mark combinations of Jeff,k and Seff,k (17) of realizations of binary trees for the indicated ensembles. The magenta symbols (circles and squares) represent all possible full binary trees with branching PMF (12) and G = 4. Black star symbols mark all possible general binary trees with branching PMF (15) and G = 3. Parameters: S = 500 (µA/cm2)2ms; J = 38.5 µA/cm2 (unfilled stars and circles) and J = 70 µA/cm2 (filled stars and squares). enables us to derive strong-coupling approximations for ensemble-averaged quantities such as the root node's firing rate. As the effective parameters in Eq. (16) solely depend on the total number of nodes and leaf nodes, ensemble averaging in Eq. (11) can equivalently be performed using the two-dimensional joint PMF of total number of nodes and leaf nodes, P2(H,N ). Then, the ensemble averages of the firing rates and CV can be calculated by using the 2-dimensional PMF, P2(N ,H), in Eqs. (11) as described in section II D before. As we restrict on finite ranges of possible branching, it is also possible to determine bounds of corresponding quantities, such as maximal and minimal firing rates and CVs of the ISI sequence. C. Onset of repetitive spiking The randomness of tree ensembles may lead to qualitatively different dynamics of individual network realizations. In the present section we consider its consequence for the threshold current setting the onset of repetitive spiking of the tree's root node in the case of deterministic input currents, i.e S = 0. We numerically calculate the threshold current as a function of coupling strength for each of 25 distinct full binary trees depicted in Fig. 4. This results in one curve for each tree network realization, each one similar to the curve shown in Fig. 2b. Besides its dependence on the coupling strength κ, the threshold current for a single tree network realization, Jth,k, is also a function of the number of nodes in the 3-rd and 4-th generation, i.e. Jth,k = Jth(κ,D3,k,D4,k). It can be expressed in units of JAH by introducing the dimensionless scaling factor R−1(κ,D3,k,D4,k), i.e. Jth,k = Jth(κ,D3,k,D4,k) = R−1(κ,D3,k,D4,k)JAH. At J = JAH a single isolated node enters the repetitive spiking regime by undergoing an Andronov Hopf bifurcation. We will refer to R−1(κ,D3,k,D4,k) as the normalized threshold current in the following. For strong coupling the threshold current of a single tree network realization approaches its limiting value, J∞,k(H,N ), given by Eq. (18). The latter depends only on the total number of nodes and leaf nodes, i.e. J∞(H,N ) = R−1 ∞ (Hk,Nk)JAH = (Nk/Hk)JAH. Here limκ→∞ R(κ,D3,k,D4,k) = R∞(Hk,Nk) = Hk/Nk. Normalized threshold currents are shown in Fig. 8(a,b). We find that it increases monotonically with the coupling strengths and finally saturates for strong coupling. For weak coupling, the actual tree structure matters as it determines the paths action potentials have to travel in order to excite the root node. Here trees with various configurations, but identical numbers of leaves and nodes, fall into tight clusters around distinct values of R−1. With the increase of coupling the clusters blur. For general binary trees, Fig. 8(b), the range of threshold currents is significantly larger than for 515253545Jeff(µA/cm2)101102Seff[(µA/cm2)2ms]r(Hz)J=38.5(General)J=70(General)J=38.5(Full)J=70(Full)5203550658012 FIG. 8. Normalized threshold current, R−1(κ,{Dk},{hk}) = Jth(κ,{Dk},{hk})/JAH, as a function of coupling strength κ for two tree network ensembles. (a) curves for all 25 full binary trees of Fig. 4, (b) same for 50 general binary trees with all possible numbers of nodes and leaf nodes, with maximal number of generations, G = 3. Curves are color-coded according to the tree's normalized threshold currents in the strong coupling limit, R−1 ∞,k = Nk/Hk. Dashed horizontal lines show the value of the applied constant current, J = 38.5 µA/cm2, used for stochastic simulations in Fig. 9. (c,d): Normalized threshold current for κ = 1000 mS/cm2 versus its theoretical strong-coupling limit, R−1 ∞,k = Nk/Hk for full (c) and general (d) binary trees. full binary trees, Fig. 8(a). Finally, for strong coupling curves for individual sample trees saturate. Their limiting values are well approximated by the theoretically predicted strong coupling limit J∞,k = (Nk/Hk)JAH, as illustrated in Fig. 8(c,d). These results indicate that for weak and moderate coupling the onset of spike generation may be strongly affected by the particular tree structure. However, for physiologically relevant coupling strengths, κ > 100 mS/cm2, threshold currents are close to their strong-coupling limits and their values mainly depend on the statistics of the total number of nodes and leaf nodes. D. Stochastic dynamics For stochastic input currents, i.e. S > 0, variability in tree structures interacts with variability caused by stochastic inputs. In order to study the stochastic dynamics we prepare ensembles of excitable trees, i.e. no spike generation at the root node if the input current had no stochastic component. To this end, we set the value of the constant input current such that it is below the sample trees' threshold currents for strong coupling, but causes non-vanishing firing rates, > 2 Hz, of all possible tree realizations. For the two types of binary trees, used in the previous section, this is achieved by using a constant input current of J = 38.5 µA/cm2. This corresponds to the normalized threshold current, R−1 = 1.325, shown by the dashed lines in Fig. 8a,b. For the coupling strengths, κ > 30 mS/cm2, all curves of the threshold current in Fig. 8a,b lie above the dashed lines, indicating that all trees are indeed excitable. Then, we simulated Eqs.(1 -- 3) for all possible non-identical tree network realizations and estimated the firing rate and CV of their root nodes, r(κ,{D},{h}) and Cτ (κ,{D},{h}), respectively. The obtained measures of spike train statistics are shown in Fig. 9. For individual tree realizations, theses measures show qualitatively similar a dependences 13 FIG. 9. Spike train statistics of ensembles of excitable trees. The upper panels (a,b,c) show results for all 25 full binary trees of Fig. 4; the lower panels (d,e,f) refer to 50 general binary trees with at most G = 3 generations and with all possible numbers of nodes and leaf nodes. Trees' firing rates (a,d) and CVs (b,e) as functions of coupling strength. Lines are color-coded according to the scaling factor, R∞,k = Hk/Nk, for the strong coupling limit; dashed horizontal lines show lower and upper limits of corresponding quantities, obtained from the strong-coupling approximation. (c,f): Firing rate as a function of the CV for κ = 1000 mS/cm2 for simulated trees (crosses) and for single root nodes (circles) with input current rescaled according to (17). The parameters for numerical simulations are: J = 38.5 µA/cm2, S = 500 (µA/cm2)2ms. on the coupling strength. In more detail, we find that firing rates become maximal at intermediate coupling strengths for both full and general binary trees. CVs attain their minimum values at similar coupling strengths. This indicates most regular firing of the root node at intermediate coupling strengths. Note that this is in qualitative agreement with previous results on regular tree networks presented in [36]. Considering the spike train statistics of the entire tree ensemble, we find that structural variability hardly affects firing rates and CVs for weak coupling. In contrast, firing rates and CVs of individual tree realizations strongly differ for physiologically relevant strong coupling. For such coupling, the spike train statistics of individual tree realizations are well-approximated by those of single isolated nodes with effective currents and noise intensities given by Eq. (17). This is further illustrated in Fig. 9(c,f), where the ISIs statistics of root nodes of tree realizations is compared to those of effective isolated nodes according to the strong-coupling approximation. Importantly, the strong-coupling approximation also allows for the prediction of lower and upper bounds of the firing rates and CVs, shown by the dashed lines in Fig. 9a -- d. The scaling parameters R∞,k and S∞,k of the effective current in Eq. (17) define the range of the trees' firing rates and CVs at strong coupling, see Fig. 7. In that sense, they provide bounds for the influence of structural variability on the spike train statistics of the root node. Since all tree realizations are excitable for strong coupling, the highest values of scaling parameters R∞,k yield the highest firing rate and lowest CV. In contrast, small values of R∞,k yield low effective currents, Eq.(17), and drive the network deep into the excitable regime. This causes Poisson-like statistics of spike generation resulting in low firing rates and CVs close to one. Next we consider the ensemble-averaged ISI statistics. It is obtained by averaging the firing rates and CVs according to Eq.(11). Averaging is simplified for the full binary trees, as the joint PMF, Eq. (13), for G = 4 depends only on the numbers of nodes in the 3-rd and 4-th generations. For the general binary trees with G = 3 the ensemble averaging is performed using the 5-dimensional joint PMF, P5(D1,D2,D3; h1, h2) which we estimated numerically. Ensemble-averaged measures of spike train statistics are shown in Fig. 10 for different values of the zero-branching probability p0. The ensemble-averaged firing rates and CVs follow curves that are qualitatively similar to those for individual trees. However, p0 strongly affects the ensemble-averaged statistics in the strong coupling regime. Low probabilities cause on average taller sample trees with smaller fractions of leaf nodes. In consequence, the effective current and noise intensity in Eq. (17) become smaller, which drives the network deep into the excitable regime, and results in low firing rates and large CVs. Larger values of p0, refer to an increased fraction of short trees with larger fractions of leaves, which receive inputs. The latter results in higher firing rates and smaller CVs. In order to quantify the influence of structural variability on the firing statistics of root nodes, we consider the 1001011021030255075100r(Hz)(a)10010110210300.250.50.751Cτ(b)R∞0.520.530.540.550.560.570.20.40.60.81010203040r(Hz)(c)ScaledsinglenodeNetwork100101102103κ(mS/cm2)0255075100r(Hz)(d)100101102103κ(mS/cm2)00.250.50.751Cτ(e)0.250.350.450.550.650.250.50.751Cτ0204060r(Hz)(f)14 FIG. 10. Ensemble-averaged ISI statistics as function of coupling strength for the indicated values of the zero-branching probability. (a,b): Ensemble averaged firing rate, (cid:104)r(cid:105), and CV (cid:104)Cτ(cid:105) for full binary tress with G = 4; (c,d): (cid:104)r(cid:105) and (cid:104)Cτ(cid:105) for general binary trees with G = 3. Dashed lines show predictions of the strong-coupling theory, obtained by ensemble averaging of corresponding values for isolated single nodes with the effective current and noise intensity according to Eq. (17). Parameters: J = 38.5 µA/cm2, S = 500 (µA/cm2)2ms. FIG. 11. Variability of the firing rate due to structural variability. Upper and lower panels show the ensemble averaged firing rate, (cid:104)r(cid:105), and its normalized SD, Cr, (11, 10) as a function of the probability of zero branching, p0, for the full binary trees and for general binary trees, respectively. Solid lines show results of direct simulations of trees; dashed blue lines show prediction of the strong-coupling theory. Other parameters are the same as in Fig. 10. normalized standard deviation, Cr, of the distribution of firing rates, see Eq. (10). We find that firing rate distributions of general binary trees show larger variability than those for full binary trees. Besides the previously noticed fact that the structural variability is most pronounced for strong coupling, we find a maximum of the Cr for a finite value 1001011021030255075100hri(Hz)(a)p0=0.05p0=0.7p0=0.910010110210300.250.50.751hCτi(b)100101102103κ(mS/cm2)0255075100hri(Hz)(c)100101102103κ(mS/cm2)00.250.50.751hCτi(d)00.250.50.75104590hri(Hz)(a)κ=1κ=10κ=1000(network)κ=1000(single)00.250.50.75100.51Cr(b)00.250.50.751p004590hri(Hz)(c)00.250.50.751p000.51Cr(d)of zero-branching probability. For full binary trees the number of possible distinct trees approaches one for p0 → 0, i.e. only the regular tree with branching two, and p0 → 1, i.e. only the regular tree with the minimal number of generations. The pronounced maximum of Cr expresses the trade-off between trees becoming more regular as p0 approaches either one of those limits. In general binary trees, however, the limit p0 → 0 still results in an ensemble of 17 possible trees with distinct pairs of number of nodes and leaf nodes, as all combinations of branching one and two are possible. Consequently, the normalized SD at p0 = 0 remains finite. 15 FIG. 12. Normalized standard deviation of the distribution of root nodes firing rates, Cr = Cr(p0, J), among ensembles of full (a) and general (b) binary tries as a function of the constant input current to leaf nodes and the zero-branching probability for the strong-coupling limit. Other parameters are the same as in Fig. 10. The variability of the root node's firing rates indeed depends on the input current to the leaves. For strong coupling a particular coupling structure is imprinted in the scaling of input current according to Eq.(17), and so the spread of effective input currents for trees in the ensemble translates into the spread of their firing rates. This can be seen in Fig.7 by comparing locations of tree realizations for two values of input currents to leaves. For J = 38.5 µA/cm2, all tree realizations are in the excitable regime and their positions in the heat map cut across a wide range of firing rates. For J = 70 µA/cm2, however, most tree realizations are in oscillatory regime, cutting across a narrower range of firing rates. A decrease of the input current shifts trees to the left in Fig. 7, i.e. deeper to the excitable regime. The increase of J moves trees to the right and trees become oscillatory. As a result, the ensemble variability of firing rate is high for small input currents and low for large currents. These results are summarized in Fig. 12, which shows the dependence of the normalized SD of firing rate, Cr, on the input current and the zero-branching probability for the strong-coupling limit. Except for small input currents, J < 30 µA/cm2, the firing rate variability of general binary trees is larger than that of full binary trees, as they cut across wider ranges of scaling factors of input current in Eq. (17). E. Non-binary trees We finally consider an example of random non-binary trees. In this example, random branching is drawn from a uniform distribution with at most 4 offspring for each node, and trees with at most G = 4 generations are allowed. To avoid short trees we set the zero branching probability to zero, pg(d) = 0, for generations g = 0, 1, 2, as in the example of full binary trees. Thus, the branching PMF is given by 4(cid:88) 4(cid:88) i=1 i=0 δd,0, pg(d) =  1 4 1 5 δd,i, 0 < g ≤ 2, δd,i, 2 < g < 4, g = 4. (20) The PMF in Eq. (20) yields a large number of non-identical trees, i.e. trees that differ in the their numbers of nodes and leaf nodes per generation. Figure 13a exemplifies 3 non-identical tree network realizations obtained from the PMF. Instead of counting distinct trees and calculating their corresponding probabilities, we analyzed tree network ensembles obtained from the PM using a brute-force approach. We generated an ensemble of 2000 tree 00.250.50.751p020406080J(µA/cm2)(a)00.250.50.751p020406080(b)Cr<0.010.23>30network realizations and then proceeded with the analyses of collective dynamics of coupled HH nodes as in previous sections. For deterministic inputs, the threshold current as a function of coupling strength possesses a similar shape as those for binary trees (data not shown). This included the strong coupling regime, where threshold currents approached the value predicted by the strong coupling theory, J∞,k = (Nk/Hk)JAH. 16 FIG. 13. Firing rate statistics of non-binary random trees with uniform branching according to the PMF Eq. (20). (a) Three examples of tree networks; pairs of total number of nodes and leaf nodes are: (49,70) for the upper panel, (4,10) for the lower left, and (24,40) for the lower right panel, respectively. (b) Heat map of the firing rate of a single HH node as a function of input current and noise intensity,(cid:101)r(Jeff, Seff). Symbols mark combinations of effective parameters, Jeff,k and Seff,k in Eq.(17), for the ensemble of 2000 network realizations. (c) Probability distributions of the firing rate of the root nodes for the indicated values of coupling strengths. For strong coupling, κ = 1000 mS/cm2, solid and dashed lines compare direct simulations of trees realizations and simulations of effective single nodes, respectively. We then set the constant input current at J = 38.5 µA/cm2, for which all network realizations resulted in excitable trees for coupling κ > 100 mS/cm2. Firing statistics of root nodes of the 2000 tree realizations showed qualitatively similar dependences on the coupling strength as for binary trees of Fig. 9 (not shown). As for binary trees, structural variability of the firing rate of non-binary trees is small for weak and intermediate coupling and large for strong coupling. This is illustrated in Fig. 13c, by the probability distribution of the firing rate of root nodes. The latter broadens as the coupling strength increases. As for binary trees, the firing statistics of non-binary tree network realizations can be predicted using the strong- coupling theory of Sec.III B. In order to compare the firing rate statistics obtained from simulations for the 2000 tree network realizations with that resulting from the strong coupling approximation, we first calculate the pairs of effective current and noise intensity for each tree realization using Eq. (17). The locations of those pairs in the current noise intensity space is shown in Fig. 13b. The figure also shows a heat map of the firing rate of a single node r as a function of current and noise intensity. Then, using Eq. (19) we obtain an approximate of the firing rate for each pair of effective currents, i.e. for each tree network realization. The resulting distribution of firing rate approximates (single) is compared to the root nodes' firing rates for different coupling strengths as obtained from simulations of network activity in Fig. 13c. 51015202530Jeff(µA/cm2)101102Seff[(µA/cm2)2ms]r(Hz)52035506580(b)(a)(c)IV. CONCLUSION 17 We studied the influence of network structure on the root node's spike train statistics in random tree networks of excitable elements in which only the leaf nodes receive stochastic inputs. This setup was motivated by the morphology of certain sensory neurons which possess branched myelinated terminals with excitable nodes of Ranvier at the branch points. Myelination ends at the so-called heminodes, representing leaves of a tree, receive sensory inputs. As inputs excite heminodes, action potentials synchronously jump over myelinated branches and ultimately fire up the root node. Branched myelinated terminals can be represented by small random tree networks which differ in height (number of generations), numbers of nodes and heminodes, as well as nodal connectivity [22, 23, 25, 41]. Thus, the resulting spike train variability may be sensitive to the network structure. We developed a probabilistic framework to study the collective response of stochastic excitable elements coupled on random trees whose structure is generated by Galton-Watson random branching processes. We investigated the variability of the spike train statistics resulting from variations of network structure within a tree ensemble. We have shown that in the physiologically relevant strong coupling regime the firing statistics of the root node is determined by the number of nodes and leaf nodes, while being hardly affected by a particular nodal connectivity. Thus, trees in the ensemble can be distinguished by the total number of nodes and leaf nodes, which simplifies the calculation of the ensemble averages significantly. Furthermore, the collective response of the tree network can be predicted from a single node with an effective input, rescaled according to the number of nodes and leaf nodes. Given a joint probability distribution of the total number of nodes and leaf nodes, this allows for the calculation of the ensemble averaged firing rate and coefficient of variation as well as for setting lower and upper bounds of firing rate statistics. Using two types of binary random trees and an example of random trees with a uniform branching we found that structural variability, resulting from different realizations of the network connectivity, strongly affects the root node's spike train statistics for strong coupling. In particular, ensembles of realizations of excitable tree networks show a wide range of firing rates and coefficients of variations, consistent with experimental findings on touch receptors [24, 42]. While we considered uniform inputs to heminodes (leaf nodes), an additional level of randomness can be introduced by non-uniform random inputs to heminodes, as in [24]. This would lead to additional variability across tree realiza- tions. Interestingly, recent work yielded experimental evidence for structural plasticity of Merkel cell touch receptor complexes in healthy skin [25]. The study documented that the number of heminodes of a touch receptor afferent adjusts to the inputs from Merkel cells, which varies over a time span of several days. In our framework such an afferent remodeling corresponds to the variation of inputs, accompanied by structural changes of corresponding tree network. Strong-coupling approximation then can be used for prediction of neuronal responses during remodeling cycles. Acknowledgements The authors thank W. Just for his valuable comments and suggestions and D.F. Russell for discussions which inspired this work. AKN and ABN acknowledge support by the Neuroscience Program and by Quantitative Biology Institute at Ohio University. LSG thanks Ohio University for hospitality and support. Appendix A: Construction of Adjacency matrix 18 The adjacency matrix was used in numerical simulations of coupled nodes, Eqs. (1-3). Throughout the appendices we drop the index k referring to a particular tree network realization. In order to construct a single tree network realization, we generate a sequence of random numbers, {dg}, according to the branching PMF, pg(d). Then, the numbers of nodes in each generation, D0,D1, . . . ,DG are calculated from Eq. (5). The total number of nodes (6) yields the dimensions of the symmetric adjacency matrix, A. To construct the adjacency matrix, nodes in a tree are indexed by j, starting from the root node j = 1 in the 0-th generation and then proceeding with the nodes' offspring. The number of nodes until generation g, Mg, is given by g(cid:88) i=0 Mg = Di, g = 0, . . . , G. In the first generation, the d0,1 nodes are indexed as j = 2, ...,M1; nodes in the second generation are indexed in the order of their parent nodes, i.e. j = M1 + 1,M1 + 2, ...,M1 + d1,2 for d1,2 offspring of node j = 2 in generation g = 1 and so on. This is illustrated in Fig. 14(a). Thus, node indexes run from j = 1 to j = MG = N . FIG. 14. Example of a random tree network with G = 4 (a) and corresponding adjacency matrix (b), constructed following Eqs. (A1,A2). As the full adjacency matrix follows from symmetry, we restrict our description to the upper triangular matrix Aup. Its first row contains all connections to the root node: (cid:40) Aup j,1 = A1,j = 1, 0, j = 2, . . . ,M1, else. (A1) Interconnections between a node j in generation g and its offspring in generation g + 1 result in a sequences of ones of lengths dg,j in the jth row of Aup, where dg,j is the number of offspring of jth node, which is part of generation g. In more detail, Aup j,i = 1, 1, 1, 0, i = (l + 1), i = (l + 2), ... i = (l + Dg) = Mg, else ... j =(cid:0) j = (Mg + 1), . . . . . . . . . ., (Mg + dg,l+1) j = (Mg + 1 + dg,l+1), . . . , (Mg + dg,l+1 + dg,l+2) Dg−1(cid:88) b=1 (cid:1), . . . ,(cid:0) Dg(cid:88) b=1 Mg + 1 + dg,l+b Mg + dg,l+b = Mg+1 (cid:1) , 1 ≤ g ≤ G − 1, (A2)  where l = Mg−1. Finally, the full adjacency matrix follows from symmetry. (a) (b)12345678Appendix B: Probability mass function of numbers of nodes and leaves for full binary trees For full binary trees with a maximum of G = 4 generations and branching PMF (12) the joint PMF of the number of nodes and leaf nodes in 3rd and 4th generations, P2(D3,D4, h2, h3) = P2(D3,D4) is 19 P2(D3,D4) = P(D3) × P(D4D3), where P(D3) is given by the binomial distribution P(D3 = 2n3) = pD2−n3 0 (1 − p0)n3, (B1) (B2) (cid:18) (cid:19) D2 n3 (cid:18) (cid:19) D3 n4 with integer values n3 = 0, 1, ..,D2, specifying the number of parent nodes in generation g = 2. Accordingly, for given D3, we find P(D4 = 2n4D3) = pD3−n4 0 (1 − p0)n4 . (B3) Here n4 = 0, 1, ..,D3 is the number of parent nodes in the 3rd generation. Applying Eq. (B1) this yields Eq. (13). For the full binary tree considered in the main text, we find H = 4 + (N − 7)/2. In consequence, the joint PMF of the total number of leaf nodes and nodes is determined by the PMF of the total number of nodes, P1(N ). The latter can be obtained by summing Eq. (13) such that, D3 +D4 = N − 7, or, equivalently, n3 + n4 = (N − 7)/2. This yields Eq. (14). Appendix C: Dynamics in the strong coupling limit In the strong coupling limit, each of the network realizations approaches a synchronized state. Its dynamics can be treated as that of a single node given by Eq. (16). In the following, we derive the effective parameters, see Eq. (17), which account for the influence of network structure on the dynamics in the strong coupling limit. In order to simplify the notations, we skip the index referring to the considered network realization. For the derivation, we follow the approach presented in [36]. We first consider the coupling term in Eq. (1). Instead of using the adjacency matrix, (see Appendix A), we can rewrite the coupling term as a sum over interconnections between adjacent nodes: (cid:88) C Vj = −Iion,j + Jj(t) + κ = −Iion,j + Jj(t) + κ(1 − δgj ,0)(cid:0)Vmj − Vj i∈gen. gj−1 (cid:1) + κ Aj,i (Vi − Vj) + κ Aj,m (Vm − Vj) (cid:0)Voj − Vj (cid:1) . (C1) (C2) (cid:88) (cid:88) m∈gen. gj +1 oj∈ offspring of j Here gj denotes the generation of node j. In the first line, the sums run over all nodes in adjacent generations. In the second line, only nodes that are connected to node j are considered, i.e. its only parent node, mj, in generation gj − 1 and all offspring in generation gj + 1. The term with the Kronecker delta accounts for the fact that the root node has no parent. Note that the relation between nodes, their offspring, and their parent nodes implies that k = oj ⇐⇒ j = mk. these differences as new variables As only the differences in membrane potentials between nodes and their offspring enter Eq. (C1), we introduce Voltage difference and the corresponding notation is illustrated in Fig. 15. Applying this to Eq. (C1), we obtain ∆Voj ,j := Voj − Vj. (C3) C Vj = −Iion,j + Jj(t) − κ(1 − δgj ,0)∆Vj,mj + κ ∆Voj ,j. (C4) (cid:88) oj∈offspirng of j From this, we can derive the dynamics of the voltage differences by subtraction of the two equations for Voj and Vj: C d dt ∆Voj ,j = −Iion,oj + Iion,j + Joj (t) − Jj(t) + κ(1 − δgj ,0)∆Vj, mj (cid:88)  + κ ∆Vo(cid:48) j ,j (cid:88) ∆Voj ,j + −κ o(cid:48) j∈offspring of j oo(cid:48) j∈offspring of oj ∆Voo(cid:48) j ,oj (C5) 20 Illustration of notations for the voltage differences, ∆Vj,mj , ∆Voj ,j, and ∆Vooj ,oj for a tree fragment. Therein FIG. 15. oj ∈ (g + 1) labels the node under consideration; the node with the index j ∈ g is the parent of oj. The node with the index oj ∈ (g + 1) is the offspring of j and is the parent of the node ooj ∈ (g + 2). Red semicircles show leaves at which branching is terminated. 1. Generation-averaged dynamics In the strong coupling limit, the nodal dynamics is synchronized and nodes within the same generation become statistically indistinguishable. To make use of this fact, we follow the approach presented by Kouvaris et al. [43] and extend it to stochastic excitable elements on random tree networks. To this end, we consider the dynamics of the generation-averaged membrane potentials Vj, g = 0, 1, 2, . . . ,G. (C6) (cid:88) (cid:104)V (cid:105)g := 1 Dg j ∈ gen.g Here the average is taken in each generation g < G of the kth random tree network realization (the realization index k is dropped), i.e. only generation that actually include nodes are considered. Applying Eq. (C6), to Eq. (C1), we obtain the generation-averaged membrane potential dynamics C d dt (cid:104)V (cid:105)g = −(cid:104)Iion(cid:105)g + (cid:104)J (t)(cid:105)g − κ(cid:104)V (cid:105)g + κ 1 Dg Vmj + κ 1 Dg j∈gen.g j∈gen.g oj∈offspring of j (cid:88) (cid:88) (cid:0)Voj − Vj (cid:1) . (C7) Note that averaging over the zeroth generation yields (cid:104)V (cid:105)0 = V1. As the first sum on the right-hand side of Eq. (C7) runs over the Dg offspring which share Dg−1 parent nodes, we find 1 Dg j∈gen. g Vmj = Dg−1 Dg 1 Dg−1 i∈gen. g−1 (cid:104)d V (cid:105)g−1 . (C8) Accordingly, the double sum in (C7) can be simplified by noting that the jth node in generation g has dg,j offspring in generation g + 1, (cid:88) (cid:88) j∈gen. g oj∈offspring of j j∈gen. g dg,jVj = (cid:104)d V (cid:105)g . (C9) The other double sum can be reduced to a single sum over the nodes in the (g + 1)th generation (cid:88) 1 Dg dg−1,i Vi = Dg−1 Dg (cid:88) (cid:88) Voj . (cid:88) (cid:88) (cid:88) (cid:88) 1 Dg Vj = 1 Dg Voj = 1 Dg j∈g oj∈gen. g+1 oj∈gen. g+1 This can be further simplified by considering the generation average membrane potential, (cid:88) 1 Dg oj∈g+1 Voj = Dg+1 Dg 1 Dg+1 Voj = Dg+1 Dg (cid:104)V (cid:105)g+1 . (cid:88) oj∈g+1 21 (C10) From the recurrent relation of the Galton-Watson process (5) it follows that the ratios of numbers of nodes in adjacent generations can be replaced by the mean branching within a generation as Dg+1/Dg = (cid:104)d(cid:105)g. Then using Eqs.(C8 -- C10) in Eq. (C7) yields (cid:32) (cid:33) (cid:16) (cid:17) C d dt (cid:104)V (cid:105)g = −(cid:104)Iion(cid:105)g + (cid:104)J (t)(cid:105)g − κ (cid:104)V (cid:105)g − 1 (cid:104)d(cid:105)g−1 (cid:104)d V (cid:105)g−1 + κ (cid:104)d(cid:105)g (cid:104)V (cid:105)g+1 − (cid:104)d V (cid:105)g . (C11) In the limit of strong coupling, we can assume that Vi ≈ (cid:104)V (cid:105)g,i for i in generation gi. Then, we can decouple the branching from the generation average, i.e. (cid:104)d V (cid:105)g ≈ (cid:104)d(cid:105)g (cid:104)V (cid:105)g. Performing similar simplification for the root node, g = 0 and the nodes in the last generation g = G, we end up with the following system for the generation-averaged membrane potentials d dt (cid:104)V (cid:105)0 = −(cid:104)Iion(cid:105)0 + (cid:104)J (t)(cid:105)0 + κ d0,1 ((cid:104)V (cid:105)1 − (cid:104)V (cid:105)0) , d dt (cid:104)V (cid:105)g = −(cid:104)Iion(cid:105)g + (cid:104)J (t)(cid:105)g − κ d dt (cid:104)V (cid:105)G (cid:104)V (cid:105)g − (cid:104)V (cid:105)g−1 (cid:104)V (cid:105)G−1 − (cid:104)V (cid:105)G (cid:16) + κ(cid:0) = −(cid:104)Iion(cid:105)G + (cid:104)J (t)(cid:105)G (cid:17) (cid:1) . C C C + κ(cid:104)d(cid:105)g (cid:16) (cid:104)V (cid:105)g+1 − (cid:104)V (cid:105)g (cid:17) , g = 1, . . . ,G − 1, (C12) 2. Dynamics of generation-averaged membrane potential differences Next we introduce the differences between generation-averaged membrane potentials from adjacent generations, ∆(cid:104)V (cid:105)g := (cid:104)V (cid:105)g+1 − (cid:104)V (cid:105)g, g = 0, 1, . . . ,G − 1. Subtraction of the corresponding equations in (C12) yields C d dt (cid:104)∆V (cid:105)g = −∆(cid:104)Iion(cid:105)g + ∆(cid:104)J (t)(cid:105)g + −κ ((cid:104)d(cid:105)0 + 1)∆(cid:104)V (cid:105)0 + κ(cid:104)d(cid:105)1 ∆(cid:104)V (cid:105)2 , (cid:16) (cid:17) g = 0,  + κ∆(cid:104)V (cid:105)g−1 − κ (cid:104)d(cid:105)g + 1 ∆(cid:104)V (cid:105)g + κ(cid:104)d(cid:105)g+1 ∆(cid:104)V (cid:105)g+1 , 1 ≤ g < G − 1, κ ∆(cid:104)V (cid:105)G−2 − (cid:104)d(cid:105)G−1 ∆(cid:104)V (cid:105)G−1 , g = G − 1, (C13) Here we introduced the differences between generation-averaged ionic currents ∆(cid:104)Iion(cid:105)g := (cid:104)Iion(cid:105)g+1 − (cid:104)Iion(cid:105)g and input currents ∆(cid:104)J (t)(cid:105)g = (cid:104)J (t)(cid:105)g+1 − (cid:104)J (t)(cid:105)g and set d0,1 = (cid:104)d(cid:105)0. The differences of generation-averaged input currents ∆(cid:104)J (t)(cid:105)g can be separated in a deterministic part ∆(cid:104)J(cid:105)g := (cid:104)J(cid:105)g+1 − (cid:104)J(cid:105)g and a stochastic one ∆ξg(t) = (cid:104)ξ(t)(cid:105)g+1 − (cid:104)ξ(t)(cid:105)g. Next, we consider the difference of the generation-averaged ionic currents ∆(cid:104)Iion(cid:105)g. Both, the difference be- tween membrane potentials of individual nodes and corresponding generation averages, and the difference between generation-averaged membrane potentials of adjacent generations ∆(cid:104)V (cid:105)g become small in the case of strong coupling. We therefore approximate the differences between generation-averaged ionic currents to the first order in ∆(cid:104)V (cid:105)g around a vanishing mean difference. We assume that it can be expanded in a Taylor expansion around ∆(cid:104)V (cid:105)g = 0. This yields (cid:104)Iion(cid:105)g+1 − (cid:104)Iion(cid:105)g ≈ ag + bg∆(cid:104)V (cid:105)g + h.o.. As we restrict on networks of identical nodes, except for leaf nodes, we assume that the coefficient ag vanishes and that the coefficients bg are small compared to the coupling 22 strength κ, i.e. bg (cid:28) κ. Using these assumptions, Eq. (C13) can be linearized and we find C d dt (cid:104)∆V (cid:105)g,k ≈ ∆(cid:104)J (t)(cid:105)g +  + (cid:17) −κ ((cid:104)d(cid:105)0 + 1) ∆(cid:104)V (cid:105)0 + κ(cid:104)d(cid:105)1 ∆(cid:104)V (cid:105)1 , (cid:16) (cid:104)d(cid:105)G−1 + 1(cid:1) ∆(cid:104)V (cid:105)G−1 , κ∆(cid:104)V (cid:105)G−2 − κ(cid:0) κ∆(cid:104)V (cid:105)g−1 − κ (cid:104)d(cid:105)g + 1 g = 0, ∆(cid:104)V (cid:105)g + κ(cid:104)d(cid:105)g+1 ∆(cid:104)V (cid:105)g+1 , 1 ≤ g < G − 1, g = G − 1, (C14) In consequence, the dynamics of the differences of the generation-averaged membrane potentials can be approximated by a multidimensional Ornstein-Uhlenbeck process, Here we introduced the G-dimensional vectors, C d dt ∆(cid:104)V(cid:105) ≈ B∆(cid:104)V(cid:105) + ∆(cid:104)J(cid:105) + ∆(cid:104)ξ(cid:105) (t). and the G × G tridiagonal matrix, ∆(cid:104)V(cid:105) = (∆(cid:104)V (cid:105)0 , ..., ∆(cid:104)V (cid:105)G−1)T , ∆(cid:104)J(cid:105) = (∆(cid:104)J0(cid:105) , ∆(cid:104)J1(cid:105) , ..., ∆(cid:104)JG−1(cid:105))T , ∆(cid:104)ξ(t)(cid:105) =(cid:0)∆(cid:104)ξ(t)(cid:105)0 , ∆(cid:104)ξ(t)(cid:105)1 , ..., ∆(cid:104)ξ(t)(cid:105)G−1 −κ ((cid:104)d(cid:105)0 + 1) ... ... −κ ((cid:104)d(cid:105)2 + 1) ... ... −κ ((cid:104)d(cid:105)1 + 1) κ(cid:104)d(cid:105)1 0 (cid:1)T , κ(cid:104)d(cid:105)2 ... 0 0 ... 0 κ −κ(cid:0) (cid:104)d(cid:105)G−1 + 1(cid:1) κ(cid:104)d(cid:105)G−1 B = κ 0 ... 0 κ ... .. (C15) (C16)  . In accordance to our notation for differences of generation-averaged quantities, we introduced the differences of generation-averaged constant and noisy current components, ∆(cid:104)J(cid:105)g = (cid:104)J(cid:105)g+1 − (cid:104)J(cid:105)g and ∆(cid:104)ξ(t)(cid:105)g = (cid:104)ξ(t)(cid:105)g+1 − (cid:104)ξ(t)(cid:105)g. In the strong coupling limit, temporal deviations of ∆(cid:104)V(cid:105) from its stationary value decay extremely fast. Hence, we can use an adiabatic approximation [44], to approximate ∆(cid:104)V(cid:105) by its stationary value plus a white Gaussian noise. Both, the stationary voltage difference and the intensity of the Gaussian white noise in the strong coupling limit can be obtained by setting the left-hand side of Eq. (C15) to zero. This yields ∆(cid:104)V(cid:105) ≈ −B−1 (∆(cid:104)J(cid:105) + ∆(cid:104)ξ(t)(cid:105)) , (C17) where B−1 is the inverse of the matrix B. 3. Single node description for strongly-coupled random tree networks In order to obtain an approximation for the dynamics of the root node, only the first component, ∆(cid:104)V (cid:105)0, of Eq. (C17) is need. Using the latter in Eq. (C12) yields Hereafter the index "1" denotes the first component of a G-dimensional vector. From Eq.(C18), we find the effective current Jeff and noise intensity Seff for the current realization of the tree network as (cid:0)B−1 (∆(cid:104)I(cid:105) + ∆(cid:104)ξ(t)(cid:105))(cid:1) C V1 = −Iion,1 + κ d0 (cid:112) (cid:0)B−1∆(cid:104)J(cid:105) (cid:1) 2Seff ξ(t) = κ d0 1 , (cid:0)B−1∆(cid:104)ξ(t)(cid:105) (cid:1) Jeff = κd0 1 . (C18) 1 . (C19) The later relation is obtained by noting that the sum of Gaussian white noises yields a Gaussian white noise with modified intensity. For a given tree realization the inverse of B can be calculated explicitly using the formula for the inverse matrix B−1 = 1 B adj(B). (C20) Here B and adj(B) refer to the determinant and adjugate of the matrix B, respectively. In the following, we present explicit formulas for the cases used in the main text, G = 3, 4. In case of G = 3, Bk is a 3 × 3 matrix. Its determinant reads 23 B = −κ3 ((cid:104)d(cid:105)0 (cid:104)d(cid:105)1 (cid:104)d(cid:105)2 + (cid:104)d(cid:105)0 (cid:104)d(cid:105)1 + (cid:104)d(cid:105)0 + 1) = −κ3N . Its adjugate matrix reads adj B = κ2 (cid:104)d(cid:105)1 (cid:104)d(cid:105)2 + (cid:104)d(cid:105)1 + 1 (cid:104)d(cid:105)1 (cid:104)d(cid:105)2 + (cid:104)d(cid:105)1 (cid:104)d(cid:105)1 (cid:104)d(cid:105)2 (cid:104)d(cid:105)2 + 1 (cid:104)d(cid:105)0 (cid:104)d(cid:105)2 + (cid:104)d(cid:105)0 + (cid:104)d(cid:105)2 + 1 (cid:104)d(cid:105)0 (cid:104)d(cid:105)2 + (cid:104)d(cid:105)2 1 (cid:104)d(cid:105)0 + 1 (cid:104)d(cid:105)0 (cid:104)d(cid:105)1 + (cid:104)d(cid:105)0 + 1 T . (C21) (C22) (C23) Using this, we can evaluate the effective parameters in Eq. (C19) and find Jeff = R∞J = H N J, Seff = S∞S = H N 2 S. Similarly, in the case of G = 4, B is a (4 × 4)-matrix with determinant B = −κ4N . Evaluation of the adjugate matrix, also yields Eq.(C23). We stress that derivations in this appendix are done for the particular tree realization. Thus, assigning the index k for tree realizations in (C23) to the total number of leaves and nodes gives the scaling relations Eq.(17,18) of the main text. 24 [1] M. Newman, Networks: An Introduction (Oxford university press, 2010). [2] A.-L. Barab´asi and M. P´osfai, Network Science (Cambridge university press, 2016). [3] S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D.-U. Hwang, Physics Reports 424, 175 (2006). [4] A. Arenas, A. D´ıaz-Guilera, J. Kurths, Y. Moreno, and C. Zhou, Physics Reports 469, 93 (2008). [5] I. Z. Kiss and J. L. Hudson, AIChE journal 49, 2234 (2003). [6] A. S. Mikhailov and G. Ertl, Engineering of Chemical Complexity (World Scientific, 2012). [7] I. Farkas, D. Helbing, and T. Vicsek, Nature 419, 131 (2002). [8] S. P. Borgatti, A. Mehra, D. J. Brass, and G. Labianca, Science 323, 892 (2009). [9] M. E. J. Newman, Physical Review E 66, 016128 (2002). [10] Y. Chen, J. K. Kim, A. J. Hirning, K. Josi´c, and M. R. Bennett, Science 349, 986 (2015). [11] M. London and M. Hausser, Annual Review of Neuroscience 28, 503 (2005). [12] O. Kinouchi and M. Copelli, Nature Physics 2, 348 (2006). [13] L. L. Gollo, O. Kinouchi, and M. Copelli, Scientific Reports 3 (2013). [14] E. Bullmore and O. Sporns, Nature Reviews Neuroscience 10, 186 (2009). [15] G. K. Ocker, Y. Hu, M. A. Buice, B. Doiron, K. Josi´c, R. Rosenbaum, and E. Shea-Brown, Current Opinion in Neurobiology 46, 109 (2017). [16] B. Lindner, J. Garcıa-Ojalvo, A. Neiman, and L. Schimansky-Geier, Physics Reports 392, 321 (2004). [17] F. Sagu´es, J. M. Sancho, and J. Garc´ıa-Ojalvo, Reviews of Modern Physics 79, 829 (2007). [18] G. K. Ocker, K. Josi´c, E. Shea-Brown, and M. A. Buice, PLoS Computational Biology 13, e1005583 (2017). [19] D. R. Chialvo, Nature Physics 6, 744 (2010). [20] D. B. Larremore, W. L. Shew, and J. G. Restrepo, Physical Review Letters 106, 058101 (2011). [21] M. A. Buice, J. D. Cowan, and C. C. Chow, Neural Computation 22, 377 (2010). [22] R. W. Banks, M. Hulliger, K. Scheepstra, and E. Otten, The Journal of Physiology 498, 177 (1997). [23] R. W. Banks, D. Barker, and M. Stacey, Philosophical Transactions of the Royal Society of London B: Biological Sciences 299, 329 (1982). [24] D. R. Lesniak, K. L. Marshall, S. A. Wellnitz, B. A. Jenkins, Y. Baba, M. N. Rasband, G. J. Gerling, and E. A. Lumpkin, Elife 3, e01488 (2014). [25] K. L. Marshall, R. C. Clary, Y. Baba, R. L. Orlowsky, G. J. Gerling, and E. A. Lumpkin, Cell Reports 17, 1719 (2016). [26] J. Besson, The Lancet 353, 1610 (1999). [27] J. Yu and J. Zhang, Neuroscience Letters 362, 171 (2004). [28] L.-Y. Lee and J. Yu, Comprehensive Physiology 4, 287 (2014). [29] D. Rogers, L. Neiman, and D. F. Russell, Bulletin of the American Physical Society 58 (2013). [30] D. Quick, W. Kennedy, and R. Poppele, Neuroscience 5, 109 (1980). [31] G. S. Bewick and R. W. Banks, Pflugers Archiv-European Journal of Physiology 467, 175 (2015). [32] J. P. Eagles and R. L. Purple, Brain Research 77, 187 (1974). [33] R. Carr, J. Gregory, and U. Proske, Brain Research 800, 97 (1998). [34] J. Kroller, O.-J. Grusser, and L.-R. Weiss, Biological Cybernetics 63, 91 (1990). [35] J. A. Kromer, L. Schimansky-Geier, and A. B. Neiman, Physical Review E 93, 042406 (2016). [36] J. Kromer, A. Khaledi-Nasab, L. Schimansky-Geier, and A. B. Neiman, Scientific Reports 7 (2017). [37] G. B. Ermentrout and D. H. Terman, Mathematical Foundations of Neuroscience (Springer Science & Business Media, 2010). [38] C. C. McIntyre, A. G. Richardson, and W. M. Grill, Journal of Neurophysiology 87, 995 (2002). [39] M. Drmota, Random Trees: An Interplay Between Combinatorics and Probability (Springer Science & Business Media, 2009). [40] T. E. Harris, The Theory of Branching Processes (Courier Corporation, 2002). [41] C. M. Walsh, D. M. Bautista, and E. A. Lumpkin, Current Opinion in Neurobiology 34, 133 (2015). [42] S. A. Wellnitz, D. R. Lesniak, G. J. Gerling, and E. A. Lumpkin, Journal of Neurophysiology 103, 3378 (2010). [43] N. E. Kouvaris, T. Isele, A. S. Mikhailov, and E. Scholl, EPL (Europhysics Letters) 106, 68001 (2014). [44] N. G. Van Kampen, Physics Reports 124, 69 (1985).
1307.5563	1	1307	2013-07-21T20:02:40	Social interactions dominate speed control in driving natural flocks toward criticality	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.PE" ]	Flocks of birds exhibit a remarkable degree of coordination and collective response. It is not just that thousands of individuals fly, on average, in the same direction and at the same speed, but that even the fluctuations around the mean velocity are correlated over long distances. Quantitative measurements on flocks of starlings, in particular, show that these fluctuations are scale-free, with effective correlation lengths proportional to the linear size of the flock. Here we construct models for the joint distribution of velocities in the flock that reproduce the observed local correlations between individuals and their neighbors, as well as the variance of flight speeds across individuals, but otherwise have as little structure as possible. These minimally structured, or maximum entropy models provide quantitative, parameter-free predictions for the spread of correlations throughout the flock, and these are in excellent agreement with the data. These models are mathematically equivalent to statistical physics models for ordering in magnets, and the correct prediction of scale-free correlations arises because the parameters - completely determined by the data - are in the critical regime. In biological terms, criticality allows the flock to achieve maximal correlation across long distances with limited speed fluctuations.	physics.bio-ph	physics	Social interactions dominate speed control in driving natural flocks toward criticality William Bialeka, Andrea Cavagnab, Irene Giardinab, Thierry Morac, Oliver Pohlb,∗ Edmondo Silvestrib, Massimiliano Vialeb, and Aleksandra Walczakd aJoseph Henry Laboratories of Physics and Lewis -- Sigler Institute for Integrative Genomics, Princeton University, Princeton, New Jersey 08544 USA bIstituto Sistemi Complessi (ISC -- CNR), Via dei Taurini 19, 00185 Roma, Italy bDipartimento di Fisica, "Sapienza" Universit´a di Roma, P.le Aldo Moro 2, 00185 Roma, Italy cLaboratoire de Physique Statistique de l' ´Ecole Normale Sup´erieure, CNRS and Universites Paris VI and Paris VII, 24 rue Lhomond, 75231 Paris Cedex 05, France, and dLaboratoire de Physique Th´eorique de l' ´Ecole Normale Sup´erieure, CNRS and University Paris VI, 24 rue Lhomond, 75231 Paris Cedex 05, France (Dated: July 30, 2018) Flocks of birds exhibit a remarkable degree of coordination and collective response. It is not just that thousands of individuals fly, on average, in the same direction and at the same speed, but that even the fluctuations around the mean velocity are correlated over long distances. Quantitative measurements on flocks of starlings, in particular, show that these fluctuations are scale -- free, with effective correlation lengths proportional to the linear size of the flock. Here we construct models for the joint distribution of velocities in the flock that reproduce the observed local correlations between individuals and their neighbors, as well as the variance of flight speeds across individuals, but otherwise have as little structure as possible. These minimally structured, or maximum entropy models provide quantitative, parameter -- free predictions for the spread of correlations throughout the flock, and these are in excellent agreement with the data. These models are mathematically equivalent to statistical physics models for ordering in magnets, and the correct prediction of scale -- free correlations arises because the parameters -- completely determined by the data -- are in the critical regime. In biological terms, criticality allows the flock to achieve maximal correlation across long distances with limited speed fluctuations. I. INTRODUCTION In a flock of birds, thousands of individuals will fly in the same direction and at the same speed, for long pe- riods of time. But this average behavior is not enough for flocking to be advantageous. The entire flock must re- spond to dangers that may be visible only to a small frac- tion of individuals, requiring information to propagate over long distances. Although it is difficult to measure this information flow directly [1], we know that attacks by predators on a flock have very low success rates [2 -- 4], and that the evasion of predators by starling flocks is associated with the triggering and propagation of waves through the flock [5]. Even in the absence of predators, we can see deviations of individual behavior from the av- erage behavior of the flock, and correlations in these fluc- tuations provide a signature of information flow through the flock. Strikingly, observations on flocks of starlings show that these correlations extend over very long dis- tances, comparable to the size of the flock itself [6]. It is generally believed that the interactions among birds in a flock are local -- each bird aligns its flight di- rection and speed to those of its near neighbors [7]. If this is correct, then we have to understand how local ∗ Present address: Institut fur Theoretische Physik, Technis- che Universitat Berlin, Hardenbergstrasse 36, D -- 10623 Berlin -- Charlottenburg, Germany. interactions can generate correlations over much longer distances. In physics, we have two very different mecha- nisms for local interactions to produce correlations that are essentially scale -- free, extending over distances com- parable to the size of the system as a whole. If the system spontaneously breaks a continuous symmetry, for exam- ple when all the spins in a magnet select a particular direction in space along which the macroscopic magne- tization will point, then the fluctuations in the system are dominated by "Goldstone modes" that do not de- cay on any fixed length scale [8]. If we can think of the alignment of flight directions in a flock as being like the alignment of spins in a magnet [9 -- 11], then we can under- stand the emergence of scale -- free correlations by analogy with Goldstone's theorem. We have shown that this is more than a metaphor [13]: the minimally structured model consistent with the observed correlations among flight directions of neighboring birds is exactly equiva- lent to a model of spins in a magnet, and the resulting (parameter -- free) prediction of long ranged correlations among fluctuations in flight direction agrees quantita- tively with the data. Not just the fluctuations in flight direction, but also the fluctuations in flight speed are correlated over long dis- tances [6]. Now there is no analogy to Goldstone modes, because choosing a speed does not correspond to break- ing any plausible symmetry of the system. But there is a second mechanism by which physical systems generate scale -- free correlations, and this is by tuning parameters to a critical point [8, 12]. As we explore the parame- 3 1 0 2 l u J 1 2 ] h p - o i b . s c i s y h p [ 1 v 3 6 5 5 . 7 0 3 1 : v i X r a ter space of a system (e.g., changing temperature and pressure), we encounter phase transitions, where small changes in parameters produce qualitative changes in be- havior of a macroscopic sample (e.g., between liquid and gas). Along the lines in parameter space where these phase transitions exist, there are special points, called critical points, where the dependence on parameters be- comes, for very large systems, singular but not discontin- uous. At these points, fluctuations (e.g., in the density of the liquid) become correlated on all length scales, from the molecular scale of the interactions to the macroscopic scale of the sample as a whole. Tuning to a critical point provides a potential expla- nation for the observed scale -- free correlations in speed of flocking birds, but this is just an analogy; the goal of this paper is to see if we can construct a quantitative theory. Our strategy follows that in Ref [13]: we will construct the least structured models that are consistent with mea- sured correlations among neighboring birds, and then see if these models can predict correctly the persistence of correlations over much longer distances, comparable to the size of the flock. We will see that this does work, and that the underlying mechanism really is the tuning of the system to a critical point. From a biological point of view, this tuning means that individuals in a flock combine individual speed control and social interactions with their neighbors so as to achieve a maximal range of influence while keeping speed variability low. II. BUILDING A MODEL FROM DATA We consider flocks of European starlings, Sturnus vul- garis, in the field. The work of Refs [14, 15] provides a detailed description of these flocks, resulting in the as- signment of three -- dimensional positions and velocities, at each moment in time, to each individual bird in flocks with up to several thousand members; for a summary see Appendix A. From these raw data, one can extract a va- riety of features that serve to characterize the nature of the ordering in the flock [6, 16], including the scale -- free correlations noted above. The positions and velocities of all the birds in the flock are stochastic -- with elements of randomness, but corre- lated. In making a model, we want to be able to predict the probability distribution out of which these random variables are drawn. One approach is to consider a de- tailed model for the dynamics of the flock, typically with many parameters to describe the interactions that cause the flock to cohere and align. In this approach, the con- nection between the model dynamics and the joint distri- bution of velocities in the flock can be complicated, and fitting the parameters of the interactions is difficult. Al- ternatively, we can take some set of observations on the flock as given and try to construct models that reproduce these observations exactly; among the (generally infinite) set of models that can do this, we want to choose the one that has the least structure. Minimizing structure 2 means that the velocities we choose out of the distribu- tion are as random as they can be while still matching the properties of the flock that we have chosen as essen- tial. As emphasized by Jaynes [17, 18], these minimally structured distributions have maximum entropy, and this provides a connection to the ideas of statistical physics, even though we are describing a system that is not in thermal equilibrium (see Appendix B). The maximum entropy approach to model building is far from new, but there has been a resurgence of interest in the use of these ideas to describe biological systems [19 -- 28]. In Ref [13], we took a first step toward a max- imum entropy description of flocks, building models for the distribution of flight directions that match the aver- age local correlation between the direction of a bird and its nearest neighbors. Surprisingly, fixing this one num- ber leads to a model that, with no free parameters, pro- vides an essentially complete, quantitative description of the propagation of directional order throughout the en- tire flock. Here we want to generalize this approach to consider not just flight directions, but also speed. As explained above, we expect that accounting for the ob- served properties of speed ordering is a qualitatively dif- ferent problem from the case of directional ordering. Given the positions of the birds in space, the state of the flock is defined by the velocity (cid:126)vi of each bird. This three dimensional vector is composed of the speed, vi ≡ (cid:126)vi, and a unit vector, (cid:126)si = (cid:126)vi/vi, that points in the direction of flight. Our intuition is that the most important interactions are local, between a bird and its immediate neighbors. If this is correct, then the essential features of the system should be captured by measuring local correlations, as in Ref [13]. We can quantify local correlations in the flock by ask- ing how similar, on average, the velocity of each bird is to its neighbors. To do this, we define N(cid:88) i=1 1 nc (cid:88) j∈Ni Qint = 1 2v2 0N (cid:126)vi − (cid:126)vj2. (1) Here Ni is the relevant neighborhood of bird i, which we take to be its first nc nearest neighbors [13, 16]. We compare a bird to each of its neighbors, average over the neighborhood, and then average over all N birds in the flock; we normalize the result by a typical speed v0 so that we have a dimensionless measure of correlation or similarity. If we take v0 to be the average speed of birds in the flock, then typical values for Qint are ∼ 10−2 (Table I in Appendix A), showing that individual birds indeed fly with velocities that are very similar to those of their neighbors. The definition of Qint quantifies the similarity of each bird's flight vector to that of its neighbors, but it can't completely specify the structure of the flock. If we add a constant to all the velocities, so that the flock flies (cid:80)N faster or slower, then Qint is unchanged. We would like to match the average speed of the birds in the flock, V = i=1 vi, to its observed value (cid:104)V (cid:105)exp. In addition, we 1 N 3 (a) (b) (c) FIG. 1. Inference of the three interaction parameters g,J and nc. (a) For fixed values of J and nc, the value of the speed control parameter g is found by equating the theoretical prediction for the variance of fractional speed fluctuations, σ2 (red line), to its experimental value (black horizontal line). (b) Once the value of g is determined for all possible values of J and nc, the interaction strength J can be set by equating the theoretical prediction for Qint (red line) to its experimental value (black line). (c) Once g and J are computed for given values of nc, the log -- likelihood of the data, (cid:104)ln P ({(cid:126)vi})(cid:105)exp becomes a function of nc only, and the interaction range nc can be evaluated by maximizing this function. All panels refer to the same snapshot (image 2) of flock 25-10, and mathematical details can be found in Appendix D. know that individual birds have speeds that vary around the mean, so we would also like to match the variance of speeds. This is equivalent to fixing not just the mean speed, but the mean square speed, V2 = 1 N (cid:80)N i=1 v2 i . The maximum entropy distribution consistent with measured values of Qint, V , and V2 has the form (see Appendix B), P ({(cid:126)vi}) = 1 Z exp − J 4v2 0 N(cid:88) N(cid:88) ij=1 i=1 + µ v0 nij(cid:126)vi − (cid:126)vj2 vi − g 2v2 0 (cid:35) v2 i , (2) N(cid:88) i=1 where Z is a constant that ensures the normalization of the probability distribution, and we have inserted factors of v0 so that other parameters are dimensionless. The matrix nij maps the connections between birds: nij = 1 if bird j is in the neighborhood of bird i (j ∈ Ni), and zero otherwise; we symmetrize to give nij = (nij + nji)/2. The parameters J, µ, and g must be adjusted so that the average values of Qint, V , and V2 computed from the probability distribution match those observed for the flock; as explained in Appendix D, these computations can be done analytically. The only remaining parameter is the number of relevant neighbors nc, which we fix by requiring that the probability of the observed velocities be as large as possible (maximum likelihood). Figure 1 shows one example of our solution to the "in- verse problem" of determining the parameters J, g, and nc. Importantly, the quantities that we are trying to match are averages over all the birds in the flock, and so they are determined with small errors even from a single snapshot of the velocities. The parameters in turn are determined very precisely, and are consistent for a single flock across time, as in Ref [13]. III. SOME INTUITION N(cid:88) i=1 N(cid:88) ij=1 Maximum entropy distributions are mathematically equivalent to the Boltzmann distribution for systems in thermal equilibrium, and we can use this identity to gain some intuition for the predictions of the model. We recall that a system described by the Boltzmann dis- tribution will occupy a state s with probability ps ∝ exp(−Es/kBT ), where Es is the energy of the state and kBT is the typical thermal energy; for our purposes we can choose units so that kBT = 1. Thus Eq (2) defines an energy function or Hamiltonian on the space of the birds' velocities, and this can be written as H({(cid:126)vi}) = J 4V 2 nij(cid:126)vi − (cid:126)vj2 + g 2V 2 (vi − V )2 , (3) where we have eliminated the parameter µ in favor of the mean speed V , which is now fixed to its experimental value (cid:104)V (cid:105)exp, and we have set the arbitrary scale v0 = V . The first term in this Hamiltonian describes the ten- dency of the individual velocities to adjust both direction and modulus to their neighbors, while the second term forces the speed to have, on average, the value V . From this perspective, we can interpret J as the stiffness of an effective "spring" that ties each bird's velocity to that of its neighbors, and g as the stiffness of a competing spring that ties each speed to the desired mean. Larger J means a tighter connection to the neighbors, and larger g means a tighter individual control over speed. There are interesting limiting cases that give us a sense for what this model predicts. If the parameter g is very large, then the speed of individual birds hardly fluctuates at all. In this limit, we can rewrite the Hamiltonian as H({(cid:126)vi}) ≈ Hdir({(cid:126)si}) = − J 2 nij(cid:126)si · (cid:126)sj, (4) N(cid:88) ij=1 0.010.1110g0.00480.00490.0050s2s2 Datas2 Model020406080J00.0050.010.0150.02QintQint ModelQint Data1020304050607080 nc4375440044254450447545004525Log-Likelihood Optimal Valuewhere (cid:126)si is the unit vector pointing in the direction that bird i is flying. This Hamiltonian describes the tendency of the directions of individual birds to align with their neighbors, and is exactly the model in Ref [13]. If there are nonzero but small fluctuations in speed, then we can write vi = V (1 + i), and expand in powers of . The result (Appendix D) is that The value of g is set by requiring that our model match the observed variance in speed across the birds in the flock, as in Fig 1a. But J and g also compete with one another to determine the distance over which speed fluc- tuations will be correlated, Eq (10). Importantly, we are not free to adjust this correlation length by some fitting procedure: either the model gets it right, or it doesn't. H({(cid:126)vi}) ≈ Hdir({(cid:126)si}) + Hsp({i}), (5) 4 where the "speed Hamiltonian" Hsp({i}) = V 2 2v2 0 N(cid:88) i,j=1 (gδij + JNij) ij, (6) where the matrix Nij has the form Nij = −nij + δij nik. (7) N(cid:88) k=1 Thus our full model breaks into two pieces, one of which describes fluctuations in flight direction, and one of which describes the fluctuations in speed. Importantly, the strength of the "springs" that tie the speed of each bird to that of its neighbors is determined by the same parameter J which enters the description of directional fluctuations in Eq (4). Thus, we have a unified model for how birds adjust their vector velocities to those of their neighbors, rather than separate models (with separate parameters) for the adjustment of direction and speed. To get a sense for the structure of Hsp it is useful to imagine a continuum limit, in which the variations in speed from bird to bird are so smooth that we can pic- ture the speed fluctuations as a continuous function of position in the flock, ((cid:126)x). In this limit (Appendix E), we have (cid:90) d3x(cid:2)Jncr2 c (∇)2 + g2((cid:126)x)(cid:3) , Hsp ≈ ρ 2 where rc is the typical distance to a neighboring bird, and ρ is the density of the flock. This model predicts that the fluctuations will behave as (cid:104)((cid:126)x)((cid:126)x(cid:48))(cid:105) ∝ exp (−(cid:126)x − (cid:126)x(cid:48)/ξbulk) , where correlation length ξbulk ∼ rc (cid:115) Jnc g IV. SCALE -- FREE CORRELATIONS Once the parameters J, g and nc are determined (Fig 1), Eq (2) provides a model for the joint distribution of velocities for all the birds in the flock; everything that we compute from this distribution is a parameter -- free prediction. We start by measuring the similarity of the vector velocities among birds that are not just nearest neighbors, but are separated by greater distances. By analogy with Eq (1), we can define Q(r) = 1 V 2(cid:104)(cid:126)vi − (cid:126)vj2(cid:105)rij=r, (11) where the average is over all pairs of birds in the flock separated by a distance rij = r. We see in Fig 2a that the predicted Q(r) matches the data very closely, out to distances comparable to the overall size of the flock, more than ten times farther than the nearest neighbors. We next decompose the relationships among velocities into contributions from direction and speed. If we take all the units vectors (cid:126)si and average, we obtain the overall polarization of the flock, N(cid:88) i=1 (cid:126)P = 1 N (cid:126)si, (12) and we can characterize the fluctuations around this over- all direction by a correlation function (cid:126)sj − (cid:126)P (cid:17) ·(cid:16) (cid:126)si − (cid:126)P (cid:17)(cid:29) (cid:28)(cid:16) Cdir(r) = . (13) In Fig 2b we compare the data with the predictions of the model, and again find that the agreement is very good, on all scales. rij=r By analogy with Eq (13), we can define correlations among the fluctuations in speed, Csp(r) = (vi − V ) · (vj − V ) . (14) rij=r (cid:28) (cid:29) (8) (9) (10) determines the distance over which the fluctuations in speed will be correlated; the subscript reminds us that we are treating the flock as a bulk material, with no bound- aries. In this simple picture, there is a critical point at g = 0 where the correlation length ξbulk becomes infinite. Thus the parameter J determines the propagation of directional order through the flock, and to describe the speed fluctuations we have only one extra parameter g. Fig 2c shows that the observed correlations are in agree- ment with the predictions of the model, again over the full range of distances. Thus, we have succeeded in con- structing a model based on local interactions that gener- ates correlations in speed fluctuations over long distances, matching the data quantitatively. The discussion in Section III suggests that long ranged correlations are associated with the approach to a criti- cal point at g = 0. To see if this intuition is correct, we 5 (a) (b) (c) FIG. 2. Correlation functions predicted by the maximum entropy model (red circles) vs. experiments (blue diamonds). (a) Similarity of velocities as a function of distance, defined in Eq (11). Dashed line indicates the size of the neighborhood defined by nc birds, within which we match the average Q exactly, by construction. (b) Correlations between fluctuations in flight direction as a function of distance, defined in Eq (13). (c) Correlations between fluctuations in speed as a function of distance, defined in Eq (14). All panels refer to the same flock and snapshot as in Fig 1; theoretical predictions from Appendix D 2. show in Fig 3a what happens to the predicted Csp(r) as we change the value of g. Large values of g correspond to small variances in speed, and to correlation functions that decay very rapidly with distance. As g becomes smaller, both the speed variance and the correlation length in- crease, until, for sufficiently small g, there really isn't a characteristic scale to the decay of the correlations, and Csp(r) is almost a straight line. This is the shape of the correlation function we observe, and the success of the theory is that the value of g that matches the observed speed variance is in this regime. We can quantify the approach to criticality by the di- mensionless ratio g/(Jnc) that enters Eq (10). From Fig 1, we see that g/(Jnc) ∼ 10−3, and this is typical. This suggests that real flocks are very close to criticality, and that this is why we observe scale -- free speed correlations. Note that g cannot be exactly zero, otherwise the vari- ance in speed would be infinite; a non -- zero (even if small) value of g is necessary to fix the flock's speed. To be more precise about the approach to criticality, we need to take into account the finite size of the flocks. Equations (9) and (10) hold only for an infinite system; for a finite system, the range of the correlation cannot in- crease indefinitely, since it is limited by the system size. As g is lowered, the behavior of the correlations is in- fluenced more and more by these finite size effects: the exponential decay in Eq (9) is modified, and the typi- cal distance over which correlations extend is no longer described by ξbulk. A more faithful estimate of the corre- lation length ξ is given instead by the zero of the correla- tion function [6], and the theoretical prediction depends in a non -- trivial way on g and the system size L. For small enough values of g, however, the system is effec- tively critical and scale -- free; we should see ξ ∝ L. In Fig 3a we show that decreasing g below the level required to match the speed variance of the real flock has essentially no effect, and curves with all smaller values of g "pile up" as shown in yellow. Repeating the analysis on flocks of different sizes (Fig 3b), we see that the correlation length does scale with size, and this pattern is captured perfectly by our maximum entropy models. We conclude that flocks do in fact exhibit critical be- havior, being close enough to the critical point to achieve maximum speed correlation length while still maintain- ing a well defined cruising speed and limited speed fluc- tuations. These conclusions are rather robust, and also (a) (b) FIG. 3. (a) Correlation function of the speed fluctuations, for different values of the control parameter g (increasing in the direction of the arrow). (b) Correlation length, defined as the point where the correlation function crosses zero [6], in flocks of different sizes, for the experimental data (blue diamonds) and for the model (red circles). 05101520253035r (m)0.000.050.100.150.20Q(r) DataModel05101520253035r (m)-0.03-0.02-0.0100.010.020.03Cdir(r)DataModel05101520253035r (m)-0.01-0.008-0.006-0.004-0.00200.0020.004Csp(r)DataModel0510152025r (m)-0.004-0.00200.0020.004Csp(r)gDataoptimal g020406080 L (m)010203040 x (m)DataModelhold when considering more general maximum entropy models where speed and flight directions are regulated by different interaction parameters (see Appendix F). V. DYNAMICAL MODEL The fact that maximum entropy models are equiva- lent to the Boltzmann distribution for a system in ther- mal equilibrium suggests a natural dynamical model, in which the various degrees of freedom in the system exe- cute Brownian motion on the energy landscape. We can describe such dynamics with a Langevin equation, γ d(cid:126)vi(t) dt = −∇iH({(cid:126)vj}) + (cid:126)ηi(t), (15) where ∇i indicates the derivatives with respect to the components of the velocity (cid:126)vi, γ is a constant to set the time scale of the dynamics, and the Langevin force (cid:126)ηi(t) is a random, white noise function of time. These dynamics are guaranteed, if we assume that the positions of the birds are fixed, to generate velocities that are drawn from the probability distribution in Eq (2). But to give a more realistic model we should add to Eq (15) forces that depend on the positions of the birds [29 -- 31], so as to fix the overall density of the flock (see Appendix G), and the velocities should drive the birds' positions, d(cid:126)xi dt = (cid:126)vi. (16) Equations (15) and (16) define a "self -- propelled particle" model of interacting birds, and is similar to the Vicsek model, so often used to describe flocking particles [32, 33]. In contrast to that model, and to most of flocking models in the literature, the speed of the individual particles is not fixed, but regulated by the control parameter g. Simulations of the dynamical model defined by Eqs (15, 16) are shown in Fig 4. As expected from the anal- ysis of the (static) maximum entropy model, the fluctu- ations in speed have a correlation length that grows as g is reduced. If g is not too small, we see correlations that decay exponentially [Eq (9)], and the correlation length varies with g/(Jnc) as expected. When g is lowered even further, the exponential decay is modified by finite size corrections, and the correlation length -- now computed as the zero -- crossing point of the correlation function -- keeps decreasing until a maximal, size dependent satu- ration value is reached. In this regime, the correlations extend over a distance determined by the system size, and ξ in fact grows linearly with L corresponding to scale -- free behavior (Fig 4b, inset). This scenario confirms that the mechanism identified in the previous section produces scale -- free correlations in the speed even when the full dy- namical behavior of the flock is taken into account. 6 (a) (b) FIG. 4. Simulations of a dynamical model (see Appendix for details). (a) Correlation function of the speed fluctuations at different values of g, in a flock of N = 16384 birds. Inset: Correlation length, measured from the exponential decay of the correlation functions at small r, as a function of g/(Jnc). (b) For smaller g, correlation lengths are measured from the zero crossing of the correlation function. For g/(Jnc) (cid:28) 1, ξ approaches a maximum value that depends on the size of the system. low-g maximum of ξ, as a function of the system size; the linear dependence of ξ on L is typical of scale -- free behavior. Inset: VI. CONCLUSIONS The understanding of collective behavior in matter at thermal equilibrium provide a touchstone for thinking about emergent phenomena in complex, biological sys- tems. Flocking seems like an especially attractive exam- ple, in which the alignment of birds in a flock reminds us of the alignment of spins in a magnet or molecules in a liquid crystal. But birds are vastly more complex than spins, and this might be nothing more than a metaphor. The goal of this paper and its companion [13] has been to show that we can go beyond metaphor, that there is a statistical mechanics description of flocks which makes quantitative, parameter -- free predictions that are in de- tailed agreement with the data. One dramatic collective phenomenon that can emerge in statistical mechanics is the existence of a critical point. At such points, distant elements of a system become cor- related with one another, far beyond the range of local 0510r (m)05e-061e-05Csp(r)g=0.05g=1.25e-2g=3.12e-3g=4.88e-50.010.1g1 x0.0010.010.1 g/(Jnc)110 x (m)N=1024, L=2N=4096, L=9N=16384, L=1551015L0246 xsatinteractions among the individual elements. At generic parameter values, correlations are expected to decay on some characteristic spatial scale ξ, so that a very large system is composed of many nearly independent pieces of volume ξ3; often, ξ is not much larger than the range of the interactions themselves. But at a critical point, the correlation length ξ becomes (formally) infinitely large, and the scale over which correlations extend becomes comparable to the linear size L of the entire system; rather than having many independent pieces, the system acts (almost) as one. The idea that biological systems might tune themselves to critical points is not new [34], but has languished for lack of detailed comparison with experiment. The emer- gence of new and more extensive data, as well as new ideas about how to connect theory and experiment, has led to a re -- examination of criticality in a wide variety of biological systems [35]. In this context, the observation of long ranged, or scale -- free correlations in the velocities of starlings in a flock [6] is very suggestive. Our results here show that these correlations are not just analogous to the correlations at a critical point: we have a very accurate description of the entire distribution of speed and direction fluctuations in the flock, this description is mathematically equivalent to a statistical mechanics model of a magnet, and the observed scale -- free correla- tions are predicted correctly because the parameters of this model are in the critical regime. Our approach is not a "fit" to the observed scale -- free correlations in the flock. Instead we take from the data a measurement of local correlations, and the variance of individual birds' speeds relative to the average over the flock, and build the least structured model that is con- sistent with these two measurements. Thus, rather than thinking of criticality as occurring in the neighborhood of a special point in the space of model parameters, we can think of it as a statement about the behavior of the flock itself. In particular, as emphasized in Fig 3, even a factor of two change in the variance of the speeds would predict correlations that decay much more rapidly with distance, inconsistent with what we see in real flocks. Biologically, birds may vary their speeds either for in- dividual reasons [36], or to follow their neighbors, par- alleling the competing forces captured in the model. In this language, the critical point is the place where social forces overwhelm individual preferences. More broadly, the critical regime is one in which is individuals achieve maximal coherence with their neighbors while still keep- ing some control over their speeds. Why do flocks organize themselves to be critical? His- torically, there has been much more speculation about the advantages of criticality for biological systems than there has been direct evidence, so we do not want to add too much here. We note, however, that in the statis- tical mechanics framework, the long ranged correlations at criticality are mathematically equivalent to the state- ment that information can propagate over similarly long distances. Away from criticality, a signal visible only to 7 one bird on the border of the flock can influence just a handful of near neighbors; at criticality, the same signal can spread to influence the behavior of the entire flock. Such susceptibility seems advantageous in terms of anti -- predatory strategies, but it would be attractive to have more direct measurements of the propagating signal [1]. The critical point is a place where many quantities are extremal; it remains to be seen which of these is most meaningful to the birds. ACKNOWLEDGMENTS We thank G Tkacik and G Parisi for many helpful discussions. Work in Princeton was supported in part by National Science Foundation Grants PHY -- 0957573 and CCF -- 0939370, and by the WM Keck Foundation; work in Rome was supported in part by grants IIT -- Seed Artswarm, ERC -- StG n. 257126, US -- AFOSR grant FA95501010250 (through the University of Maryland); work in Paris was supported by grant ERC -- StG n. 306312. Our collaboration was facilitated by the Initia- tive for the Theoretical Sciences at the Graduate Center of the City University of New York. Appendix A: Data The data that we analyze here were obtained from ob- servations on large flocks of starlings, Sturnus vulgaris, in the field. Using stereometric photography and inno- vative computer vision techniques [14, 15] the individual 3D coordinates and velocities were measured in cohesive groups of up to 4268 individuals [6, 16]. As summarized in Table I, we have data from 21 distinct flocking events, with sizes ranging from 122 to 4268 individuals and linear extensions from 9.1 to 85.7 m. Each event consists of up to 40 consecutive 3D configurations (individual positions and velocities), at time intervals of 1/10 s. All events correspond to strongly ordered flocks, with polarization [from Eq (12)] between (cid:126)P = 0.844 and (cid:126)P = 0.992. The border of each flock at each instant of time has been computed using the α -- shape algorithm [42], as explained in detail in [15]. Appendix B: The maximum entropy approach The concept of entropy has its roots in thermodynam- ics, roughly 150 years ago. The idea that we can use max- imum entropy as a strategy to construct simplified mod- els outside of equilibrium thermodynamics is now more than 50 years old [17]. Here, so that our discussion is self -- contained, we review this general strategy. See also Ref [13], and Appendix A.7 of Ref [37]. We assume that the state of the system can be de- scribed by a set of variables that we shall call v ≡ {(cid:126)v1, (cid:126)v2 ··· , (cid:126)vN}, by analogy with the velocities of birds Event N P 17 -- 06 552 0.935 21 -- 06 717 0.973 25 -- 08 1571 0.962 25 -- 10 1047 0.991 25 -- 11 1176 0.959 28 -- 10 1246 0.982 29 -- 03 440 0.963 31 -- 01 2126 0.844 32 -- 06 809 0.981 42 -- 03 431 0.979 49 -- 05 797 0.995 54 -- 08 4268 0.966 57 -- 03 3242 0.978 58 -- 06 442 0.984 58 -- 07 554 0.977 63 -- 05 890 0.978 69 -- 09 239 0.985 69 -- 10 1129 0.987 69 -- 19 803 0.975 72 -- 02 122 0.992 77 -- 07 186 0.978 (cid:104)V (cid:105)exp (m/s) L (m) Qint 9.96 12.06 12.47 12.57 10.07 11.22 10.75 8.13 9.99 10.68 14.02 19.17 14.38 10.13 10.81 10.24 11.97 12.04 14.16 13.24 9.50 51.8 1.29e-01 32.1 1.22e-02 59.8 2.63e-02 33.5 8.36e-03 43.3 6.27e-02 36.5 6.43e-03 37.1 1.43e-02 76.8 5.50e-02 22.2 1.52e-02 29.9 1.62e-02 19.2 6.49e-03 78.7 4.29e-02 85.7 1.53e-02 23.1 1.34e-02 19.1 1.35e-02 52.9 1.86e-02 17.1 2.68e-02 47.3 2.35e-02 26.4 3.65e-02 10.6 1.12e-02 9.1 4.27e-02 TABLE I. Summary of experimental data. Flocking events are labelled according to experimental session number and to the position within the session to which they belong. The number of birds N is the number of individuals for which we obtained a 3D reconstruction of positions in space. The polarization P is the global degree of alignment, as defined in the text. The linear size L of the flock is defined as the maximum distance between two birds belonging to the flock. The speed (cid:104)V (cid:105)exp is the average of the individual speeds over all the individuals in the flock, and Qint is as defined in Eq (1). All values are averaged over several snapshots during the flocking event. in a flock. Although we can measure, for example, the velocity of every bird in a flock, we typically can't col- lect enough data to make reliable estimates of very com- plicated quantitates. As an example, with N variables describing the state of the system, we need more than N independent measurements to be sure that the co- variance matrix of these variable is not artificially sin- gular. What does seem reasonable is to assume that there is a much smaller set of observables, {Oµ(v)} with µ = 1, 2, ··· , K, that we can extract from the sys- tem, and that we have enough data to make reliable statements about the average values of these obervables, {(cid:104)Oµ(v)(cid:105)exp}. Our task is to build a probability distribution P (v) such that we reproduce, exactly, the expectation values of the K observables, that is (cid:104)Oµ(v)(cid:105)P ≡(cid:88) P (v)Oµ(v) = (cid:104)Oµ(v)(cid:105)exp, (B1) 8 v for all µ = 1, 2, ··· , K; it is useful to phrase the nor- malization of the distribution as a similar constraint, the statement that the average of the "function" O0(v) = 1 must equal the "experimental" value of 1. The problem is that that there are infinitely many dis- tributions that can satisfy the constraints in Eq (B1). Out of all these distributions, we want to find the one that has as little structure as possible, so that we can derive the minimal consequences of the experimental ob- servations on {(cid:104)Oµ(v)(cid:105)exp}. Asking for a probability dis- tribution P (v) that has as little structure as possible is equivalent to asking that the variables v that we draw out of this distribution be as random as possible. Shan- non proved that the only measure of (lack of) structure or randomness that is consistent with several simple con- straints is the entropy of the distribution [38, 39], P (v) ln P (v). (B2) S [P ] = −(cid:88) v Thus, we are looking for the distribution P (v) that max- imizes the entropy in Eq (B2) while obeying the experi- mental constraints from Eq (B1). Such constrained op- timization problems can be solved using the method of the Lagrange multipliers [40]: we introduce a generalized entropy function, S [P ;{λν}] = S [P ] − K(cid:88) µ=0 λµ [(cid:104)Oµ(v)(cid:105)P − (cid:104)Oµ(v)(cid:105)exp] , (B3) where a multiplier λµ appears for each constraint to be satisfied, and then we maximize S both with respect to the probability distribution P (v) and with respect to the parameters {λµ}. Maximizing with S respect to P (v) gives P (v) = 1 Z({λν}) exp λµOµ(v) , (B4) where Z({λν}) = exp(−λ0 − 1). Since optimizing with respect to λ0 will enforce normalization of the distribu- tion, we can write, explicitly, (cid:34) − K(cid:88) µ=1 (cid:35) (cid:35) Z({λν}) = exp λµOµ(v) . (B5) Maximizing with respect to {λν} gives us the set of K simultaneous equations in Eq (B1), which we can now write more explicitly as (cid:88) (cid:34) − K(cid:88) (cid:35) Oµ(v) exp v ν=1 λνOν(v) . (B6) (cid:104)Oµ(v)(cid:105)exp = 1 Z({λν}) (cid:88) (cid:34) − K(cid:88) v µ=1 tion with an effective energy K(cid:88) µ=1 E(v) kBT = 9 λµOµ(v). (B8) We note that this energy is the sum of several terms, one for each of the observables whose expectation value we fix based on experimental data. It also is useful to note the connection of the maximum entropy approach to more conventional model building. If we take the form of the probability distribution in Eq (B4) as given, then our problem is only to "fit" the pa- rameters {λν}. A standard method is maximum likeli- hood. If we have Ns independent samples of the system's state, v(1), v(2), ··· , v(Ns), then the probability that the model generates these data is given by Pmodel(data) = P (v(i)). (B9) Ns(cid:89) i=1 We note that, in general, this is a very nonlinear set of equations for the parameters {λν}, and very hard to solve. In the next section we exploit special features of the flock problem -- in particular, the strong polarization of the flock -- to simplify this problem so that we can make analytic progress. Maximum entropy distributions are mathematically equivalent to the Boltzmann distribution in statistical physics. We recall that if a physical system in state v has energy E(v), then when it comes to equilibrium at temperature T the probability that is in any particular state is given by (cid:20) (cid:21) PBoltz(v) = 1 Z exp − E(v) kBT , (B7) where kB is Boltzmann's constant, and serves to convert between conventional units of temperature and energy. Comparing with Eq (B4), we see that the maximum en- tropy distribution is equivalent to a Boltzmann distribu- Pmodel(data) = 1 Z Ns ({λν}) Then we can form the normalized log probability, Substituting from Eq (B4) we can make this more ex- plicit, (cid:35) (cid:34) = i=1 µ=1 exp λµOµ(v(i)) − K(cid:88) Ns(cid:89) ln Pmodel(data) = − lnZ({λν}) − K(cid:88) = − lnZ({λν}) − K(cid:88) µ=1 1 Ns 1 Z Ns({λν}) (cid:34) λµ 1 Ns exp Ns(cid:88) i=1 λµ(cid:104)Oµ(v)(cid:105)exp, µ=1 (cid:34) − K(cid:88) µ=1 Ns(cid:88) i=1 λµ (cid:35) Oµ(v(i)) (cid:35) Oµ(v(i)) . (B10) (B11) (B12) where in the last step we recognize the normalized sum over samples as the experimental expectation value. Now if we want to maximize the probability, or likelihood, we should differentiate with respect to the parameters and set the result to zero: = 0 ⇒ ∂ lnZ({λν}) But with the explicit expression for Z in Eq (B5), we can compute: ∂ ln Pmodel(data) ∂λµ ∂λµ = −(cid:104)Oµ(v)(cid:105)exp. (cid:34) − K(cid:88) ν=1 (cid:35) λνOν(v) ∂ lnZ({λν}) ∂λµ = 1 Z({λν}) ∂Z({λν}) 1 ∂ Z({λν}) ∂λµ ∂λµ (cid:88) v 1 Z({λν}) = − = −(cid:88) = −(cid:88) v 1 Z({λν}) exp P (v)Oµ(v). = (cid:34) − K(cid:88) (cid:34) − K(cid:88) ν=1 ν=1 exp λνOν(v) λνOν(v) Oµ(v) (cid:88) exp v Oµ(v) (cid:35) (cid:35) (B13) (B14) (B15) (B16) (B17) v We recognize this as the expectation value of Oµ(v) with respect to the probability distribution P (v). Thus we 10 have ∂ lnZ({λν}) ∂λµ = −(cid:104)Oµ(v)(cid:105)P , and hence Eq (B13) becomes (cid:104)Oµ(v)(cid:105)P = (cid:104)Oµ(v)(cid:105)exp. where vi = (cid:126)vi is the speed of bird i. (B18) (B19) Equation (B8) tells us that the effective energy func- tion or Hamiltonian for a maximum entropy model is composed of one term for each of the observables whose expectation values we match to the data. Thus we should have That is, once we have the form of the maximum entropy distribution in Eq (B4), maximizing the likelihood of the data with respect to parameters is equivalent to imposing the constraints in Eq (B1). Appendix C: Maximum entropy model for flocks H(v) = λ1Qint + λ1V + λ3V2, (C6) and the probability distribution P (v) = e−H(v) Z (J, g, µ) . (C7) Let us now apply the maximum entropy approach to the case of bird flocks. The state of the system is charac- terized by the set v ≡ {(cid:126)v1, (cid:126)v2 ··· , (cid:126)vN} of the individual bird velocities. As discussed in the main text, we consider observables that measure the local correlations between birds and their neighbors, and the mean and variance of flight speeds. When we look at a snapshot of the flock, we can iden- tify bird j as being in the neighborhood of bird i (j ∈ Ni) if it is one of the closest nc neighbors. Then we measure the mean -- square difference in velocity between a bird and those in its neighborhood, N(cid:88) (cid:88) j∈Ni 1 nc 1 2N v2 0 i=1 Qint = (cid:126)vi − (cid:126)vj2, (C1) where we have normalized by a scale v0 to obtain a di- mensionless measure; in solving the model we shall see that it is natural to set this scale equal to the observed mean speed of the birds. It will be convenient to write this in a slightly different form, so we introduce matrix nij = 1 if j ∈ Ni and nij = 0 otherwise. Then we have It will be useful to absorb factors of N so that the ef- fective energy becomes "extensive," that is proportional (on average) to the number of birds in the flock, while the parameters of the model remain formally independent of N . Similarly, we would like to separate the choice of units for velocity from the dimensionless parameters of our model, so we introduce a scale v0 as in the main text. Thus we write N(cid:88) i=1 N(cid:88) i=1 vi. H(v) = J 4v2 0 nij(cid:126)vi − (cid:126)vj2 + g 2v2 0 i − µ v2 v0 (C8) With P (v) ∝ exp[−H(v)], we obtain Eq (2) of the main text. Appendix D: Solving the model N(cid:88) i,j=1 N(cid:88) N(cid:88) i=1 j=1 N(cid:88) N(cid:88) N(cid:88) N(cid:88) i=1 i=1 vi v2 i , V = V2 = 1 N 1 N Qint = 1 2N v2 0 1 nc nij(cid:126)vi − (cid:126)vj2. (C2) The first step in using the maximum entropy model is to compute the partition function Z (J, g, µ). Since the role of Z (J, g, µ) is to enforce normalization, we have We notice that the indices i and j appear symmetrically, but the matrix nij is not symmetric, since "being in the neighborhood" is not a symmetrical relationship (if you are my nearest neighbor, I might not be your nearest neighbor). Only the symmetric part survives the sum- mation, so we can write Z(J, g, µ) = dv e−H(v), (D1) (three -- dimensional) velocities, dv =(cid:81) where dv is the volume element in the space of all the i d3(cid:126)vi. (cid:90) Qint = 1 2N v2 0 1 nc where nij = (nij + nji)/2. nij(cid:126)vi − (cid:126)vj2, (C3) In addition to Qint, we chose as observables the mean speed and the mean -- square speed across the flock, i=1 j=1 1. Computation with free boundary conditions We begin by treating all birds as equivalent, without regard to their location in the interior or on the boundary of the flock, and we return to this below. It will be useful to think of the velocity as being composed of a speed and a direction, (cid:126)vi = vi(cid:126)si, where (cid:126)si = 1. Translating into these variables, we obtain from Eq (C8): (C4) (C5) i,j=1 N(cid:88) N(cid:88) N(cid:88) i,j=1 i,j=1 H(v) = = J 4v2 0 J 4v2 0 = − J 2v2 0 (cid:2)v2 nijvi(cid:126)si − vj(cid:126)sj2 + g 2v2 0 nij i − 2vivj(cid:126)si · (cid:126)sj + v2 N(cid:88) j nijvivj(cid:126)si · (cid:126)sj + 1 2v2 0 vi i=1 N(cid:88) (cid:3) + (cid:32) N(cid:88) i − µ v2 N(cid:88) v0 N(cid:88) g 2v2 0 i=1 i=1 g + J i=1 k=1 N(cid:88) i=1 i − µ (cid:33) v2 v0 nik i − µ v2 v0 vi N(cid:88) i=1 11 (D2) (D3) (D4) vi. Notice that the term controlling the mean -- square speed now has two contributions, one from the "direct" control parameter g and one from the social interactions with neighbors, ∝ J. In addition to rewriting the Hamiltonian, we also need to express the volume element dv in terms of the new direction and speed variables. For each bird, d3vi = v2 i dvid3(cid:126)siδ((cid:126)si − 1), (D5) where the delta function enforces the constraint that (cid:126)si is a unit vector, and the factor v2 is the Jacobian of the i transformation. In the limit that speed fluctuations are small -- which they are in the flock -- the effect of the Jacobian can always be absorbed into a redefinition of the parameters µ and g, so we drop this term here. Thus we have  J 2v2 0 N(cid:88) i,j=1 (cid:32) N(cid:88) (cid:33) N(cid:88) i=1 k=1  N(cid:88) i=1 µ v0 g + J nik v2 i + vi (D6) Z (J, g, µ) = dvid3(cid:126)siδ((cid:126)si − 1) exp nijvivj(cid:126)si · (cid:126)sj − 1 2v2 0 (cid:90) N(cid:89) i=1 Now we want to use the fact that fluctuations are small in order to simplify our calculation; we can verify, at the end, that the fluctuations predicted by the model really are small, and hence that our approximations are consis- tent. This is a now classical approximation scheme in the theory of magnetism [41], but we go through the details here in the hopes of making the calculation accessible to a broader audience. We can write the speeds as vi = V (1 + i), (D7) where V is the mean speed over the flock from Eq (C4), vi, (D8) and i is the fractional fluctuation around this mean; we expect i (cid:28) 1. Notice that with this definition we have i = 0. (D9) obtain the polarization of the flock as in Eq (12), N(cid:88) i=1 (cid:126)P = 1 N (cid:126)si. (D11) This polarization has a magnitude P and a direction that we will denote by the unit vector n, so that (cid:126)P = P n. We expect that flight directions of individual birds will be close to n, so we can write (cid:126)si = sL i n + (cid:126)πi, (D12) where (cid:126)πi is a (small) vector perpendicular to n, and the "longitudinal" term sL is necessary to be sure that (cid:126)si i remains a unit vector. As with the i above, not all N of these variables are independent, since the definition of the polarization in Eq (D11) requires that N(cid:88) i=1 P = 1 N sL i , (D13) N(cid:88) i=1 (cid:126)πi = 0. (D14) N(cid:88) i=1 V = 1 N N(cid:88) i=1 (cid:32) N(cid:89) N(cid:89) Transforming from integrating over speeds to integrating over their fluctuations, we have (cid:33)  N(cid:88)  . and (D10) dvi = V N dV di δ j i=1 i=1 j=1 To say that fluctuations in direction are small requires a bit more care. We can average the unit vectors (cid:126)si to Thus we have N(cid:89) i=1 d3(cid:126)siδ((cid:126)si − 1) = (cid:90) d2(cid:126)n (cid:90) 4π (cid:34) N(cid:89) i=1 dP (cid:18)(cid:113) d2πidsL i δ Now, if we substitute into Eq (D4), we have nij(1 + i)(1 + j)(cid:0)sL j + (cid:126)πi · (cid:126)πj i sL N(cid:88) i,j=1 H(v) = − JV 2 2v2 0 (cid:19)(cid:35) 12 (cid:32) P − 1 N δ (cid:33) sL i δ (cid:32) N(cid:88) i=1 N(cid:88) i=1 (cid:33) (cid:126)πi (D15) (cid:33) N(cid:88) k=1 nik (1 + i)2 − N µ v0 V (D16) g + J i ]2 + (cid:126)πi2 − 1 [sL (cid:32) N(cid:88) (cid:1) + V 2 2v2 0 i=1 Although we have changed variables in a way that makes it easy to make the approximation that fluctuations are small, we haven't actually used this approximation yet in simplifying the Hamiltonian. We notice that one set of delta functions in Eq (D15) enforces (D17) where the approximation is that (cid:126)πi is small. If we substitute this into Eq (D16), then to be consistent we should keep only terms up to second order in (cid:126)πi and i. The result is (1 + i)2 − N µ v0 V (D18) sL i =(cid:112)1 − (cid:126)πi2 ≈ 1 − (cid:126)πi2/2 + ··· , N(cid:88) (cid:1) . (cid:0)−(cid:126)πi2/2 − (cid:126)πj2/2 + (cid:126)πi · (cid:126)πj N(cid:88) V 2 2v2 0 (cid:32) g + J N(cid:88) nij k=1 i=1 nij(1 + i)(1 + j) + i,j=1 N(cid:88) i,j=1 − JV 2 2v2 0 (cid:33) nik H(v) = − JV 2 2v2 0 (cid:18) gV 2 2v2 0 N(cid:88) N(cid:88) i,j=1 i,j=1 N(cid:88) A crucial simplification is that the terms related to speed fluctuations (i) are decoupled from those related to di- rectional fluctuations ((cid:126)πi). Thus we have, as in Eq (5), H(v) = Hdir({(cid:126)πi}) + Hsp({i}) + E0(V ), (D19) where E0(V ) is the effective energy when all i = 0, E0(V ) = N − µV . (D20) (cid:19) Collecting terms, and dropping constants independent of {(cid:126)πi} and {i}, we find that Hdir({(cid:126)πi}) = JV 2 2v2 0 Nij(cid:126)πi · (cid:126)πj Hsp({i}) = V 2 2v2 0 (gδij + JNij) ij, (D21) (D22) where the matrix Nij has the form Nij = −nij + δij nik. (D23) k=1 In trying to compute the partition function, we will need to integrate not just over the "local" variables {i, (cid:126)πi}, but also -- as can be seen from the volume ele- ments in Eqs (D10) and (D15) -- over the global variables V , P , and n. The integral over the direction of polar- ization is simple because there is no dependence of the integrand on n; this is a consequence of the overall rota- tional invariance in our formulation of the problem. The integral over the magnitude of the polarization is also simple, since the delta function just gives us N(cid:88) i=1 N(cid:88) i=1 P = 1 N i ≈ 1 − 1 sL 2N (cid:126)πi2. (D24) (cid:90) ZV ≈ The integral over V is more interesting, since the V de- pendence of the integrand is dominated by E0(V ). Thus we need to do an integral of the form dV e−E0(V ). (D25) The key point is that E0 ∝ N , and so the integrand is very sharply peaked around some V∗. But the average of V is one of the quantities that we are fixing from the data, so we must have V∗ = (cid:104)V (cid:105)exp, and this serves to set the parameter µ, as explained in the main text. Importantly, the factor of N insures that the variations in V around V∗ will be very small in large flocks, and hence we can replace V → V∗ = (cid:104)V (cid:105)exp everywhere else in our calculations. We are also free to choose the scale v0 = (cid:104)V (cid:105)exp, and then we can simplify Hdir({(cid:126)πi}) = Hsp({i}) = J 2 1 2 Nij(cid:126)πi · (cid:126)πj, (D26) (gδij + JNij) ij. (D27) N(cid:88) N(cid:88) i,j=1 i,j=1 This separation of direction and speed variables in the Hamiltonian means that the partition function can be factorized, Z (J, g, µ) ∝ Zdir(J)Zsp(J, g)eN g/2, (D28) where Zdir(J) = Zsp(J, g) = (cid:90) (cid:34) N(cid:89) (cid:90) (cid:34) N(cid:89) i=1 (cid:35) δ (cid:32) N(cid:88)  N(cid:88) i=1 j e−Hdir({(cid:126)πi}) (cid:126)πi (cid:33)  e−Hsp({i}). d2πi (cid:35) di δ i=1 j=1 (D29) (D30) Now we have to do the integrals in Eqs (D29) and (D30), but these are not so difficult because they are Gaussians. The behavior of these integrals is determined the structure of the matrix Nij. To understand this struc- ture, imagine that the birds are in a line, and the relevant 13 neighborhood is just the two nearest neighbors along the line. Then we can see that Nij is the discrete approxima- tion to the (negative) second derivative along the line. In higher dimensions this becomes the Laplacian operator, and so Nij is called a Laplacian matrix. As with the neg- ative Laplacian, the eigenvalues {Λa} of Nij are positive, except for the smallest one, which exactly zero (Λ1 = 0). If we define the eigenvectors of Nij by wa i such that N(cid:88) j=1 Nijwa j = Λawa i , (D31) then the eigenvector associated with the zero eigenvalue is the "uniform" mode, w1 i = constant. But displace- ments along this direction are fixed to zero by the delta functions that appear in the integrals of Eqs (D29) and (D30), and this is crucial for doing the integrals. We recall that, for a general N × N matrix Mij, where λn(M ) are the eigenvalues of M . In the case of Zsp, we have (cid:90) dN x exp N(cid:88) − 1 (cid:90) (cid:34) N(cid:89) i,j=1 2 xiMijxj (cid:35) det M  = (cid:20) (2π)N (cid:21)1/2 ∝ exp  N(cid:88)  exp − 1 N(cid:88) j 2 i,j=1 Zsp(J, g) = di δ i=1 j=1 (cid:32) N(cid:88) a=1 − 1 2 (cid:33) ln[λa(M )] , (D32)  . (D33) i(gδij + JNij)j The relevant matrix is now Mij = gδij + JNij, and the eigenvalues are λa(M ) = g +JΛa, where again Λa are the eigenvalues of the Laplacian matrix Nij. We note that the integral runs over N dimensions, but the delta function fixes one combination of the {i} to be zero, and as noted above this combination is parallel to the first eigenvector. So, up to constant factors, the effect of the delta function is to exclude the first (zero) eigenvalue from the sum in Eq (D32), so that Zsp(J, g) ∝ exp ln[g + JΛa] . (D34) (cid:32) N(cid:88) a=2 − 1 2 (cid:33) rotate our coordinates into the eigenvectors of the matrix Mij, we note that in this basis fluctuations along each coordinate are independent with variance 1/Λn(M ), and then to recover the correlations in the original basis we rotate back. Again we have to be careful to respect the delta function, which serves to eliminate the fluctuations along w1 i . The end result is that N(cid:88) a=2 (cid:104)ij(cid:105) = wa i wa j g + JΛa . (D36) Since the effective Hamiltonian for speed fluctuations in Eq (D27) is a quadratic function of the {i}, the prob- ability distribution of the speed fluctuations is Gaussian,  N(cid:88) j=1  exp − 1 2 j N(cid:88) i,j=1  . iMijj P ({i}) = 1 Zsp(J, g) δ (D35) Thus we can calculate the correlations between the values of for different birds i and j in a standard way: we This result, or more precisely its generalization to the case where we treat the birds on the boundary of the flock separately, Eq (D63), is the basis for our prediction of the speed correlations as a function of the distance between birds, in Fig 2c. We can carry through the same calculation for the di- rection fluctuations. The only differences are that the vector (cid:126)πi has two components, so there are twice as many variables, and that the matrix which controls the fluctu- 14 As noted at the end of Appendix B, imposing the constraint that expectation values of observables in our model be equal to those found in the data is equivalent to maximum likelihood inference. Thus, to complete our calculation and find the parameters of our model, we should compute the probability of the data in the model, as function of the parameters J, g, and nc. Putting to- gether the results in this section, we can write the log of the full probability distribution as ations is now simple Mij = JNij. The results are (cid:32) − d − 1 2 N(cid:88) a=2 (cid:33) Zdir(J) ∝ exp ln[JΛa] , (D37) and (cid:104)(cid:126)πi · (cid:126)πj(cid:105) = (d − 1) N(cid:88) a=2 wa i wa j JΛa , (D38) where we give the result for motion in d dimensions; here d = 3. Φ ≡ ln P (datamodel) = − lnZ − (cid:104)H(v)(cid:105)exp (cid:28) J N(cid:88) 4V 2 i,j=1 (cid:29) − exp (cid:28) g 2V 2 N(cid:88) i=1 (vi − V )2 nij(cid:126)vi − (cid:126)vj2 = − lnZdir(J) − lnZsp(J, g) − N(cid:88) a=2 = N(cid:88) a=2 ln[JΛa] + 1 2 ln[g + JΛa] − N (cid:104)Qint(cid:105)exp − N Jnc 2 (cid:104)σ2(cid:105)exp, g 2 (cid:29) (D39) (D40) exp (D41) where (cid:104)···(cid:105) denotes an average over the data, we iden- tify Qint from Eq (1) of the main text, and σ2 is the fractional variance of individual birds' speeds around the flock mean. The result for Φ in Eq (D41) is simple enough that we can maximize to give explicit equations that determine the parameters. Thus N(cid:88) a=2 ⇒ 1 N ∂Φ ∂g = 0 1 g + JΛa = (cid:104)σ2(cid:105)exp, and similarly ∂Φ ∂J = 0 + 1 2 Λa g + JΛa = N (cid:104)Qint(cid:105)exp nc 2 ⇒ (N − 1) (cid:18) J N(cid:88) (cid:19) a=2 d 1 − 1 N − g(cid:104)σ2(cid:105)exp = Jnc(cid:104)Qint(cid:105)exp. 2. Computation with fixed boundary conditions So far, we have assumed free boundary conditions, cor- responding to the ideal situation where speed and orien- tations of all individuals in a flock can fluctuate in the same manner, exploring the whole accessible space of pos- sible fluctuations, given the interaction between birds. In natural flocks this is not very realistic: individuals on the boundary are constantly subject to environmental stim- uli, so that they will adjust their direction and speed not only in response to neighboring birds, but also in response to external cues. To cope with this fact, we now perform the computation of the partition functions and of the likelihood using "fixed boundary conditions," where the velocities of the birds on the boundary of the flock are held fixed at their observed values. We note that for large systems, such as the flocks we are consid- ering, boundary individuals are a negligible fraction of all individuals. As discussed more fully in Ref [13], the values of the inferred parameters do not change much with changing the boundary conditions. Fixed boundary conditions are however necessary to adequately take into account the effects of boundary on the correlations. To perform the computations with fixed conditions on the border, it is convenient to divide the birds in two groups: internal birds i, j ∈ I and birds belonging to the border a, b ∈ B. Then, Eq (C8) becomes (D42) (D43) (D44) (D45) (D46) Finally, we can substitute the solutions to these equa- tions, J∗ and g∗, back into Eq (D41) and maximize with respect to nc, as in Fig 1c. (cid:88) (cid:16) i,j∈I (cid:17) Nij + g J δij (cid:126)vi · (cid:126)vj − J v0 (cid:88) i∈I H(v) = J 2v2 0 (cid:126)hi · (cid:126)vi + HB(J, g) − µ v0 N(cid:88) i=1 vi, (D47) where (cid:126)hi = 1 v0 (cid:88) (cid:88) a∈B a,b∈B nia(cid:126)va (cid:16) HB(J, g) = J 2v2 0 Nab + g J δab (cid:17) (D48) (cid:126)va · (cid:126)vb. (D49) We can see from these expressions that holding velocities (cid:126)va fixed on the border of the flock is equivalent to consid- ering a flock in presence of a field (cid:126)hi acting on those birds who see the border birds as their neighbors. Note that birds deep in the interior do not couple directly to the field, but may feel its influence if it propagates through the flock. It will be useful to decompose these fields in relation to the mean flight direction n, as in Eq (D12), (cid:126)hi = hL i n + (cid:126)h⊥ i . (D50) The computation of the partition function now pro- ceeds exactly as in the previous subsection. The only difference is that integrations must now be performed on internal variables only; the algebra is slightly more com- plicated, but the conceptual are the same. Corresponding to Eq (D28) we have Z(J, g; nc) = e−HB(J,g)Zdir(J)Zsp(J, g)eN g/2, (D51) and in place of Eqs (D29) and (D30) we have (cid:90) (cid:34)(cid:89) (cid:90) (cid:34)(cid:89) i∈I i∈I (cid:35) e−Hdir({ (cid:126)πi∈I}) δ (cid:32) N(cid:88)  N(cid:88) i=1 j (cid:126)πi (cid:33)  e−Hsp({i∈I}), d2πi (cid:35) di δ (D52) Zdir(J) = Zsp(J, g) = j=1 (cid:88) i,j∈I J 2 (D53) (A−1)ij (cid:126)h⊥ i j − d − 1 · (cid:126)h⊥ 2 lnZdir(J) = 15 where we note that the integration is only over internal variables, but the delta function constraints involve all the variables. As in the case of free boundaries, we first integrate over global variables, which has the effect of pinning the mean velocity to its observed value, and then we can choose the scale v0 = (cid:104)V (cid:105)exp, simplifying all the expressions. The reduced Hamiltonians for the internal variables, analogs of Eqs (D26) and (D27), then become Hdir({(cid:126)πi∈I}) = Hsp({i∈I}) = (cid:88) (cid:88) i,j∈I i,j∈I J 2 J 2 (cid:88) i∈I (cid:126)h⊥ i ij − J Nij(cid:126)πi · (cid:126)πj − J (cid:16) Nij + g J δij (cid:17) · (cid:126)πi (cid:88) i∈I (D54) bii, (D55) where i −(cid:88) a∈B (cid:88) a∈B bi = hL nia = niaa (D56) is the fluctuating part of the longitudinal component of border field. Although we have same matrix Nij in these equations as in the previous section, the indices ij are restricted to the interior of the flock, and on this restricted space the matrix has different properties. To remind us of this fact, it is convenient to introduce the two matrices Aij = Nij and Bij = Nij + (g/J)δij, with indices that refer only to birds internal to the flock, i ∈ I. Then the partition functions that we need to evaluate are again Gaussian integrals, controlled by the properties of these matrices. We find, corresponding to Eqs (D34) and (D37), − d − 1 2 ln det A − J 2 i,j∈I(A−1)ij , (D57) and lnZsp(J, g) = (cid:88) i,j∈I J 2 (B−1)ijbibj − 1 2 (NI − 1) log (J) − 1 2 log (NI − 1) ln (J) − d − 1 ln (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:88) a∈B (cid:88) a∈B 2 (cid:88) i,j∈I (cid:126)πa + (A−1)ij (cid:88) i,j∈I (cid:88) i,j∈I (cid:88) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 (cid:126)h⊥ i,j∈I i (A−1)ij  1(cid:80)  1(cid:80) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 (B−1)ij − 1 2 ln det B − J 2 a + (B−1)ijbi i,j∈I(B−1)ij . (D58) Similarly, the probability distributions of the variables {i, (cid:126)πi} again are Gaussian, and we can find, by analogy with Eqs (D36) and (E14), the correlation functions. One new feature is that birds in the interior can have nonzero averages of these fluctuations, since they are responding to the birds on the boundary. Instead of rotating to the basis of eigenvectors, it is useful to define the matrices (cid:101)Aij = (A−1)ij − (cid:101)Bij = (B−1)ij − (cid:80) (cid:80) l∈I(A−1)il (cid:80) (cid:80) l∈I(B−1)il (cid:80) m∈I(A−1)jm (cid:80) m∈I(B−1)jm l,m∈I(A−1)lm l,m∈I(B−1)lm , (D59) .(D60) correlations in these fluctuations to be (cid:104)(cid:126)πi · (cid:126)πj(cid:105) = (cid:104)(cid:126)πi(cid:105) = d − 1 J (cid:88) (cid:101)Aij + (cid:104)(cid:126)πi(cid:105) · (cid:104)(cid:126)πj(cid:105), (cid:101)Aij j − (cid:126)h⊥ (cid:80) (cid:80) j∈I(A−1)ij l,m∈I(A−1)lm j∈I 16 (cid:88) a∈B (D61) (cid:126)πa. (D62) Similarly, we find the mean speed fluctuation and corre- lations to be (cid:104)i · j(cid:105) = (cid:104)i(cid:105) = 1 J (cid:101)Bij + (cid:104)i(cid:105) · (cid:104)j(cid:105), (cid:80) (cid:88) (cid:101)Bijbj − (cid:80) j∈I(B−1)ij l,m∈I(B−1)lm j∈I (cid:88) a∈B (D63) a. (D64) Then we find the mean directional fluctuation and the The correlation functions that we present in Figs 2 and 3 are based on these expressions. Finally, we need to find the conditions that set the values of the parameters. By analogy with Eqs (D43) and (D46), we find (cid:18) d 1 J NI − 1 N − g(cid:104)σ2(cid:105)exp (cid:19) = nc(cid:104)Qint(cid:105)exp + 1 N (A−1)ij (cid:126)h⊥ i · (cid:126)h⊥ j + (B−1)ijbibj (cid:88) (cid:88) i,j∈I i,j∈I (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:88) a∈B − 1 N (cid:126)πa + (cid:88) a∈B − 1 N (A−1)ij (cid:126)h⊥ j (cid:88) (cid:88) i,j∈I a + (B−1)ijbj (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 i,j∈I 1 N (cid:88) 1(cid:80) 2 1(cid:80) i,j∈I(A−1)ij − 1 2N(cid:104)V (cid:105)2 exp a,b∈B nab(cid:126)va − (cid:126)vb2 + i,j∈I(B−1)ij g N J(cid:104)V (cid:105)2 exp (cid:88) a∈B (va − (cid:104)V (cid:105)exp)2,(D65) and (cid:88) i∈I (cid:101)Bii + 1 N (cid:88) a∈B 2 a + 1 N (cid:80) (cid:80) i,j∈I(B−2)ij i,j∈I(B−1)ij (cid:32)(cid:88) a∈B (cid:33)2 a (cid:104)σ2(cid:105)exp = 1 N J Finally, the optimal value of nc can be found by maximizing the log-likelihood Φ(J, g; nc) = − lnZdir(J) − lnZsp(J, g) + HB(J, g) − JncN 2 (cid:104)Qint(cid:105)exp − N (cid:104)σ2(cid:105)exp, g 2 (D66) (D67) where we substitute for J and g the (nc dependent) solutions of Eqs (D65) and(D66). An example of the likelihood as a function of nc is given in the main text. Appendix E: Goldstone modes and the continuum limit In this Appendix we would like to make more explicit some of the mathematics behind the intuitions described in Section III of the main text. Our discussion is for the case (Appendix D 1) with free boundary conditions. We start by looking at the effective Hamiltonian for the directional variables {(cid:126)πi}, in Eq (D26), Hdir({(cid:126)πi}) = J 2 Nij(cid:126)πi · (cid:126)πj. N(cid:88) i,j=1 As explained in the discussion leading up Eq (D31), the matrix Nij has a zero eigenvalue, but in fact the whole eigenvalue spectrum has a special structure. To see this, it is useful to imagine that the birds are arranged along a line, and that the neighborhood is only the very nearest neighbor. Then we can label the birds by n, and the bird n+1 is the neighbor of bird n; we can rearrange the terms in the sum to give N(cid:88) Hdir({(cid:126)πi}) = J 2 (cid:126)πn − (cid:126)πn+12. (E1) Now suppose that the direction of flight varies only very slowly, so that we can picture a continuous function of position x in the flock, despite the fact that the birds are located at discrete positions xn = nrc, where rc is the typical distance between the nearest birds. Then we have (cid:126)π(x), and 17 Suppose that we have a function φ(x), with zero mean. If all points x are equivalent, we can characterize the statistics of fluctuations in φ(x) using the correlation function, Cφ(x − x(cid:48)) = (cid:104)φ(x)φ(x(cid:48))(cid:105). (E5) It is also useful to consider the Fourier transform of the correlation function, the power spectrum, Importantly, we can write the entire probability distri- bution for the functions φ(x) using the power spectrum, (cid:34) (cid:90) dk 2π − 1 2 φ(k)2 Sφ(k) (cid:35) , (E7) P [φ(x)] = 1 Z exp (cid:90) φ(k) = n=1 Sφ(k) = dx e+ikxCφ(x). (E6) (cid:90) (cid:90) (cid:20) (cid:20) (cid:90) . ∂x (cid:12)(cid:12)(cid:12)(cid:12) ∂(cid:126)π(x) (cid:12)(cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12)(cid:12) ∂(cid:126)π(x) (cid:12)(cid:12)(cid:12)(cid:12)2 ∂x N(cid:88) n=1 (cid:90) (cid:90) Hdir({(cid:126)πi}) ≈ Jr2 2 c (E2) where Since we are assuming that variations are smooth, we can turn the sum into an integral, Hdir({(cid:126)πi}) = Jr2 c 2 ρ dx , (E3) where ρ is the density of birds along the line. If we do the same calculation not with birds along a line, but on a regular lattice in three -- dimensional space, we find dx e+ikxφ(x) (E8) is the Fourier transform of the function φ(x). Since we have P ∝ exp[−H], Eq (E4) tells us that (cid:90) (cid:21) P [(cid:126)π(x)] = 1 Z exp − Jncr2 c ρ 2 d3x∇(cid:126)π((cid:126)x)2 . (E9) We can also write this in terms of the Fourier transforms, Hdir({(cid:126)πi}) = Jncr2 c 2 ρ d3x∇(cid:126)π(x)2, (E4) π((cid:126)k) = dx e+i(cid:126)k·(cid:126)x(cid:126)π((cid:126)x), (E10) where we also include the more realistic possibility that the "neighborhood" is not just one neighbor but a group of nc neighbors. The crucial point about Eq (E4) is that if we consider variations in flight direction on a scale (cid:96), such as (cid:126)π(x) ∼ A sin(2πx/(cid:96)), then we have Hdir ∝ A2/(cid:96)2. Thus, as the length scale of variations becomes large ((cid:96) → ∞), the "stiffness" which resists the variations goes to zero. This vanishing stiffness at long wavelengths is the signature of a "Goldstone mode," which arises because the original model allowed flight in any direction, but the actual state of the flock breaks this symmetry by selecting a particular direction [8]. If the stiffness that opposes variations (in the Hamil- tonian) goes down, then the variance of these fluctua- tions (in the probability distribution) goes up. Thus in the presence of Goldstone modes we will see a large vari- ance of fluctuations corresponding to variations over long length scales. In other words, we will see long -- ranged correlations. It is important the these are not just "long ranged," but they are genuinely scale -- free. To see this it is useful to remember some mathematical facts about Gaussian random functions (see, for example, Appendix A.2 of Ref [37]). and then Eq (E9) becomes P [(cid:126)π(x)] = 1 Z exp − Jncr2 c 2 (cid:90) ρ (cid:21) . d3k (2π)3 (cid:126)k2 π((cid:126)k)2 (E11) But now we can read off the power spectrum, by com- paring Eqs (E11) and (E7); we see that Sπ((cid:126)k) = 1 Jncr2 c ρ · 1 (cid:126)k2 . (E12) If we transform back to give the correlation function, we have Cπ((cid:126)x) = d3k (2π)3 e−i(cid:126)k·(cid:126)xSπ((cid:126)k) (cid:90) = 1 Jncr2 c ρ d3k (2π)3 e−i(cid:126)k·(cid:126)x 1 (cid:126)k2 . (E13) (E14) The key point about this result is that there is nothing in the integral to set a characteristic scale for (cid:126)x. In fact, if we double the value of (cid:126)x we make up for this by cutting the value of (cid:126)k in half so that (cid:126)k · (cid:126)x stays fixed, but since we are integrating over all possible values of (cid:126)k, all that happens is that the whole integral is reduced by a factor of two. This dimensional analysis argument tells us that Cπ((cid:126)x) ∝ 1 (cid:126)x . (E15) This is a "power -- law" decay of correlations with distance (here the power is 1), and it has no characteristic scale. Thus, scale -- free correlations in directional fluctuations are a consequence of the Goldstone modes. The predictions for speed fluctuations are very differ- ent than for directional fluctuations. In taking the limit of smooth, continuous variations for directional varia- tions, we found Hdir({(cid:126)πi}) = J 2 N(cid:88) i,j=1 Nij(cid:126)πi · (cid:126)πj (cid:90) → Jncr2 c 2 ρ d3x∇(cid:126)π(x)2. (E16) (cid:19) The same argument for speed fluctuations starts with Eq (D22), and gives i,j=1 N(cid:88) (cid:90) (cid:90) ρ ρ v2 0 (cid:18) gV 2 d3x(cid:2)Jncr2 (cid:104) d3k (2π)3 Hsp({i}) = 1 2 → 1 2 = 1 2 δij + JNij ij c∇((cid:126)x)2 + g2((cid:126)x)(cid:3) (cid:105)((cid:126)k)2, c(cid:126)k2 + g Jncr2 (E17) (E18) where in the last step we transform to the Fourier repre- sentation. By the same argument that leads to Eq (E12), we recognize the predicted power spectrum for fluctua- tions in the speed, S((cid:126)k) = 1 Jncr2 c ρ · 1 (cid:126)k2 + g/(Jncr2 c ) . (E19) than a characteristic scale kc = 1/ξ =(cid:112)g/(Jncr2 Thus, where S(cid:126)π grows without bound as the wavevector (cid:126)k becomes small, S((cid:126)k) stops growing once (cid:126)k is smaller c ). We note that ξ has the dimensions of a length, and we ex- pect that this will set the scale over which correlations extend. Indeed, if we transform back to get the correla- tion function, we have (cid:90) C((cid:126)x) = = (cid:90) d3k (2π)3 e−i(cid:126)k·(cid:126)xS((cid:126)k) (2π)3 e−i(cid:126)k·(cid:126)x d3k 1 Jncr2 c ρ ∝ e−(cid:126)x/ξ, (E20) 1 (cid:126)k2 + g/(Jncr2 c ) (E21) (E22) corresponding to Eq (9) of the main text. 18 From these results we can see that, for generic values of g/J, the maximum entropy model predicts very dif- ferent kinds of correlations for directions and speeds. In the case of directions, the correlations have a dominant contribution from long wavelength modes, there is no in- trinsic length scale, and we see scale -- free behavior. On the contrary, in the case of speed fluctuations the con- tribution of the long wavelength modes is cut off by the 'mass' term (by analogy with field theory [8]) g/J, re- sulting in correlations that decay exponentially with the distance between birds. However, when g/J goes to zero, or, more precisely, when the predicted correlation length ξ becomes comparable to the linear dimensions of the flock as whole, our analysis breaks down. We have de- scribed an essentially infinite system, with no boundaries. When g/Jnc is small enough that ξ ∼ rc then the whole flock is effective correlated, and a more detailed analysis is needed. We shall see that, in this "critical" regime, it is possible for the speed fluctuations also to be scale -- free. (cid:112)Jnc/g ∼ L, Appendix F: Decoupling speeds and flight directions The approach we have taken thus far is to build the least structured models that are consistent with the ob- served similarity of velocities between birds and their near neighbors. Importantly, we treat the velocities as vectors, and use a measure of similarity that is a rotation- ally invariant, analytic function of these vectors, Qint in Eqs (1) and (C1). One could imagine, however, that real birds do not obey these symmetries. In particular, they could have very separate mechanisms for adjusting their speeds and directions in relation to those of their neigh- bors, or their perceptual apparatus for estimating speeds and directions may introduce errors that are not equiva- lent to an isotropic vector error. Under these conditions, it would make more sense to build models that have sep- arate constraints for the observed degree of speed and direction similarity among neighbors, and this is what we explore in this Appendix. We can measure the degree of similarity or correlation among directions in the same way that we did in Ref [13], defining N(cid:88) i=1 1 ndir c (cid:88) j∈N dir i Cint = 1 N (cid:126)si · (cid:126)sj, (F1) where we allow that the neighborhood for measuring di- rectional similarity may have a size ndir that differs from the corresponding neighborhood for measuring speed similarity, nsp c . We can also define a (dis)similarity mea- sure for the speeds, by analogy with Qint, c N(cid:88) i=1 (cid:88) j∈N sp i Qsp int = 1 2N 1 nsp c (vi − vj)2. (F2) If we build the maximum entropy model consistent with measured values of these quantities, plus the mean and variance of individual speeds across the flock, we obtain, instead of Eq (C8), N(cid:88) i,j=1 N(cid:88) ij (vi − vj)2 − J dir nsp N(cid:88) N(cid:88) 2v2 0 i,j=1 vi, + g 2v2 0 i − µ v2 v0 i=1 i=1 ij (cid:126)si · (cid:126)sj ndir (F3) H(v) = J sp 4v2 0 where nsp ij is defined as with nij above, but with neighbor- c , and similarly for ndir ij hoods of size nsp . Notice that we now have two different coupling strengths, J sp and J dir, controlling speed and directional ordering, respectively. 19 Because our original model breaks into separate pieces for directional and speed fluctuations, we can carry over all the calculations, being careful about the values of the If we set J sp = J dir we are back to our parameters. original model. With the two separate parameters we find the log -- likelihood, by analogy with Eq (D67), Φ = − lnZdir(J dir; ndir c ) − lnZsp(J sp, g; nsp c ) + HB − J spnsp c N 2 (cid:104)Qsp int(cid:105)exp + J dirndir c N 2 (cid:104)Cint(cid:105)exp − N (cid:104)σ2(cid:105)exp. g 2 (F4) We can then infer, independently for speed and orien- tation, the interaction parameters, and compare them to see how different they are. We can also check whether and how much the predictions for the correlation func- tions are better than in the simpler, unified model. Re- sults are shown in Fig 5. We can see that for most flocks the global interaction strength Jnc for the speed and di- rectional degrees of freedom are very similar to each other (Fig 5a): in this case the unified model discussed in the previous section is basically equivalent to this more gen- eral model, both in terms of values of the inferred pa- rameters and in terms of predictions for the correlation functions. For a few flocks, however, we observe a decou- pling between flight directions and speeds. This typically occurs when the fractional speed fluctuations are on a different scale from the directional fluctuations. In these cases, the model that fixes the local similarities of speed and direction separately provides better predictions for the speed correlations than the unified model (Fig 5b), although these differences are not huge. Building a model that fixes the local similarities of speed and direction separately must provide a more ac- curate description of the system, since it imposes two different ways in which our model distribution P (v) has to match the real distribution of (vector) velocities. The fact that the gain in accuracy usually is small seems sig- nificant, and suggests that those rare instances where differences are larger should have biological meaning. Indeed, in most of the events where the decoupling is stronger (to the right in Fig 5a) the flocks are turn- ing. Recent findings [1] show that additional conservation laws must be taken into account to explain the dynamics during the turn. Even if such conservation laws do not modify the form of the probability distribution we are investigating in the present work, they might give rise to different effective parameters for directions and speeds. Appendix G: Dynamical model In this section we describe more in detail the dynamical model introduced in Eqs (15) and (16), and its numerical implementation. We have γ d(cid:126)vi(t) dt (cid:88) = −∇iH({(cid:126)vj}) + (cid:126)ηi(t) = − J 2v2 0 nij((cid:126)vi − (cid:126)vj) − g v2 0 (cid:88) j∈Ni j 1 nc (cid:126)fij + (cid:126)ηi(t) + d(cid:126)xi dt = (cid:126)vi, (G1) (vi − v) (cid:126)vi vi (G2) (G3) where we have added, as described in the text, forces (cid:126)fij If we write the that serve to hold the flock together. vector components of (cid:126)ηi(t) as ην i (t), with ν = 1, 2, 3, then (cid:104)ην i (t)ηµ j (t(cid:48))(cid:105) = 2γT δijδµνδ(t − t(cid:48)), (G4) where T is an effective temperature for the noisy dy- namics. We can chose our units of time so that γ = 1, and from the discussion in Appendix D, we can chose v = v0 = (cid:104)V (cid:105)exp, the desired mean speed of the flock. In this form, the model that we are considering de- scribes "self -- propelled particles" (SPP), and is very sim- ilar to the Vicsek model with attraction, which has been studied extensively in the literature [29, 30, 32, 33]. An attraction term is required to keep the flock cohesive in open space and prevent fluctuations and/or pertur- bations to disrupt the group. It has been shown that these effects are remarkably less important in models with topological interactions [16, 30, 45], which are much more robust in cohesion than SPP models with metric in- teractions. Nevertheless, even in the topological case, an 20 butions of the forces (cid:126)fij, we include only birds within a limited neighborhood, j ∈ Ni. As in the measure of sim- ilarity Qin, this neighborhood is defined topologically, so that each bird feels the effect of nc closest neighbors, rather than all the birds within a fixed physical distance. In addition, for these simulations we introduced a balanc- ing criterion, according to which a bird considers inter- acting neighbors homogeneously around it to coordinate with. This mimics the idea of a shell of relevant topolog- ical neighbors, and is similar to using Voronoi neighbors, as in Ref [45], but is much easier to implement numeri- cally. A balanced interaction enhances the stability of the flock [30], increasing the range of parameters where Eqs (G2) and (G3) give rise to realistic behavior. However, we checked also the simple topological case, obtaining qualitatively similar results. Despite its similarity with other SPP models, the model we are considering has a crucial new ingredient, namely that the speeds of the individual birds are not fixed but can change in time. Accordingly, Eq (G2) de- scribes the evolution of the full velocity (rather than the flight direction, as in Ref [32]), with a term ∝ g that sets the scale of the speed fluctuations. In addition, existing SPP models are usually defined as discrete dynamical update equations, which do not have a well defined con- tinuum limit. In contrast, we have defined our model as a stochastic differential equation. We simulate our model using a finite interval (Euler) discretization, and we checked that macroscopic proper- ties of the flock (e.g., the mean speed) remained the same if the size of the time step was decreased. Parameters J and nc can be taken from the discussion of real flocks, and the temperature T adjusted until the polarization is in the range seen in the data (Table I). We simulated flocks of different sizes, and checked that the flock had come to a stationary state before making measurements. With all other parameters fixed, we varied g, with the results shown in Fig 4. Long ranged correlations can arise through one other mechanism that we have not discussed, and this is the emergence of "hydrodynamic modes;" it has been argued that such modes are an essential feature of self -- propelled particle models on the largest spatial and temporal scales [9, 10]. The simulations described here suggest, however, that such effects become dominant only on much larger scales in space and especially in time, and thus cannot explain the scale free speed correlations that we observe at equal times in real flocks. We know that both metric and topological SPP/Vicsek models exhibit giant density fluctuations on large scales [45], yet we have seen that as long as g is finite, speed correlations are short range and a critical value of g is necessary to make them scale -- free. (a) (b) FIG. 5. Model with independent interactions for speed and flight directions. (a) The inferred global interaction strength Jnc for the orientational degrees of freedom (vertical axis) vs the speed degrees of freedom (horizontal axis). The straight line corresponds to y = x, i.e. to the global model where the interaction parameters are the same for speed and flight directions. (b) Prediction for the speed correlation function of the unified model Eq (C8) and for the decoupled model based on Eq (F3), for flock 28-10, corresponding to the point most on the right in panel (a). attraction force is the most controlled way to fix the den- sity of the group to a stationary value, therefore we will include it. We choose the forces rij−re ra−rhc if rij < ra otherwise, (G5) 4 1 (cid:40) 1 (cid:126)fij = α (cid:126)rij rij where (cid:126)rij is the vector from bird i to bird j, rij = (cid:126)rij is its length; re is the equilibrium distance between birds where the force vanishes, while ra and rhc set spatial scales for the extent of the force. In our simulations we choose re = 0.5, ra = 0.8, and rhc = 0.2, which sets our units of length, and α = 0.95. An important point is that, when we sum the contri- [1] A Attanasi, A Cavagna, L Del Castello, I Giardina, TS Grigera, A Jelic, S Melillo, L Parisi, O Pohl, E Shen, and M Viale, Superfluid transport of information in turn- 0200400600800Jnc - Speed0100200300400500600700800Jnc - Directions05101520253035r (m)-0.00200.0020.0040.0060.008Csp(r)DataUnified ModelDecoupled Model21 ing of flocks. arXiv.org:1303.7097 [cond -- mat.stat -- mech] (2013). [2] HR Pulliam, On the advantages of flocking. J Theor Biol 38, 419 -- 422 (1973). [3] W Cresswell, Flocking is an effective anti -- predation strategy in redshanks, Tringa totanus. Anim Behav 47, 433 -- 442 (1994). [4] J Krause and GD Ruxton, Living in Groups (Oxford Uni- versity Press, Oxford, 2002). [5] A Procaccini, A Orlandi, A Cavagna, I Giardina, F Zo- ratto, D Santucci, F Chiarotti, CK Hemelrijk, E Alleva, G Parisi, and C Carere, Propagating waves in starling, Sturnus vulgaris, flocks under predation. Anim Behav 82, 759 -- 765 (2011). [6] A Cavagna, A Cimarelli, I Giardina, G Parisi, R Santa- gati, F Stefanini, and M Viale, Scale -- free correlations in starling flocks. Proc Natl Acad Sci (USA) 107, 11865 -- 11870 (2010). [7] ID Couzin and J Krause, Self -- organization and collec- tive behavior in vertebrates. Adv Study Behav 32, 1 -- 75 (2003). [21] TR Lezon, JR Banavar, M Cieplak, A Maritan, and NV Federoff, Using the principle of entropy maximization to infer genetic interaction networks from gene expression patterns. Proc Nat'l Acad Sci (USA) 103, 19033 -- 19038 (2006). [22] A Tang, D Jackson, J Hobbs, W Chen, A Prieto, JL Smith, H Patel, A Sher, A Litke, and JM Beggs, A max- imum entropy model applied to spatial and temporal cor- relations from cortical networks in vitro. J Neursoci 28, 505 -- 518 (2008). [23] M Weigt, RA White, H Szurmant, JA Hoch, and T Hwa, Identification of direct residue contacts in protein -- protein interaction by message passing. Proc Natl Acad Sci (USA) 106, 67 -- 72 (2009). [24] N Halabi, O Rivoire, S Leibler, and R Ranganathan, Protein sectors: Evolutionary units of three -- dimensional structure. Cell 138, 774 -- 786 (2009). [25] T Mora, AM Walczak, W Bialek, and CG Callan, Max- imum entropy models for antibody diversity. Proc Natl Acad Sci (USA) 107, 5405 -- 5410 (2010) (2009). [26] GJ Stephens and W Bialek, Statistical mechanics of let- [8] G Parisi, Statistical Field Theory (Addison -- Wesley, Red- ters in words. Phys Rev E 81, 066119 (2010). wood City CA, 1988). [9] J Toner and Y Tu, Long -- range order in a two -- dimensional XY model: How birds fly together. Phys Rev Lett 75, 4326 -- 4329 (1995). [10] J Toner and Y Tu, Flocks, herds, and schools: A quan- titative theory of flocking. Phys Rev E 58, 4828 -- 4858 (1998). [11] S Ramaswamy, The mechanics and statistics of active [27] G Tkacik, O Marre, T Mora, D Amodei, MJ Berry II, and W Bialek, The simplest maximum entropy model for collective behavior in a neural network. J Stat Mech P03011 (2013); arXiv.org:1207.6319 (2012). [28] G Tkacik, O Marre, D Amodei, E Schneidman, W Bialek, and MJ Berry II, Searching for collective behavior in a network of real neurons. arXiv.og:1306.3061 [q -- bio.NC] (2013). matter. Annu Rev Cond Matt Phys 1, 323 -- 345 (2010). [29] G Gr´egoire and H Chat´e, Onset of collective and cohesive [12] KG Wilson, Problems in physics with many scales of motion. Phys Rev Lett 92, 025702 (2004). length. Sci Am 241, 158 -- 179 (1979). [13] W Bialek, A Cavagna, I Giardina, T Mora, E Silvestri, M Viale, and AM Walczak, Statistical mechanics for natural flocks of birds Proc Natl Acad Sci (USA) 109, 4786 -- 4791 (2012). [14] A Cavagna, I Giardina, A Orlandi, G Parisi, A Procac- cini, M Viale, and V Zdravkovic, The STARFLAG hand- book on collective animal behaviour: 1. Empirical meth- ods, Anim Behav 76, 217 -- 236 (2008). [15] A Cavagna, I Giardina, A Orlandi, G Parisi, and A Pro- caccini, The STARFLAG handbook on collective animal behaviour: 2. Three -- dimensional analysis, Anim Behav 76, 237 -- 248 (2008). [16] M Ballerini, N Cabibbo, R Candelier, A Cavagna, E Cis- bani, I Giardina, V Lecomte, A Orlandi, G Parisi, A Pro- caccini, M Viale, and V Zdravkovic, Interaction ruling animal collective behavior depends on topological rather than metric distance: Evidence from a field study. Proc Natl Acad Sci (USA) 105, 1232 -- 1237 (2008). [17] ET Jaynes, Information theory and statistical mechanics. Phys Rev 106, 620 -- 630 (1957). [18] DJC Mackay, Information Theory, Inference, and Learn- ing Algorithms (Cambridge University Press, Cambridge, 2003) [19] E Schneidman, MJ Berry II, R Segev, and W Bialek, Weak pairwise correlations imply strongly correlated net- work states in a neural population. Nature 440, 1007 -- 1012 (2006). [20] J Shlens, GD Field, JL Gauthier, MI Grivich, D Petr- usca, A Sher, AM Litke, and EJ Chichilnisky, The struc- ture of multi -- neuron firing patterns in primate retina. J Neurosci 26, 8254 -- 8266 (2006). [30] M Camperi, A Cavagna, I Giardina, G Parisi, and E Sil- vestri, Spatially balanced topological interaction grants optimal cohesion in flocking models. Interface Focus 2, 715 -- 725 (2012). [31] O Pohl, Analyse und Simulation eines stochastischen Modells zur Schwarmdynamik (Diplomarbeit, Rheinis- chen Friedrich -- Wilhelms -- Universitat Bonn, 2011). [32] T Vicsek, A Czir´ok, E Ben -- Jacob, I Cohen, and O Shochet, Novel type of phase transition in a system of self -- driven particles. Phys Rev Lett 75, 1226 -- 1229 (1995). [33] G Gr´egoire, H Chat´e, and Y Tu, Moving and staying to- gether without a leader, Physica D 181, 157 -- 170 (2003). [34] P Bak, How Nature Works: The Science of Self -- Organized Criticality (Copernicus, New York, 1996). [35] W Bialek and T Mora, Are biological systems poised at criticality? J Stat Phys 144, 268 -- 302 (2011). [36] JMV Rayner, PW Viscardi, S Ward, and JR Speak- man, Aerodynamics and energetics of intermittent flight in birds. Amer Zool 41, 188 -- 204 (2001). [37] W Bialek, Biophysics: Searching for Principles (Prince- ton University Press, Princeton, 2012). [38] CE Shannon, A mathematical theory of communica- tion. Bell Sys. Tech. J. 27, 379 -- 423 & 623 -- 656 (1948). Reprinted in CE Shannon and W Weaver, The Mathe- matical Theory of Communication (University of Illinois Press, Urbana, 1949). [39] TM Cover and JA Thomas, Elements of Information Theory (Wiley, New York, 1991). [40] CM Bender and SA Orszag, Advanced Mathematical Methods for Scientists and Engineers (McGraw -- Hill, New York, 1978). [41] FJ Dyson, General theory of spin -- wave interactions. 20, 1491 -- 1510 (2010). Phys Rev 102, 1217 -- 1230 (1956). [42] H Edelsbrunner and EP Mucke, Three -- dimensional al- pha shapes. ACM Trans Graphics 13, 43 -- 72 (1994). [43] A Cavagna, A Cimarelli, I Giardina, G Parisi, R Santa- gati, F Stefanini, and R Tavarone. From empirical data to inter-individual interactions: unveiling the rules of col- lective animal behaviour, Math Models Methods Appl Sci [44] A Cavagna, A Cimarelli, I Giardina, A Orlandi, G Parisi, A Procaccini, R Santagati, and F Stefanini, New statis- tical tools for analyzing the structure of animal groups Math. Biosc. 214, 32-34 (2008). [45] F Ginelli and H Chat´e, Relevance of metric -- free interac- tions in flocking phenomena. Phys Rev Lett 105, 168103 (2010). 22
1712.03901	1	1712	2017-12-05T21:20:30	Label-free optical vibrational spectroscopy to detect the metabolic state of M. tuberculosis cells at the site of disease	[ "physics.bio-ph", "physics.optics" ]	Tuberculosis relapse is a barrier to shorter treatment. It is thought that lipid rich cells, phenotypically resistant to antibiotics, may play a major role. Most studies investigating relapse use sputum samples although tissue bacteria may play an important role. We developed a non-destructive, label-free technique combining wavelength modulated Raman (WMR) spectroscopy and fluorescence detection (Nile Red staining) to interrogate Mycobacterium tuberculosis cell state. This approach could differentiate single 'dormant' (lipid rich, LR) and 'non-dormant' (lipid poor, LP) cells with high sensitivity and specificity. We applied this to experimentally infected guinea pig lung sections and were able to distinguish both cell types showing that the LR phenotype dominates in infected tissue. Both in-vitro and ex-vivo spectra correlated well, showing for the first time that Mycobacterium tuberculosis, likely to be phenotypically resistant to antibiotics, are present in large numbers in tissue. This is an important step in understanding the pathology of relapse supporting the idea that they may be caused by M. tuberculosis cells with lipid inclusions.	physics.bio-ph	physics	Label-free optical vibrational spectroscopy to detect the metabolic state of M. tuberculosis cells at the site of disease Vincent O. Baron1, ‡, Mingzhou Chen2, ‡, , Simon O. Clark3, Ann Williams3, Robert J. H. Hammond1, Kishan Dholakia2 and Stephen H. Gillespie1, 1 School of Medicine, University of St Andrews, St Andrews, UK, KY16 9TF 2 SUPA, School of Physics and Astronomy, University of St Andrews, KY16 9SS, St Andrews, UK 3 Public Health England, Porton Down, Salisbury, Wiltshire, SP4 0JG, UK ‡ These authors contributed equally to this work *Corresponding authors: [email protected] or [email protected] Abstract Tuberculosis relapse is a barrier to shorter treatment. It is thought that lipid rich cells, phenotypically resistant to antibiotics, may play a major role. Most studies investigating relapse use sputum samples although tissue bacteria may play an important role. We developed a non- destructive, label-free technique combining wavelength modulated Raman (WMR) spectroscopy and fluorescence detection (Nile Red staining) to interrogate Mycobacterium tuberculosis cell state. This approach could differentiate single "dormant" (lipid rich, LR) and "non-dormant" (lipid poor, LP) cells with high sensitivity and specificity. We applied this to experimentally infected guinea pig lung sections and were able to distinguish both cell types showing that the LR phenotype dominates in infected tissue. Both in-vitro and ex-vivo spectra correlated well, showing for the first time that Mycobacterium tuberculosis, likely to be phenotypically resistant to antibiotics, are present in large numbers in tissue. This is an important step in understanding the pathology of relapse supporting the idea that they may be caused by M. tuberculosis cells with lipid inclusions. Introduction Tuberculosis is now established as the most important cause of death due to infectious disease, yet treatment has not improved in fifty years. Relapse after successful treatment is the major barrier to shorter therapy for tuberculosis as has been confirmed by recent tuberculosis clinical trials where more bactericidal regimens have failed due to higher relapse rate 1-3. Despite its importance, we know very little about the bacteriology of relapse even though this outcome is possible with patients who clear their sputum rapidly 4. We need improved and non-destructive methods to study mycobacteria at the site of the disease. It is often speculated that relapse is linked to bacteria that survive treatment. Several studies demonstrate the accumulation of lipids in intracellular bodies that are associated with a lower metabolic rate 5-8. It has been shown recently that these lipid body positive cells are up to 40 times more resistant to key components of the treatment regimen such as isoniazid 9. An attractive approach to investigate the lipid content of Mycobacterium tuberculosis would be to use an all-optical label-free method to examine bacterial cells in tissue in a way that would, for example, allow additional immunology studies. Optical interrogation can lead to inelastic scattering of light and distinct vibrational bands in subsequent spectra that can be used to distinguish the molecular content of the bacteria under investigation. In particular, Raman spectroscopy has been used previously as a means of identifying bacterial taxonomy at a single cell level in a range of species including mycobacteria 10,11. These studies, however, have only previously been performed on isolated cultured cells 12 and isolated cells from sputum 13. This letter reports, for the first time, the metabolic cell state of mycobacteria by exploring the lipid content of individual cells in tissue using wavelength modulated Raman (WMR) spectroscopy. WMR analysis uses a scan of the laser wavelengths, rather than a single wavelength, for Raman excitation combined with subsequent multivariate statistical analysis that removes all background fluorescence 14,15. WMR spectroscopy shows an increase in signal- to-noise ratio compared to five methods including standard Raman spectroscopy 16. Using this approach, we were able to detect LR M. tuberculosis cells in infected tissue with very high sensitivity and specificity. This is a major step towards understanding the pathogenesis of tuberculosis at the infection site and provides a paradigm for its application to a broad range of infectious diseases. Results Single cell discrimination between lipid rich and lipid poor bacteria We separated mixed cultures of M. smegmatis, M. bovis (Bacillus Calmette-Guérin, BCG) and M. tuberculosis into lipid rich (LR) and lipid poor (LP) fractions with greater than 90% purity using our previously published method 9, and performed WMR spectroscopy on the separated cells. We recorded Raman spectra from ~60 individual lipid rich (LR) and lipid poor (LP) mycobacteria cells (M. smegmatis, BCG and M. tuberculosis) in-vitro. The spectra from the mycobacterial species are illustrated in the Supplementary Section 1. In wavelength modulated Raman spectra (WMR spectra) zero-crossings are equivalent to peak positions in standard Raman spectra and the peak-to-valley corresponds to the peak intensity in standard Raman spectra. The two phenotypes mainly differ in two lipid peaks at 1300 cm-1 (designated lipid band A) and at 1440-1450 cm-1 (designated lipid band B), see Supplementary Fig. 1 (see Supplementary Section 4 for the peak assignment). LR cells showed higher Raman peak intensity in both lipid bands (A and B) compared to LP cells. The two phenotypes are distinguished with high specificity and sensitivity, M. smegmatis (93.8%/96.8%), BCG (100%/96.8%) and M. tuberculosis (92.6%/96.1%) (see Supplementary Fig. 1). Identification of M.tuberculosis in unstained infected lung tissue sections Having created a tool to determine the lipid status of mycobacterial cells non-destructively we applied the technique to unstained, formaldehyde-fixed guinea pig lung tissue infected by M. tuberculosis to understand the cell state of the bacteria at the site of disease (see Methods section for more details). Guinea pig tissue was used in the analyses because the lesions which develop during pulmonary tuberculosis in guinea pigs are histologically similar to the disease seen in humans 17. In total, 107 single M. tuberculosis cells were interrogated by WMR spectroscopy from the lung alveoli (Fig. 1c and Fig. 2b). An example of a single bacillus in an alveoli is shown in Fig. 1a. The tissue section is attached to a quartz coverslip. The background of the lung alveoli was compared to the signal from a clean quartz slide. The two signals were found very comparable and no important signal is coming from the lung alveoli itself (Fig. 1b). The Fig. 1c shows the average WMR spectrum of the bacteria acquired in the alveoli with the two background signals. Figure 1 investigation of the impact of the lung alveoli to the bacterial Raman spectrum. The Fig. 1a shows an example of single bacillus in a lung alveoli interrogated by WMR spectroscopy. The scale bar corresponds to 5 μm. The Fig. 1b presents the average Raman spectrum of the quartz slide surface (red line) and the average Raman spectrum of the lung alveoli (background)(green line). The tissue section like any other preparation interrogated with WMR spectroscopy is on a quartz coverslip. In the Fig. 1c the average spectrum of M. tuberculosis cells acquired in tissue are added (blue line) and compared with the lung alveoli and the quartz slide background. In both Fig. 1b and Fig. 1c the shaded coloured areas represent the standard deviation. In Fig. 1b and Fig. 1c the x-axis is in wavenumbers the y-axis represents the differential intensity in arbitrary units. It was not possible to relate our in-vitro and ex-vivo data directly as it was noted that lipid bands A and B were impacted differently by formalin fixation and freezing 11,18 (Fig. 2a and 2b). In M. tuberculosis lipid band A, intensity is much lower when acquired in tissue as compared to in-vitro. In contrast, there is less variation in lipid band B intensity between in-vitro and ex-vivo samples. Lipid band B was therefore used in subsequent examinations (Fig 2c). To compare in-vitro with ex-vivo data we needed to establish a ratio between a band that varies and one that varies very little. Moreover, when in-vitro data are compared (Fig. 2a) similar peak intensities are found in the band between 1050 cm-1 and 1070 cm-1 (see Supplementary Section 4, peak assignments) for both M. tuberculosis LR and LP phenotypes. This was also true for M. tuberculosis in ex-vivo samples. Thus, this band is ideal as an internal reference and we, therefore calculated the peak- to-peak ratio (R band B/Ref band) between the maximum peak intensity of lipid band B and the maximum peak intensity in the internal reference band (1050 cm-1 and 1070 cm-1) for each WMR spectrum acquired from M.tuberculosis in lung tissue (Fig. 2b) and also for each WMR spectrum acquired from M.tuberculosis in-vitro (Fig. 2a). A higher ratio indicates a higher lipid concentration in the cell. Using this technique it is notable that the distribution of values for LR and LP cells in-vitro overlaps, but the peak of the distributions are clearly separated (Fig. 2c). We are able to observe a lipid ratio value R band B/Ref band analogous to both LP and LR among cells in tissue. The R band B/Ref band obtained from WMR spectra of cells in infected lung tissue shows a wide distribution confirming the presence of both LR and LP phenotypes. It is notable that the distribution of cell state in tissue is skewed towards LR phenotype (Fig. 2c). A similar result was obtained when (R band A/Ref band), with band A adjusted for the losses associated with fixation and freezing was used (see Supplementary Section 3). (a) (b) (c) Figure 2 (a) WMR spectra of LR and LP M. tuberculosis cells. Mean spectra of both LR (red curves) and LP (green curves) cells are calculated from ~60 WMR spectra are taken from single bacteria cells. The shaded area represents one standard deviation. The x-coordinate corresponds to the Raman shift (in wavenumber, cm-1) and the y-coordinate the differential Raman intensity in arbitrary units. (b) WMR spectra from M. tuberculosis cells in the infected lung tissue. The solid red line shows the mean WMR spectrum averaged over from all spectra of 107 cells taken at the single bacterium cell level. The colour-shaded area represents the associated single standard deviation. Ratio of two peak intensity (band B / Ref band)0.511.522.53Percentage of cell population (%)05101520253035M.tuberculosis LPM.tuberculosis LRM.tuberculosis in TissueThe acquisition time for each single bacterium was 150 seconds in total (see Methods section for more details). (c) Percentage of M. tuberculosis in-vitro LR, LP and form tissue with a given peak- to-peak ratio (R band B/Ref band) calculated by dividing the lipid band B intensity value by an internal reference band (Ref band: 1050 cm-1 to 1070 cm-1) peak intensity for each in-vitro M. tuberculosis WMR spectra (Fig. 2a) and each M. tuberculosis acquired in the lung tissue (Fig. 2b). The x- coordinate corresponds to R band B/Ref band between 0.5 and 3 and the y-coordinates represent the percentage of the bacterial population for each specific lipid ratio value. The blue, green and yellow bars correspond to the in-vitro LP, the in-vitro LR and the ex-vivo populations respectively. The lines in Fig. 2c are included to illustrate the shape of the distribution. Identification of LR and LP single bacterium in the M.tuberculosis infected tissue To ensure that the match between in-vitro and ex-vivo cells that we demonstrate is not an in-vitro artefact, we obtained an alternative measure of lipid cell state from tissue in parallel with WMR spectroscopy. To do this we used our previously published staining method 9. As shown in Fig. 3, all LP cells appear only red caused by excitation from cell wall polar lipids. LR bacteria are red from polar cell wall lipids and are distinguished from LP by green fluorescence emission caused by the non-polar lipid body. We studied in-vitro stained cells and stained cells in guinea pig infected lung. Nile red staining does not impact significantly the WMR spectrum of bacteria as shown in Supplementary Section 6. We scored the Nile Red stained bacilli as either LR or LP as previously described and recorded the Raman spectrum from these defined cells (Fig. 4). This showed that, for both in-vitro cells and cells in tissue that the main source of variability is found in bands A and B. We also showed that in-vitro LR and LP cells could be discriminated by WMR spectroscopy (sensitivity (84.0%) and specificity (80.2%)) (Fig. 4a and 4b). LP cells cluster closely but LR are more dispersed. The presence of green fluorescence emission coming from intracellular lipid bodies correlates with higher Raman lipid peaks (in both lipid band A and B). Polar lipid stain Non-polar lipid stain Combined picture Bright field (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) s i s o l u c r e b u t . M o r t i v - n I s i s o l u c r e b u t . M o v i v - x E ) n o i t c e s e u s s i t g n u l ( 2𝑢m Figure 3 M. tuberculosis in-vitro and from frozen lung tissue section stained with Nile red and observed on the fluorescence-Raman combined system (more details in the method section). (a, b, c, d) show in-vitro M. tuberculosis and (e, f, g, h, i, j, k and l) show M. tuberculosis in lung tissue. (a, e, i) show the red fluorescence (polar lipids) and (b, f, j) show the green fluorescence emission (non- polar lipids). (c, g, k) represent the pseudo images made by combining red and green fluorescence emissions images. Bright field images (White light) are shown in d, h and l. Both red and green fluorescence emissions can be observed in LR cells encircled in blue, while only red fluorescence emission is presented in LP cells encircled in yellow. The size of bacteria can be calculated from the scale bar shown in l (typically between 0.5 – 1μm in width and 2 - 4μm in length). (a) (b) (c) (e) (d) (f) Figure 4 M.tuberculosis in-vitro and from infected lung tissue WMR spectra and PCA clusters. LR and LP bacteria are pre-assigned by fluorescence signals from Nile red stained bacteria. (a) Mean spectra of both LR (red curve) and LP (green curve) cells averaged over 100 WMR spectra taken in- vitro. The light coloured area represents the associated single standard deviation. The first three PCs for both LR and LP cells form two 3D clusters in (b). The LR cluster shows a broader cluster compared to LP group. (c, d) M. tuberculosis from lung tissue WMR spectra and PCA clusters. (c) Mean spectra of both LR (red curve) and LP (green curve) cells. The first three PCs for both LR and LP cells form two 3D clusters in (d). The LR cluster shows a broader cluster compared to LP group similarly than in-vitro. (e) Comparison of in-vitro and lung tissue M.tuberculosis LR and LP cells. In- vitro LR (red) and in-vitro LP (green) were compared to the ex-vivo LR (blue) and ex-vivo LP (yellow) bacilli acquired in infected lung tissue. (e) Mean spectra of both lipid rich and lipid poor cells acquired with WMR spectroscopy. (f) 3D clusters are plotted with the first three PCs for LR and LP in-vitro cells and the cells from infected tissue. We then applied this method to M. tuberculosis cells (93 in total) in infected lung tissue sections. 56 cells were classified LR by Nile Red staining and 37 as LP (Fig. 3e - l). Using PCA LR and LP cells clustered together although there is a wider distribution for the LR cells as had been noted in-vitro (Fig. 4d). As in the case of our in-vitro data, green fluorescence correlates with higher lipid peaks in lipid bands A and B. A broader cluster in LR group observed in Fig. 4d also shows higher variability among the LR population than the LP population in tissue. Thus, it was possible to discriminate between LR and LP cells with high sensitivity (92.6%) and specificity (84.6%) (Fig. 4c and 4d). This is illustrated by combing the figures to demonstrate the overlap between the differing cell types in-vitro and ex-vivo which cluster together (Fig. 4e and 4f). Using this measure, we are able to demonstrate that M. tuberculosis cells in the lung are predominately LR (60%). Discussion Preventing relapse is the main reason for prolonged treatment in tuberculosis, yet the causes are still not fully understood 1,4. In particular we have shown that some patients who become sputum culture negative rapidly still go on to relapse 4. Our research provides a new tool to unravel this problem. This is the first time the label-free vibrational spectroscopy has been used to study the pathogenesis of tuberculosis in situ. Whilst Raman has been used previously to distinguish bacterial genus and species 10,11, our work makes a major leap forward giving us the ability to distinguish bacterial cell state at the site of disease. Reducing the Raman signal window size may affect the results slightly but does so within a small standard deviation of 0.044 in the obtained sensitivity and specificity (see in Supplementary Section 2 for more details). The possibility of this window reduction with a similar performance actually gives us a way to make our system more compact and to have higher resolution, i.e. using a grating only over a range of 1400-1500cm-1 and a shorter acquisition time. Future developments of our work therefore could lead to wide field video-rate acquisition in-vivo with non-linear vibrational spectroscopy approaches using suitable tuneable lasers for coherent anti-Stokes Raman scattering or stimulated Raman scattering 19,20. This confirms that vibrational spectroscopy can be a powerful methodology that can be applied to pathogenesis and pharmacodynamics studies. The study of the physiological status of M. tuberculosis bacteria has been limited to the sputum while may not fully represent the cells responsible for relapse. The physiological state of the bacteria that never get into the sputum are missed by such approaches 21. Using WMRS we have been able to differentiate lipid rich from lipid poor mycobacteria in tissue and consequently our data bridges the gap between tissue and sputum. Importantly, current methods, for example Nile Red staining (see in Supplementary Section 5 for more details), require destructive techniques, which means that no additional studies can be performed 22. The preparations that we have used could be re-examined with, for example, immunohistochemical methods to differentiate macrophage activation near to M. tuberculosis for which the metabolic state is now known. This opens up a wide range of new research possibilities that we will explore in future experiments. The distinction between LR and LP is important as the LR phenotype has been associated with a reduced metabolic activity and down regulation of a series of enzymes 6. More importantly, such LR cells have been shown to be phenotypically resistant to anti-tuberculosis antibiotics 7,23. Antibiotic susceptibility testing of purified populations of LR and LP cells suggest that the resistance can increase from between 3-5 times (ciprofloxacin) to more than 20 times for isoniazid and rifampicin 9. We have now shown that in tissue the majority of the cells that we examined had a LR phenotype that implies a significant degree of phenotypic resistance. Our confidence in this result is increased by the concordance of in-vitro and ex-vivo clustering as demonstrated in Fig. 4f; confirming as well that no significant signal from the lung is participating to the WMR spectra (Fig 1). Thus, our observation may throw some light on the mismatch between standard anti-tuberculosis susceptibility testing and the outcome of treatment in patients and the overall result of clinical trials. Many patients relapse after apparently successful treatment and this is thought to be due to dormant bacteria able to survive treatment and regrow 4. The presence of lipid bodies positive bacteria in TB infected patients' sputum has been linked to a higher risk of poor outcome 24. However sputum data might not be representative of the bacterial population living in patient lungs. Our data derived from lung tissue supports the idea that relapses may be caused by LR M. tuberculosis with phenotypic resistance to the prescribed treatment. It should be noted, however, that we characterised bacteria in the alveoli and not in the complex structure within solid tissue. Future developments will enable us to achieve this. Thus, using Raman spectroscopy we have shown that lipid rich bacteria are present in the lung at the site of disease and are the predominant cell type (Fig. 2c, Fig. 4c and 4d). This observation suggests that this characteristic may allow LR cells to survive antibiotic chemotherapy. Methods Bacterial culture conditions M. bovis (BCG, NCTC 5692), M. smegmatis (NCTC 8159), M. tuberculosis (NCTC7416) were grown at 37°C in Middlebrook 7H9 medium (Sigma-Aldrich) supplemented with 4 ml of 50% glycerol (for 450 ml) (Sigma-Aldrich) and 0.05% tween80 (Fisher BioReagents). BCG and M. tuberculosis culture medium were also supplemented with 1 vial of Middlebrook ADC enrichment (Sigma- Aldrich). Animal infection The lung sample analysed in these studies were from specific pathogen-free guinea pigs infected with an aerosol dose of 10–50 CFU (retained dose in the lung) of M. tuberculosis H37Rv (NCTC cat. no. 7416) 25. Nose-only aerosol challenge was performed using a fully contained Henderson apparatus as previously described 26 and 27 in conjunction with the AeroMP (Biaera) control unit 28. Guinea pig experimental work was conducted according to UK Home Office legislation for animal experimentation and was approved by the UK Home Office. All animals were weighed weekly and observed daily in order to monitor any adverse effects. Four weeks post-challenge, animals were killed by an overdose of sodium pentobarbital. At necropsy, the lung was immediately excised and sections of the left and right cranial lobes and right caudal lobes were fixed in 10% neutral buffered formalin for up to 18 months. Frozen section Following fixation, the lung tissue was placed into OCT (optimal cutting temperature) solution (30% sucrose in PBS) to embed and freeze the tissue on dry ice. The fixed and frozen tissue was sectioned into 5µm slices using a cryostat by Amsbio Ltd, UK (AMS Biotechnology (Europe) Limited / Registered office: 184 Park Drive, Milton Park, Abingdon OX14 4SE / Registered in England & Wales: Company number 2117791 / ISO 9001:2008 registered firm). In this study three serial lung sections coming from one guinea pig were interrogated. Nile Red staining of Bacterial culture The bacteria were stained using a Nile Red stock solution at 250 μg/ml diluted in DMSO (Sigma- Aldrich) and a 1 μl aliquot added to 100 μl of bacterial suspension, vortexed and left for ten minutes in the dark at room temperature. The tubes were spun down at 20,000 g for 3 minutes the supernatant was discarded. The bacteria were washed twice using PBS. The bacterial pellet was resuspended in 20 μl of PBS and 10 μl heat fixed on a glass slide for light microscope or a quartz slide for Raman spectroscopy. Nile Red staining of guinea pig lung tissue section A 5μm thick guinea pig frozen lung section mounted on quartz coverslip (SPI Supplies, 01015T- AB) was stained using 2 μl of Nile red diluted in PBS, 25 μg/ml final concentration. The 2 μl were placed on the tissue section and the coverslip was then placed on top of a thick quartz slide (SPI Supplies, 01016-AB). The mount was sealed as noted previously. The sample was then interrogated by WMR spectroscopy. Figure 5. System setup for a combined Raman-fluorescence microscope: SF: single-mode fibre; MF: multi-mode fibre (200um core diameter); LF: Laser line filter (Semrock LL01-785); EF: Edge filter (Semrock LPD02-785RU); NF: Notch filter (Semrock NF03-785E); TRITC/FITC: Fluorescence Filter cubes (FITC: excitation 475-490nm/emission 500-540nm, TRITC: excitation 545-565nm/emission 580-620nm); FM：flip mirror; CCD1: Hamamatsu ORCA_ER; CCD2:Imaging Source USB camera (DFK 42AUC03); CCD3:Andor Newton Camera (cooled at -70 oC); L1~L6: lenses; F/#: F number matcher. Combined Raman-fluorescence spectroscopy setup WMR spectra were recorded by a combined confocal Raman-fluorescence imaging system based on a Nikon microscope (Nikon TE2000-E). A tunable Ti:Sa laser (SolsTis M Squared lasers, 1W@785nm) was focused by a microscope objective (Nikon Plano, 40x, oil) onto a single cell using the arrangement illustrated in Figure 5. Excited Raman photons were then collected by a spectrometer formed with a monochromator (Andor Shamrock SR303i, 400 lines/mm grating @850nm) and a cooled CCD camera (Andor Newton, CCD3). With this configuration, the confocal diameter and depth of the system are 5μm and 4.68μm respectively. In order to get enough strong Raman signal from a single cell, we apply a laser dosage at 150mW onto the sample plane. Over a long period of exposure time, this focused laser power didn't show any damage to the cells. The system is able to switch from Raman spectroscopy to fluorescence imaging by placing FITC/TRITC cubes into the microscope and flipping the flip mirror (FM) to CCD1 (Hamamatsu ORCA-ER). Fluorescence images were taken via standard Nikon fluorescence cubes (FITC and TRITC) using a Nikon fluorescence white light source (Lamp2). Red fluorescence emission (580~620nm) of Nile red stained samples excited by a green wavelength (545~565nm) light shows the signals from the polar lipids whereas green fluorescence emission (500~540nm) excited by a blue wavelength (475~490nm) light shows the signals from the non-polar lipids (lipid bodies). WMR spectra In order to obtain the WMR spectrum from a single cell, five spectra were recorded continuously over 2.5 minutes with a 30 seconds integration time for each spectrum. During acquisition, the laser line was tuned continuously over a total modulation range of Δλ=1nm. A single WMR spectrum can be produced from these five spectra with all background fluorescence being removed essentially. In the WMR spectrum, all Raman peaks will locate at the zero crossings while their peak intensity will be reflected by the peak-to-valley value around the zero crossing. Raman calibration and spectra processing Raman spectra were taken from polystyrene beads (1um in diameter) as a control to calibrate the laser line and the drift in the system. The known Raman peak position (1001.4 cm-1) of polystyrene bead was used for calculate the laser line. In this way, the drift in the laser line can be monitored during the whole data acquisition procedure. Suppose the drift is very slow (typically <0.2nm over a day), we can calibrate the laser line used for each Raman spectrum from cells through an interpolation. Each Raman spectrum was also normalized by its total intensity (i.e. the integration over the area covered by the spectrum) in order to avoid any fluctuation in the laser power during wavelength modulation. The spectral region between 1000 cm-1 and 1800 cm-1 was used for the data analysis. Principal component analysis (PCA) WMR spectra were collected from 60~100 cells from each cell phenotype. PCA was then applied to these training dataset in order to reduce the dimensionality. To distinguish between each two different cell phenotypes, the first seven principal components were taken into account, as they accounted for more than 70% of variances in these training dataset. This algorithm was written in Matlab codes. Leave-one-out cross validation (LOOCV) The ability of distinguishing between each two different cell subsets was estimated by the method of leave-one-out cross validation. A multiple-dimensional space was defined by the principle components from the training dataset without one Raman spectrum. This leave-out spectrum was then classified in this space with the nearest neighbor algorithm. With this LOOCV method, correct or incorrect predictions for lipid rich and lipid poor cells in the training dataset were then used to estimate the specificity and sensitivity. Confusion matrix can also be estimated from correct and incorrect predictions for more than three cell subsets in the training dataset. This algorithm was written in Matlab codes. References 1 2 3 4 5 6 7 8 9 10 11 12 13 Gillespie, S. H. et al. Four-month moxifloxacin-based regimens for drug-sensitive tuberculosis. N. Engl. J. Med. 371, 1577-1587, doi:10.1056/Nejmoa1407426 (2014). Merle, C. S. et al. A Four-month gatifloxacin-containing regimen for treating tuberculosis. N. Engl. J. Med. 371, 1588-1598, doi:10.1056/NEJMoa1315817 (2014). Jindani, A. et al. High-dose rifapentine with moxifloxacin for pulmonary tuberculosis. N. Engl. J. Med. 371, 1599-1608, doi:10.1056/NEJMoa1314210 (2014). Phillips, P. P. et al. Limited role of culture conversion for decision-making in individual patient care and for advancing novel regimens to confirmatory clinical trials. BMC Med. 14, 19, doi:10.1186/s12916-016-0565-y (2016). Daniel, J. et al. Induction of a novel class of diacylglycerol acyltransferases and triacylglycerol accumulation in Mycobacterium tuberculosis as it goes into a dormancy- like state in culture. J. Bacteriol. 186, 5017-5030, doi:10.1128/Jb.186.15.5017-5030.2004 (2004). Garton, N. J. et al. Cytological and transcript analyses reveal fat and lazy persister-like bacilli in tuberculous sputum. PLoS Med. 5, e75, doi:10.1371/journal.pmed.0050075 (2008). Deb, C. et al. A novel in vitro multiple-stress dormancy model for Mycobacterium tuberculosis generates a lipid-loaded, drug-tolerant, dormant pathogen. PLoS One 4, e6077, doi:10.1371/journal.pone.0006077 (2009). Baek, S. H., Li, A. H. & Sassetti, C. M. Metabolic regulation of mycobacterial growth and antibiotic sensitivity. PLoS Biol. 9, e1001065, doi:10.1371/journal.pbio.1001065 (2011). Hammond, R. J., Baron, V. O., Oravcova, K., Lipworth, S. & Gillespie, S. H. Phenotypic resistance in mycobacteria: is it because I am old or fat that I resist you? J. Antimicrob. Chemother. 70, 2823-2827, doi:10.1093/jac/dkv178 (2015). Maquelin, K. et al. Identification of medically relevant microorganisms by vibrational spectroscopy. J. Microbiol. Methods 51, 255-271, doi:10.1016/S0167-7012(02)00127-6 (2002). Buijtels, P. C. A. M. et al. Rapid identification of mycobacteria by Raman spectroscopy. J. Clin. Microbiol. 46, 961-965, doi:10.1128/Jcm.01763-07 (2008). Pahlow, S. et al. Isolation and identification of bacteria by means of Raman spectroscopy. Adv. Drug Deliv. Rev. 89, 105-120, doi:10.1016/j.addr.2015.04.006 (2015). Kloss, S. et al. Destruction-free procedure for the isolation of bacteria from sputum samples for Raman spectroscopic analysis. Anal. Bioanal. Chem. 407, 8333-8341, doi:10.1007/s00216-015-8743-x (2015). 14 15 16 17 18 19 20 21 22 23 24 25 26 De Luca, A. C., Mazilu, M., Riches, A., Herrington, C. S. & Dholakia, K. Online fluorescence suppression in modulated Raman spectroscopy. Anal. Chem. 82, 738-745, doi:10.1021/ac9026737 (2010). Chen, M. et al. The use of wavelength modulated Raman spectroscopy in label-free identification of T lymphocyte subsets, natural killer cells and dendritic cells. PLoS One 10, e0125158, doi:10.1371/journal.pone.0125158 (2015). Mazilu, M., De Luca, A. C., Riches, A., Herrington, C. S. & Dholakia, K. Optimal algorithm for fluorescence suppression of modulated Raman spectroscopy. Opt. Express 18, 11382- 11395, doi:10.1364/OE.18.011382 (2010). Clark, S., Hall, Y. & Williams, A. Animal models of tuberculosis: Guinea pigs. Cold Spring Harb. Perspect. Med. 5, a018572, doi:10.1101/cshperspect.a018572 (2015). Galli, R. et al. Effects of tissue fixation on coherent anti-Stokes Raman scattering images of brain. J. Biomed. Opt. 19, 071402, doi:10.1117/1.JBO.19.7.071402 (2014). Evans, C. L. et al. Chemical imaging of tissue in vivo with video-rate coherent anti-Stokes Raman scattering microscopy. Proc. Natl. Acad. Sci. U.S.A. 102, 16807-16812, doi:10.1073/pnas.0508282102 (2005). Saar, B. G. et al. Video-rate molecular imaging in vivo with stimulated Raman scattering. Science 330, 1368-1370, doi:10.1126/science.1197236 (2010). Mukamolova, G. V., Turapov, O., Malkin, J., Woltmann, G. & Barer, M. R. Resuscitation- promoting factors reveal an occult population of tubercle bacilli in sputum. Am. J. Respir. Crit. Care Med. 181, 174-180, doi:10.1164/Rccm.200905-0661oc (2010). Davies, A. P. et al. Resuscitation-promoting factors are expressed in Mycobacterium tuberculosis-infected human tissue. Tuberculosis 88, 462-468, doi:10.1016/j.tube.2008.01.007 (2008). Gomez, J. E. & McKinney, J. D. M. tuberculosis persistence, latency, and drug tolerance. Tuberculosis 84, 29-44, doi:10.0.3.248/j.tube.2003.08.003 (2004). Sloan, D. J. et al. Pharmacodynamic modeling of bacillary elimination rates and detection of bacterial lipid bodies in sputum to predict and understand outcomes in treatment of pulmonary tuberculosis. Clin. Infect. Dis. 61, 1-8, doi:10.1093/cid/civ195 (2015). James, B. W., Williams, A. & Marsh, P. D. The physiology and pathogenicity of Mycobacterium tuberculosis grown under controlled conditions in a defined medium. J. Appl. Microbiol. 88, 669-677, doi:10.1046/J.1365-2672.2000.01020.X (2000). Lever, M. S., Williams, A. & Bennett, A. M. Survival of mycobacterial species in aerosols generated from artificial saliva. Lett. Appl. Microbiol. 31, 238-241, doi:10.1046/J.1365- 2672.2000.00807.X (2000). Clark, S. O., Hall, Y., Kelly, D. L. F., Hatch, G. J. & Williams, A. Survival of Mycobacterium tuberculosis during experimental aerosolization and implications for aerosol challenge models. J. Appl. Microbiol. 111, 350-359, doi:10.1111/j.1365-2672.2011.05069.x (2011). Hartings, J. M. & Roy, C. J. The automated bioaerosol exposure system: preclinical platform development and a respiratory dosimetry application with nonhuman primates. J. Pharmacol. Toxicol. Methods 49, 39-55, doi:10.1016/j.vascn.2003.07.001 (2004). 27 28 Acknowledgements This work was supported by the PreDiCT-TB consortium [IMI Joint undertaking grant agreement number 115337, resources of which are composed of financial contribution from the European Union's Seventh Framework Programme (FP7/2007-2013) and EFPIA companies' in kind contribution (www.imi.europa.eu)]. M.C. and K.D. thank the UK Engineering and Physical Sciences Research Council (Grant code EP/J01771X/1) and a European Union FAMOS project (FP7 ICT, 317744) for funding. K.D. acknowledges support from a Royal Society Leverhulme Trust Senior Fellowship and the loan of a laser from M Squared Lasers. This work was supported by the Department of Health, UK. The views expressed in this publication are those of the authors and not necessarily those of the Department of Health. We thank the staff of the Biological Investigations Group at PHE Porton for assistance in conducting the study and thank Dr. Peter Caie and Mrs In Hwa Um from School of Medicine in University of St Andrews for their help with the immuno-staining. Author contributions V.O.B. and M.C. contributed equally to this work. M.C. designed the optical system. V.O.B. prepared the samples. V.O.B. and M.C. performed the experiments and data analysis. V.O.B., M.C., K.D. and S.H.G. contributed to the development and planning of the project, interpretation and discussion of the data and the writing of the manuscript. S.O.C. and A.W. provided TB in-vivo modelling expertise and in-vivo samples and contributed to the writing of the manuscript. R.J.H.H. provided the mouse tissue infected with BCG and characterized the LP/LR cells using Nile red staining and helped with tissue sectioning and contributed to the writing of methods. Additional information Animal experimentation work was conducted in Public health England and the licensing of the experimental protocols was approved by the UK Home Office. Correspondence and requests for materials should be addressed to S.H.G. ([email protected]) or M.C. (mingzhou.chen@st- andrews.ac.uk). Competing financial interest The authors declare no competing financial interests.
1111.6454	1	1111	2011-11-28T14:46:56	Label-free bacteria detection using evanescent mode of a suspended core terahertz fiber	[ "physics.bio-ph", "physics.med-ph", "physics.optics" ]	We propose for the first time an E. coli bacteria sensor based on the evanescent field of the fundamental mode of a suspended-core terahertz fiber. The sensor is capable of E. coli detection at concentrations in the range of 104-109 cfu/ml. The polyethylene fiber features a 150 {\mu}m core suspended by three deeply sub-wavelength bridges in the center of a 5.1 mm-diameter cladding tube. The fiber core is biofunctionalized with T4 bacteriophages which bind and eventually destroy (lyse) their bacterial target. Using environmental SEM we demonstrate that E. coli is first captured by the phages on the fiber surface. After 25 minutes, most of the bacteria is infected by phages and then destroyed with ~1{\mu}m-size fragments remaining bound to the fiber surface. The bacteria-binding and subsequent lysis unambiguously correlate with a strong increase of the fiber absorption. This signal allows the detection and quantification of bacteria concentration. Presented bacteria detection method is label-free and it does not rely on the presence of any bacterial "fingerprint" features in the THz spectrum.	physics.bio-ph	physics	Label-free bacteria detection using evanescent mode of a suspended core terahertz fiber Anna Mazhorova,1 Andrey Markov,1 Andy Ng,2 Raja Chinnappan,2 Mohammed Zourob2,* and Maksim Skorobogatiy1,* 1École Polytechnique de Montréal, Génie Physique, Québec, Canada 2 Institut National de la Recherche Scientifique, Varennes, Québec, Canada *[email protected]; Abstract: We propose for the first time an E. coli bacteria sensor based on the evanescent field of the fundamental mode of a suspended -core terahertz fiber. The sensor is capable of E. coli detection at concentrations in the range of 104-109 cfu/ml. The polyethylene fiber features a 150 µm core suspended by three deeply sub-wavelength bridges in the center of a 5.1 mm-diameter cladding tube. The fiber core is biofunctionalized with T4 bacteriophages which bind and eventually destroy (lyse) their bacterial target. Using environmental SEM we demonstrate that E. coli is first captured by the phages on the fiber surface . After 25 minutes, most of the bacteria is infected by phages and then destroyed with ~1m-size fragments remaining bound to the fiber surface. The bacteria-binding and subsequent lysis unambiguously correlate with a strong increase of the fiber absorption. This signal allows the detection and quantification of bacteria concentration. Presented bacteria detection method is label-free and it does not rely on the presence of any bacterial “fingerprint” features in the THz spectrum. References and links 5. 4. 1. M. Zourob, S.Elwary, A. P. F. Turner (Eds.), Principles of Bacterial Detection: Biosensors, Recognition Receptors and Microsystems (Springer Science+Business Media, LLC, 2008). 2. N. Laman, S. SreeHarsha, D. Grischkowsky and J. S. Melinger, “High -resolution waveguide THz spectroscopy of biological molecules,” Biophys. J. 94, 1010-1020 (2008). 3. C. Markos, W. Yuan, K. Vlachos, G. E. Town, O. Bang, “Label-free biosensing with high sensitivity in dual-core microstructured polymer optical fibers,” Opt. Express 19, 7790-7798 (2011). http://www.opticsinfobase.org/abstract.cfm?URI=oe-19-8-7790 J. R. Ott, M. Heuck, C. Agger, P. D. Rasmussen, O. Bang, “Label-free and selective nonlinear fiber-optical biosensing,” Opt. Express 16, 20834-20847 (2008). http://www.opticsinfobase.org/oe/abstract.cfm?URI=oe-16-25-20834 J. F. O'Hara, R. Singh, I. Brener, E. Smirnova, J. Han, A. J. Taylor and W. Zhang, “Thin-film sensing with planar terahertz metamaterials: sensitivity and limitations,” Opt. Express 16, 1786-1795 (2008). http://www.opticsinfobase.org/oe/abstract.cfm?URI=oe-16-3-1786 6. T.Globus, T. Khromova, D. Woolard and A. Samuels, “THz resonance spectra of Bacillus Subtilis cells and spores in PE pellets and dilute water solutions,” Proc. SPIE 6212, 62120K (2006). 7. A. Bykhovski, X. Li, T. Globus, T. Khromova, B. Gelmont, D. Woolard, A. C. Samuels and J. O. Jensen, “THz absorption signature detection of genetic material of E. coli and B. subtilis,” Proc. SPIE 5995, 59950N (2005). 8. M. Walther, M. R. Freeman, F. A. Hegmann, “Metal-wire terahertz time-domain spectroscopy,” Appl. Phys. Lett. 87, 261107 (2005). 9. L. Cheng, S. Hayashi, A. Dobroiu, C. Otani, K. Kawase, T. Miyazawa, Y. Ogawa, “Terahertz-wave absorption in liquids measured using the evanescent field of a silicon waveguide,” Appl. Phys. Lett. 92, 181104 (2008). 10. B. You, T.-A. Liu, J.-L. Peng, C.-L. Pan, J.-Y. Lu, “A terahertz plastic wire based evanescent field sensor for high sensitivity liquid detection,” Opt. Express 17, 20675-20683 (2009). http://www.opticsinfobase.org/abstract.cfm?URI=oe-17-23-20675 11. P. Arora, A. Sindhu, N. Dilbaghi, A. Chaudhury, “Biosensors as innovative tools for the detection of food borne pathogens,” Biosens. Bioelectron. (to be published). 12. M.N. Velasco-Garcia, “Optical biosensors for probing at the cellular level: A review of recent progress and future prospects,” Seminars in Cell & Developmental Biology 20, 27–33 (2009). 13. S. K. Arya, A. Singh, R. Naidoo, P. Wu, M. T. McDermott, S. Evoy, “Chemically immobilized T4- bacteriophage for specific Escherichia coli detection using surface plasmon resonance,” Analyst 136, 486- 492 (2011). 14. Y.-S. Jin, G.-J. Kim, S.-G. Jeon, “Terahertz Dielectric Properties of Polymers,” Journal of the Korean Physical Society 49, 513 (2006). 15. A. Hassani, A. Dupuis and M. Skorobogatiy, “Porous polymer fibers for low-loss Terahertz guiding,” Opt. Express 16, 6340–6351 (2008). http://www.opticsinfobase.org/oe/abstract.cfm?URI=oe-16-9-6340 16. M. Rozé, B. Ung, A. Mazhorova, M. Walther, M. Skorobogat iy, “Suspended core subwavelength fibers: towards practical designs for low-loss terahertz guidance,” Opt. Express 19, 9127-9138 (2011). http://www.opticsinfobase.org/abstract.cfm?URI=oe-19-10-9127 17. A. Leung, P.M. Shankar, R. Mutharasan, “A review of fiber-optic biosensors,” Sensors and Actuators, B: Chemical 125 (2), 688-703 (2007). 18. J. Zhang, D. Grischkowsky, “Waveguide terahertz time-domain spectroscopy of nanometer water layers,” Opt. Lett. 29, 1617-1619 (2004). http://www.opticsinfobase.org/abstract.cfm?URI=ol-29-14-1617 19. L. Duvillaret, F. Garet, J.-L. Coutaz, “A reliable method for extraction of material parameters in terahertz time-domain spectroscopy,” IEEE Journal of Selected Topics in Quantum Electronics 2, 739-746 (1996). 20. Larry R. Engelking, Textbook of Veterinary Physiological Chemistry (Teton NewMedia, 2004) Chap.1. 21. W. Shua, J. Liu, H. Ji, M. Lu, “Core structure of the outer membrane lipoprotein from Escherichia coli at 1.9 å resolution,” Journal of Molecular Biology 299, 1101-1112 (2000). 22. A. Bykhovski, B. Gelmont, “The Influence of Environment on Terahertz Spectra of Biological Molecules,” J. Phys. Chem. B 114 (38), 12349–12357 (2010). 23. M. van Exter, Ch. Fattinger and D. Grischkowsky, “Terahertz time-domain spectroscopy of water vapor,” Opt. Lett. 14, 1128-1130 (1989). http://www.opticsinfobase.org/abstract.cfm?URI=ol-14-20-1128 24. R. A. Cheville and D. Grischkowsky, “Far-infrared foreign and self-broadened rotational linewidths of high-temperature water vapor,” J. Opt. Soc. Am. B 16, 317-322 (1999). http://www.opticsinfobase.org/abstract.cfm?URI=josab -16-2-317 25. J. R. Birch, J. D. Dromey and J. Lesurf, “The optical constants of some common low-loss polymers between 4 and 40 cm-1,” Infrared Phys. 21(4), 225–228 (1981). 26. J.E.K. Laurens, K.E. Oughstun, “Electromagnetic impulse response of triply-distilled water” in Proceedings of IEEE Conference on Ultra-Wideband Short-Pulse Electromagnetics 4 (Tel-Aviv , Israel, 1998), 243-264. 1. Introduction The selectivity and sensitivity of optical biosensors have put them into the class of most popular biosensors [1-4]. A large number of methods for highly sensitive chemical or biological detection that operate in the THz range were developed. Among these methods, metamaterials and frequency selective surfaces with a resonant frequency response that is tunable by design have been employed. It has been shown that even a small quantity of material deposited on THz metamaterial can shift its resonance frequency [5]. In addition, techniques based on THz vibrational spectra characterization have become available for biological materials in the form of the diluted water solutions and pressed pellets [6,7]. Recently, molecular spectroscopy of lactose dispersed on the top of metal wire was demonstrated [8]. A related technique [9] measured the absorption properties of liquids in the THz wave range, based on the interaction between the liquid and the evanescent wave covering the surface of a cylindrical waveguide made of high-resistivity silicon. A similar label-free THz sensing based on the plastic wires was also reported [10]. The present work is motivated by the recent reports on evanescent field or surface plasmon resonance (SPR) biosensors in the visible and near -IR regions [10,12]. Most of the investigations in these spectral regions focus on the detection of relatively low concentrations of proteins and small molecules with sizes about/less than tens of nanometers. The reason is that the probing length of an evanescent field of an optical waveguide, or the size of the tail of a plasmon that are used to detect changes in the optical properties at the inter faces, is deeply subwavelength. For example, a typical size of a plasmon tail in the visible range is smaller than 100 nm, which allows us to track the dynamics of a biochemical event in the immediate vicinity of the interface. However, when one tries to measure directly the presence of a bacteria on a metallic surface, the standard SPR method runs into a difficulty as the size of the probing field (plasmon) is not only smaller than the size of a bacteria, but is also comparable to the size of a phage molecule (~100 nm) that connects the bacteria to the surface. Therefore the detection of large target such as cells and bacteria by SPR methods remain a challenge. This results in the probing of a phage layer rather than a bacteria layer. Therefore, it is desirable to match the size of the measured object to that of a probing field. In the case of bacteria detection with a typical bacteria size of 1 -10 µm, one can pursue two approaches. One is to use short wavelengths (such as in the visible or near-IR) [13], and then use well delocalized evanescent field of a waveguide mode ( for example, mode of a subwavelength fiber) to cover the object. Another alternative is to use longer wavelengths (such as in the mid- or far-IR) and then match the extent of the modal evanescent field by the waveguide design. In this paper we explore the possibility of using THz fibers operating at wavelength longer than the size of the studied object. We use the high refractive index combination of core (polyethylene) and cladding (air) in order to significantly reduce the size of the probing evanescent tail. This allows us to develop a label-free technique (target bacteria are not labeled or altered and are detected in their natural forms) that is based on the interaction between the evanescent wave localized around the core of the fiber and a thin layer of bacteria covering it. This technique is then used for the detection and quantification of E. coli bacteria in the THz spectral range. In addition, we established a measurement protocol that allows measurement of biological properties in the aqueous environment, while using THz waves that are normally unsuitable for this purpose because of the strong absorption of water. This so-called “frozen dynamics” approach allows all the biological processes to take place in water; while the system is robust enough to allow drying for THz measurements at the intermediate stage without affecting the biological subcomponents (such as phages). 2. THz subwavelength fiber The major complexity in the design of the terahertz waveguides is the fact that most dielectric materials are highly absorbing in this spectral region [14] with a typical loss in excess of ~1 cm-1. To combat the material losses, a novel approach based on the introduction of porosity in the fiber core has been recently introduced by our group [15]. We have proposed that since the absorption loss is lowest in dry air, one way to reduce waveguide propagation loss is to maximize the fraction of power guided in the air. Fig. 1. (a, b) The THz fiber featuring a 150 µm core suspended by three 20 µm-thick bridges in the center of a 5.1 mm diameter tube, (c) 4 cm-long fiber piece used in the experiments. One of the waveguides that operates on this principle was recently described in [16] and presents a subwavelength fiber suspended on thin bridges in the middle of a larger protective tube (see Fig. 1). Large channels formed by the bridges and a tube make a convenient opto - microfluidic system that is easy fill with liquid analytes or purge with dry gases. Particularly, the THz subwavelength waveguide used in our experiments features a 150 µm core fiber suspended by three 20 µm-thick bridges in the center of a 5.1 mm diameter tube (Fig. 1. (a, b)) of 4 cm in length. This waveguide design presents several important advantages for bio- sensing applications. First, the waveguide structure allows direct and convenient access to the fiber core and to the evanescent wave guided around it. Second, the outer cladding effectively isolates the core-guided mode from the surrounding environment, (e.g., fiber holders), thereby preventing the undesirable external perturbations of the terahertz signal. All the data in our experiments was acquired using a modified THz-TDS (Time-Domain Spectroscopy) setup. The setup consists of a frequency-doubled femtosecond fiber laser (MenloSystems C-fiber laser) used as a pump source and identical GaAs dipole antennae used as source and detector yielding a spectrum ranging from ~0.1 to 3 THz. Contrary to the standard THz-TDS setup where the configuration of parabolic mirrors is static, our setup has mirrors mounted on translation rails. This flexible geometry facilitates mirrors placement, allowing measurement of waveguides up to 45 cm in length without realigning the setup. Fig. 2(a) illustrates the experimental setup where the fiber is fixed between the two irises and placed between the focal points of the two parabolic mirrors. Fig. 2. (a) Schematic of the experimental setup: fiber is placed between the focal points of the two parabolic mirrors, (b) schematic presentation of phages adsorbed onto the fiber core. The capsid adsorbed onto the fiber, while the tail (which is specific to the bacteria) faces towards the cladding for bacteria capturing. 3. Materials and methods 3.1 Bacteria culture Bacteria solution was prepared as following. A frozen stock of E. coli B strain was used to seed an overnight culture in LB media (liquid media for growing bacteria, contains 10g Tryptone, 5g yeast extract, 10 NaCl in 1L H2O). Bacteria were harvested by centrifugation at 3000xg for 10 min, followed by washing in PBS (Phosphate Buffer Saline). Dilutions of the overnight culture were spread on LB agar plate to determine the titre, expressed in colony forming unit (cfu). Appropriate dilutions of bacteria stock with a concentration 10 4 – 109cfu/ml were made in PBS for subsequent experiments. 3.2 Phage production 100uL of E.coli B log-phase culture was added to 3ml cooled top agar and poured onto LB agar plate until solidified. 100uL of phage stock specific to E.coli B was added onto the solidified top agar and incubated at 37ºC overnight. A macroplaque was developed on the lawn of bacteria after incubation. 3ml of lambda buffer was added onto the top agar, which is then scraped off and collected in a 50ml centrifuge tube, followed by an additional washing of the LB agar plate with 3ml lambda buffer. 3 drops of chloroform were added into the suspension, vortexed and centrifuge at 3500xg for 10min to pellet pieces of agar and bacteria. The supernatant was then passed through a 0.22um filter to remove the remaining bacteria. The filtrate is the stock phage solution and is prepared fresh for each experiment. The titre of the phage stock was determined by serial dilution of the stock, followed by infection to E.coli as described above. Plaques were counted and the titre of the phage stock is expressed in plaque forming unit (pfu). 4. Characterization The light propagating through an optical fiber consists of two components: the field guided in the core and the exponentially decaying evanescent field in the cladding. For evanescent sensing one typically removes the fiber cladding and introduces analyte into the immediate proximity of the fiber core [17]. In the presence of strong absorption or scattering in the analyte, the intensity of guided wave would decrease, which can then be then used to interpret various analyte properties. In application to THz evanescent sensing, both analyte refractive index and analyte absorption can be deduced by comparing the signals from the empty waveguide with that from the waveguide containing a layer of the measured material. By measuring the changes in both amplitude and phase caused by the addition of the analyte layer even very small amounts of analyte can be detected and characterized (see, for example, [18] where nanometer-size aqueous layers were characterized in THz). An immediate complication for evanescent biosensing in THz is that natural environment for a majority of bacteria is water and water is highly absorbing in THz spectral range. To avoid this problem, we have developed a protocol where all the bio-reactions take place in the natural aqueous environment, while the THz measurement is done on dried samples . Fig. 3. SEM images illustrating each step of the experiment (a) step 1- phages are immobilized on the fiber core surface, (b) step 2 – capturing of E.coli bacteria by the phages, and lysis of the bacteria (c) step 3 - fiber is washed with PBS, bacteria chunks remain bound to the core surface. In what follows all the SEM images were taken with a Quanta 200 FEG scanning electron microscope operating in high vacuum mode. First, samples were fixed with formalin for 12 hours and then metalized with several nm-thick layer of gold. To follow the dynamics of the bacteria binding we performed several experiments according to the three-step procedure described below. Before the measurement, the suspended core fiber (4 cm in length) was cleaned with isopropyl alcohol and distillated water. The fiber was incubated overnight (12 hours) with T4 -bacteriophage solution at 1010 pfu/ml concentration. During this incubation, phages are adsorbed non-specifically onto the plastic fiber surface. The T4 phages recognize and bind to specific cell surface protein on the E.coli bacteria using their tail proteins (scheme of the process is shown at Fig. 2 (b)). This recognition is highly specific and has been extensively used for typing of E. coli. After phage adsoprtion the fiber was washed 3 times with distillated water and PBS. Only 10% of the fiber’s bridges surface and core were covered by phages. The exposed fiber surface was therefore blocked with bovine serum albumin (BSA) for 30 min in order to reduce non-specific adsorption of the bacteria, and washed again with the buffer solution and dried out. Blocking with BSA is required because it is not possible to achieve 100% coverage of the fiber by the phages. Without BSA blocking, exposed fiber surface could cause non -specific adsorption of the bacteria and other components present in the sample, causing false positives. SEM imaging was performed to confirm the adsorption of the phages onto the fiber core as well as the stability of the phage layer, and the extent of fiber surface coverage. Fig. 3 (a) illustrates phages adsorbed onto the fiber (dark spots within the area define by dashed line). At the same time, THz transmission through the fiber was measured. Fiber transmission spectrum is shown in Fig. 4 (a), black line. Fig. 4. (a) Transmission spectrum of the fiber during each step of the experiment: step 1- black line, only phages, step 2 – red line, transmission of the fiber decreased due binding of E. coli bacteria to the phages, step 3 – blue line, fiber is washed with PBS. Fiber transmission increased but not up to the level of the first step, suggesting that some bacteria (or parts of it) remain bound to the fiber via specific interaction with the phages.(b) difference in transmission between each step of the experiment. After addition of the bacteria and subsequent binding of the phage on the bacte rial surface, the phage infection process will start. Phage progeny will accumulate and eventually the bacteria host is ruptured followed by the release of the phage. The period between infection and bacteria lysis is known as the latent period and usually last for 25 minutes. The second measurement of the THz transmission through the fiber was done during the first 15 min after the introduction of bacteria (Fig. 4 (a), red line). It could be clearly seen that the transmission of the fiber decreases considerably due to binding of the bacteria to the phages. A zoom-in of the waveguide core after some reaction time ~5-10min is shown in Fig. 5. There, the dashed line circumscribes the part of a core covered with the bacteriophages; E. coli Bacteria are clearly visible bound to the region covered by the phages and not to the rest of the core surface blocked with BSA. Fig. 5. Zoom-in of the waveguide core, dashed line marks the area covered with bacteriophages. Bacteria are clustered in the phage-covered surface with the rest of the surface blocked by BSA. In the last step, after 25 min of reaction time, the fiber was washed with the PBS in order to remove the unbound bacteria on the fiber surface and then again dried out. Fiber transmission (Fig. 4 (a), blue line) increased compared to the previous step, but not up to the level of the first step, suggesting that some bacteria (or parts of them created by lysis and rapture) remain bound to the fiber via specific interaction with the phage. The following SEM images (Fig. 6) were prepared at different moments of time during the process of specific recognition and binding of the E. coli bacteria with phages in order to understand the process that take place during the last 10 min of the 2nd step of the measurements. Fig. 6. SEM images of the fiber core (a) after 20 min since the beginning of the 2nd step; (b) after 30 min before washing with PBS (end of the 2nd step); (c) after 30 min after washing with PBS. The bacteria shape changed from a uniform rod shape to a random shape. Eventually, the bacteria cell wall ruptures and releases intracellular components with only the cell membrane left on the fiber. Fig. 6 (a) shows fiber core after 20 min of interaction of E. coli bacteria with the phages. The bacteria are bound to the phages and have well defined rod shapes. Fig. 6 (b) shows the fiber core at the end of the 2nd step, after 30 min of bacteria interaction with pages. It is clearly seen that by the end of the 2nd experiment the bacteria change its shape form the rod-like to a highly distorted rod shape which is a result of bacteria lysis . During lysis the bacteria cell wall eventually ruptured and releases intracellular components with only the cell membrane left on the fiber. Fig. 6 (c) illustrates the core of the fiber after 30 min of interaction of the bacteria with the phages and after washing with PBS. After the fiber was washed, traces of the bacteria (membrane fragments) can be clearly seen still bound to the phages. Visually, washed fibers corresponding to Fig. 6 (c) look cleaner than the same fibers before washing. In order to explain a considerable increase in the fiber transmission loss due to bacteria binding (during the course of the 2nd measurement) we note that E. coli consists largely of water. Therefore, it is expected that its mobilization on the core surface would leads to a significant loss increase due to water absorption. Particularly, cell dry weight ("dry weight" is the remaining material after all the water is removed) of the E. coli bacteria is only 30% and can be subdivided into protein (55%), RNA (22.5%), DNA (3.1), lipids (9.1%), peptidoglycan (2.5%), polysaccharides (3.4%) and Glycogen (2.5%), while remaining 70% of the bacteria weight is water [20]. Additionally, we expect that presence of the large and irregular bacteria clusters on the fiber surface would lead to scattering loss. Similarly, after washing the fiber at the end of the 2 nd measurement, fiber transmission is expected to increase as only the dry chunks of bacteria membrane remain on the surface. Moreover, their surface coverage is smaller than this of the bacteria before washing, therefore scattering loss is expected to be smaller in washed fibers. For completeness, we note that the outer membrane of the gram-negative bacteria such as E. coli is a complex structure composed of phospholipids, proteins, and polysaccharides that controls the permeability and helps maintain the shape and rigidity of the cell. The protein complement of the outer membrane includes a Murein-lipoprotein that is one of the most abundant proteins in bacteria: there are about 7.2 105 molecules per cell. Finally, from Fig. 4 we observe that there are two spectral regions in the vicinity of 0.7 THz and in the range of 1.5-2.0 THz where changes in the fiber absorption are especially pronounced. These particular spectral regions of high sensitivity were consistently identified in each of the several repetitions of our experiments. The physical nature of high absorption in these spectral regions is still unclear to us, while there is an indication that it can be related to spectral signatures of specific macromolecules abundant in the E. coli bacteria composition. In particular, in the THz spectral range large tandem repeats peptides or nucleosides have some observable spectral features, whereas nonperiodic biological molecules such as DNA and proteins have relatively featureless spectra [2]. Optimal regions for the THz detection of E. coli bacteria found in our work correlate with the position of the absorption peaks of the Murein-lipoprotein or Broun’s lipoprotein (BLP), that is one of the most abundant proteins in bacteria [21,22]. Additionally, there is a 0.68 THz absorption peak of Thioredoxin protein which a commonly occurring electron transport protein in a bacteria. For the sake of clarity we also note that all the measurements were done in the nitrogen purged cage at 12% of humidity, therefore all the narrow dips in the transmission spectrum as observed in Fig. 4 (for example the 0.752 THz peak) are caused by the water vapor [23,24] and has to be distinguished from the broad absorption features due to bacteria . 4. Results and Discussion In this section we present and interpret changes in the fiber absorption loss during bacteria binding process. Fiber absorption loss is defined as where is the fiber length, is the transmitted power through the setup without fiber, and is a transmitted power through a setup with fiber . To see the influence of bacteria binding on the THz transmission of a suspended core fiber in dynamics we measure absorption losses of the fiber (with a dry sample) as a function of time. Particularly, during each step of the experiment we did at least 10-15 scans of the transmitted spectrum (one scan takes ~1 min), as a result, we have absorption losses of the fiber as function of time. Also, as it was mentioned before, the THz measurement is done with dried samples, while the bio -reactions take place in the natural aqueous environment. In particularly, after 1 st step of the experiment fiber was incubated with bacteria aqueous solution and then after 15 min (during this time phages have already captured bacteria) fiber was dried out for the transmission measurements. T his gives us ~15 min gap in the sensogram where it was impossible to measure losses due to high water absorption in THz spectral range. Fig. 7. Absorption losses of the fiber. (a) reference sensorgram; (b) sensorgram for bacteria concentration at 106 cfu/ml as a function of time; (c) Correlation between the changes in the fiber absorption losses and bacteria concentration: difference between base level (absorption losses of the fiber with phages immobilized on the fiber core) and losses of the “washed” fiber is shown in blue triangles, as a function of a concentration. Black dots correspond to the difference between absorption losses of the fiber after 2nd and 3rd steps (before and after fiber washing). Figure 7 illustrates effect of the bacteria binding on the THz transmission of a suspended core fiber at 0.7 THz. First the reference sensorgram was acquired (Fig. 7(a)) for which the fiber was prepared exactly the same way as described in the characterization section, however, no phages were used (phage solution was simply substituted with buffer during the 1st step), while the fiber surface was blocked with BSA to prevent non-specific bacteria binding to the core surface. Absorption losses returned to the initial level after bacteria were washed away with PBS. This transmission measurement and SEM images (not showed here), prove that the fiber surface was indeed blocked with Bovine Serum Albumin and no non-specific adsorption of the bacteria took place on the fiber surface. In Fig. 7(b), results of the experiments with phages are presented where despite extensive washing, some bacteria are always retained by specific binding to the phage -coated fiber, and as a consequence, fiber absorption loss does not return to the original level. More importantly, ln()refPPLLrefPPwe are able to show the correlation between the fiber absorption losses and the bacteria concentration of the samples, further suggesting the specificity of the detection method . Thus, in Fig. 7 (c) difference between the base level (absorption losses of the fiber with phages immobilized on the fiber core) and losses of the “washed” fiber s are shown in blue triangles, as a function of bacteria concentration in the solution used during the 2nd experiment. Also, in black circles we show difference between absorption losses of the fiber after 2nd and 3rd steps (before and after washing of the fiber). This difference is related to a signal due to both specific bacteria binding to the phages and nonspecific temporary adsorbed bacteria onto a fiber surface so by itself it cannot be used for sensing. From the presented data (blue triangles) we see that the detection of limit of this method is around 104 cfu/ml, which is comparable with existing commercially available methods [13] The commercial methods usually employ antibodies (Abs) as recognition elements instead of bacteriophages. Production of the specific Abs, however, is difficult, expensive and very time -consuming. Also, antibodies easily lose their activity when subjected to changing environmental conditions. Bacteriophages are alternative method of bacteria capture that offer many advantages compared to antibodies. Particularly, phages are easy to produce in large quantities at a relatively low cost, they have long storage life and unlike antibodies can be physically adsorbed onto the fiber surface. Considerable sensitivity enhancement of our method is possible by increasing the amount of the initial phage coverage of the fiber surface by better anchoring the phages via covalent immobilization onto the fiber surface, rather than physical adsorption. This however implies considerable effort in fiber surface biofunctionalisation chemistry which is beyond the scope of this paper. 5. Simple theoretical model to explain changes in the fiber absorption loss We believe that strong changes in the fiber transmission loss during the 2 nd phase of the experiment (before washing the fiber) can be explained by the presence of water containing bacteria on the fiber core due to non-permanent adsorption or sedimentation during rapid drying of samples. Fig. 8. SEM images of the fiber core with bacteria concentration of (a) 109 cfu/ml and 10 min interaction time (b) with a concentration of 106 cfu/ml, 10 min interaction time. As seen in Fig. 7(c) these losses are proportional to the bacteria concentration. Moreover, from the direct observation of the SEM images, bacteria coverage of the fiber core is also proportional to the bacteria concentration. For example, at 109 and 106 cfu/ml, the average coverage was correspondingly 4.35% and 1.52% (Fig. 8. (a,b). These values were achieved after analyzing presented SEM images via custom LabVIEW software. First we determined the ratio of the area covered with the bacteria (in pixels) to the total area of the picture in percentage. At the same time total number of the bacteria in the field of view was also counted, which divided by the area of the SEM image yields the coverage ratio of 45 cells/100 µm2 for the concentration 109 cfu/ml and 15 cells/100 µm2 for the concentration 106 cfu/ml. Despite the fact that at concentration 109 cfu/ml we have 1000 times more bacteria then at the concentration 106 cfu/ml, we don’t have a significant increasing of the bacteria coverage ratio due to limitation of the initial phage coverage of the fiber surface (only 10%) and number of phages required for the bacteria binding and subsequent lysis. As we will see in the following, it is the presence of the water containing bacteria on the fiber surface that leads to strong increase in the fiber absorption before washing. After washing, only dry bacteria chunks remain bound at the fiber surface and, therefore, fiber loss decreases. Thus, the additional loss after washing is most probably due to scattering on the bacteria remains. To create theoretical model of the experiment we first studied the modal structure of our fiber. We started by importing the fiber cross-section as captured by the optical microscope Fig. 1. (a) into COMSOL Multiphysics Finite Element software. Complex effective refractive indices and field profiles of both core guided and cladding modes were then studied. Then, the power coupling coefficients have been computed from the overlap integrals of the modal fields with those of the Gaussian beam of a source as described in [16]. Inspection of the coupling coefficients confirms that the fundamental mode is predominantly excited in the 0.2 – 2.0 THz frequency range. Transverse distribution of the longitudinal power flux in the fundamental mode of a fiber at 0.71THz is shown in Fig. 9 (a) and a typical penetration depth of the evanescent field into the air cladding is determined to be which is in the good agreement with a theoretical estimate . Next, absorption losses of a suspended core fiber with a bacteria layer have been investigated numerically using COMSOL. Simplified fiber geometry has been used to facilitate the task of distribution of bacteria on the core and characterization of the bacteria coverage ratio. We have reduced the transverse shape of the core to a circle and we have assumed that the bridges are straight and of constant width. The values of the waveguide core radius and the bridges width were chosen to make the changes in the effective refractive index, the absorption losses and the modal size minimal compared to the microscope imported . design, and, hence, for simulation we used and Fig. 9. Transverse distribution of power flow for the fundamental mode of (a) the real waveguide and of (b) the simplified design of the waveguide profile. The field is confined in the central solid core and is guided by total internal reflection. E. coli are typically rod-shaped, about long and in diameter. To simulate bacteria we have approximated their shape as an infinite cylinder of diameter positioned along the fiber length. Considering that bacteria are mostly made of water (70% by total wet wt. [20]) we considered their refractive index and losses as those of the bulk water. The complex dielectric permittivity of water has been calculated using the full Rocard- Powles-Lorentz model with the parameters taken from [26]. For 0.71 THz the water refractive index is 2.19+0.65i resulting in absorption loss of 191 cm-1. For low density polyethylene the real part of the refractive is about 1.514 in THz frequency range and the material loss are is the where computed from the measurements [25] frequency in THz. 255m0.71/429corecladdingTHzm175coreDm30bridgesdm2.0m0.5m1bacteriadm120.750.10.2cmFig. 90. (a) Randomly distributed 1 µm water cylinders on the waveguide core’s surface for a given value of surface coverage. (b) Absorption loss of the fundamental mode for the fiber with a bacteria layer as a function of the bacteria coverage ratio. The crossing of the dash lines in the figure corresponds to the experimentally measured value of the propagation losses minus the theoretically estimated coupling loss for the 109 cfu/ml concentration of bacteria. From the SEM images of the fiber surface with bacteria (see Fig. 8) we see that bacteria are randomly positioned on the core surface with relatively low concentrations. To take this into account in our simulations the water cylinders (bacteria) were randomly distributed (see Fig. 10 (a)) on the fiber core surface with a fixed value of a coverage ratio (i.e., the core’s surface occupied by the water cylinders to the overall core’s surface). Then, the fundamental mode and its losses were found using FEM software, and the modal losses were then averaged over 10 random bacteria distributions. In Fig. 10 (b) we present thus calculated propagation losses of the fiber fundamental mode as a function of the bacteria coverage ratio. We observe that increase in the bacteria coverage ratio leads to a linear growth of the modal absorption losses. Due to much higher absorption losses of water compared to those of a polyethylene core, even a small coverage ratio of the bacteria results in large increase in the modal absorption loss. As an example, in Fig. 10 (b) we show at the intersection of the dashed lines the case of bacteria of concentration 109 cfu/ml that corresponds to the experimental fiber loss of ~1.2cm-1 at 0.71THz. Our theoretical analysis predicts that this loss is produced by the absorption loss of the fundamental modes due to 4% coverage ratio of water on the core’s surface multiplied by the modal coupling coefficient from the Gaussian beam to the fundamental mode, which corresponds well with the experimental value of the bacteria coverage ratio 4.35% found from the SEM observations Another important feature found in Fig. 10 (b) is an observation that for a given water coverage ratio, fiber absorption loss is small at low frequencies, then shows a rapid increase in the vicinity of 0.5THz, and, then, at frequencies above 0.7THz it simply follows the absorption loss of the polyethylene material with an almost constant frequency independent loss contribution from the water layer. This is easy to rationalize by noting that at low frequencies (<0.3THz in our case) mode of a subwavelength fiber is strongly delocalized, and, therefore, its presence in the thin water layer around the core is minimal. When frequency increases (~0.5THz in our case), fiber mode shows rapid localization in the vi cinity of a core region, and, therefore, its relative presence in the water layer and, hence, losses also rapidly increase. This also explains why in Fig. 4, there is little difference in the measured transmission losses of bare and bacteria activated fibers at frequencies below 0.5THz. Finally, at even higher frequencies (above 0.7 THz in our case) fundamental mode becomes mostly localized in the fiber core. Absorption loss contribution due to a small water layer on the fiber surface decreases and becomes almost independent of the operation wavelength because the decreasing ratio of power inside water layer is compensated by the increasing with frequency water absorption losses. Finally, at even higher frequencies (above 0.7 THz in our case) fundamental mode becomes mostly localized in the fiber core of radius R with a fraction of power in thin layer (of thickness d) on the core surface being . Here we assume that the layer thickness is smaller than the evanescent field penetration depth into the air cladding 23~/dR, which is a true assumption all the way up to 10 THz. Considering that the bulk absorption loss of water increases with frequency [25] we conclude that at higher frequencies absorption loss contribution due to a small water layer on the fiber surface becomes a slowly decreasing function of the operation frequency , which should be compared to the behavior of the fiber loss due to polyethylene absorption (see Fig. 10). 6. Conclusion We demonstrate the possibility of using suspended core (d=150 µm) polyethylene THz fibers for sensing of the E. coli bacteria with detection limit of 104 cfu/ml. Structure of the suspended core fiber allows convenient access to the fiber core and to the evanescent part of the guided wave. Outer cladding effectively isolates the core-guided mode from interacting with the surrounding environment such as fiber holders and humid air, thus making the fiber convenient to handle and to operate. It was shown that selective binding of the E. coli bacteria to fiber surface bio- functionalized with specific phages unambiguously influences the THz transmission properties of the suspended core fiber. Moreover, changes in the fiber absorption loss can be correlated with the concentration of bacteria samples. Thus, our setup allows not only detection of E. coli, but also quantitative measurement of its concentration. Presented bacteria detection method is label-free and it does not rely on the presence of any bacterial “fingerprint” features in the THz spectrum. Simple theoretical model was also developed in order to explain observed changes in the fiber absorption losses. It was found that strong increase in the fiber loss can be explained by absorption of a thin water layer corresponding to the bound bacteria. 4corecladd~3~/dR2~
1105.2517	1	1105	2011-05-12T16:03:52	X-ray phase contrast imaging of biological specimens with tabletop synchrotron radiation	[ "physics.bio-ph", "physics.acc-ph", "physics.med-ph", "physics.plasm-ph" ]	Since their discovery in 1896, x-rays have had a profound impact on science, medicine and technology. Here we show that the x-rays from a novel tabletop source of bright coherent synchrotron radiation can be applied to phase contrast imaging of biological specimens, yielding superior image quality and avoiding the need for scarce or expensive conventional sources.	physics.bio-ph	physics	X-ray phase contrast imaging of biological specimens with tabletop synchrotron radiation S. Kneip1,2, C. McGuffey2, F. Dollar2, M.S. Bloom1, V. Chvykov2, G. Kalintchenko2, K. Krushelnick2, A. Maksimchuk2, S.P.D. Mangles1, T. Matsuoka2, Z. Najmudin1, C.A.J. Palmer1, J. Schreiber1, W. Schumaker2, A.G.R. Thomas2, V. Yanovsky2 1The Blackett Laboratory, Imperial College London, London, SW7 2BZ, UK 2Center for Ultrafast Optical Science, University of Michigan, Ann Arbor, MI, 48109, USA Since their discovery in 1896, x-rays have had a profound impact on science, medicine and technology. Here we show that the x-rays from a novel tabletop source of bright coherent synchrotron radiation can be applied to phase contrast imaging of biological specimens, yielding superior image quality and avoiding the need for scarce or expensive conventional sources. The x-ray tube is still the most common source of x-rays1, but cutting-edge medical applications demand high quality beams of x-rays, the production of which often requires large, expensive synchrotron x-ray facilities2. Our scheme deploys a recently demonstrated tabletop source of 1-100 keV synchrotron radiation3. Here, in order to produce high quality x-rays, electrons are accelerated and wiggled analogously to a conventional synchrotron, but on the millimeter rather than tens of meter scale. We use the scheme to record absorption and phase contrast images of a tetra fish, damselfly and yellow jacket, in particular highlighting the contrast enhancement achievable with the simple propagation technique of phase contrast imaging. X-rays have a much shorter wavelength than visible light, can penetrate matter and image the interior of solid objects. Image contrast is obtained through preferential absorption of x-rays in dense regions of the sample (absorption contrast)1, or through bending of the wavefront in regions of the sample with differing refractive index decrement (phase contrast)4. For medical radiography, an absorption contrast contact image is recorded of the specimen, which is in contact with the detector. High contrast is limited to dense tissue such as bones. Image contrast from low-density tissue can be enhanced using radioopaque and radiolucent agents, but administration is invasive and/or limited to certain applications5. Phase sensitive radiography sidesteps the need for radioopaque agents to visualize soft tissue, and has the additional benefit that the dose can be reduced by using harder, more transmissive x- rays6. The huge potential of phase contrast imaging for medical diagnostics has been realized for quite some time, documented by numerous animal studies7-9. Progress has been held back due to the lack of suitable x-ray sources with the necessary spatial coherence properties and/or due to the requirement for cumbersome imaging techniques. Interferometric and refractive techniques require optics, are limited in their field of view, e.g. by requiring monolithic crystals, and/or require scanning data techniques6, 10, 11. Spatial coherence Ltrans increases proportionally with the distance from the source u and with the inverse of the source size w. Conventional x-ray tubes must therefore be apertured, increasing the required exposure time. Development in microfocal x-ray tubes offers improvements, but brightness is limited due to anode material and cooling. With the advent of high power lasers, various schemes have been studied for phase contrast imaging12-14. Conventional synchrotrons continue to be the ideal source for phase contrast imaging2 but access is limited. As shown in figure 1a, the x-ray source discussed here is based on focusing a pulsed high power laser into a millimeter-sized plume of helium gas, which is immediately ionized and turned into a plasma (see methods). Within the plasma, electrons are accelerated and wiggled analogously to a conventional synchrotron. The observation of an x-ray beam, originating from the interaction and pointed along the laser direction is correlated with the electron beam. Depending on experimental parameters, the x-ray beam divergence is measured to be 5−15 mrad, the 1/e2 x-rays intensity source size is 1 − 3 µm and the spectrum is synchrotron like with average photon energy (critical energy) of Ecrit ≃ 10-40 keV. Each laser pulse delivers a 30 fs flash of ≃106 photons mrad−2. This corresponds to a peak brightness of 1022 photons per (second mm2 mrad2 0.1%BW), which is comparable to conventional 3rd generation synchrotrons and makes possible high contrast imaging in a single shot. More details of the experiment are described elsewhere3. Due to the small source size, the x-ray beam has an appreciable degree of spatial coherence. This enables phase contrast imaging with a simple propagation technique (see methods). To achieve contrast enhancement with the propagation technique, the image distance (specimen to detector) has to be increased2. With figure 1b-d we demonstrate that this x-ray source can produce detailed radiographs of biological specimens with a single 30 fs exposure. Contact images of a tetra fish (figure 2b) and damselfly (figure 2c-d) were taken at a distance of v=2.79 m from the source, where the specimen is in contact with the detector at u=v=2.79 m from the source. Absorption contrast from the absorbing skeleton of the tetra is good but absorption contrast from the non-absorbing thin wings and uniformly absorbing exoskeleton of the damselfly is poor. Using the propagation technique, with the damselfly at u=0.44 m and the detector at v=1.83 m from the source, the path- integrated phase change of the spatially coherent x-rays through the sample leads to enhanced edge contrast, revealing the fine structure of the wing, the legs and exoskeleton, as shown in figure 1d. The ratio of the cross section for phase shift to absorption is greater than 100 for 18 keV x-rays for Z<156. The transverse coherence length Ltrans is approximately 10 µm for 10 keV x-rays 0.5 m from a 1 µm source. The figure 3b shows a phase contrast enhanced image of a yellow jacket taken in a single shot 30 fs exposure in the same configuration as figure 1d. Resolution of the contact radiographs is limited by the 13 µm pixel size of the ccd detector which is much larger than the x-ray source size. Fine details of the skeleton and fins of the fish can be noticed in figure 1b and an intensity lineout across the caudal fin (shaded yellow box in figure 1b) is plotted in figure 2a. The spines and rays of the fin are resolved, demonstrates imaging resolution of at least ≃120 µm. Due to the increased image distance necessary for propagation phase contrast imaging, the ccd detector is recording a M=v/u≃4.2 magnified image of the specimen, when compared to the contact images. This has two consequences. Firstly, to capture the entire specimen, multiple sub-images have to be joined, as indicated by the red dashed lines in figure 2. Secondly, the effective detector pixel size 13 µm /4.2=3.1 µm is now on the order of the x-ray source size, permitting a greater image resolution. Thus, to allow for a fair comparison of image quality obtained with absorption and phase technique, we have artificially increased the pixel size of the phase image of figure 1d to 13 µm. Intensity lineouts across the leg of the damselfly (shaded yellow box in figure 1c,d) are plotted in figure 2a. This demonstrates clear edge enhancement and imaging resolution at the few pixel level (≃70 µm) for the phase contrast image 1d. When printing a single shot exposure raw data image with pixel size left unaltered, even greater detail of the specimen, reminiscent of the compound eye, can be seen in figure 2b. The demonstrated tabletop source of synchrotron radiation may also be suitable for lensless imaging and tomographic reconstruction of 3D phase and absorption information9. This would to be the broad synchrotron spectrum require monochromatized. Laser developments proposed in the near-future (e.g. diode pumping) may enable the repetition rate of our system to be increased from 0.1 to 100 Hz. This can compensate any loss of average brightness incurred through monochromatization. The flexibility of a laser-based source combined with a precise narrow x-ray beam would facilitate scanning imaging or multiple views without moving the object. The demonstrated scalability to hard x-rays (>50 keV, ref [3] and therein) complements existing sources and opens the possibility to even do phase contrast imaging of dense objects such as bones15. Another unique benefit of this source is the ultrashort pulse duration which can freeze motion blur (the heart beat of a mouse is 10 Hz) and also allow direct study on the timescale of molecular interactions. The absolute time synchronization can enable optical pump-probe experiments. Hence with further development, this tabletop synchrotron source will offer a cheap and compact route to make advanced imaging 1. Roentgen, W.C. Nature 53, 274-‐276 (1896). schemes more commonly available. Suortti, P. & Thomlinson, W. Phys. Med. Biol. 48, R1 (2003). 2. 3. Kneip, S. et al. Nat. Phys. 6, 980-‐983 (2010). 4. Bonse, U. & Hart, M. Appl. Phys. Lett. 6, 155 (1965). 5. Speck, U. X-‐ray Contrast Media: Overview, Use and Pharmaceutical Aspects, Edn. 2nd. (Springer, 1991). 6. Momose, A. Jpn. J. Appl. Phys. Part 1 -‐ Regul. Pap. Brief Commun. Rev. Pap. 44, 6355-‐6367 (2005). Momose, A., Takeda, T., Itai, Y. & Hirano, K. Nat Med 2, 473-‐475 (1996). 7. 8. Wu, J. et al. Kidney Int 75, 945-‐951 (2009). 9. Dierolf, M. et al. Nature 467, 436-‐U482 (2010). 10. Davis, T.J., Gao, D., Gureyev, T.E., Stevenson, A.W. & Wilkins, S.W. Nature 373, 595-‐598 (1995). 11. Pagot, E. et al. Phys. Med. Biol. 50, 709-‐724 (2005). 12. Chen, M.C. et al. Phys. Rev. Lett. 105, 173901 (2010). Toth, R., Kieffer, J.C., Fourmaux, S., Ozaki, T. & Krol, A. Rev. Sci. Instrum. 76, 13. 6 (2005). Chen, L.M. et al. Appl. Phys. Lett. 90 (2007). 14. 15. Mori, K. et al. Proceedings of IUPAC International Congress on Analytical Sciences 2001, p.i1427 (2001). 16. Mangles, S.P.D. et al. Nature 431, 535-‐538 (2004). 17. Geddes, C.G.R. et al. Nature 431, 538-‐541 (2004). Faure, J. et al. Nature 431, 541-‐544 (2004). 18. Rousse, A. et al. Phys. Rev. Lett. 93, 135005 (2004). 19. 20. Wilkins, S.W., Gureyev, T.E., Gao, D., Pogany, A. & Stevenson, A.W. Nature 384, 335-‐338 (1996). Acknowledgements This work was partially supported by the US National Science Foundation through the Physics Frontier Center FOCUS Grant No PHY-0114336 and the NSF/DNDO Grant No. 0833499. Author Contributions The experiment and analysis was carried out in main by SK, CM and FD with contributions from MSB, SPDM, TM, CAJP and WS. VY, GK and VC operated the laser, SK, CMG, KK, ZN, JS and AGRT contributed to planning, interpretation and manuscript preparation. Competing Interests The authors declare that they have no competing financial interests. Correspondence Correspondence should be addressed to SK (email: [email protected]). Figure 1: Schematic of the experimental setup and results. (a) A high power laser is focused into a tenuous gas jet, creating a miniature plasma accelerator and wiggler, analogous to the conventional accelerator and wiggler of a synchrotron x-ray light source. The emerging electron and x-ray beams are separated with a magnet. (b-d) The x-ray beam can be used to image biomedical specimens. X-ray contact radiograph (absorption contrast) yields good contrast for the case of a tetra fish (b), du to its highly absorbing skeleton, but poor contrast of a damselfly (c), due to poorly absorbing wings (d) Exploiting the coherent quality of our tabletop synchrotron x-rays, the image quality of the damselfly can be improved by propagation the x-rays 1.4 m from the specimen to the detector (phase contrast). The phase contrast images are taken in a single shot 30 fs exposure. Yellow boxes are explained in the text. Figure 2: Improved image quality with phase contrast imaging. (a) Single shot 30 fs exposure of the head of the damselfly. The details of the compound eye (1), exoskeleton (2) and leg (3) evidence the enhanced image quality obtained using the propagation scheme for phase contrast imaging. (b) The figure depicts lineouts taken from the yellow shaded areas in figure 2b (solid black), c (dashed gray) and d (solid gray). Improved contrast and resolution at the few pixel level (≃70 µm) is achieved through edge enhancement in the phase contrast geometry (compare gray lines). Figure 3: (a) A point source emits spherical x-ray wavefronts which are distorted on passing through a phase object. Propagating the distorted wavefront onto the detector, can lead to local focusing and defocusing. (b) Phase contrast image of a yellow jacket taken in a single shot 30 fs exposure of synchrotron radiation from the tabletop source, using the propagation technique of phase contrast imaging (a). Methods: Wakefield Acceleration and Radiation Generation The x-ray source discussed here is based on focusing a pulsed high power laser into a millimeter-sized plume of helium gas, which is immediately ionized and turned into a plasma. As the laser propagates through the plasma, it drives an electron density oscillation (plasma wave) with phase velocity near the speed of light in vacuum. The ponderomotive force of the laser displaces electrons from the almost stationary ions, setting up large accelerating fields. Electrons can be trapped by these fields, resulting in Gigaelectronvolt per centimeter energy gain16-18. At the same time, the electrons are oscillating transversely due to radial electrostatic fields of the plasma wave, emitting a bright beam of synchrotron-like x-rays19 with appreciable degree of spatial coherence due to its micrometer source size3. Laser The experiments were carried with the high power HERCULES laser at the Center for Ultrafast Optical Science at the University of Michigan, Ann Arbor. A schematic is of the experimental setup is shown in figure 1a. Laser pulses with a pulse duration of tL=32 fs and an energy of EL=(2.2±0.1) J were focused to dfwhm =(10.8±0.5) µm (full width at half maximum) onto the front edge of a supersonic Helium gas jet with 3 mm diameter, reaching intensities of (2.0±0.4) Wcm-2 and fully ionized plasma densities of 3 to 8×1018 cm-3. Quasi mono- and polyenergetic electron beams of ≃100 pC charge and peak energy of ≃120 MeV are deflected, and dispersed with a permanent magnet for measurement. Phase Contrast Imaging In x-ray imaging, spatial contrast is a consequence of changes of the thickness and refractive index of the specimen n=1−δ−iβ, where δ and β are the real and imaginary part of the refractive index. Changes of β and δ integrated along the x-ray propagation direction and thickness of the specimen result in absorption and phase contrast respectively. Figure 3a indicates how phase contrast is achieved. The local propagation direction of an electromagnetic wave is perpendicular to the phase front (arrows). On passing through a phase object, spatial variations in δ will distort an initially spherical x-ray wavefront from a point source. Propagating the distorted wavefront a sufficient distance v will lead to local focusing (converging arrows) of the x-rays, which can be observed as edge enhancement on an area detector. Monochromatic x-rays are not required20. To benefit from phase contrast enhancement, sufficiently spherical wavefronts, i.e. sufficient transverse coherence Ltrans = "u /(2#w ) scales with the distance Ltrans is required. Transverse coherence from the source u and the inverse of the source size w, where " is the wavelength of the radiation. In our case, Ltrans is approximately 10 µm for 10 keV x-rays and the 1 ! µm source at only u=0.5 m from the source, allowing for phase contrast enhancement ! in a very compact geometry. Figure 1: Figure 2: Figure 3:
1205.6619	1	1205	2012-05-30T10:21:25	Modified extended tanh-function method and nonlinear dynamics of microtubules	[ "physics.bio-ph" ]	We here present a model of nonlinear dynamics of microtubules (MT) in the context of modified extended tanh-function (METHF) method. We rely on the ferroelectric model of MTs published earlier by Satari\'c et al [1] where the motion of MT subunits is reduced to a single longitudinal degree of freedom per dimer. It is shown that such nonlinear model can lead to existence of kink solitons moving along the MTs. An analytical solution of the basic equation, describing MT dynamics, was compared with the numerical one and a perfect agreement was demonstrated. It is now clearer how the values of the basic parameters of the model, proportional to viscosity and internal electric field, impact MT dynamics. Finally, we offer a possible scenario of how living cells utilize these kinks as signaling tools for regulation of cellular traffic as well as MT depolymerisation.	physics.bio-ph	physics	Modified extended tanh-function method and nonlinear dynamics of microtubules Slobodan Zdravkovića,, Louis Kavithab,c, Miljko V. Satarićd, Slobodan Zekovića, Jovana Petrovića a Institut za nuklearne nauke Vinča, Univerzitet u Beogradu, Poštanski fah 522, 11001 Beograd, Serbia b Department of Physics, Periyar University, Salem-636 011, India c The Abdus Salam International Centre for Theoretical Physics, Trieste, Italy d Fakultet tehničkih nauka, Univerzitet u Novom Sadu, 21000 Novi Sad, Serbia A B S T R A C T We here present a model of nonlinear dynamics of microtubules (MT) in the context of modified extended tanh-function (METHF) method. We rely on the ferroelectric model of MTs published earlier by Satarić et al [1] where the motion of MT subunits is reduced to a single longitudinal degree of freedom per dimer. It is shown that such nonlinear model can lead to existence of kink solitons moving along the MTs. An analytical solution of the basic equation, describing MT dynamics, was compared with the numerical one and a perfect agreement was demonstrated. It is now clearer how the values of the basic parameters of the model, proportional to viscosity and internal electric field, impact MT dynamics. Finally, we offer a possible scenario of how living cells utilize these kinks as signaling tools for regulation of cellular traffic as well as MT depolymerisation. 1. Introduction Microtubules are major cytoskeletal proteins. They are hollow cylinders formed by protofilaments (PF) representing series of proteins known as tubulin dimers [2,3]. There are usually 13 longitudinal PFs covering the cylindrical walls of MTs. The inner and the nm15 nm25 outer diameters of the cylinder are , while its length may span and dimensions from the order of micrometer to the order of millimetre. Each dimer is an electric dipole whose length and longitudinal component of the electric dipole moment nm8=l are [5], respectively. The constituent parts of the [2-4] and =p 337 Debye dimers are α and β tubulins, corresponding to positively and negatively charged sides, respectively [2-4]. Corresponding author. E-mail addresses: [email protected] (S.Zdr.), [email protected] (L.K.), [email protected] (M.V.S.), [email protected] (S.Zek.), [email protected] 1 In this paper we demonstrate how METHF method [6-10] can be used in the study of nonlinear dynamics of MTs. The paper is organized as follows. In Section 2 we explain the well known model for MTs we rely on [1]. The modification of the model presented in this paper is a generalization of the original one and will be referred to as u-model. The model brings about a crucial nonlinear differential equation, describing nonlinear dynamics of MTs. In Section 3 we briefly describe METHF method. Then we solve the basic nonlinear differential equation, mentioned above. We show that its solution is a kink-like solitonic wave. This result is compared with numerical solutions in Section 4. Finally, in Section 5, we give concluding remarks. In particular, we emphasize the biological importance of the studied kink-like solitons. 2. U-model of MTs The model we rely on assumes only one degree of freedom of dimers motion within nu the PF [1]. This is a longitudinal displacement of a dimer at a position n denoted as and thus we call the model as u-model. The overall effect of the surrounding dimers on a dipole at a chosen site n can be described by a double-well potential [1] uV ( d where A and B are positive parameters that should be estimated. As an electrical dipole, a dimer in the intrinsic electric field of the MT acquires the additional potential energy given by [1] uV ( el where E is the magnitude of the intrinsic electric field at the site n, as the dimer n exists in the electric field of all other dimers, and represents the excess charge within the q 0>E and . dipole. It is assumed that 0>q The Hamiltonian for one PF is represented as follows H (1) , (2) , (3) Cu −=) , uV ( C = −= Bu Au qE 1 2 1 4 + ) n = + 2 n 4 n n n − u n 2 ) n 1 + um & 2 ⎡ ⎢⎣ ∑ n 2 n + uk ( 2 ) n ⎤ ⎥⎦ m k where dot means the first derivative with respect to time, is is a mass of the dimer, an intra-dimer stiffness parameter and the integer n determines the position of the considered dimer in the PF [1]. The first term represents a kinetic energy of the dimer, the second one is a potential energy of the chemical interaction between the neighbouring dimers belonging to the same PF and the last term is the combined potential uV ( . (4) −= Cu Bu Au + − ) n n 4 n 1 2 2 n 1 4 2 , l + 1 2 2 l um & n u ∂ x ∂ u 2 ∂ x 2 ∂ u ±→± 1 (5) It is obvious that the nearest neighbour approximation is used. However, this does not mean that the influence of the neighbouring PFs is completely ignored as the value of the electric field E depends also on the dipoles belonging to the neighbouring PFs. nq np u q = p By using the generalized coordinates and , defined as and = n n n txu tu n → applying a continuum approximation and making a series expansion ),( )( u n we can straightforwardly obtain an appropriate dynamical equation of motion. In order to derive a realistic equation, the viscosity of the solvent should also be taken into Fv u consideration. This can be achieved by introducing a viscosity force into the &γ−= obtained dynamical equation of motion, where γ is a viscosity coefficient [1]. All this brings about the following nonlinear partial differential equation um 2 ∂ t 2 ∂ It is well known that, for a given wave equation, a travelling wave x t and only through a unified variable ξ which depends upon x ωκξ t − ≡ where κ and ω are constants. Substitution of following ordinary differential equation (ODE) ( ) m u 2 ω By introducing a dimensionless function ψ through the relation , (7) . (8) . (6) by ξ transforms Eq. (6) into the is a solution −′′ u γω −′ u ∂ γ t ∂ u 2 2 ∂ x ∂ x and t − lk 2 2 κ − qE − qE )(ξu − lk 2 3 Bu Au + 3 Bu Au + = 0 = 0 − + u = , (9) ψ A B a much more convenient equation can be obtained. This final ODE reads 3 −′′ −′ σψψψρψα + − and contains the following new parameters: = 0 , (10) m 2 ω α = 2 2 κ kl − A , (11) 3 . u ≡′ du ξd It was already mentioned that this approach represented a certain improvement of the original model, explained in [1]. If we compare Eq. (10) with the appropriate one in [1] 1−=α . Therefore, this approach is more general. We we can see that they are equal for treat the parameters ρ and σ as an input and will determine values of dinamical parameters of the system. We will see that the final result depends on ρ and σ only, i.e. on the parameters that determine their values. The crucial equation (10) will be solved in the next section. Before we proceed we want to discuss the potential energy , defined by Eq. (4). This step is very important )(uV to understand the physics behind Eq. (10) and its solutions. Using the procedure explained above we can easily obtain the following convenient expression for this potential V where , (14) A 2 B ) ( ψ 1 1 4 2 ψψσψ −= + − 2 4 ) ( ψ . (15) f 0=σ the The function is shown in Fig. 1 for two values of the parameter σ. For (ψf ) function , is symmetric (curve a) while for and, consequently, the potential (ψf (ψV ) ) the increasing σ the right minimum becomes deeper and the left one becomes shallower and elevated. To find the values of ψ for which reaches a maximum and minima (ψf ) we should solve the equation ) ( ψ = f ψψσψf ( ) 3 = ′ +−−= =σ ρ= and , (12) qE AA B γω (13) A 0 . (16) 4 0.0 -0.2 f (ψ ) -0.4 -0.6 a a b -1.0 -0.5 0.0 0.5 1.0 ψ (ψf ) 0=σ and (b) 3.0=σ for: (a) . Fig. 1. The function According to the procedure explained in Appendix we can easily obtain the following roots of Eq. (16): R =ψ ψ max , (17) , (18) cos F sin3 F Fcos 2 3 + = ( − 1 3 ) ( cos 1 3 F + sin3 F ) , (19) ψ L −= where F = 1 3 ⎛ arccos ⎜⎜ ⎝ σ σ 0 ⎞ ⎟⎟ ⎠ . (20) Lψ correspond to the right and left minimum of the function Rψ and The functions and a critical value of the parameter σ has the value (ψf ) 5 2 33 . (21) 0 =σ This means that the three real roots of Eq. (16) exist for . 0σσ< The dependence Rψ , maxψ and exψ stands for (ex σψ is shown in Fig. 2, where ) Lψ . Rψ is a slowly increasing function of σ while Lψ and maxψ approach each We see that other for the increasing σ. This means that the right minimum moves to the right for the higher σ while the maximum and the left minimum of the function become (ψf ) closer. All this can be noticed in Fig. 1. For the left minimum and the maximum 0σσ= disappear, coalescing in the saddle point. ΨR ΨL Ψex(σ) 1.0 0.5 0.0 -0.5 -1.0 σ0 Ψmax 0.0 0.1 0.2 0.3 0. 4 σ (ψV ) as a Fig. 2. The values of exψ corresponding to the extrema of the potential function of the parameter σ. Rψ and maxψ for a couple of values of σ are shown in Table 1. Of course, The values maxψ does not exist for . 0σσ= 6 Table 1 σ Rψ maxψ 0 1 0 0.1 1.047 -0.101 0.2 1.088 -0.209 0.3 1.125 -0.339 0σ 1.155 n a All this suggests that the higher σ, i.e. the bigger value of , increases stability of qE MTs dynamics around the right minimum which becomes deeper. This requires further research and should be checked by stability analysis which will be a topic of a separate publication. 3. Modified extended tanh-function method In what follows we briefly outline the basic features of METHF method. The method will be applied for solving Eq. (10). According to this procedure we look for the possible solution in the form [6,7] b i −Φ+Φ i i i ) , (22) ψ = a 0 + M (∑ a i 1 = )(ξΦ=Φ is a solution of the well known Riccati equation (23) where the function b 2Φ+=Φ ′ and b are real constants ib ia 0a and , , is the first derivative [6,7]. The parameters Φ ′ that should be determined as well as an integer M . The possible solutions of (23) depend b on the parameter as follows: 1) If 2) If 0>b 0=b then then 3) If 0<b then =Φ −=Φ b tan 1 ξ −−=Φ b cot −=Φ )ξb ( ( )ξb , or , (25) ( )ξb ( )ξb , (24) . (26) −−=Φ , or tanh − b coth − b Balancing the orders of ψ′′ and we find 3ψ with respect to the new function Φ . Namely, the highest orders of the function Φ in the expressions for ψ′′ and 3ψ 1=M M3Φ 2+Φ M are . respectively and they are equal for and 1=M Eqs. (22) and (23) yield the expressions for ψ′ , ψ′′ and 3ψ as functions of ψ is example, the second derivative of . For Φ 7 + 3 a 0 + a ρ− 1 bb 2 1 2 3 − +Φ −= a 0 bb 2 Φ 1 1 − 2 ba 3 1 1 = 2 aa 3 10 = 2 ba 3 10 3 2 aa 1 0 bb ρ+ 1 + bb 2 α 1 + 2 ba 3 11 a 2 = α 1 + 6 baa 110 ba 2 α 1 + b +−= 1 2 ba 3 0 1 a +−= 1 − ba b σρρ + − 1 1 3ψ brings about the crucial 0 , (28) . (27) , (29) A A B B B 1 2 2 3 3 − − − +Φ+Φ+Φ+Φ+Φ+Φ 1 2 2 3 3 , (30) , (31) , (33) , (32) baψ a 2 2 3 =′′ +Φ+Φ 1 1 Eq. (10) in terms of the expressions for ψ, ψ′ , ψ′′ and equation A 1 where the following set of abbreviations is introduced: A 0 A 1 B 1 A 2 B A 3 and B 3 Of course, Eq. (28) is satisfied if all these coefficients are simultaneously equal to zero which renders a system of seven equations. Before embarking on solving such a system, we examine its behaviour under the conditions given by Eqs. (24)-(26). The solutions expressed through tangents and cotangents cannot be biophysically tractable as these functions diverge. The same can be said for hyperbolic cotangent. This means that we are 1 ≠a 1 =b 0<b 0 0 looking for the acceptable solutions for which , and , which reduces the system mentioned above. Hence, the system can be reduced to the following system of four equations a a + − 0 a 31 +− 0 aa 3 = ρ 10 2 a 2 −= α 1 Its solutions, i.e. the values of the parameters b , (34) . (35) (36) and α, are given through ba − σρ 1 b 2 ,0 = α bb 2 2 = α 1 − + , . = ,0 ⎫ ⎪ ⎪ ⎬ ⎪ ⎪ ⎭ + 3 a 1 + 3 b 1 2 3 0 2 A 0 = 0a , 1a 8 σa + 0 = 0 , (37) , (38) (39) 3 0 2 − a 8 a ρ = 1 3a 0 2 1a 2 −=α and b = 2 a 3 0 a 1 1 − 2 . (40) 0 1 3 0<b >aa 10 0σσ< 0<α . Also, and , which was discussed holds for the inequality . (41) Notice that 2 0 <a It is easy to check that the requirement (41) is equivalent to earlier. It was mentioned above that this extended version of the model reduces to the model 1−=α . To calculate the value for α we have to know the values of the from [1] if parameters determining σ and ρ. The values of these parameters have not been determined yet. It suffices now to discuss a negative sign of α, as can be seen from (39). This parameter comes from the first two terms in Eq. (6). These are inertial and elastic terms. A question is which contribution is higher and the negative value of α suggests that the elastic term in (6) is bigger than the inertial one. Using the expression for wave speed =v Eq. (11) can be written as ω (42) κ 2 κ A 2 2 − = kl ( mv ) . (43) α k 0<α may indicate a small velocity of the wave and/or a big The fact that , i.e. strong chemical bond between the neighbouring dimers in PF, as can be seen from (3). The importance of α and our intention to get as much information about it as possible is reason that we did not perform rescaling of Eq. (10), which would remove α from the 9 equation. Notice that we cannot a priori exclude the value rescaling. Eq. (37) looks like (16) and Appendix should be used again. The requirement for the existence of the three real solutions is again. Hence, the roots of (37) are 0σσ< ), (44) a 0=α , which prevents ( cos sin3 F F + = 01 = 02 ( cos 1 32 F − sin3 F ) , (45) 1 32 a a −= 03 1 3 cos F (46) 0σ are given by (20) and (21). Of course, all these and where the expressions for F three values depend on σ and they are shown in Fig. 3. One can see that (37) has three 03a ) for real solutions for and only one ( . To obtain Eqs. (44)-(46) a 0σσ< 0σσ> relationship cos was used. )(ξψ . Suppose that 1ψ describes the system in A next step is to construct the function the state corresponding to the right minimum in Fig. 1. Using Eqs. (26), (38) and (40) we easily obtain ) (47) πβ = − ) ( β cos + 1 − ( − 1 −−=Φ b tanh ( ) b ξ − −= a 3 01 ρ a 31 − 2 01 tanh ⎛ ⎜⎜ ⎝ a 3 01 ρ a 31 − 2 01 ⎞ ⎟⎟ ξ ⎠ . (48) 1=M and (48) yield the following expression for the function Finally, Eqs. (22) with 1ψ : a 3 01 ρ a 31 − 2 01 ⎛ ⎜⎜ ⎝ ⎞ ⎟⎟ ξ ⎠ )( ξψ 1 = a 01 − a 31 − 2 01 tanh , (49) )(1 ξψ depends on ρ 01a where is determined by (44), (20) and (21). We can see that )(1 ξψ from 01a and σ only as is a function of σ. Notice that the jump of the function to depends on σ only while the solitonic width, i.e. slope, depends on both ρ ∞+ ∞− and σ. It is obvious that the solitonic width is proportional to viscosity. 10 a01 a03 a02 a0(σ) 1.0 0.5 0.0 -0.5 -1.0 σ0 0.0 0.1 0.2 0.3 0.4 σ 0a 2 01 2 01 + 01 ) = a a 31 − as a function of the parameter σ. (50) Fig. 3. The values of The function (49) is shown in Fig. 4 (analytic curves) for the two values of the parameter σ and for 1=ρ . Obviously, this is an anti-kink soliton. The parameter ρ affects its slope, i.e. the width of the soliton, but not the character of the wave. 1ψ that is According to Eq. (49) we can calculate the initial and the final value of ( −∞ψ 1 and ( +∞ψ 1 Using Eq. (44) we can straightforwardly prove that ( ψ 1 as sin . (51) (52) is negative. , ψ 1 = ψ max 3 cos F a 31 − = ψ R ( +∞ ) ) = a ) −∞ − 01 F − 11 . 0=σ and (b) 3.0=σ 1=ρ and: (a) )(1 ξψ for Fig. 4. The function Therefore, nonlinear dynamics of MTs is described by antikink solitons. In other words, the system exists at the right minimum of Fig. 1 and the value of )(ξψ is ) ( . When some energy is supplied, released by hydrolysis of guanosine ψ ψ = −∞ 1 R triphosphate (GTP), the system jumps to the maximum and ψ becomes maxψ , as explained by (52). The way how this transition is performed is explained by the solitonic wave (49). Of course, sooner or later, the system spontaneously returns to its minimum and so on. These transitions represent displacements of the dimer, described by u , which is defined by Eq. (9). Obviously, for a quantitative treatment we should know the values of the parameters. Some estimates have already been done [1] but this is still an open question. We should keep in mind that the model described in this paper is somewhat different from the original one, introduced in [1]. For example, Eqs. (11), (38) and (39) represent a relationship between some parameters and will help in their estimations. Anyway, the parameter selection requires further research and is not the topic of this work. )(1 ξψ has been studied in this paper. A careful reader may ask why only the function )(ξψ . There are three solutions of (37) and, consequently, three possible functions )(2 ξψ and 02a However, it does not seem that the functions )(3 ξψ , corresponding to 03a , are physically relevant. For example, one can easily show that and ψ 3 , (53) , ψ 3 ψ max ( −∞ ) = ψ L ( +∞ ) = 12 ) ( ( = = −∞ ψ L ψ R ) +∞ , ψ 2 , (54) )(1 ξψ , corresponding to the deeper which could be expected. Obviously, the soliton energy minimum, is more relevant. Also, we can show that ψ 2 . Therefore, we study in some more which may be physically relevant only for 0σσ< details only one out of the three possible solutions as this one is physically important. )(1 ξψ is relevant for triggering mechanism, which will Also, we believe that the solution be explained in the last section. We do not study the case when only one real solution of (37) exists. Such the case corresponds to the positive b , which does not have physical meaning, as was explained above. Due to the large uncertainty in the values of the parameters A and B of the potential (4), we concentrate on the characteristics of the soliton that do not depend on these parameters. An important example is the wave velocity. After a simple algebra with Eqs. (11), (13), (36) and (42), it can be expressed as kl 2 v . (55) = 2 a = a We consider the solution given by Eqs. (44), (20) and (21). 01 0 )(1 ξψ given by (49) we can write According to the function a 3 0 ρ a 3 0 ρ a 31 − a 31 − ( κ ( χ ξ vt vt − ≡ − = x x ) 2 2 0 0 ) . (56) m + 2 γ 2 Aa 18 0 Hence, Aa 3 κ 0 χ = a 31 − γω 2 0 . (57) 2 . (58) π Λ = Obviously, χ determines the solitonic width Λ through [11] χ It is convenient to write Λ as an integer of dimer’s length l , i.e. Nl=Λ All this brings about the following expression for v : . (59) 13 2 0 . (60) 0 (61) 2 0 . (62) a 3 0 v = ANl a 31 − πγ2 2 We can eliminate A from (55) and (60) and obtain mv where vC 0 kl + − = 2 C 0 = lN a 31 γ − a 12 π 0 Hence, the expression for Cv 0 = m 2 mkl 41 + 2 C 1 +− 2 0 ⎛ ⎜ ⎜ ⎝ ⎞ ⎟ ⎟ ⎠ v reads . (63) It is obvious that the speed v does not depend on A and B but does depend on some other parameters. Using (62) one can see that v is the decreasing functon of viscosity γ. v =m but the values of To estimate we can use [1]: 10 11 and kg 103.5 skg 23− 11− ⋅ =γ ⋅ N k and are not known. Also, of special importance is the intensity of electric field E 0a as this determines through σ. Namely, Eq. (12) can be written as =σ , (64) = = qE AA B qdE A B dA pE A B dA is the longitudinal component of the electric dipole moment and the where qd p ≡ nm4≈d distance between the centres of positive and negative charges in any dimer is [5]. Therefore, the estimates of E , A and B are of crucial importance. An attempt to estimate these parameters is in progress even though some estimates were done in [1]. Some values of the velocity v for and four values of k are given in Table 2 for 30=N different values of σ (different strengths of the intrinsic electric field E ). 14 0.3 10.2 50.5 100.1 197.0 0σ 6.3 31.5 62.8 124.8 mN1.0=k mN5.0=k mN1=k mN2=k 0.2 12.8 63.4 125.1 243.9 Table 2 σ ( )smv ( )smv ( )smv ( )smv 0 18.9 92.1 179.2 341.4 0.1 15.6 76.6 150.2 290.0 4. Numerical solutions In order to test the described analytical procedure and to study behaviour of the soliton for those parameters of the MT for which analytical solutions do not exist, we have solved Eq. (10) numerically. The standard shooting method with the 4th order Runge- Kutta integrator was used. Particular solutions are determined by their asymptotic 0=x behaviour and are centred at . The numerical step was . 510 − Numerical solutions were first compared to the analytical solutions obtained by METHF method. Excellent agreement shown in Fig. 4 validates our analytical approach. We further generated numerical solutions of Eq. (10) for those parameters α for which our analytical procedure does not give solutions. To enable a direct comparison, we kept the parameters ρ and σ constant and again considered dimers within a MT with the 0=σ ) and 1=ρ , and with a non-zero electric field intrinsic electric field zero ( 3.0=σ ( ) and 1=ρ . The results are shown in Fig. 5 and Fig. 6, respectively. The values 01a for were calculated according to (37). As the parameter α multiplies only a derivative in Eq. (10) it does not influence asymptotic solutions. These are determined by σ only. For small absolute values of α the soliton is narrow and has a shape similar to tanh function, meaning that the transition between the initial and the final position of the dimer is fast and smooth. On the other hand, for big α the dimmer experiences oscillation before it stabilizes in the final value. The period and amplitude of oscillations increase with α . These results indicate that there is an optimal parameter α for a given process to occur. 15 )(1 ξψ for Fig. 5. The function 0=σ for different values of α 1=ρ and )(1 ξψ for Fig. 6. The function 3.0=σ 1=ρ and for different values of α 16 5. Conclusions In this paper we thoroughly revisited an earlier work [1] dedicated to nonlinear dynamics of tubulin dimers within the same PF of a stable cellular MT. The basis of the model is the quartic nonlinear potential describing the mean field of chemical bonds between neighbouring tubulin dimers. This potential does dominantly depend on nu longitudinal displacements of the tubulin dimers along MT axis, as shown by Eq. (1). Inclusion of dipole-dipole interaction between tubulin dimers as well as the impact of viscous damping of cytosol has led to dimensionless nonlinear ODE (10). The effective potential (15) was discussed respecting the parameter σ which is proportional to the magnitude of intrinsic electric field created by MT itself. To solve Eq. (10) we have relied on METHF method [6-10]. The physically tractable solutions have the shape of anti-kink soliton (49), well known in nonlinear ferroelectrics and ferromagnetics. These solutions for the two values of σ are compared with the numerical solutions and a perfect agreement was demonstrated. It might be important to point out that the Hamiltonian is only what was taken from Ref. [1]. In addition, the way how the viscosity term was introduced in the dynamical equation of motion is widely accepted and can be found in many references. The procedure explained in this paper is completely different from the one in Ref. [1]. It turned out that the METHF method is both elegant and useful. It enables us to estimate the maximum of electric field E . Namely, the highest value of σ is 0σ , given by (21). p d A According to Eq. (64), we see that this is proportional to E. As the values of , , and B can be found in literature we can calculate the maximum of E . This is not the topic of this paper as we believe that the values of the parameters are still an open problem, requiring further research. It was mentioned above that the crucial ODE for ψ in this paper and in Ref. [1] are 1−=α , which certainly means that this approach is more general. equal only if One can say that the final result of the both papers, this one and Ref. [1], is the )(1 ξψ describing dynamics of MT when function )(ξψ . We showed only the function the system is in the deeper minimum of the potential well. This is shown in Figs. 4-6. There is one very important advantage of the method applied here over the one used in )(ξψ does not depend neither on Ref. [1]. If we look at Eq. (3.11) of Ref. [1] we see that ρ nor on σ, which is a big problem. In fact, Eq. (3.18) was an attempt to introduce the dependence of electric field but it turned out that the term with σ was negligible. As for )(1 ξψ , existing in this paper, it depends on both parameters as 01a depends on σ. This is the big advantage as ρ and σ, being proportional to viscosity and internal electric field, are internal parameters and we certainly expect that the final result, i.e. dynamics of MT, )(1 ξψ shows that the solitonic should depend on them. In addition, the expression for width is proportional to viscosity, which is physically plausible. Also, Figs. 4-6 show how σ affects the final result. Kink velocities are estimated in both in this paper and in Ref. [1]. We can see that these estimations are different. The estimated values of solitonic speed here are in the 10m to a few hundreds of s sm . This interval may be comparable to range from about 17 s 30m , but is much greater than the speed the speed of nerve axon potential, spanning to µm1 s of MT motor protein which is, in a case of kinesin, around , or ionic waves within s µm ). the cell (a few We expect that these stable solitons could be elicited for example by the energy released in hydrolysis of tubulin-nucleotide GTP, or by the interaction with other MT associated proteins. We here offer the ideas of how these solitons could be utilized by a cell for some important mechanisms underlying its functional dynamics. First, it is plausible to expect that anti-kink soliton reaching the MT tip could cause the detachment of last tubulin dimer. This is one of triggering mechanisms responsible for depolymerisation of MTs. Second, it is well known that MTs serve as a “road network” for motor proteins (kinesin and dynein) dragging different “cargos” such as vesicles and mitochondria to different sub-cellular locations. Kinesin motors move along MT towards MT plus-end and dynein motors move in opposite direction. Both classes of motors are present on the same “cargo” [12] and it is still unclear which mechanism decides which motor will take the action and drag the cargo in the proper direction. We argue that the cell’s compartment which needs the specific cargo will launch the soliton of above kind sending it as a signal along the closest MT. The soliton will activate proper motors being close to MT along which the soliton propagates. For example, if the need for mitochondria is expressed in the regions close to the cell membrane, the solitons created in these MTs will propagate down MTs towards cell nucleus. These solitons will turn on kinesin motors in order to attach to the same MTs and to carry mitochondria towards MT plus-ends which are close to cell membrane. It might be useful to mention that some similar ideas were elaborated earlier [13,14]. Appendix A polynomial x has three real roots for where [15] qD 2 = 4 For the negative x k where . (A.2) (A.1) or one real and two complex conjugate roots for , (A.3) D the solutions of (A.1) are [15] πϕ k 2 ⎛ + ⎞ cos ⎜ ⎟ 3 ⎝ ⎠ , 2,1,0=k px q =+ 0 , 0>D , = 2 3 r 0<D + p 3 27 3 + 0<p 18 (A.4) . (A.5) D the solution of (A.1) is [15] (A.6) − p 3 )2 ϕ sin( x −= where tan = ϕ 3 tan ⎛ ⎜ ⎝ w 2 ⎞ ⎟ ⎠ (A.7) r = 3p − 27 and cos −=ϕ For the positive q r 2 and w = arcsin Acknowledgements 3 2 2 q ⎛ − ⎜ ⎝ p 3 ⎞ ⎟ ⎠ ⎡ ⎢ ⎢ ⎣ . (A.8) ⎤ ⎥ ⎥ ⎦ This research was supported by funds from Serbian Ministry of Sciences, grants III45010 and OI171009. L. Kavitha gratefully acknowledges the financial support from UGC in the form of major research project, BRNS, India in the form of Young Scientist Research Award and ICTP, Italy in the form of Junior Associateship. S. Zdravković gratefully acknowledges the hospitality of the Abdus Salam International Centre for Theoretical Physics, Trieste, Italy, where a big part of this work was done. References [1] Satarić MV, Tuszyńsky JA, Žakula RB. Kink like excitations as an energy- transfer mechanism in microtubules. Phys. Rev. E 1993; 48:589-597. [2] Dustin P. Microtubules. Berlin: Springer; 1984. 19 [3] Tuszyńsky JA, Hameroff S, Satarić MV, Trpisová B, Nip MLA. Ferroelectric Behavior Information for Implications in Microtubule Dipole Lattices: Processing, Signaling and Assembly/Disassembly. J. Theor. Biol. 1995; 174:371- 380. [4] Satarić MV, Tuszyńsky JA. Nonlinear Dynamics of Microtubules: Biophysical Implications. J. Biol. Phys. 2005; 31:487-500. [5] Schoutens JE. Dipole–Dipole Interactions in Microtubules. J. Biol. Phys. 2005; 31:35–55. [6] Ali AHA. The modified extended tanh-function method for solving coupled MKdV and coupled Hirota–Satsuma coupled KdV equations. Phys. Lett. A 2007; 363:420-425. [7] El-Wakil SA, Abdou MA. New exact traveling wave solutions using modified extended tanh-function method. Chaos, Solitons Fractals 2007; 31:840-852. [8] Kavitha L, Prabhu A, Gopi D. New exact shape changing solitary solutions of a generalized Hirota equation with nonlinear inhomogeneities. Chaos, Solitons Fractals 2009; 42:2322–2329. [9] Kavitha L, Akila N, Prabhu A, Kuzmanovska-Barandovska O, Gopi D. Exact solitary solutions of an inhomogeneous modified nonlinear Schrödinger equation with competing nonlinearities. Math. Comput. Modelling. 2011; 53:1095–1110. [10] Kavitha L, Srividya B, Gopi D. Effect of nonlinear inhomogeneity on the creation and annihilation of magnetic soliton. J. Magn. Magn. Mater. 2010; 322:1793– 1810. [11] Zdravković S, Satarić MV. Transverse interaction in DNA molecule. BioSystems 2011; 105:10–13. [12] Gross SP. Hither and yon: a review of bi-directional microtubule-based transport. Phys. Biol. 2004; 1:R1-R11. [13] Satarić MV, Budimski-Petković L, Lončarević I, Tuszyńsky JA. Modeling the role of intrinsic electric fields in microtubules as an additional control mechanism of bi-directonal intracellular transport. Cell, Biochem. Biophys. 2008; 52:113- 124. [14] Satarić MV, Kozmidis-Luburić U, Budimski-Petković L, Lončarević I. Intrinsic electric field as a control mechanism of intracellular transport along microtubules. J. Comput. Theor. Nanosci. 2009; 6:721-731. [15] Smirnov VI. Kurs vishey matematiki, tom 1. Moscow: Nauka 1965. (In Russian). 20
1305.1154	1	1305	2013-05-06T11:53:58	Ion-mediated RNA structural collapse: effect of spatial confinement	[ "physics.bio-ph", "cond-mat.soft", "physics.chem-ph", "q-bio.BM" ]	RNAs are negatively charged molecules residing in macromolecular crowding cellular environments. Macromolecular confinement can influence the ion effects in RNA folding. In this work, using the recently developed tightly bound ion model for ion fluctuation and correlation, we investigate the confinement effect on the ion-mediated RNA structural collapse for a simple model system. We found that, for both Na$^+$ and Mg$^{2+}$, ion efficiencies in mediating structural collapse/folding are significantly enhanced by the structural confinement. Such an enhancement in the ion efficiency is attributed to the decreased electrostatic free energy difference between the compact conformation ensemble and the (restricted) extended conformation ensemble due to the spatial restriction.	physics.bio-ph	physics	Ion-mediated RNA structural collapse: effect of spatial confinement Zhi-Jie TAN1∗ and Shi-Jie CHEN2∗ 1Department of Physics and Key Laboratory of Artificial Micro & Nano-structures of Ministry of Education, School of Physics and Technology, Wuhan University, Wuhan, P.R. China 430072 2Department of Physics and Astronomy and Department of Biochemistry, University of Missouri, Columbia, MO 65211 Abstract RNAs are negatively charged molecules residing in macromolecular crowding cellular environments. Macromolecular confinement can influence the ion effects in RNA folding. In this work, using the recently developed tightly bound ion model for ion fluctuation and correlation, we investigate the confinement effect on the ion-mediated RNA structural collapse for a simple model system. We found that, for both Na+ and Mg2+, ion efficiencies in mediating structural collapse/folding are significantly enhanced by the structural confinement. Such an enhancement in the ion efficiency is attributed to the decreased electrostatic free energy difference between the compact conformation ensemble and the (restricted) extended conformation ensemble due to the spatial restriction. -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- Running title: Ion-mediated RNA collapse and crowding effect ∗Emails: [email protected] (to S.J.C.); [email protected] (to Z.J.T.). Keywords: RNA folding, ion electrostatics, macromolecular crowding, tightly bound ion model 3 1 0 2 y a M 6 ] h p - o i b . s c i s y h p [ 1 v 4 5 1 1 . 5 0 3 1 : v i X r a Introduction Cellular functions of nucleic acids are intrinsically related to their folding in cellular environments (1 -- 3). Since DNAs/RNAs are highly negatively charged molecules, folding into compact native structures requires metal ions to overcome the strong Coulombic repulsions (1)-(13). Furthermore, RNAs in vivo are surrounded by many other molecules (14 -- 16), and the volume percentage of macromolecules in cells can reach 40% (14 -- 16). The presence of other macromolecules around an RNA can cause spatial confinement for DNA/RNAs folding. The macromolecular crowding may greatly influence the RNA folding (17 -- 21). However, the problem of how the spatial constraint affects the ion-induced RNA folding is not fully solved. In this work, we will construct a simplified structural model to investigate the macromolecular crowding effect on ion-mediated RNA collapse/folding. For RNA secondary structure folding (1 -- 3), the thermodynamic parameters at standard salt condition (namely, 1M Na+), have been experimentally determined. Parameters for other ionic conditions have also been fitted from experimental data and theoretical calculations (22)-(33). These parameters have led to many accurate predictions for RNA/DNA secondary structures and folding kinetics (22)-(33). For tertiary structure folding, however, our understanding of the role of ions, especially the effects of multivalent ions such as Mg2+, remains incomplete (1)-(13). Experimental and computational studies showed that metal ions with higher charge density are more efficient in promoting RNA collapse and stronger ion-correlations could potentially contribute to the efficient role of Mg2+ ions (34)-(40). Another series of experiments for short DNA helices suggested the existence of helix-helix attractive force at high multivalent ion concen- tration and repulsive force at low ion concentration (41, 42). Furthermore, experiments on a paradigm system consisting of two helices tethered by a loop suggested that Mg2+ is >10-fold more efficient than the predictions from the Poisson-Boltzmann (PB) theory in promoting structural collapse (43 -- 45). In the cells, the presence of other molecules can cause an crowded environment that can cause spatial confinement for the conformations of RNAs and consequently influence the free energy landscape of RNAs. For proteins, macromolecular crowding has been shown to play an important role in folding stability, struc- ture, and kinetics (14, 15, 46 -- 51). For RNAs, however, the study of macromolecular crowding effect on folding is relatively new (52), especially on its impact on the role of ions in folding (19). RNA tertiary structure collapse involves high charge density, quantitative description for the role of metal ions is challenging, especially for multivalent ions. Molecular dynamics studies for ion-RNA in- teractions have provided novel insights into ion and structure/sequence-specific forces in RNA folding, especially the force arising from site-specific binding of ions (53)-(58). The two classic polyelectrolyte theories: the counterion condensation (CC) theory (59) and the PB theory (60)-(65) have been successful in predicting electrostatic properties of biomolecules (59)-(65). However, for the complex RNA tertiary struc- tures, the line-charge structural model used in the CC theory could be oversimplified (59). Furthermore, RNAs with high charge density generally induce high ion concentration and consequently could possibly cause ion-ion correlations for multivalent ions in the vicinity of RNA surface. The PB theory ignores such potentially important effects for multivalent ion solutions (66 -- 68). In order to take into account the effects of ion correlations and ion-binding ensemble, the tightly bound ion (TBI) model was developed (66, 67); see the Supplementary Material for an introduction of the TBI model. Experimental comparisons sug- 2 gest that the TBI model may offer improved predictions on the ion effect of thermodynamic stability of DNA/RNA helices/hairpins (31 -- 33), DNA helix assembly (69 -- 71), and ion-binding properties and tertiary folding stability of RNAs in the presence of Mg2+ (72 -- 74). In this work, we will employ the newly refined TBI model (72, 73) to investigate a paradigm system with two nucleic acid helices tethered by a loop. We will focus on the effect of conformational constraint on the role of ions in RNA folding. The conformational constraint can arise from the macromolecular crowding effect. Moreover, the study goes far beyond the previous studies which only involves planar con- figurations (70). Specifically, we will consider the conformational fluctuation of helices at the all-atom level with 3-dimensional rotations and compare our predictions with the recent experimental data as well as the predictions from the PB theory (44), Methods Structural model The model system consists of two helices tethered by a flexible loop; see Fig. 1. In the model, we allow for three-dimensional rotation of the helix axes. We note that in a previous study, only symmetric co-planar model rotations were allowed (70). As shown in Fig. 1, a configuration of the system can be constructed through the following operations. First, the two parallel helices are separated by a axis-axis distance x; Second, the two helices are symmetrically rotated around the respective ends (O and O') of the helix axis in the axis-axis plane with an angle θ; Third, the two helices are rotated so that the projections of two helix axes in the plane perpendicular to line of O-O' are separated by an angle γ. The configuration of the system can be described by three parameters: the distance x, the angles θ and γ. Here, the parameter γ is used to produce non-planar configurations. The (DNA) helices here are assumed to adopt the B-form structure (75). We use the intervals of (3Å, 20◦, 20◦) to sample the conformational space of (x ∈ [5Å, 56Å], θ ∈ [0, 180◦], γ ∈ [0, 180◦]). To reduce the computational time, we randomly select ∼600 configurations from the full 3D conformation ensemble for calculations. Tests with the different sampling show that our results are quite robust. In the structural model, the three parameters (x, θ, γ) are used to produce the major portion of the full conformational ensemble, including the non-planar/planar and compact/extended conformational ensembles. Although this approximation may be valid as we focus on the low-resolution energy landscape of the system, it would be useful in further studies to examine the effects of other degrees of freedoms such as the spin of helices and asymmetrical rotations of helix axes. Free energy For a given configuration of the system described by (x,θ,γ), the total free energy G(x, θ, γ) can be calculated as G(x, θ, γ) = GE(x, θ, γ) + Gloop(x, L) + Gcs + G3◦, (1) where GE(x, θ, γ) is the electrostatic free energy for the two helices in an ionic solution. Gloop(x, L) is the free energy of the loop. Gcs is the coaxial stacking free energy between the two helices, and G3◦ is the 3 tertiary contact energy as the two helices are in close contact. In the study, we calculate GE(x, θ, γ) using the all-atom TBI model (72, 73); see the Supplemental Information for a detailed description of the model. We calculate Gloop(x, L) from Gloop = −kBT lnPloop(x, L), (2) where Ploop(x, L) is the probability of a loop of length L with end-to-end distance x. Ploop(x, L) can be 4(1−(x/L)2)(cid:17), where A estimated from an approximate analytical expression (76): Ploop(x, L) = is the normalization constant (76 -- 78) and lp is the persistence length of the loop. In our calculations, in order to make direct comparison with the experiment, we assume that the loop is a PEG chain (44) and the persistence length is 3.8Å (44, 70). (1−(x/L)2)9/2 exp(cid:16) −3L/lp 4πA(x/L)2 The coaxial stacking free energy term Gcs is an important energetic factor for nucleic acid helix align- ment (28, 79). We model this term based on the thermodynamic parameters (23, 79) and the RNA structure analysis (28) Gcs = G0 for x < 10Å and θ > 150◦; = 0 else, (3) (4) − δg. Here, XX XX GA (at where G0 = G XX 25◦C) in the experimental system (44). δg(≃ −1 kcal/mol) is an offset between the coaxial stacking energy and the corresponding canonical base pairs (79). XX is the coaxial stacked base pairs, and G XX = −1.5 kcal/mol for CT XX G3◦ is added to account for the possible tertiary contacts (45) G3◦ = g3◦ for Rcc ≤ 28Å; = 0 else, (5) (6) where g3◦ is a constant negative energy. In the calculation, we use -10kBT, -12kBT, and -14kBT to investigate the effect of the tertiary contact, respectively. We use the condition Rcc ≤ 28Å, a typical inter-axial distance for ion-induced DNA aggregates (80 -- 83), to simulate a close contact where tertiary contact can occur. Our control tests show that the small changes in the criteria around 28Å would not cause significant changes in our quantitative predictions or qualitative conclusions. For the purpose of visualizing the free energy landscape, we will also use the inter-axis angle Θ to represent the two angles θ and γ (shown in Fig. 1) to show the electrostatic free energy landscape in 2D (x, Θ) plane instead of in 3D (x, θ, γ) space. Θ = arccos(cid:18)cos2( θ 2 )cos2( γ 2 )(cid:19) , (7) Spatial confinement As shown in Fig. 1, we use Rcc, the distance between the centers of the two axes of the helices, to describe the compactness of the system. Furthermore, we use the distance between the outer axial ends 4 RPP′ ≤ Rmax, to quantify the spatial constraint of the system. A smaller Rmax corresponds to a stronger spatial confinement due to, for example, macromolecular crowders. To quantify the compactness of the system for an ensemble of conformations, we use a Boltzmann- weighted averaged value Rcc for the distance Rcc between the centers of the helix pair (70) Results and Discussions Rcc = P(x,θ,γ) Rcce−G(x,θ,γ)/kBT P(x,θ,γ) e−G(x,θ,γ)/kBT . (8) To investigate how the different ionic conditions affect the folding properties of the model system, we apply the TBI model to calculate the electrostatic free energy landscape for a wide range of Na+ and Mg2+ conditions: [Mg2+] ∈ [0.01mM, 0.1M] and [Na+] ∈ [0M, 2M]. To directly simulate the experiments (44), we assume that the system is always immersed in a 16mM Na+ background (from the Na-MOPS buffer). For convenience, we define the following two specific states for the model system: (i) the random relaxation state where the two helices can fluctuate randomly, corresponding to the state with fully neutralized helices (43, 44); (ii) the (electrostatically) folded (collapsed) state with Rcc ≤ 30Å (3, 69, 80 -- 83). In the following, we will first present the results on the electrostatic free energy landscape. We will then discuss the role of ions in the structural collapse (without the possible tertiary contact). Finally, we will discuss the effects of spatial confinement and tertiary contacts on the ion role in structural folding. Electrostatic free energy landscape In Na+ solutions. Figs. 2A-D show the electrostatic free energy landscape GE for the different added [Na+]. At low added [Na+], the helices tend to avoid each other with the largest x and Θ (see Fig. 2A) due to the weak charge neutralization. An increased [Na+] would enhance the ion binding and charge neutralization due to a lower entropic penalty for ion-binding, thus causing a weakened inter-helix repulsion. As shown in Fig. 2B, the helices fluctuate around less extended conformations with smaller x and Θ values. When [Na+] becomes very high (> 0.3M), the helices become nearly fully neutralized, thus the helices can fluctuate randomly in the full conformational space (see Figs. 2CD), except for very compact configurations (very small x and Θ) where helix-helix and ion-helix exclusions play a dominate role. In Mg2+ solutions. Figs. 2E-H show the electrostatic free energy landscapes GE for the helices with the different added [Mg2+]. At low [Mg2+], the helices form extended conformations with large x and Θ. With the increase of [Mg2+], the two helices switch to the conformations with smaller Θ and x (Figs. 2FG). For high [Mg2+] (> 3 mM), the helices tend to adopt the conformations with very small Θ (Θ . 40◦) and intermediate x (32Å . x . 47Å); see Fig. 2H. The above transition with increasing [Mg2+] can be attributed to the divalent ion-mediated helix-helix interactions (43, 69, 70). When [Mg2+] is very low, background Na+ of low concentration in buffer only gives weak neutralization to helices, thus the helices repel each other and 5 the conformations with largest x and Θ would be favorable. With the increase of [Mg2+], Mg2+ will com- pete with background Na+, and the self-organization of Mg2+ could induce (slight) attractive force between helices (69, 70). Such attraction would become stronger for the helix-helix near-parallel configurations at intermediate separation and at higher [Mg2+] (70). As a result, the helices would transform to the confor- mations with small Θ and intermediate x (69). Compared with the results for 2D (co-planar) configurations (Fig. 2 in Ref (70)), more extended con- formations (large Θ) are shown in the current 3D free energy landscapes, causing an increase in the con- formational entropy of the two helices and weakening the effect of the possible Mg2+-mediated helix-helix attractive force in the structural collapse. Ion-mediated structural collapse: without spatial confinement First, we discuss the role of ions in the structural collapse without spatial confinement and tertiary con- tact by setting G3◦ = 0 in Eq. 1. In the following, we present general features for the system and the comparisons with the experimental data (44). General features As shown in Fig. 3A, with the increase of [Na+] to 2M, Rcc decreases from a large value to that of the random relaxation (fully neutralized) state. Such a trend of Rcc is a results of the electrostatic free energy landscapes shown in Fig. 2A-D and can be attributed to the aforementioned enhanced Na+-binding at higher [Na+]. Also shown in Fig. 3A is the loop length dependence of the structural compaction at low and high [Na+]. A PEG chain is quite flexible (lp ≃ 3.8Å). The loop (length L ≫ lp) tends to minimize the loop entropic loss upon structure formation. Therefore, the loop drives the system to fluctuate around a relatively smaller end-to-end distance x since a larger x corresponds to a smaller loop entropy. For example, for PEG loops with L = 56Å and 32Å, x ≃ 20Å and 15Å, are the most favorable according to Eq. 2, respectively. The loop effect is slightly stronger for low [Na+] where the favorable configurations are extended and the effect of the (short) tether can become more important. In a solution with added [Mg2+], at low [Mg2+], the favorable conformations are those with large x and Θ, thus a loop length can contribute to the compaction of the system. In contrast, at a high [Mg2+], Mg2+ induces (slightly) attractive force between helices at small inter-axis angles Θ (see Fig. 2H). When the helices are driven by the loop to a smaller x (to gain loop conformation entropy), the low free energy conformations (in the small x regime in Figs. 2GH) correspond to a range of Θ. As a result, a small x does not necessarily lead to compaction of the system (characterized by a small Rcc). We note that a high [Mg2+] can induce a state that is slightly more compact than the random relaxation (fully neutralized) state yet much looser than the upper limit of the electrostatically folded state (3, 80 -- 83), as shown in Fig. 3B. Comparisons with experiments To make direct comparison with the experimental system (44), we assume the loop length to be L = 32Å. We calculated the Boltzmann-weighted averaged SAXS profiles I(Q) over conformation ensemble for each ionic condition. Here Q is the scattering vector which is equal to 4πsin(ϑ)/λ, where ϑ is the Bragg angle 6 and λ is the wave length of the radiation. Figs. 3C&D show the SAXS profiles for the different added [Na+] and [Mg2+], respectively. As the added [Na+] is increased from 0 to 2M or Mg2+ is added from 0 to 0.1M, the SAXS profile changes from that of the extended state to that of the nearly fully neutralized state. In addition, the profiles indicate that the system at high [Mg2+] (∼ 0.1M) is slightly more compact than that at high [Na+] (∼ 2M). Since the experimental SAXS profiles are in arbitrary unit and with vertical shifts (44), we cannot make direct comparisons for the SAXS profiles between the experiments and our predictions. Based on the calculated SAXS profiles and compactness Rcc, we can estimate the fraction "relaxed" χ for the system by (44) Ω (Na+ or Mg2+) = (1 − χ) Ω (no added salt) + χ Ω (high salt), (9) where Ω stands for a structural parameter, such as I(Q) or Rcc. Ω (no added salt) is that in 16mM Na+ background from the Na-MOPS buffer without added salt, and Ω(high salt) is taken as that of the fully- neutralized state in our estimation. Based on the equation, we estimate χ from the calculated I(Q) and Rcc respectively, and compare our predictions with experiments for the cases with added Na+ and Mg2+. Figs. 3E&F show the fraction "relaxed" χ as a function of added [Na+] and [Mg2+], respectively. When [Na+] is added, χ increases from zero to ∼ 0.9 at 2M [Na+], suggesting that the system is close to the random relaxation (fully neutralized) state. The lines estimated from I(Q) and Rcc are almost identical, and the predictions agree well with the experimental data, as shown in Fig. 3E. As shown in Fig. 3F, when Mg2+ is added, χ increases from zero to ∼ 0.9 at ∼8mM [Mg2+], and may slightly exceeds 1 when [Mg2+] & 10mM, suggesting slightly more compact state induced by high [Mg2+]. In addition, the predicted χ from Rcc is slightly larger than that from I(Q) at high [Mg2+] (shown in Fig. 3F), which is in accordance with the above described relations of Rcc and I(Q) with [Na+] and [Mg2+]. Fig. 3F also shows that the predicted χ's from the TBI model are in good accordance with the experimental data, while PB theory predicts >10-fold higher midpoint of [Mg2+] for the Mg2+-mediated collapse. This result may be attributed to the ignored ion-ion correlations in the PB. The inclusion of ion correlations would allow ions to self-organize to form low-energy state, and thus the TBI model predicts a more efficient role of Mg2+ than the PB (67, 72, 73). It is important to note that although the system has the similar global compactness at high [Na+] and [Mg2+], the free energy landscapes are very different as shown in Fig. 2. The PEG loop used in the above calculation is electrically neutral, while a realistic RNA loop is a polyanionic chain. Recent experiments suggest that the hinge stiffness can become an energetic barrier for RNA folding (39). As shown previously (70), a nucleotide loop may enhance the sharpness of ion-mediated structural collapse due to the ion-dependence of the persistence length of the loop (e.g., (70)). Ion-mediated folding: with spatial confinement In this section, we investigate the influence of spatial restriction as well as a tertiary contact on the ion- mediated collapse of the above two-helix system. 7 Electrostatic free energy landscape with spatial confinement. With the conformational confinement defined by the limit Rmax for the distance RPP′ between the outer ends of the two helices (see Fig. 1), the conformational space of the system is confined. The electrostatic free energy landscape of the confined system GE(Rcc, Rmax) is given by GE(Rcc, Rmax) = −kBT ln RPP′ ≤Rmax X Rcc e−GE (x,θ,γ)/kBT , (10) where Rcc is the distance between the two centers of the two helices (see Fig. 1). As shown in Fig. 4, the spatial restriction influences the free energy landscape significantly for both Na+ and Mg2+ solutions. For large Rmax, the free energy landscape is close to that without spatial restriction. At low [Na+], the favorable conformations are clustered in the region of large Rcc. At high [Na+] (e.g., 2M), the helices can fluctuate nearly in the whole region with all the different Rcc values. With the decrease of Rmax, the conformations of the two helices are confined by the restriction and the extended conformations are severely limited. This causes the free energy GE(Rcc, Rmax) at high [Na+] to become close to that at low [Na+]. In the limit of strong spatial confinement such as Rmax ≤ 40Å, the free energy GE(Rcc, Rmax) at the different Na+ concentrations can become nearly identical; see Figs. 4A-D in the small Rmax regime. For Mg2+ solutions, the spatial confinement effect on GE(Rcc, Rmax) is similar to that for Na+ solutions, except for the much more efficient role of the Mg2+ ions in promoting the collapse of the system. The electrostatic free energy landscapes show that with the spatial confinement, the extended conformations are much more significantly restricted than the compact conformations, causing the destabilization of the ex- tended state. In the ion-promoted collapse, the spatial confinement would effectively enhance the ion effect in folding. Ion-mediated folding: effects of spatial confinement and tertiary contacts To investigate the role of tertiary contacts in ion-induced structural collapse, we add a negative constant energy to model the effect of the tertiary contacts. Specifically, we choose three values -10kBT, -12kBT, and -14kBT for g3◦ (see Eq. 6). Helix-helix attraction arising from the tertiary contacts would cause a shift in the conformational dis- tribution toward the compact state and lower the ion concentration required to induce the collapse of the structure. Furthermore, the tertiary contact results in a much sharper ion-induced structural collapse (than the tertiary contact-free case). The effect is more pronounced for stronger tertiary contacts. As shown in Figs. 5ACE for Na+, the midpoint for the folding transition occurs at [Na+]∼ 0.5M for g3◦ = −10kBT and [Na+]∼ 0.08M for g3◦ = −14kBT , respectively, and the stronger tertiary contacts cause a slightly more compact state. As shown in Figs. 5BDF, the midpoints of of Mg2+-dependent folding are ∼0.8mM and ∼0.2mM for g3◦ = −10kBT and g3◦ = −14kBT , respectively. Again, we note that much lower [Mg2+] (∼ mM) is required to cause folding than [Na+] (∼ M), which is in accordance with the experimental data (13, 34, 72, 73). To further examine the effect of the spatial confinement in the presence of tertiary contacts, we calculated Rcc with a given Rmax constraint (the maximum distance of P-P' shown in Fig. 1). As shown in Figs. 5A-F, the spatial confinement significantly enhances the ion efficiency in mediating the structural collapse/folding 8 in both Na+ and Mg2+ solutions. For example, for g3◦ = −10kBT , with the decrease of Rmax from 120Å to 70Å, the [Na+] and [Mg2+] at the midpoint of the folding transition decreases from ∼0.5M to ∼ 0.1M and from ∼1mM to ∼0.3mM, respectively. For a stronger tertiary contact, e.g., g3◦ = −14kBT , the spatial confinement would further enhance the ion efficiency for both of Na+ and Mg2+, as shown in Figs. 5A-F. Such higher efficiency of Mg2+ than Na+ is attributed to the higher valence of Mg2+ which corresponds to the less entropy for ion binding and stronger ion-ion correlation, causing higher Mg2+ binding affinity (72) and consequently higher efficiency in stabilizing RNA (73). The above spatial confinement-promoted ion (Na+ and Mg2+) efficiency in mediating structural folding is in accordance with the recent experiment which showed that a lower Mg2+ is required to fold Azoarcus ri- bozyme in PEG environments (19). The spatial confinement can be caused by molecular crowding or other in vivo effects. The spatial confinement can destabilize the low-salt (extended) state by reducing the con- formational space and consequently decrease the electrostatic free energy difference between the compact conformations and the (restricted) extended conformations (52). In addition, the tertiary contacts would further stabilize the compact state. These two effects combined together cause a reduced ion concentration required to induce the folding transition. Therefore, Na+ and Mg2+ are more efficient in RNA folding in a crowded environment. Conclusions and Discussions In this work, we employed the newly refined TBI model (72, 73) to investigate how ions cause com- paction for a system of loop-tethered helices. The main advantage of the TBI model is its ability to account for the possible ion correlation and fluctuation effects. The present study distinguishes from the previous studies (70) on similar systems in several novel aspects. First, the present study is based on the 3D (instead of planar) rotational degrees of freedom in the conformational sampling of the system. In contrast to co-planar (2D) helix orientations, the 3D random orientation of the helices would reduce the (ensemble averaged) ion correlation effect and the associated force between the helices. Second, the main focus of the present study is the influence of the spatial confinement and the ter- tiary contacts upon the ion effect and the comparisons with the the PB-based predictions and the recent experimental results. The study leads to the following main conclusions: 1. Na+ can induce a transition from an ensemble of extended conformations to an ensemble of random relaxed conformations with a flat energy landscape. In contrast, Mg2+ can induce a transition from an extended state to a conformation ensemble that is slightly more ordered and slightly more com- pact than the random relaxation conformational ensemble. However, such a Mg2+-induced state is far looser than the (electrostatically) folded state. The predictions from the TBI model are in accor- dance with the experimental data for both Na+ and Mg2+ solutions, while the PB theory significantly underestimates the Mg2+ efficiency. 2. A tertiary contact would make the ion (Na+ and Mg2+)-dependent folding transition much sharper than the tertiary contact-free case. Furthermore, stronger tertiary contacts can fold RNAs at lower ion 9 concentrations for both of Na+ and Mg2+. 3. Spatial confinement (e.g., due to macromolecular crowding) significantly promotes the ion efficiency in mediating RNA folding for both Na+ and Mg2+. The effect of the spatial confinement can be attributed to the decreased electrostatic free energy difference between the compact ensemble and the (restricted) extended ensemble. The current theory gives overall good agreement with the experiment results, suggesting an overall reli- able analysis and predictions for the ion-induced compaction and the effects of spatial confinement. Further investigation of the problem requires improvements of the model in several aspects. First, in the structural modeling, although we have taken in account the non-planar conformations, we have ignored conformations generated from the spin of the helices and the nonsymmetric rotations of the helix axes. Including these conformations may cause changes in the free energy landscape, though a detailed study for such changes requires exceedingly high computational cost. Second, the current form of the TBI model cannot treat the sequence preference in ion-binding (71, 85). The specific binding could make important contributions to the ion effect, although for the PEG-tethered helix system here, the contribution of the specific binding may be weak. Finally, we have neglected the interference from the helix on the loop conformations except for the constraint of the end-end distance x. The effect of the loop-helix interference may not be significant for a PEG loop. However, such a effect can be important for a polynucleotide loop which is highly charged. Acknowledgments This research was supported by the National Science Foundation (grants MCB0920067 and MCB0920411), National Institutes of Health (grant GM063732) (to S.-J.C.), National Science Foundation of China (grants 11074191, 11175132, and 10844007), Program for New Century Excellent Talents (NCET 08-0408), Fundamental Research Funds for the Central Universities, the National Key Scientific Program (973)- Nanoscience and Nanotechnology (No. 2011CB933600) and the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry (to Z.-J.T.). 10 References [1] Bloomfield, V.A., D.M. Crothers, and I. Tinoco,Jr. 2000. Nucleic Acids: Structure, Properties and Functions. University Science Books, Sausalito, CA. [2] Tinoco, I., and C. Bustamante. 1999. How RNA folds. J. Mol. Biol. 293:271-281. [3] Bloomfield, V.A. 1997. DNA condensation by multivalent cations. Biopolymers 44:269-282. [4] Brion, P., and E. Westhof. 1997. Hierarchy and dynamics of RNA folding. Annu. Rev. Biophys. Biomol. Struct. 26:113-137. [5] Sosnick, T.R. and T. Pan. 2003. RNA folding: models and perspectives. Curr. Opin. Struct. Biol. 13:309-316. [6] Draper, D.E. 2008. RNA folding: thermodynamic and molecular descriptions of the roles of ions. Biophys. J. 95:5489-5495. [7] Chu, V.B., and D. Herschlag. 2008. Unwinding RNA's secrets: advances in the biology, physics, and modeling of complex RNAs. Curr. Opin. Struct. Biol. 18:305-314. [8] Chen, S.J. 2008. RNA Folding: conformational statistics, folding kinetics, and ion electrostatics. Annu. Rev. Biophys. 37:197-214. [9] Woodson, S.A. 2010. Compact intermediates in RNA folding. Annu. Rev. Biophys. 39:61-77. [10] Wong, G.C., and L. Pollack. 2010. Electrostatics of strongly charged biological polymers: ion-mediated inter- actions and self-organization in nucleic acids and proteins. Annu. Rev. Phys. Chem. 61:171-189. [11] Vander Meulen, K.A., S.E. Butcher. 2012. Characterization of the kinetic and thermodynamic landscape of RNA folding using a novel application of isothermal titration calorimetry Nucleic Acids Res. 40:2140-2151. [12] Stellwagen, E., J.M. Muse, N.C. Stellwagen. 2011. Monovalent cation size and DNA conformational stability. Biochemistry 50:3084-94. [13] Tan, Z.J., and S.J. Chen. 2011. Importance of diffuse metal ion binding to RNA. Met. Ions Life Sci. 9:101-124. [14] Minton, A.P. 2000. Implications of macromolecular crowding for protein assembly. J. Biol. Chem. 10:34-39. [15] Zhou, H.X., G. Rivas, and A.P. Minton. 2008. Macromolecular crowding and confinement: biochemical, bio- physical, and potential physiological consequences. Annu. Rev. Biophys. 37:375-97. [16] Burz, D.S., and A. Shekhtman. 2009. Structural biology: Inside the living cell. Nature 458:37-8. [17] Lambert, D., D. Leipply, and D.E. Draper. 2010. The osmolyte TMAO stabilizes native RNA tertiary struc- tures in the absence of Mg2+: evidence for a large barrier to folding from phosphate dehydration. J. Mol. Biol. 404:138-57. 11 [18] Pincus, D.L., C. Hyeon, and D. Thirumalai. 2008. Effects of trimethylamine N-oxide (TMAO) and crowding agents on the stability of RNA hairpins. J. Am. Chem. Soc. 130:7364-72. [19] Kilburn, D., J.H. Roh, L. Guo, R.M. Briber, S.A. Woodson. 2010. Molecular crowding stabilizes folded RNA structure by the excluded volume effect. J. Am. Chem. Soc. 132:8690-8696. [20] Zheng, K.W., Z. Chen, Y.H. Hao, and Z. Tan. 2010. Molecular crowding creates an essential environment for the formation of stable G-quadruplexes in long double-stranded DNA. Nucleic Acids Res. 38:327-38. [21] Rajendran, A., S. Nakano, and N. Sugimoto. 2010. Molecular crowding of the cosolutes induces an intramolec- ular i-motif structure of triplet repeat DNA oligomers at neutral pH. Chem. Commun. 46:1299-301. [22] Serra, M.J., and D.H. Turner. 1995. Predicting thermodynamic properties of RNA. Methods Enzymol. 259:242- 261. [23] SantaLucia, J.,Jr. 1998. A unified view of polymer, dumbbell, and oligonucleotide DNA nearest-neighbor ther- modynamics. Proc. Natl. Acad. Sci. USA 95:1460-1465. [24] Xia, T., J. SantaLucia, M.E. Burkard, R. Kierzek, S.J. Schroeder, X. Jiao, C. Cox, and D.H. Turner. 1998. Ther- modynamic parameters for an expanded nearest-neighbor model for formation of RNA duplexes with Watson- Crick base pairs. Biochemistry 37:14719-14735. [25] Chen, S.J., and K.A. Dill. 2000. RNA folding energy landscapes. Proc. Natl. Acad. Sci. USA. 97:646-651. [26] Zhang, W.B., and S.J. Chen. 2002. RNA hairpin-folding kinetics. Proc. Natl. Acad. Sci. USA. 99:1931-1936. [27] Zuker, M. 2003. Mfold web server for nucleic acid folding and hybridization prediction. Nucleic Acids Res. 31:3406-3415. [28] Tyagi, R., and D.H. Mathews. 2007. Predicting helical coaxial stacking in RNA multibranch loops. RNA 13:939- 951. [29] Blake, R.D., and S.G. Delcourt. 1998. Thermal stability of DNA. Nucleic Acids Res. 26:3323-3332. [30] Owczarzy, R., Y. You, B.G. Moreira, J.A. Manthey, L. Huang, M. A. Behlke, and J. A. Walder. 2004. Effects of sodium ions on DNA duplex oligomers: improved predictions of melting temperatures. Biochemistry 43:3537- 3554. [31] Tan, Z.J., and S.J. Chen. 2008. Salt dependence of nucleic acid hairpin stability. Biophys. J. 95:738-752. [32] Tan, Z.J., and S.J. Chen. 2007. RNA helix stability in mixed Na+/Mg2+ solution. Biophys. J. 92:3615-3632. [33] Tan, Z.J., and S.J. Chen. 2006. Nucleic acid helix stability: effects of salt concentration, cation valency and size, and chain length. Biophys. J. 90:1175-1190. [34] Heilman-Miller, S.L., D. Thirumalai, and S.A. Woodson. 2001. Role of counterion condensation in folding of the Tetrahymena ribozyme. I. equilibrium stabilization by cations. J. Mol. Biol. 306:1157-1166. [35] Koculi, E, C. Hyeon, D. Thirumalai, and S.A. Woodson. 2007. Charge density of divalent metal cations deter- mines RNA stability. J. Am. Chem. Soc. 129:2676-2682. 12 [36] Takamoto, K., Q. He, S. Morris, M.R. Chance, and M. Brenowitz. 2002. Monovalent cations mediate formation of native tertiary structure of the Tetrahymena thermophila ribozyme. Nature Struct. Biol. 9:928-933. [37] Soto, A.M., V. Misra, and D.E. Draper. 2007. Tertiary structure of an RNA pseudoknot is stabilized by "diffuse" Mg(2+) ions. Biochemistry 46:2973-2983. [38] Kim, H.D., G.U. Nienhaus, T. Ha, J.W. Orr, J.R. Williamson, and S. Chu. 2002. Mg2+-dependent conforma- tional change of RNA studied by fluorescence correlation and FRET on immobilized single molecules. Proc. Natl. Acad. Sci. USA 99:4284-4289. [39] Schlatterer, J.C., L.W. Kwok, J.S. Lamb, H.Y. Park, K. Andresen, M. Brenowitz, and L. Pollack. 2008. Hinge stiffness is a barrier to RNA folding. J. Mol. Biol. 379:859-870. [40] Lipfert, J., A.Y. Sim, D. Herschlag, S. Doniach. 2010. Dissecting electrostatic screening, specific ion binding, and ligand binding in an energetic model for glycine riboswitch folding. RNA 16:708-719. [41] Qiu, X., K. Andresen, L.W. Kwok, J.S. Lamb, H.Y. Park, L. Pollack. 2007. Inter-DNA attraction mediated by divalent counterions. Phys. Rev. Lett. 99:038104. [42] Qiu, X., K. Andresen, J.S. Lamb, L.W. Kwok, L. Pollack. 2008. Abrupt transition from a free, repulsive to a condensed, attractive DNA phase, induced by multivalent polyamine cations. Phys. Rev. Lett. 101:228101. [43] Bai, Y., R. Das, I. S. Millett, D. Herschlag, and S. Doniach. 2005. Probing counterion modulated repulsion and attraction between nucleic acid duplexes in solution. Proc. Natl. Acad. Sci. USA 102:1035C1040. [44] Bai, Y., V.B. Chu, J. Lipfert, V.S. Pande, D. Herschlag, S. Doniach. 2008. Critical assessment of nucleic acid electrostatics via experimental and computational investigation of an unfolded state ensemble. J. Am. Chem. Soc. 130:12334-12341. [45] Chu, V.B., J. Lipfert, Y. Bai, V.S. Pande, S. Doniach, D. Herschlag. 2009. Do conformational biases of simple helical junctions influence RNA folding stability and specificity? RNA 15:2195-2205. [46] Cheung, M.S., D. Klimov, D. Thirumalai. 2005. Molecular crowding enhances native state stability and refolding rates of globular proteins. Proc. Natl. Acad. Sci USA. 102:4753-8. [47] Kudlay, A., M.S. Cheung, D. Thirumalai. 2009. Crowding effects on the structural transitions in a flexible helical homopolymer. Phys. Rev. Lett. 102:118101. [48] Wang, W., W.X. Xu Y. Levy, E. Trizac, P.G. Wolynes. 2009. Confinement effects on the kinetics and thermody- namics of protein dimerization. Proc. Natl. Acad. Sci. USA 106:5517-22. [49] Qin, S., and H.X. Zhou. 2009. Atomistic of macromolecular crowding predicts modest increases in folding and binding stability. Biophys. J. 97:12-19. [50] Jiao, M., H.T. Li, J. Chen, A.P. Minton, Y. Liang. 2010. Attractive protein-polymer interactions markedly alter the effect of macromolecular crowding on protein association equilibria. Biophys. J. 99:914-23. [51] Wang, Q., K.-C. Liang, A. Czader, M. N. Waxham, M. S. Cheung. 2011. The effect of macromolecular crowding, ionic strength and calcium binding on calmodulin dynamics. PLoS Comput. Biol. 7:e1002114. 13 [52] Denesyuk, N., and D. Thirumalai. 2011. Crowding Promotes the Switch from Hairpin to Pseudoknot Confor- mation in Human Telomerase RNA. J. Am. Chem. Soc. 133:11858-11861. [53] Chen, A.A., M. Marucho, N.A. Baker. and R. Pappu. 2009. Simulations of RNA Interactions with Monovalent Ions. Methods Enzymol. 469:411-432. [54] Chen, A.A., D.E. Draper. and R.V. Pappu. 2009. Molecular simulation studies of monovalent counterion-me diated interactions in a model RNA kissing loop. J. Mol. Biol. 390:805-819. [55] Auffinger, P., L. Bielecki, and E. Westhof. 2003. The Mg2+ Binding Sites of the 5S rRNA Loop E Motif as Investigated by Molecular Dynamics Simulations. Chem. Biol. 10:551-561. [56] Dong, F., B. Olsen. and N.A. Baker. 2008. Computational Methods for Biomolecular Electrostatics. Methods Cell Biol. 84:843-870. [57] Joung, I. and T.E. Cheatham. 2009. Molecular dynamics simulations of the dynamic and energetic properties of alkali and halide ions using water-model specific ion parameters. J. Phys. Chem. B. 113:13279-13290. [58] Kirmizialtin, S., Pabit, S.A., Meisburger, S.P., Pollack, L., Elber, R. 2012 RNA and its ionic cloud: Solution scattering experiments and atomically detailed simulations Biophy. J. 102:819-828. [59] Manning, G.S. 1978. The molecular theory of polyelectrolyte solutions with applications to the electrostatic properties of polynucleotides. Q. Rev. Biophys. 11:179-246. [60] Gilson, M.K., K.A. Sharp, and B. Honig. 1987. Calculating the electrostatic potential of molecules in solution: method and error assessment. J. Comput. Chem. 9:327-335. [61] You, T.J., and S.C. Harvey. 1993. Finite element approach to the electrostatics of macromolecules with arbitrary geometries. J. Comput. Chem. 14:484-501. [62] Baker, N.A., D. Sept, S. Joseph, M.J. Holst, and J.A. McCammon. 2000. Electrostatics of nanosystems: Appli- cation to microtubules and the ribosome. Proc. Natl. Acad. Sci. USA 98:10037-10041. [63] Boschitsch, A.H., and M.O. Fenley. 2007. A new outer boundary formulation and energy corrections for the nonlinear Poisson-Boltzmann equation. J. Comput. Chem. 28:909-921. [64] Chen, D., Z. Chen, C. Chen, W. Geng, and G.W. Wei. 2011. MIBPB: A software package for electrostatic analysis. J. Comput. Chem. 32:756-70. [65] Lu, B., X. Cheng, J. Huang, and J.A. McCammon. 2010. AFMPB: An adaptive fast multipole Poisson- Boltzmann solver for calculating electrostatics in biomolecular systems. Comput. Phys. Commun. 181:1150- 1160. [66] Tan, Z.J. and S.J. Chen. 2009. Predicting electrostatic forces in RNA folding. Methods Enzymol. 469:465-487. [67] Tan, Z.J., and S.J. Chen. 2005. Electrostatic correlations and fluctuations for ion binding to a finite length polyelectrolyte. J. Chem. Phys. 122:044903. [68] Grochowski, P., J. Trylska. 2008. Continuum molecular electrostatics, salt effects and counterion binding. A review of the Poisson-Boltzmann theory and its modifications. Biopolymers 89:93-113. 14 [69] Tan, Z.J., and S.J. Chen. 2006. Ion-mediated nucleic acid helix-helix interactions. Biophys. J. 91:518-536. [70] Tan, Z.J., and S.J. Chen. 2006. Electrostatic free energy landscape for nucleic acid helix assembly. Nucleic Acids Res. 34:6629-6639. [71] Tan, Z.J., and S.J. Chen. 2008. Electrostatic free energy landscapes for DNA helix bending. Biophys. J. 94:3137- 3149 [72] Tan, Z.J., and S.J. Chen. 2010. Predicting ion binding properties for RNA tertiary structures. Biophys. J. 99:1565- 1576. [73] Tan, Z.J., and S.J. Chen. 2011. Salt contribution to RNA tertiary folding stability. Biophys. J. 101:176-187. [74] Chen, G., Z.J. Tan, and S.J. Chen. 2010. Salt dependent folding energy landscape of RNA three-way junction. Biophys. J. 98:111-120. [75] Lu, X.J., and W.K. Olson. 2003. 3DNA: a software package for the analysis, rebuilding and visualization of three-dimensional nucleic acid structures. Nucleic Acids Res. 31:5108-5121. [76] Thirumalai, D. and B.Y. Ha. 1998. Statistical mechanics of semiflexible chains In Grosberg, A. (Ed.). Theoretical and mathematical models in polymer research, San Diego, CA Academic Press. Vol.1, pp.135. [77] Hyeon, C., R.T. Dima, D. Thirumalai D. 2006. Size, shape, and flexibility of RNA structures. J. Chem. Phys. 125:194905. [78] Murphy, M.C., I. Rasnik, W. Cheng, T.M. Lohman, and T. Ha. 2004. Probing single-stranded DNA conforma- tional flexibility using fluorescence spectroscopy. Biophys. J. 86:2530-2537. [79] Walter, A.E., and D.H. Turner. 1994. Sequence dependence of stability for coaxial stacking of RNA helixes with Watson-Crick base paired interfaces. Biochemistry 33:12715-12719. [80] Rau, D.C. and Parsegian, V.A. 1992. Direct measurement of the intermolecular forces between counterion- condensed DNA double helices. Evidence for long range attractive hydration forces. Biophys. J. 61:246-259. [81] Raspaud, E., Durand, D., Livolant, F. 2005. Interhelical spacing in liquid crystalline spermine and spermidine- DNA precipitates Biophys. J. 88:392-403. [82] Todd, B.A., V.A. Parsegian, A. Shirahata, T.J. Thomas, and D.C. Rau. 2008. Attractive forces between cation condensed DNA double helices. Biophys. J. 94:4775-4782. [83] Varnai, P., and Y. Timsit. 2010. Differential stability of DNA crossovers in solution mediated by divalent cations. Nucleic Acids Res. 38:4163-4172. [84] D. Svergun, C. Barberato and M. H. J. Koch. 1995. CRYSOL-a program to evaluate X-ray solution scattering of biological macromolecules from atomic coordinates. J. Appl. Cryst. 28:768-773. [85] Stellwagen, E., Q. Dong, and N.C. Stellwagen. 2007. Quantitative analysis of monovalent counterion binding to random-sequence, double-stranded DNA using the replacement ion method. Biochemistry 46:2050-2058. 15 FIGURE CAPTION FIGURE 1 The conformation of two (DNA) helices tethered by a loop can be characterized by three struc- tural parameters: the end-to-end distance x (O-O'), the angles θ and γ illustrated in the figure. The freedom of γ (rotation of the axis into the paper) is used to produce non-planar configurations. The P-P' distance RPP′ ≤ Rmax represents the spatial restriction for the two helices due to, e.g., the macro- molecular crowding effect. Rcc between the centers of the helices is used to quantify the compactness of the model. Tertiary contact energy can be added for conformations with a small Rcc. The DNA helices are constructed with the software X3DNA (75) and are displayed with the software PyMol (http://pymol.sourceforge.net/). FIGURE 2 The electrostatic free energy landscapes GE(x, θ, γ) as a function of [Na+] (A-D) and [Mg2+] (E-F). Here x is the end-to-end distance and Θ is the angle between the two helix axes. Θ is deter- mined by the (θ,γ) angles (see Eq. 7). The color scale represents the energy difference between the configuration (x, Θ) and the state with the minimum free energy. From 0 to 5 kBT , the color changes gradually from Red to Blue. Note that the system is in a 16mM Na+ background due to the Na-MOPS buffer (44). FIGURE 3 Ion-mediated structural collapse in the absence of the tertiary contact and spatial confinement. (A, B) Ion concentration-dependence of the compactness Rcc (the mean distance between the he- lix centers) as a function of Na+ (A) and Mg2+ (B). Dashed lines denote the random relaxation (fully-neutralized) state and the dotted lines denote the upper limit of the electrostatically folded state (Rcc ∼ 30Å) (69, 70, 80 -- 82). (C,D) The SAXS profiles calculated for the different added Na+ (C) and Mg2+ (D). Lines are the reference SAXS profiles. Dashed lines: the upper limit of the elec- trostatically folded state (Rcc ≃ 30Å); Solid lines: the random relaxation (fully-neutralized) state; Dotted lines: the extended state (with no added salts). The SAXS profiles are calculated from the software CRYSOL (84) with the statistical weight determined from the free energy G(x, θ, γ). (E,F) The "relaxed" fraction χ as a function of [Na+] (E) and [Mg2+] (F), respectively; see Eq. 9. Symbols: the experimental data by Bai et al (44); Solid lines: the calculated results with the SAXS profiles I(Q) determined from the TBI model; Dotted lines: calculated with the compactness Rcc determined from the TBI model; Dashed lines: calculated with the SAXS profiles determined from the PB theory by Bai et al (44). The "relaxed" state is taken as the fully neutralized state. Note that the system is in 16mM Na+ background due to the Na-MOPS buffer (44). FIGURE 4 The electrostatic free energy landscapes GE(Rcc, Rmax), where Rmax represents the spatial re- striction for the helices (e.g., due to the macromolecular crowding); see Eq. 10. The color scale denotes the energy difference between GE(Rcc, Rmax) and the minimum value for each Rmax: ∆GE = GE(Rcc, Rmax) − GE(Rcc, Rmax)min. From 0 to 5kBT , the color changes uniformly from Red to Blue. FIGURE 5 Ion-mediated structural folding in the presence of tertiary contact and spatial restriction. (A,B) The compactness Rcc of the loop-tethered helices as functions of [Na+] (A) and [Mg2+] (B) for the different Rmax values (from the right to the left: 120Å, 82Å, 78Å, and 70Å) and the different tertiary 16 contact energy g3◦ (solid lines: -10kBT , and dashed lines: -14 kBT ). (C,D) The folded fraction ffolded as a function of [Na+] (C) and [Mg2+] (D) for the different Rmax values (from the right to the left: 120Å, 82Å, 78Å and 70Å) and the different tertiary contact energy g3◦ (solid lines: -10kBT , and dashed lines: -14 kBT ). (E,F) The ion concentration midpoints for the folding transition as a function of Rmax and the different tertiary contact energies g3◦. From the top to the bottom: g3◦=-10kBT , -12kBT , and -14kBT , respectively. 17 γ P' P θ R PP' x O O' Figure 1: 18 180 Θ 0 5 180 Θ 0 5 A 0M Na+ B 0.08M Na + C 0.3M Na+ D 2M Na+ E 0.05mM Mg2+ 56 5 F 0.6mM Mg2+ 56 5 G 2mM Mg2+ 56 5 H 10mM Mg 2+ 56 Ax 556 Ax 56 5 Ax 56 5 Ax 56 Figure 2: 19 1.2 1 0.8 0.6 0.4 0.2 ' ' d e x a l e r ' ' n o i t c a r f 0 0.001 1.2 1 ' ' d e x a l e r ' ' n o i t c a r f 0.8 0.6 0.4 0.2 0 1e−05 expt data TBI with SAXS profiles TBI with R ' s cc E Na+ 0.01 0.1 + [Na ] (M) 1 expt data TBI with SAXS profiles TBI with R 'scc PB from Bai et al F Mg 2+ 0.01 0.1 0.0001 0.001 2+ [Mg ] (M) L=56A L=32A 60 50 A Na+ ) A ( c c R 40 30 0.001 electrostatically folded state 0.01 0.1 [Na ] (M) + L=56A L=32A 60 50 B Mg2+ ) A ( c c R C Na+ 3 5 0 1 * ) Q 2 ( I * Q 1 + 2M Na 0.3M Na+ + 0.08M Na + 0M Na 1 0 0 0 0.1 Q (1/A) 0.2 D Mg 2+ 3 5 0 1 * ) Q 2 ( I * Q 1 2+ 0.1M Mg 2+ 0.01M Mg 2+ 2mM Mg 2+ 0.6mM Mg 0.05mM Mg 2+ 40 30 1e−05 electrostatically folded state 0.0001 0.001 2+ [Mg ] (M) 0.01 0.1 0 0.1 Q (1/A) 0.2 Figure 3: 20 A 0M Na+ B 0.08M Na + C 0.3M Na+ D 2M Na+ A c c R A c c R E 0.05mM Mg2+ F 0.6mM Mg2+ G 2mM Mg2+ H 10mM Mg 2+ Rmax A Rmax A Rmax A Rmax A Figure 4: 21 50 ) A ( c c R 40 30 0.001 50 ) A 40 ( c c R 30 1e−05 A Na+ 1 0.8 0.6 d e d l o f f 0.4 0.2 0.01 0.1 [Na ] (M) + 1 0 0.001 B Mg2+ 1 0.8 0.6 d e d l o f f 0.4 0.2 0.0001 0.001 2+ [Mg ] (M) 0.01 0.1 0 1e−05 + C Na 0.01 0.1 [Na ] (M) + 1 1 ) M ( m ] + a N 0.1 [ 0.01 1 ) M 0.1 ( m ] + 2 g M [ g 3 B = − 1 0 k T − 1 2 k T B − 1 4 k T B E Na+ 82 86 70 74 78 max R (A) g 3 B = − 1 0 k T − 1 2 k T B − 1 4 k T B 0.0001 0.001 2+ [Mg ] (M) D Mg 0.01 2+ 0.1 0.01 70 74 F Mg2+ 78 max R (A) 82 86 Figure 5: 22
1803.11396	2	1803	2018-06-13T10:33:38	Distribution of label spacings for genome mapping in nanochannels	[ "physics.bio-ph" ]	In genome mapping experiments, long DNA molecules are stretched by confining them to very narrow channels, so that the locations of sequence-specific fluorescent labels along the channel axis provide large-scale genomic information. It is difficult, however, to make the channels narrow enough so that the DNA molecule is fully stretched. In practice its conformations may form hairpins that change the spacings between internal segments of the DNA molecule, and thus the label locations along the channel axis. Here we describe a theory for the distribution of label spacings that explains the heavy tails observed in distributions of label spacings in genome mapping experiments.	physics.bio-ph	physics	Distribution of label spacings for genome mapping in nanochannels D. Odman,1 E. Werner,1 K. D. Dorfman,2 C. R. Doering,3 and B. Mehlig1 1)Department of Physics, University of Gothenburg, 41296 Gothenburg, Sweden 2)Department of Chemical Engineering and Materials Science, University of Minnesota, Minneapolis, MN 55455, USA 3)Center for the Study of Complex Systems, University of Michigan, Ann Arbor, MI 48109-1042, USA (Dated: 27 August 2018) In genome mapping experiments, long DNA molecules are stretched by confining them to very narrow channels, so that the locations of sequence-specific fluorescent labels along the channel axis provide large-scale genomic information. It is difficult, however, to make the channels narrow enough so that the DNA molecule is fully stretched. In practice its conformations may form hairpins that change the spacings between internal segments of the DNA molecule, and thus the label locations along the channel axis. Here we describe a theory for the distribution of label spacings that explains the heavy tails observed in distributions of label spacings in genome mapping experiments. 8 1 0 2 n u J 3 1 ] h p - o i b . s c i s y h p [ 2 v 6 9 3 1 1 . 3 0 8 1 : v i X r a 1 I. INTRODUCTION Long DNA molecules in ionic solution adopt random conformations. In equilibrium, the size of such DNA blobs is determined by a balance between entropic forces and excluded- volume interactions.1 In order to experimentally study local properties (melting patterns, DNA-protein reactions, DNA-sequence information), it has been suggested to stretch the DNA molecule, either by applying a force to both ends of the molecule,2,3 or by confining it to a nanochannel.4 -- 7 In next-generation genomics, for example, the locations along the channel axis of sequence-specific fluorescent labels attached to the DNA molecule can be read by microscopy,8 -- 12 providing a genetic fingerprint. This genome mapping technique requires the molecule to assume an effectively linear conformation, so that its global extension is close to its contour length. This can be achieved by making the channel very narrow, of the order of the persis- tence length (cid:96)P of the DNA molecule or smaller.13 -- 18 However, if the channel is not narrow enough, turns ('hairpin' bends) reduce the DNA extension along the channel axis.15,19 -- 28 In genome mapping experiments, such hairpins may cause errors by changing the spacings X between the fluorescent labels. Larger hairpins may even change their order. The distribution of label spacings has been measured in experiments.29,30 It exhibits a heavy tail at small spacings. The origin of this tail is not understood; it could be due to hairpins or small deflections from locally straight conformations.29 There is neither a theory for this distribution, nor are there simulation results that quantify the microscopic DNA conformations. Even the most efficient algorithms simulating steady-state conformations of discrete wormlike chains31 have not reliably computed the tails of the distribution for such narrow channels. Here we derive a theory for the distribution of label spacings in narrow channels with channel widths of the order of the persistence length. We model the correlated process of hairpin bends by the telegraph model introduced in Ref. 32, and account for deflections from straight polymer segments using a Gaussian approximation.33 We find a closed expression for the distribution in the limit where hairpin distances are large. The theory explains why the distribution is significantly skewed: the heavy left tail is caused by many relatively short S- shaped hairpins, which do not arise due to the cooperative mechanism proposed elsewhere.22 The right tail is approximately Gaussian. The theory predicts that the distribution depends 2 sensitively on the channel width D. Our results are in good agreement with measured label- spacing distributions for strongly confined DNA29 for wide channels (D = 50 nm), and the theory says that hairpins are frequent. For narrower channels (D = 41 nm) the theory predicts that hairpins are rare. In the experiment, short label spacings are somewhat more frequently observed than predicted. We discuss reasons for this discrepancy. II. TELEGRAPH MODEL To compute the distribution of label spacings along the channel axis, we use the model derived in Ref. 32, projecting the three-dimensional DNA configurations x(t) onto the chan- nel axis x. The model consists of two parts: an ideal correlated random walk, and a bias that takes into account self avoidance. Note that this model is distinctly different from the accelerated-particle method.34,35 This method does not take into account self avoidance. It applies therefore when there are no hairpins, that is in the extreme Odijk limit. We write the probability P1[x(t)] of observing the one-dimensional, projected configura- tion x(t) as P1[x(t)] = P0[x(t)] A [x(t)] . (1) Here P0 is the distribution of the ideal telegraph process describing the position x(t) at time t of a particle moving with speed v0. The particle changes its velocity ±v0 randomly at rate r. Random changes in the sign of the velocity give rise to S-hairpins and C-hairpins (Fig. 1). The process lasts from t = 0 to t = T . The parameters of the telegraph process map to those of the three-dimensional problem by letting T be the contour length of the confined polymer and determining the parameters v0 and r by comparing correlations. The velocity correlations of the telegraph model decay exponentially36 (cid:104)v(t)v(0)(cid:105) = v2 0 exp(−2rt). The tangent correlations of the confined polymer have the same form:32 (cid:104)vx(t)vx(0)(cid:105) = a2 exp(−t/g), where the contour-length coordinate t corresponds to time in the telegraph process. Further, vx is the x-component of the tangent vector v at t, the parameter a characterises the degree to which the tangent vectors align with the channel direction,37 and g is the global persistence length.19 One concludes that v0 = a and r = (2g)−1. The factor A in Eq. (1) equals the fraction of three-dimensional polymer configurations that contain no colliding segments. For narrow channels, an expression for A was derived 3 FIG. 1. (a) Conformation of a confined DNA molecule. The distance along the channel axis between the two fluorescent labels (green) is denoted by X. The conformation shown exhibits two S-hairpins and a C-hairpin at the left end. (b) Representation within the telegraph model. in Ref. 32: A [x(t)] = exp(cid:2) − (cid:90) dx pcoll(x)(cid:3) with pcoll = Ns(x)[Ns(x) − 1] . ε 2v2 0 (2) Here Ns(x) is the number of strands at location x, and ε parametrises the penalty for overlaps of the process. This parameter depends on the persistence length (cid:96)P and upon the effective width w of the confined DNA molecule, as well as on the channel width D. Assuming that a segment of length λ = ((cid:96)PD2)1/3 (the Odijk deflection length13) is unlikely to overlap with another segment, an expression for ε was derived in Ref. 32: ε = (cid:104)δ(y − y(cid:48))δ(z − z(cid:48))vex(cid:105)ideal/(cid:96)2 . (3) Here y and z are transverse channel coordinates of a polymer segment of length (cid:96). Primed coordinates label a second, independent segment, and vex denotes the excluded volume between these two segments. It depends upon the orientation of the segments. If (cid:96) (cid:29) w, then vex = 2w(cid:96)2 sin θ, where θ is the angle between the two segments.38 The average (cid:104)···(cid:105)ideal is over the conformations of the confined ideal polymer. We can write the integral over the collision probability as (cid:90) dx pcoll(x) = εX v2 0 ρ(0) k k(k−1) 2 . (4) ∞(cid:88) k=2 ∞(cid:88) k=2 Here ρ(0) k is the fraction of channel positions x with k ideal DNA strands. The k = 1-term does not contribute to the sum. Putting these results together we obtain: A [x(t)] = exp(cid:2) − εX v2 0 (cid:3) . k(k−1) 2 ρ(0) k (5) In summary, Eq. (5) describes conformation fluctuations on contour-length scales larger than the Odijk deflection length λ, and it is assumed that self avoidance is sufficiently weak so 4 1 (a)(b)XXthat a hairpin of contour length λ is unlikely to involve collisions. The bias A leads to a penalty against configurations of the ideal process with significant overlaps. In this model the effect of self avoidance is determined by the dimensionless parameter combination α = ε 2r v0 = εg a . (6) In Ref. 32 the mean and variance of DNA extension in a nanochannel was computed numerically as a function of α by simulations of the telegraph model. It was shown that the results of direct numerical simulations of DNA extension in a nanochannel fall on universal scaling curves when plotted as a function of α, as predicted by the telegraph model. The small-α behaviour of the mean extension and its variance was analysed in detail, and it was also argued that the extension variance scales as ∝ α−3 for large α, in the telegraph model. In the following we show how to compute the distribution of label spacings in the limit of large α and large rT . We obtain this distribution in closed form, consistent with the prediction that the variance is proportional to α−3, and it allows to calculate the prefactor. III. DISTRIBUTION OF LABEL SPACINGS FOR rT (cid:29) 1 AND α (cid:29) 1 A. Ideal process Consider first the ideal process. We can disregard the DNA-contour pieces to the left and to the right of the labels in Fig. 1. Assume that the process starts at x(0) = 0 with ±v0 with equal probability. In the limit of T → ∞ for fixed X(cid:48) = X/(v0T ) the distribution P0(X, T ) of X = x(T ) is derived in Appendix A: √ (cid:114) (7) P0(X(cid:48), T ) ∼ 1 2 1 − X(cid:48)2 1 + (1 − X(cid:48)2)3/4 erT ( rT 2π √ 1−X(cid:48)2−1) , normalised to unity on −1 ≤ X(cid:48) ≤ 1, for large rT . B. Self avoidance To take into account self avoidance for α (cid:29) 1, it suffices to consider the large-X tail of P0(X, T ), where X ≈ v0T . In this tail, hairpins are short and rare, so that they do not overlap. In the large-T limit, the contribution of C-hairpins to X is negligible in the ideal process, because it is O(1), whereas the contribution of S-hairpins is O(T ). We can therefore 5 FIG. 2. Distribution P1(X(cid:48), T ) of dimensionless labels spacings X(cid:48) along the channel axis, for the telegraph model. The blue lines show the theory, Eq. (9). Symbols show data from simulations of the telegraph model (see Appendix B). Parameters left panel: r = 0.00275, T = 7000, ε = 0.01. Right panel: r = 0.0055, T = 20000, ε = 0.02. Both panels: v0 = 1 so that α = 1.81. approximate Eq. (5) as A [x(t)] = exp(3ρ(0) Contour length and extension are related by v0T = X(ρ(0) 3 εX/v2 A [x(t)] = exp(cid:2) 3ε (T v0 − X)(cid:3) . 2v2 0 0). Normalisation implies: ρ(0) 1 + ρ(0) 3 = 1. 1 + 3ρ(0) 3 ). Solving for ρ(0) 3 yields (8) Using Eqs. (1), (7) and (8) we obtain an expression for the large-X tail of the spacing distribution in the weakly self-avoiding telegraph model: √ P1(X(cid:48), T ) ∼ N 1+ (1−X(cid:48)2)3/4 e−rT S(X(cid:48)) , 1−X(cid:48)2 with normalisation factor N , and with S(X(cid:48)) = 3α(1 − X(cid:48)) + 1 − √ 1 − X(cid:48)2 . (9a) (9b) The distribution has a heavy left tail resulting from conformations shortened by hairpins. It depends on two dimensionless parameters, α and rT . The maximum of P1(X(cid:48), T ) is at X(cid:48) ∼ 1 − 1/(18α2). Expanding the action S(X(cid:48)) around this point we find S(X(cid:48)) ∼ Smin + 27 2 α3 δX(cid:48)2. This determines the variance: 0T ∼ 2 2r v2 σ2 1 27α−3 . (10) The numerical prefactor ≈ 0.09 found in Ref. 32 is in reasonable agreement with 2/27 ≈ 0.074. Fig. 2 shows results for P1(X, T ) from computer simulations of the telegraph model (see Appendix B). We observe excellent agreement. 6 1rT=19.25rT=110P1(X0,T)X0X0The model predicts microscopic conformation properties, such as the distribution of hair- pin lengths xH along the channel axis. In the ideal process, hairpin turns are Poisson distributed with rate r. The probability for two strands to overlap for a length xH is ∝ exp(−2rαxH/v0), and there are three pairs of strands to check for overlaps. So the proba- bility of surviving the self-avoidance check is ∝ A ∼ exp(−6rαxH/v0). Thus the distribution of xH is exponential with rate 6rα/v0 P (xH) = (6rα/v0) exp(−6rαxH/v0) , (11) and the mean hairpin length is (cid:104)xH(cid:105) = v0/(6rα). In the limit D/(cid:96)P → 0, the global persistence length g = 1/(2r) diverges as D/(cid:96)P → 0.32 But since r cancels out in (cid:104)xH(cid:105) [see Eq. (6)], the mean hairpin length (cid:104)xH(cid:105) depends only weakly on D/(cid:96)P. We conclude that overlaps become rare as D/(cid:96)P → 0 because there are fewer and fewer hairpins, not because their length tends to zero. IV. COMPARISON WITH EXPERIMENT Reinhart et al. 29 report experimental measurements of distributions of fluorescent label spacings. We now show that Eq. (9) explains the shape of the measured distributions, and that it reveals the mechanism causing the substantial skewness of the measured distributions. Up to this point we have neglected the effect of small deflections from straight contours. The telegraph model takes into account the fact that the DNA segments need not align with the channel direction, but neglects the effect of small deflections on the extension. In narrow channels (D (cid:28) (cid:96)P), Odijk's theory says that deflections cause the extension to slightly contract, and that the fluctuations of the extension around its mean are Gaussian in the limit of large contour-length separations (L (cid:29) λ), with variance33 σ2 Odijk = 0.0096 LD2 eff/(cid:96)P . (12) Here and in the following, we express all results in terms of the physical variables contour length L, alignment factor a, and global persistence length g, instead of T , v0, and r. For wider channels, the variance is expected to be larger than that given by Eq. (12), but a central-limit argument shows that the distribution remains Gaussian. We model the extension fluctuations due to small deflections as Gaussian. Since the 7 product of Gaussians is Gaussian we find for the label-spacing distribution: dδX δ(X − X1 + δX) P1(X1, L) ρ(δX) P(X, L) = dX1 (cid:90) (cid:90) aL (cid:90) aL 0 (13) (14) = dX1P1(X1, L)ρ(X1 − X) . Here P1(X, L) is the distribution of spacings X between fluorescent labels with contour- 0 length distance L as obtained from the telegraph model, and 0)−1/2 exp[−δX 2/(2σ2 ρ(δX) = (2πσ2 0)] , (15) σ2 = σ2 with free parameter σ2 0, the variance due to small deflections (no hairpins). Eq. (15) gives 1 is the extension variance in the telegraph model, asymptotic to Eq. (10). The label spacing distribution P(X, L) depends on three 0 for the extension variance, where σ2 1 + σ2 dimensionless parameters: α (self avoidance), L/(2g) (number of hairpin turns), and σ0/(aL) (effect of deflections). The parameter σ0 in Eq. (15) could in principle be obtained by computer simulations of short 3D confined ideal wormlike chains, discarding all conformations that have hairpins. Here we take a different approach. For the experimental parameters given in Table I, the distribution P1(X, L) looks qualitatively like the distributions shown in the left panels of Fig. 2: P1(X, L) increases monotonously as a function of X until aL, and it is zero for X > aL. As a consequence, Eq. (13) predicts that the right tail of P(X, L) (for aL < X < L) is caused by deflections in the absence of hairpins. We can therefore determine σ0 by fitting TABLE I. Parameters for comparison with experiments in Fig. 3. Deff = D − w is the effective channel width (see main text). Other parameters: (cid:96)P is the persistence length of the DNA molecule, w is its effective width, g is the global persistence length, a is the alignment factor in the telegraph model. The parameters (cid:96)P, w, g, and a are obtained from the Supplemental Material from Ref.32 The parameters α and σ0 are defined in Eqs. (3) and (11) in the text. D (nm) Deff (nm) (cid:96)P (nm) w (nm) g (nm) a α σ0(28125 bp) (nm) σ0(53125 bp) (nm) 40 42 51 53 34.4 36.4 45.4 47.4 52 52 52 52 5.6 5.6 5.6 5.6 940 743 362 321 0.85 8.53 0.84 6.13 72 72 0.81 2.10 112 0.80 1.74 99 117 117 176 137 8 FIG. 3. Comparison between the experimental data from Reinhart et al. (2015),29 triangles, and theory, solid lines. The theory is based on Eq. (13) and on simulations of the telegraph model for P1(X, L). The hashed region indicates the contribution of hairpin-conformations. The solid red lines correspond to the effective channel widths Deff quoted in Table I. The solid green lines were obtained using Deff that are 2 nm larger (see text). The blue dashed line shows the Gaussian ρ(δX), with σ0 fitted (see text). the right tail of P(X, L) to the experimental data. The fitted values of σ0 (Table I) are somewhat larger than the Odijk prediction in Eq. (12) -- by a factor of two, roughly. Given the value of σ0, Eq. (13) yields a parameter-free prediction for the effect of hairpins upon the distribution P(X, L) of fluorescent label spacings X at contour-length separation L. The result is shown in Fig. 3. The Gaussian right tails (blue dashed lines in Fig. 3) are good fits to the experimental data29 (triangles). Also shown is the theory for P, based on Eqs. (13,15) and telegraph simulations for P1(X, L), red solid lines. The simulation parameters were obtained using the mapping derived in Ref. 32, see Table I. The parameter values in this table are based on the estimate Deff = D − w for the 'effective channel width', argued to take into account screened electrostatic interactions between the DNA and the channel walls.39,40 However, the quoted expression for Deff is just a rough estimate. We have therefore run a second set of telegraph simulations for slightly larger values of Deff, namely 36.4 nm and 47.4 nm (solid green lines). The parameters used are also given in Table I. 9 1(a)(b)(c)(d)D=51nmD=40nmlog10P(XhXi)XhXi(nm)-6-4-2-6-4-2-6-4-2-6-4-2-100001000-100001000-100001000-100001000L=53125bpL=28125bpPanels (a) and (b) in Fig. 3 show the comparison for the widest channel measured, for contour-length separations L = 53125 bp and 28125 bp. We estimate that these values correspond to L = 18 µm and 9.6 µm, assuming that the low intercalation used in these experiments does not affect L.41 We observe excellent agreement between theory and exper- iment. The hashed regions indicate the difference between the Gaussian approximation and the full theory, it corresponds to the contributions of hairpins shortening the conformations. We see that the effect is substantial. We observe also that the extension distributions in Fig. 3(a,b) are relatively insensitive to the precise channel size; the telegraph results are almost the same for the two different values of Deff. V. DISCUSSION AND CONCLUSIONS Are the heavy left tails in Fig. 3(a,b) caused by few long hairpins or by many small ones? Using the parameters from Table I we find that the mean hairpin length is (cid:104)xH(cid:105) ≈ 47 nm. This means that the typical hairpins are quite short, typically much shorter than the 2500 bp (850 nm) lower bound for resolving nearest-neighbor fluorescent labels in experiments.41 This is an important observation because the experimental data were conditioned on the sequence of fluorescent labels.29 Conformations that did not agree with the order of labels obtained from a reference genome were discarded. If hairpins frequently changed the order of fluorescent labels, the conditioning would cause a bias in the extension distribution. Here we can conclude that this effect is likely to be small, because the hairpins are quite short. This also means that expected hairpin contour length is of the order of the deflection length λ, which implies that the assumptions of the theory are only marginally met for the data from Ref. 29. Now consider the narrow-channel data in Fig. 3(c,d). The right tail of the extension is approximately Gaussian (blue dashed lines), and the left tails are heavy (solid red lines). But the experiments give a larger probability of very small spacings than the theory. In other words, the experimental distribution is more skewed than the theory predicts. The theory, in fact, predicts that the effect of hairpins is negligible for D = 40 nm, as opposed to the D = 51 nm channel. We do not know the reason for this discrepancy. It might be that the most important hairpins are too short for our theory to apply, but this is not likely as (cid:104)xH(cid:105) is of the same order as in the wide channel. We cannot exclude that there 10 are other reasons that bias the experimental data to smaller spacings. Using a different experimental method, Sheats et al. 30 suggested that the distributions are somewhat less skewed in narrow channels. However, we believe that the exposure time in the analysis by Sheats et al. 30 was too short to allow for a reliable estimate of the tails of the distribution and too few independent conformations were sampled. Comparing the theoretical predictions for slightly different channel widths (D = 40 nm and 42 nm) we infer that the theoretical results are quite sensitive to the value of the effective channel width -- which is not known precisely. The theory for 42 nm channels is much closer to the experimental data. The sensitive dependence is of interest because it reflects the fact that hairpin formation is an activated process. What remains to be done? First, the Gaussian model for the right-hand tail of the spacing distribution P(X, L) is highly idealised, and the parameter σ0 was fitted to the data. The exact dependence of σ0 upon L could be obtained from simulations of confined discrete wormlike chains,31 by measuring the fluctuations of the end-to-end distance of short chains conditional on no hairpins, or using propagator methods.28 It may be possible, but much more difficult, to simulate the full distribution to determine how our theory breaks down as (cid:104)xH(cid:105) becomes smaller. Second, the sensitive dependence of the predicted distribution upon channel size can test theories for the effective channel width,40 an important open question for genome mapping experiments. We further expect that the tails are sensitive to flexibility of the DNA backbone, providing a probe for sequence-specific effects on confined DNA conformations, an emerging area of interest.41 Finally, an entirely open question is to understand the conformational dynamics.42 -- 46 While the results described here show that the equilibrium conformation statistics of strongly confined DNA molecules are now well understood, much less is known about the dynamics. ACKNOWLEDGMENTS This work was supported by VR grants 2013-3992 and 2017-3865, by NIH grant R01- HG006851, and by NSF grant DMS-1515161. Computational resources were provided by C3SE and SNIC. 11 REFERENCES 1Z. G. Wang, "50th Anniversary Perspective: Polymer Conformation -- A Pedagogical Review," Macromolecules 50, 9073 -- 9114 (2017). 2T. Strick, J.-F. Allemand, V. Croquette, and D. Bensimon, "Twisting and stretching single DNA molecules," Progress in Biophysics and Molecular Biology 74, 115 -- 140 (2000). 3C. Bustamante, J. C. Macosko, and G. J. L. Wuite, "Grabbing the cat by the tail: manipulating molecules one by one," Nature Reviews Molecular Cell Biology 1, 130 EP (2000). 4Y. M. Wang, J. O. Tegenfeldt, W. Reisner, R. Riehn, X.-J. Guan, L. Guo, I. Golding, E. C. Cox, J. Sturm, and R. H. Austin, "Single-molecule studies of repressor-DNA interactions show long-range interactions." Proc. Natl. Acad. Sci. USA 102, 9796 -- 9801 (2005). 5R. Metzler, W. Reisner, R. Riehn, R. Austin, J. O. Tegenfeldt, and I. M. Sokolov, "Diffu- sion mechanisms of localised knots along a polymer," Europhys. Lett. 76, 696 -- 702 (2006). 6C. Zhang, P. G. Shao, J. A. van Kan, and J. R. C. van der Maarel, "Macromolecu- lar crowding induced elongation and compaction of single DNA molecules confined in a nanochannel," Proc. Natl. Acad. Sci. USA 106, 16651 -- 16656 (2009). 7E. Werner and B. Mehlig, "Confined polymers in the extended de Gennes regime," Phys. Rev. E 90, 062602 (2014). 8K. Jo, D. M. Dhingra, T. Odijk, J. J. de Pablo, M. D. Graham, R. Runnheim, D. For- rest, and D. C. Schwartz, "A single-molecule barcoding system using nanoslits for DNA analysis," Proc. Natl. Acad. Sci. USA 104, 2673 -- 2678 (2007). 9S. K. Das, M. D. Austin, M. C. Akana, P. Deshpande, H. Cao, and M. Xiao, "Single molecule linear analysis of DNA in nano-channel labeled with sequence specific fluorescent probes," Nucleic Acids Res. 38, e177 (2010). 10E. T. Lam, A. Hastie, C. Lin, D. Ehrlich, S. K. Das, M. D. Austin, P. Deshpande, H. Cao, N. Nagarajan, M. Xiao, and P. Y. Kwok, "Genome mapping on nanochannel arrays for structural variation analysis and sequence assembly," Nat. Biotechnol. 30, 771 -- 776 (2012). 11Y. Michaeli and Y. Ebenstein, "Channeling DNA for optical mapping," Nat. Biotechnol. 30, 762 -- 763 (2012). 12K. L. Kounovsky-Shafer, J. P. Hernandez-Ortiz, K. Potamousis, G. Tsivd, M. Place, P. Ravindarn, K. Jo, S. Zhou, T. Odijk, J. J. de Pablo, and D. C. Schwartz, "Elec- 12 trostatic confinement and manipulation of DNA molecules for genomic analysis," Proc. Natl. Acad. Sci. USA 114, 13400 -- 13405 (2017). 13T. Odijk, "On the Statistics and Dynamics of Confined or Entangled Stiff Polymers," Macromolecules 16, 1340 -- 1344 (1983). 14D. Huh, K. L. Mills, X. Zhu, M. A. Burns, M. D. Thouless, and S. Takayama, "Tunable elastomeric nanochannels for nanofluidic manipulation," Nat. Mater. 6, 424 -- 428 (2007). 15T. Odijk, "Scaling theory of DNA confined in nanochannels and nanoslits," Phys. Rev. E 77, 060901(R) (2008). 16Y. Kim, K. S. Kim, K. L. Kounovsky, R. Chang, G. Y. Jung, J. J. DePablo, K. Jo, and D. C. Schwartz, "Nanochannel confinement: DNA stretch approaching full contour length," Lab Chip 11, 1721 -- 1729 (2011). 17L. D. Menard and J. M. Ramsey, "Electrokinetically-driven transport of DNA through focused ion beam milled nanofluidic channels," Anal. Chem. 85, 1146 -- 1153 (2013). 18C. Zhang, A. Hernandez-Garcia, K. Jiang, Z. Gong, D. Guttula, S. Y. Ng, P. P. Malar, J. A. Van Kan, L. Dai, P. S. Doyle, R. de Vries, and J. R. C. Van Der Maarel, "Amplified stretch of bottlebrush-coated DNA in nanofluidic channels," Nucleic Acids Res. 41, e189 (2013). 19T. Odijk, "DNA confined in nanochannels: Hairpin tightening by entropic depletion," J. Chem. Phys. 125, 204904 (2006). 20P. Cifra, "Channel confinement of flexible and semiflexible macromolecules," J. Chem. Phys. 131, 224903 (2009). 21T. Su, S. Das, M. Xiao, and P. Purohit, "Transition between two regimes describing internal fluctuation of DNA in a nanochannel," PLoS ONE 6, e16890 (2011). 22L. Dai, S. Y. Ng, P. S. Doyle, and J. R. C. van der Maarel, "Conformation Model of Back-Folding and Looping of a Single DNA Molecule Confined Inside a Nanochannel," ACS Macro Lett. 1, 1046 -- 1050 (2012). 23P. Cifra and T. Bleha, "Detection of chain backfolding in simulation of DNA in nanofluidic channels," Soft Matter 8, 9022 -- 9028 (2012). 24Y. L. Chen, "Electro-entropic excluded volume effects on DNA looping and relaxation in nanochannels," Biomicrofluidics 7, 054119 (2013). 25Y. L. Chen, Y. H. Lin, J. F. Chang, and P. K. Lin, "Dynamics and confinement of semiflexible polymers in strong quasi-1D and -2D confinement," Macromolecules 47, 1199 -- 13 1205 (2014). 26J. Shin, A. G. Cherstvy, and R. Metzler, "Polymer looping is controlled by macromolecular crowding, spatial confinement, and chain stiffness," ACS Macro Lett. 4, 202 -- 206 (2015). 27A. Muralidhar, D. R. Tree, and K. D. Dorfman, "Backfolding of wormlike chains confined in nanochannels," Macromolecules 47, 8446 -- 8458 (2014). 28J. Z. Y. Chen, "Conformational Properties of a Back-Folding Wormlike Chain Confined in a Cylindrical Tube," Phys. Rev. Lett. 118, 247802 (2017). 29W. F. Reinhart, J. G. Reifenberger, D. Gupta, A. Muralidhar, J. Sheats, H. Cao, and K. D. Dorfman, "Distribution of distances between DNA barcode labels in nanochannels close to the persistence length," J. Chem. Phys. 142, 064902 (2015). 30J. Sheats, J. G. Reifenberger, H. Cao, and K. D. Dorfman, "Measurements of DNA bar- code label separations in nanochannels from time-series data," Biomicrofluidics 9, 064119 (2015). 31D. R. Tree, Y. Wang, and K. D. Dorfman, "Extension of DNA in a nanochannel as a rod-to-coil transition," Phys. Rev. Lett. 110, 208103 (2013). 32E. Werner, G. K. Cheong, D. Gupta, K. D. Dorfman, and B. Mehlig, "One-parameter scaling theory for DNA extension in a nanochannel," Phys. Rev. Lett. 119, 268102 (2017). 33T. W. Burkhardt, Y. Yang, and G. Gompper, "Fluctuations of a long, semiflexible polymer in a narrow channel," Phys. Rev. E 82, 041801 (2010). 34T. W. Burkhardt, "Free energy of a semiflexible polymer in a tube and statistics of a randomly-accelerated particle," J. Phys. A.: Math. Gen. 30, L167 -- L172 (1997). 35D. J. Bicout and T. W. Burkhardt, "Simulation of a semiflexible polymer in a narrow cylindrical pore," J. Phys. A.: Math. Gen. 34, 5745 -- 5750 (2001). 36V. Balakrishnan and S. Chaturvedi, "Persistent diffusion on a line," Physica A 148, 581 (1988). 37E. Werner, F. Persson, F. Westerlund, J. O. Tegenfeldt, and B. Mehlig, "Orientational correlations in confined DNA," Phys. Rev. E 86, 041802 (2012). 38L. Onsager, "The effects of shape on the interaction of colloidal particles," Ann. N. Y. Acad. Sci. 51, 627 -- 659 (1949). 39Y. Wang, D. R. Tree, and K. D. Dorfman, "Simulation of DNA extension in nanochan- nels," Macromolecules 44, 6594 -- 6604 (2011). 40W. Reisner, J. N. Pedersen, and R. H. Austin, "DNA confinement in nanochannels: 14 Physics and biological applications," Rep. Prog. Phys. 75, 106601 (2012). 41H.-M. Chuang, J. G. Reifenberger, H. Cao, and K. D. Dorfman, "Sequence-dependent persistence length of long DNA," Phys. Rev. Lett. 119, 227802 (2017). 42S. L. Levy, J. T. Mannion, J. Cheng, C. H. Reccius, and H. G. Craighead, "Entropic Unfolding of DNA Molecules in Nanofluidic Channels," Nano Lett. 8, 3839 -- 3844 (2008). 43G. O. Ib´anez-Garc´ıa, P. Goldstein, and A. Zarzosa-P´erez, "Hairpin polymer unfolding in square nanochannels," J. Polym. Sci. B Polym. Phys. 51, 1411 -- 1418 (2013). 44M. Alizadehheidari, E. Werner, C. Noble, M. Reiter-Schad, L. K. Nyberg, J. Fritzsche, B. Mehlig, J. O. Tegenfeldt, T. Ambjornsson, F. Persson, and F. Westerlund, "Nanocon- fined circular and linear DNA: Equilibrium conformations and unfolding kinetics," Macro- molecules 48, 871 -- 878 (2015). 45E. Werner, A. Jain, A. Muralidhar, K. Frykholm, T. St Clere Smithe, J. Fritzsche, F. West- erlund, K. Dorfman, and B. Mehlig, "Hairpins in the conformations of a confined polymer," Biomicrofluidics 12, 024105 (2018). 46J. Krog, M. Alizadehheidaria, E. Werner, S. K. Bikkarolla, J. O. Tegenfeldt, J. Fritzsche, B. Mehlig, M. A. Lomholt, F. Westerlund, and T. Ambjornsson, "Stochastic model and experiments of the circular-to-linear unfolding dynamics of nanoconfined DNA and asso- ciated Bayesian parameter estimation framework," (2018). 47L. Dai, J. van der Maarel, and P. S. Doyle, "Extended de Gennes regime of DNA confined in a nanochannel," Macromolecules 47, 2445 -- 2450 (2014). 48T. S. C. Smithe, V. Iarko, A. Muralidhar, E. Werner, K. D. Dorfman, and B. Mehlig, "Finite-size corrections for confined polymers in the extended de Gennes regime," Phys. Rev. E 92, 062601 (2015). 49P. Grassberger, "Pruned-enriched Rosenbluth method: Simulations of θ polymers of chain length up to 1 000 000," Phys. Rev. E 56, 3682 -- 3693 (1997). 50T. Prellberg and J. Krawczyk, "Flat histogram version of the pruned and enriched Rosen- bluth method," Phys. Rev. Lett. 92, 120602 (2004). Appendix A: Ideal distribution Assume that the polymer starts at x = 0 at t = 0 with ±v0 with equal probability. Denote the distribution of x(t) by p(x, t). Decompose p(x, t) = ρ+(x, t) + ρ−(x, t), where 15 ρ+(x, t) is the probability to find the process at x at time t with velocity v0 > 0, and ρ−(x, t) is the probability to find the process at x at time t with velocity v0 < 0. Then:  ρ+(x, t)  =  −r − v0∂x ∂ ∂t r −r + v0∂x r  ρ+(x, t) ρ−(x, t)  . ρ−(x, t) Let q(x, t) = ρ+(x, t) − ρ−(x, t): (16) (17) ∂tp = −v0∂xq , and ∂tq = −2rq − v0∂xp . To solve (17) we use the Laplace transform. Denote the Laplace transforms L of p and q by P = L (p) and Q = L (q). Since the process starts at x = 0 with ±v0 with equal probability, the initial conditions are q(x, 0) = 0 and p(x, 0) = δ(x). As a consequence the Laplace transforms obey: sP (x, s) − δ(x) = −v0∂xQ(x, s) and sQ(x, s) = −2rQ(x, s) − v0∂xP (x, s) . (18) The solution of Eq. (18) reads: Q(x, s) = −sv0 σ2 d dx P (x, s) and P (x, s) = C1eσx/v0 + C2e−σx/v0 − H(x) (cid:0)eσx/v0 − e−σx/v0(cid:1) , σ 2sv0 where H(x) is the Heaviside function and σ = p(−∞, t) = 0 gives C2 = 0, and p(∞, t) = 0 yields C1 = σ/(2sv0). Inserting these √ s2 + 2sr. The boundary condition (19) FIG. 4. Contour for inverse Laplace transform for 0 < x < v0t. There is a branch cut at [−2r, 0]. 16 1ResIms2rCCexpressions for C1 and C2 into Eq. (19) we find: (cid:2)H(x)e−σx/v0 + H(−x)eσx/v0(cid:3) and Q(x, s) = P (x, s) = σ 2sv0 (cid:2)H(x)e−σx/v0 − H(−x)eσx/v0(cid:3) . 1 v0 Now assume x > 0 and evaluate the inverse Laplace transform e−σx/v0+st p(x, t) = (cid:90) γ+i∞ γ−i∞ 1 2v0 ds 2πi σ s (20) (21) using contour integration. For x > v0t, we close the contour in the right half plane. This contour contains no singularities, so P (x, t) = 0. When 0 < x < v0t, we close the contour in the left half plane as shown in Fig. 4. The integral along C and C vanishes, the only contribution comes from integration around the branch cut: (cid:90) 2r 1 π ds 0 √ 2sr − s2 sv0 √ cos( 2sr − s2x/v0)e−st . (22) (cid:90) 2r √ 2sr − s2 sv0 When x = v0t then the integrals along C and C diverge. This gives a contribution propor- tional to δ(x− v0t) to P (x, t). Since Eq. (22) is normalised to 1− e−rt when integrating from x = 0 to x = v0t we have: √ 2sr − s2x/v0)e−st , cos( 1 π p(x, t) = e−rtδ(x − v0t) + H(x − v0t) ds (23) normalised on [0−,∞]. Numerical evaluation of the smooth part of Eq. (23) for v0 = 1 and different values of r and t shows that this result is equivalent to an expression in Ref.:36 p(x, t) = e−rtδ(x − v0t) + H(x − v0t) (cid:104) I0(rt(cid:112)1 − x2/(v0t)2) + I1(rt(cid:112)1 − x2/(v0t)2) (cid:112)1 − x2/(v0t)2 re−rt v0 (cid:105) . 0 (24) In the limit of T → ∞ for fixed x(cid:48) = x/(v0T ), stationary-phase evaluation of the integral in Eq. (23) yields (cid:114) P0(x(cid:48), T ) ∼ 1 2 √ 1 − x(cid:48)2 1 + (1 − x(cid:48)2)3/4 erT ( √ rT 2π 1−x(cid:48)2−1) . (25) This is Eq. (7) in the main text, noting that the label spacing is given by X = x(T ) since x(0) = 0. We remark that Eq. (6) is normalised to unity for −1 ≤ X(cid:48) ≤ 1. Appendix B: Description of Monte Carlo algorithm for simulation of telegraph model with self avoidance This Section describes the implementation of an algorithm to simulate the telegraph model with self avoidance. The telegraph process is implemented as described in the Sup- 17 plemental Material of Ref.,32 and the telegraph simulations use a modified version47,48 of the PERM algorithm.49,50 The algorithm grows an ensemble of N polymers, each represented by a discrete telegraph process. Initial conditions: x0 = 0 and v0 = −1 or +1 with equal probability. The polymers grow in the direction of v0. After each time step t, each polymer has a chance r of changing sign of v0, so that it continues to grow in the opposite direction. If the polymer reaches a site that it has already visited n times before, it has a chance e−n of surviving to the next time step. In the simplest form of the algorithm, the polymer is discarded if it does not survive. Attrition renders this algorithm inefficient. Therefore we used the modified version,47,48 where Nb batches each containing Np/b polymers are grown simultaneously (N = NbNp/b). Every time a polymer fails the survival check, it is replaced by a polymer randomly sampled from its batch. The replacement of polymers creates a bias in the statistics conformation statistics,47 which must be corrected for. This can be done by introducing weights as described in Ref..47 Every polymer in a given batch is assigned the same initial weight w0 = N−1 each time step the weights of all polymers in a given batch are updated as wt = wt−1 Ns Np/b where Ns is the number of polymers in the batch that survived the check. This rule decreases p/b. After , the weights of batches where many polymers have been replaced. This process continues for T time steps. For each polymer, the extension XT along the channel axis is measured. Mean µ and variance σ2 of the extension are calculated as (cid:80)N (cid:80)N (cid:80)N (cid:80)N k=1 w(k) µ = σ2 = k=1 w(k) T X (k) T k=1 w(k) T T − µ]2 T [X (k) k=1 w(k) T , (26) (27) where k labels all polymers, across all batches, and w(k) are the corresponding weights. 18
1106.0448	1	1106	2011-06-02T15:17:04	Range of ion specific effects in the hydration of ions	[ "physics.bio-ph" ]	Within the quasichemical approach, the hydration free energy of an ion is decomposed into a chemical term accounting for ion specific ion-water interactions within the coordination sphere and nonspecific contributions accounting for packing (excluded volume) and long range interactions. The change in the chemical term with a change in the radius of the coordination sphere is the compressive force exerted by the bulk solvent medium on the surface of the coordination sphere. For the Na+, K+, F-, and Cl- ions considered here this compressive force becomes equal for similarly charged ions for coordination radii of about 0.39 nm, not much larger than a water molecule. These results show that ion specific effects are short ranged and arise primarily due to differences in the local ion-water interactions.	physics.bio-ph	physics	Range of ion specific effects in the hydration of ions Department of Chemical and Biomolecular Engineering and The Institute of NanoBioTechnology, Safir Merchant and Dilip Asthagiri∗ Johns Hopkins University, Baltimore, MD 21218 (Dated: October 23, 2018) Within the quasichemical approach, the hydration free energy of an ion is decomposed into a chemical term accounting for ion specific ion-water interactions within the coordination sphere and nonspecific contributions accounting for packing (excluded volume) and long range interactions. The change in the chemical term with a change in the radius of the coordination sphere is the compressive force exerted by the bulk solvent medium on the surface of the coordination sphere. For the Na+, K+, F−, and Cl− ions considered here this compressive force becomes equal for similarly charged ions for coordination radii of about 3.9 A, not much larger than a water molecule. These results show that ion specific effects are short ranged and arise primarily due to differences in the local ion-water interactions. Keywords: Hofmeister effect, potential distribution theorem, Monte Carlo, simulations Numerous aqueous phase processes including, for ex- ample, the stability of biological macromolecules and col- loidal suspensions, are strongly influenced by the chem- ical type of the inorganic ions dissolved in the aqueous phase1. However, despite their ubiquity, a molecular level understanding of ion specific effects remains elusive1,2. Ideas such as distortion of hydrogen bond network of the bulk solvent3, balance between ion-water and water- water interactions4, and influence of ion-specific disper- sion forces5 have been proposed to rationalize these ef- fects. Understanding how the ion affects the solvent ma- trix remains of first interest in understanding ion spe- cific effects. Recent spectroscopic measurements on salt solutions6 -- 8 suggest that the ion only affects hydro- gen bonds of water molecules within its first hydration shell. An earlier theory and simulation study by us9 also showed that only a small subset of water molecules in the first hydration shell of the ion are sensitive to the type of the ion. These results taken together suggest that ion specific ion-solvent interactions are limited only to the ion's local neighborhood. In this Communication, we address the question of how far in the liquid the chem- ical type of the ion is felt and if the point at which ion- specificity is lost is also the point where continuum mod- els of hydration begin to take hold. To address these questions, we first define a coordina- tion sphere of radius r around the ion to separate the ion interaction with water molecules within the coordination shell from the longer-range interaction of the ion with the solvent outside the coordination sphere. With such a spatial partitioning, the hydration free energy of the ion, µex, can be written as10 -- 13 µex = kBT ln x0(r) − kBT ln p0(r) + µex outer(r) . (1) where kB is Boltzmann's constant and T is the temper- ature, kBT ln x0(r) is the chemical contribution to the hydration free energy, −kBT ln p0(r) is the packing con- outer(r) is the contribution to the hydra- tribution, and µex tion free energy due to ion interaction with the solvent outside the coordination shell. x0(r) is the probability of observing no water molecules within the coordination sphere of radius r around the ion; thus kBT ln x0(r) is the free energy gained by allowing water molecules to flood the empty coordination sphere, justifying the iden- tification of this term with local, chemical interactions. p0(r) is the probability of observing an empty coordina- tion sphere, but in the absence of the ion; −kBT ln p0(r) is the free energy required to create a cavity of radius r outer(r) is the free energy gained in the bulk solvent. µex in placing the ion in an empty cavity; it accounts for the interaction of the ion with the bulk medium outside the coordination sphere. For a sufficiently large r, we ex- pect µex outer(r) to be composed of a large number of small non-specific contributions14,15. Since Eq. 1 is a tautol- ogy, the point where µex outer(r) becomes insensitive to ion type is also the range to which kBT ln x0(r) captures all the ion-specific effects. To identify the range of ion specific effects, we studied ion-water systems under NVT conditions using Metropo- lis Monte Carlo simulations16,17. Water was modeled us- ing the SPC/E model18 and parameters for Na+, K+, F−, Cl− ions were obtained from an earlier study19. The chemical term was estimated by growing the coordination sphere in steps of 0.05 A. At any r, the probability of ob- serving zero (0) water molecules in a coordination sphere of radius r + 0.05 A was obtained. From this data, the entire x0(r) versus r profile was reconstructed. A simi- lar strategy was followed for p0(r) versus r, but with a step size of 0.1 A. At each r value, 3 × 105 sweeps of equilibration were followed by 3 × 105 sweeps of produc- tion. Each sweep comprised one attempted move of each particle in the system. The maximum allowed transla- tion and rotation were adjusted during the equilibration phase to target an acceptance ratio of 0.3. In the pro- duction phase these maximum displacement values were held fixed. Configurations are saved every 10 sweeps for analysis. From Fig. 1 we observe that for r > 3.9 A the change in the chemical term depends only on the charge of the ion and is independent of its chemical nature. Fig. 1 emphasizes that, for the ions studied here, ion-specific ) l o m / l a c k ( ) r ( 0 x n l T B k ∆ 0 −3 −6 −9 −12 −15 2 ) A / l o m / l a c k ( F 0 −10 −20 −30 −40 −50 −60 Na+(aq) K+(aq) F−(aq) Cl−(aq) Packing 5 3 4 r (A) 2 3 4 r (A) 2 Na+(aq) K+(aq) F−(aq) Cl−(aq) Born Model 5 FIG. 1: The change in the chemical term, ∆kBT ln x0, with change in the radius, r, of the coordination shell for various ions. The dashed line is the result for a solute that does not interact with the solvent; it thus accounts for the change in the packing contribution. Observe that the change in the chemical term approaches the change in the packing term as the coordination shell is increased. For r > 3.9 A the change in the chemical term is the same between identically charged ions. In particular, for r > 3.9 A, effects are short ranged. the increase in the chemical term is due to non-specific contributions. To elucidate the non-specific nature of contributions for r > 3.9 A, we first note that kBT ∂ ln x0(r)/p0(r) ∂r = − ∂µex outer(r) ∂r . (2) The left hand side of the above equation is the compres- sive force of the bulk on the coordination sphere of the ion relative to the compressive force on an ion-free coor- dination sphere20. For a sufficiently large r, we expect the interaction of the ion with the bulk medium outside the coordination sphere to be well-described by a Gaussian14,15. In this case, for an ion of charge q, we have µex outer = qhφi0 − β 2 · q2 · h(φ − hφi0)2i0 , (3) where h. . .i0 denotes averaging in the presence of an uncharged ion, hφi0 is the potential at the center of an uncharged ion, and h(φ − hφi0)2i0 is the fluctuation in the electrostatic potential19. By defining 1/rBorn = βh(φ − hφi0)2i0/2, it is readily seen that the fluctuation contribution is just the Born model of hydration. (For clarity, in Eq. 3 we do not explicitly indicate the cor- rections due to electrostatic self-interaction19, as these depend solely on q and not on the ion type.) In the radius range of interest we find that ∂hφi0/∂r is about a kBT /A. (A similar estimate is obtained based ∂r FIG. 2: F = kBT ∂ ln x0(r)/p0(r) , Eq. 2, the compressive force exerted by the solvent on the surface of the coordination shell containing the ion, kBT ∂[ln x0(r)]/∂r, relative to the com- pressive force on an empty cavity, kBT ∂[ln p0(r)]/∂r for var- ious coordination radii, r. The black dashed line is the es- timate of the relative compressive force based on the Born model of hydration (Eq. 4). on the data in Ref. 21 for a cavity.) Neglecting this con- tribution, we thus expect that asymptotically kBT ∂ ln x0/p0 ∂r ∼ − q2 2r2 (1 − 1 ǫ ) , (4) where ǫ is the dielectric constant for water. Figure 2 shows that for r > 3.9 A the relative compres- sive force between similarly charged ions becomes equal, but is different for anions and cations. The large r value is also different from the value obtained using the Born model, where the Born-radius is set equal to the coordi- nation radius. The discrepancy in the asymptotic limit between sim- ulation data and that based on the Born model is not particularly surprising, since the coordination radius is not the same as the Born radius19,21. For example, the position of the water molecule is denoted by the position of its oxygen atom and thus the coordination shell strictly excludes only the oxygen atoms of the water molecules but not their corresponding hydrogen atoms, which can enter the coordination sphere. Since this effect will be more pronounced for the anions than the cations, we ex- pect that the equivalent Born radius for anions to be smaller than that for the cations. What is interesting, however, is the observation that by empirically reduc- ing the cavity radius by 1.3 A for anions and 0.5 A for cations (Fig. 3), the relative compressive force for both anions and cations are similar for large coordination radii. Further, the large-r curvature of the radially translated curves compares reasonably well with that based on the the Born model of hydration. (We maintain the differ- ence of 0.8 A, as earlier studies on such shifts find a sim- ) A / l o m / l a c k ( F 0 −10 −20 −30 −40 −50 −60 Na+(aq) K+(aq) F−(aq) Cl−(aq) Born Model 4 5 1 2 3 r (A) FIG. 3: The data in Fig. 2 have been shifted along r by −1.3 A for the anions and by −0.5 A for the cations. With this empirical shift, the compressive force F = kBT ∂ ln x0(r)/p0(r) , Eq. 2, follows the trend predicted by the Born model (Eq. 4). ∂r ∗ Corresponding author: Fax: +1-410-516-5510; Email: [email protected] 1 W. Kunz, P. L. Nostro, and B. W. Ninham, Curr. Opin. Colloid Interface Sci. 9, 1 (2004). 2 Y. Zhang and P. S. Cremer, Curr. Opin. Chemical Biology 10, 658 (2006). 3 Y. Marcus, Chem. Rev. 109, 1346 (2009). 4 K. D. Collins, Methods 34, 300 (2004). 5 B. W. Ninham and V. Yaminsky, Langmuir 13, 2097 (1997). 6 A. M. Omta, M. F. Kropman, W. Sander, and H. J. Bakker, Science 301, 347 (2003). 7 M. F. Kropman and H. J. Bakker, Chem. Phys. Lett. 370, 741 (2003). 8 H. J. Bakker, M. F. Kropman, and A. W. Omta, Journal of Physics: Condensed Matter 17, 3215 (2005). 9 S. Merchant and D. Asthagiri, J. Chem. Phys. 130, 195102 (2009). 10 L. R. Pratt and S. B. Rempe, in Simulation and Theory of Electrostatic Interactions in Solution. Computational Chemistry, Biophysics, and Aqueous Solutions, edited by L. R. Pratt and G. Hummer (American Institute of Physics, Melville, NY, 1999), vol. 492 of AIP Conference Proceedings, pp. 172 -- 201. 11 M. E. Paulaitis and L. R. Pratt, Adv. Prot. Chem. 62, 283 ilar difference for anions and cations22,23.) 3 In summary, for the ions studied here, we find that a coordination radius of about 3.9 A, not much larger than the size of a water molecule, is sufficient to account for all ion specific ion-water interactions. Our results are in ac- cordance with experimental results6 -- 8 that suggest that the influence of the ion on hydrogen bonding between wa- ter molecules is minimal for water molecules beyond the first hydration shell. The short-range of ion specific ion- water interactions further suggests that any framework for modeling ion-specific effects needs to acknowledge the molecular characteristics of local ion-water interactions. tial distribution theorem and models of molecular solutions (Cambridge University Press, 2006). 13 L. R. Pratt and D. Asthagiri, in Free energy calculations: Theory and applications in chemistry and biology, edited by C. Chipot and A. Pohorille (Springer, 2007), vol. 86 of Springer series in Chemical Physics, chap. 9, pp. 323 -- 351. 14 D. Asthagiri, H. S. Ashbaugh, A. Piryatinski, M. E. Paulaitis, and L. R. Pratt, J. Am. Chem. Soc. 129, 10133 (2007). 15 J. K. Shah, D. Asthagiri, L. R. Pratt, and M. E. Paulaitis, J. Chem. Phys. 127, 144508 (2007). 16 N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, J. Chem. Phys. 21, 1087 (1953). 17 M. P. Allen and D. J. Tildesley, Computer simulation of liquids (Clarendon Press, Oxford, 1987). 18 H. J. C. Berendsen, J. R. Grigera, and T. P. Straatsma, J. Phys. Chem. 91, 6269 (1987). 19 G. Hummer, L. R. Pratt, and A. E. Garcia, J. Phys. Chem. 100, 1206 (1996). 20 D. M. Rogers and T. L. Beck, J. Chem. Phys. 129, 134505 (2008). 21 H. S. Ashbaugh, J. Phys. Chem. B 104, 7235 (2000). 22 W. M. Latimer, K. S. Pitzer, and C. M. Slansky, J. Chem. Phys. 7, 108 (1939). 23 H. S. Ashbaugh and D. Asthagiri, J. Chem. Phys. 129, (2002). 12 T. L. Beck, M. E. Paulaitis, and L. R. Pratt, The poten- 204501 (2008).
1706.06327	1	1706	2017-06-20T09:03:37	Femtosecond laser induced surface modification for prevention of bacterial adhesion on 45S5 bioactive glass	[ "physics.bio-ph" ]	Bacterial attachment and biofilm formation on implant surface has been a major concern in hospital and industrial environment. Prevention of bacterial infections of implant surface through surface treatment could be a potential solution and hence this has become a key area of research. In the present study, the antibacterial and biocompatible properties of femtosecond laser surface treated 45S5 bioactive glass (BG) have been investigated. Adhesion and sustainability of both gram positive S. aureus and gram negative P.aeruginosa and E. coli nosocomial bacteria on untreated and laser treated BG samples has been explored. An imprint method has been used to visualize the growth of bacteria on the sample surface. We observed complete bacterial rejection potentially reducing risk of biofilm formation on laser treated surface. This was correlated with surface roughness, wettability and change in surface chemical composition of the samples before and after laser treatment. Biocompatibility of the laser treated BG was demonstrated by studying the anchoring and growth of human cervix cell line INT407. Our results demonstrate that, laser surface modification of BG enables enhanced bacterial rejection without affecting its biocompatibility towards growth of human cells on it. These results open a significantly potential approach towards use of laser in successfully imparting desirable characteristics to BG based bio-implants and devices.	physics.bio-ph	physics	Femtosecond laser induced surface modification for prevention of bacterial adhesion on 45S5 bioactive glass Shazia Shaikh1, Deepti Singh2,5 , Mahesh Subramanian2,5 , Sunita Kedia1,, Anil Kumar Singh3, Kulwant Singh4, Nidhi Gupta6, Sucharita Sinha1 1Laser & Plasma Surface Processing Section, Bhabha Atomic Research Centre, Mumbai 400085 India 2 Bio-Organic Division, Bhabha Atomic Research Centre, Mumbai 400085 India. 3 Laser and Plasma Technology Division, Bhabha Atomic Research Centre, Mumbai 400085 India. 4 Materials Science Division, Bhabha Atomic Research Centre, Mumbai 400085 India. 5 Homi Bhabha National Institute, Training School complex, Anushaktinagar, Mumbai 400094 India. 6Technical Physics Division, Bhabha Atomic Research Centre, Mumbai 400085 India. Corresponding authors, Email: [email protected] 1 Abstract Bacterial attachment and biofilm formation on implant surface has been a major concern in hospital and industrial environment. Prevention of bacterial infections of implant surface through surface treatment could be a potential solution and hence this has become a key area of research. In the present study, the antibacterial and biocompatible properties of femtosecond laser surface treated 45S5 bioactive glass (BG) have been investigated. Adhesion and sustainability of both gram positive S. aureus and gram negative P.aeruginosa and E. coli nosocomial bacteria on untreated and laser treated BG samples has been explored. An imprint method has been used to visualize the growth of bacteria on the sample surface. We observed complete bacterial rejection potentially reducing risk of biofilm formation on laser treated surface. This was correlated with surface roughness, wettability and change in surface chemical composition of the samples before and after laser treatment. Biocompatibility of the laser treated BG was demonstrated by studying the anchoring and growth of human cervix cell line INT407. Our results demonstrate that, laser surface modification of BG enables enhanced bacterial rejection without affecting its biocompatibility towards growth of human cells on it. These results open a significantly potential approach towards use of laser in successfully imparting desirable characteristics to BG based bio- implants and devices. Key words: Femtosecond laser texturing; 45S5 bio-glass; Antibacterial; Biocompatible; Surface treatment 2 1. Introduction: Increasing number of premature implant failures has triggered development of techniques to reliably minimize such occurrences. Two major causes for implant failures are aseptic loosening and infections [1, 2]. Among these two, higher percentages of unsuccessful biomaterial implantation are attributed to infections [2]. Infections at implant site can be described in two stages. Stage one is the initial attachment i.e. interaction between the bacterial cell surface and implant material surface. Stage two involves specific and non-specific interactions between proteins on the bacterial surface structures and binding molecules on biomaterial surface. Microorganisms predominantly grow as communities on animate and on inanimate surfaces and once they colonize, rapid multiplication of species occurs with formation of biofilm. Growth of biofilm enhances inherent resistance of the bacterial cells and leads to inflammation and consequent implant failure through poor biocompatibility and material degradation. This has serious health implications and often requires revisions of costly implant replacement surgeries. Attachment of bacteria on surfaces is a complicated process influenced by many factors, including the inherent bacterial property, the material surface characteristics and the environmental conditions under which growth takes place. It has been shown that the initial adhesion process is governed by physical interactions between the cell and the substrate, while biological processes involves longer time scale [3]. Since the initial interaction between bacteria and the surface of the BG plays a crucial role in adhesion of the bacteria a highly characterized surface is desirable while studying this interaction process [3, 4]. Different surface properties of biomaterials such as, morphology, surface chemistry, porosity, wettability and surface roughness determines initial bacterial adhesion and growth. To prevent formation of biofilm and infections on implants it is therefore very important to address and limit bacterial adhesion in its very early stage. Among varieties of available biomaterials, 45S5 Hench bioactive glass (BG) (45% SiO2, 24.5% Na2O, 24.5% CaO and 6% P2O5) was the first biomaterial capable of bonding to bone, invented by Prof. Larry Hench at University of Florida in 1969 [5]. Since invention, BG has shown promising applications in dental and orthopedic fields. BG bonds to both soft tissue, as well as, hard tissue [5]. 3 The bone bonding mechanism has been attributed to the formation of bone like apatite i.e. hydroxyapatite on the surface of BG in simulated body fluid. To prevent bacterial infection of BG implants it is essential to discourage initial adhesion of bacterial cells on implant surfaces. For this, several approaches including impregnation of implants with antibiotics, coating with antibacterial metals such as copper or silver [6,7], chemical treatments [8] , plasma-assisted modification [9], coating with self-assembled monolayer [10], and manipulating surface topography [11], have been undertaken. Hu et. al. studied antibacterial property of Hench BG against three types of bacteria [12]. Bellantone et. al. doped BG with Ag2O to improve its antibacterial property [13]. Gorriti et. al. incorporated B2O3 into BG to enhance antimicrobial activity [14]. Dineva et. al. chemically treated BG and evaluated bacterial adhesion [8]. Raphel et. al. discussed possibility of inhibiting microbial cell growth on BG by multifunctional orthopedic coating [6]. Begum et. al. correlated antibacterial activity of BG with change in environmental pH [15]. Recently, Prabhakar et. al. used powdered dentin with BG and increased antibacterial efficiency of the sample [16]. Despite various techniques tried for bacterial rejection [9], complete inhibition of bacterial adherence to BG surfaces remains an unachieved goal making this an active area of research. Improvements over existing techniques are desired that can completely prohibit bacterial growth under a wide range of conditions. In this present work, antibacterial and biocompatibility properties of femtosecond (fs) laser treated BG have been investigated. Surface topography of BG samples synthesized by melt-quench technique was modified using Titanium: Sapphire femtosecond pulsed laser by direct laser writing technique [17]. In addition to having several advantages such as, being a single step and chemical free process, ultra short pulsed laser based surface modification allows surface treatment localized both, in time and space. As reported elsewhere, femtosecond laser induced surface texturing is free from heat affected zones and modification is localized within the focal volume [18]. However, when BG surface was treated using a nanosecond pulsed laser collateral damage induced due to accumulation of heat has been reported [19]. Femtosecond laser treated BG samples have shown enhanced bio-integration in terms of superior and faster growth of crystalline hydroxyapatite layer in comparison to the sample treated using nanosecond laser beam. Here, we report our results on a number of BG samples which were surface textured using femtosecond laser. Laser treatment 4 modified surface topography, increased roughess and hence the effective surface area of the BG sample. Attachment of three most common bacteria: Staphylococcus aureus (S. aureus), Pseudomonas aeruginosa (P. aeruginosa) and Escherichia coli (E. coli) has been investigate on untreated and laser treated BG samples. Interaction between the bacterial cells and BG surface primarily depends on micron and sub-micron levels of surface roughness. Hence, surface roughness parameters of the laser treated BG samples have been measured using 3D optical profilometer. In our study, samples with highest achieved average surface roughness (~ 6.52 μm) completely inhibited attachment of all three bacteria. X-ray Photoelectron Spectroscopy (XPS) results revealed the presence of calcium hydroxide on femtosecond laser modified BG surface, whereas no such signatures were found on untreated BG surface. Our X-ray Diffraction studies also confirmed formation of calcium hydroxide phase post laser treatment of BG samples. Wettability of the sample surface increased after laser treatment. Biocompability of BG post laser treatment has also been evaluated employing human cervix cell line INT407. Adhesion of INT407 on laser treated surface was equivalent to that of untreated surface of BG sample. This indicates that while bacterial attachement could be inhibited, biointegration efficiency of the BG remained intact after laser surface modification. Hence, femtosecond laser treatment can serve as an efficient and facile technique to modify surface of BG, successfully preventing bacterial adhesion. Femtosecond laser treatment can thus serve as a potential technique in generating antibacterial surface for implant devices, without compromising the bio-compatibility and bio-integration of such BG implants and devices. 2. Experiment: 2.1 Materials and Methods - All the materials used in the study were of analytical grade or higher. Yeast extract, Tryptone and Agar were from Becton-Dickinson and Company, Churchton, MD, USA. Sodium chloride was obtained from Chemco Fine Chemicals, Mumbai, India. The animal cell culture media, trypsin and antibiotics were from Invitrogen, Carlsbad, USA. Hoechst 33342 was from Sigma Aldrich, St. Louis, USA. 5 2.1.1. Bio-active glass preparation - Using melt-quench process 45S5 BG having nominal composition SiO2 (45 wt %), Na2O (24.5 wt %), CaO (24.5 wt %), and P2O5 (6 wt %) was synthesized [12]. For 100 gm glass batch, SiO2, CaCO3, NH4H2PO4 and Na2CO3 were mixed and calcined at 900oC for 12 h. The calcined charge was melted in ambient at 1500°C in a Pt-Rh crucible. The fabricated glass was cut into 2 mm thick circular disks having a diameter of 1.5 cm. Flat surfaces of the sample were mechanically polished and used for laser treatment. 2.1.2. Femtosecond laser induced surface modification - NDF L Sample Ti:S Laser 800 nm, 45 fs, 3 kHz Figure 1: Illustration of the experimental arrangement used for laser treatment of BG. NDF is neutral density filter, L is 5 cm focal length lens, and sample was fixed on a computer controlled XY-stage Fig. 1 illustrates the experimental arrangement used for laser surface texturing of BG. For surface modification, beam from a femtosecond Titanium: Sapphire laser delivering pulses of pulse width 45 fs and at repetition rate 3 kHz, at a wavelength of 800 nm was focused on to the polished surface of the BG using a 5cm focal length lens. Typical diameter of the focal spot was ~ 90 µm and value of optimized average laser power employed for surface treatment ranged between 180mW to 200mW. This corresponds to an average laser fluence of 0.94 J/cm2 to 1.0 J/cm2 incident on the BG sample. Surface modification or laser writing speed was controlled using a computer controlled XY- translational stage on which the BG sample was mounted. The sample was scanned in a plane perpendicular to the direction of laser beam. Sample scanning speed has been varied between 5 6 µm/sec to 55 µm/sec thereby generating samples with different surface roughness. Typically, an area of 5 mm x 5 mm was surface modified through femtosecond laser treatment of each BG sample. Results of untreated BG referred as S1 and samples, laser treated at scan speeds 35μm/sec (S2) and 15μm/sec (S3) are discussed here. 2.1.3. Surface analysis - Surface topography of BG samples before and after laser treatment and after bacterial tests was observed under scanning electron microscope (SEM) (Carl Zeiss EVO 40 SEM). Surface roughness parameters of untreated and laser treated BG sample were measured using a 3D-optical profilometer (Taylor-Hobson). Standard surface roughness parameters such as, arithmetic mean (Ra), root-mean square deviation (Rq) and peak to valley distance (Rt) also called maximum roughness were measured [4, 20]. The wettability of BG surface before and after laser treatment was studied by water contact angle measurement. The measurements were repeated at least 3 times for each sample. XPS analysis of the BG samples before and after laser surface treatment was performed using a VG make model CLAM-2 hemispherical analyser with Al Kα source (1486.6eV) having a step size of 0.3 eV. X-ray diffraction (XRD) analysis of samples was carried out using PANalytical MRD system. The measurements were done using CuKα radiation of wavelength 1.54 Ao in out of plane geometry over the 2θ range of 200-350 at angle scanning speed of 2o/min. 2.2. Antibacterial test - 2.2.1. Preparation of medium - Bacterial growth medium (Nutrient broth/agar) was prepared by dissolving Yeast extract (0.5%), Peptone (0.5%) and sodium chloride (0.5%) in ion free water. The pH was adjusted to 7.4. This was sterilized by autoclaving at 121 °C at 15 psi for 15 min. 7 2.2.2. Bacterial cultures - The bacterial cultures used in the study were obtained from the repository for microorganisms, Institute of Microbial Technology, Chandigarh, India. The organisms used in this investigation are Staphylococcus aureus ATCC 6538P, Pseudomonas aeruginosa ATCC 19154, and Escherichia coli K Strain BW25113. 2.2.3. Evaluation of bacterial adherence - Bacteria were grown overnight in Nutrient broth at 37 °C. The bacterial solution for the experiment was prepared by pelleting 1 mL overnight growth by centrifugation. The cell pellet was washed once with sterile water and re-suspended in 20 mL sterile water in a 100 mm sterile petri dish inside a laminar flow cabinet. This solution contained ~ 5×107 cells/mL. The BG discs that had been surface modified (area = 5 x 5 mm) using laser irradiation were wiped free of any contaminants with 70% ethanol, air dried inside the laminar hood and immersed in the bacterial solution and incubated for 2 h. After incubation the discs were carefully picked up with a sterile forcep, mildly rinsed in sterile water to remove the loosely adhered bacteria. The excess liquid was drained off. Imprint of the surface of the discs were taken on nutrient agar plates by precisely placing and pressing the discs gently once. The discs were removed with utmost care immediately. The agar plates that had the imprint of the circular discs (modified replica plate technique) were incubated for 24 h at 37 °C. The growth of the adhered bacteria was observed visually and documented by a Kodak Gel logic imaging system revealing the adherence pattern on the BG samples being tested. 2.3. Growth of mammalian cells on BG surface - INT 407 cells (cervix cells of human origin) were grown in DMEM medium with antibiotic (Penicillin 100 units/mL and Streptomycin 100 µg/mL) and antimycotic (amphotericin B 0.25 µg/mL). The medium was supplemented with 10% FBS (Fetal Bovine Serum). The cells were grown at 37 °C under 5% CO2. The circular 45S5 disc was placed in a 60 mm tissue culture dish. This BG disc had half its surface untreated, while remaining half had its surface modified by laser. The disc was placed in such a way that the surface to be tested for adherence of the cells was facing upward 8 and not towards the bottom of the dish. Growth medium containing 1x 106 cells was added till the disc was immersed in the medium (10 mL). The cells were allowed to attach and grow overnight in a 37 ºC incubator with 5% CO2 atmosphere. Next day DNA binding dye Hoechst 33342 was added directly in to the medium (5 µM final concentration) and incubated for 10 min. The disc was lifted with a sterile forcep placed face down on a glass slide and observed under a fluorescence microscope (Carl Zeiss LSM 780) equipped with UV filters for the presence of cells. 3. Results : (b) (a) Figure 2: SEM image of (a) untreated and (b) laser treated BG Figs. 2a and 2b are typical SEM images of pristine and laser treated BG, respectively. Microstructures generated on the surface of the BG after laser treatments are clearly visible in Fig. 2b. Surface roughness and hence the effective surface area of the BG samples distinctly increased after laser treatment. 9 Figure 3: Optical profilometer image of BG (a) untreated (S1), laser treated at fluence 1.0 J/cm2 and sample scanning speed of (b) 35 μm/sec (S2), and (c) 15 μm/sec (S3) Figs. 3a, 3b and 3c are surface scanned optical images of sample S1, S2 and S3, respectively. Color variation in the images indicates modified topography of the BG after laser treatment. Roughness parameters of the untreated and laser treated BG are summarized in table-I. Sample Ra (μm) Rq (μm) Rt (μm) S1 S2 S3 0.42 1.88 6.25 0.32 2.54 7.74 7.50 23.5 43.7 Table-I: Roughness parameters of different BG samples Roughness of the sample enhanced significantly when BG was translated at lowest scanned speed, referred to as sample S3. Slower movement allowed longer interaction time to the laser pulses with the BG. Multiple laser pulses deposited their energy on BG surface and when temperature exceeded melting point of the sample, melting occurred on BG surface. Associated hydrodynamic instability in the melt volume resulted in modification of surface [21]. Subsequent interaction of the laser pulses with the roughened surface of the BG caused non-uniform energy deposition. Higher laser energy was deposited in the valley regions in comparison to peaks formed on the surface. This non-uniform energy distribution resulted in a temperature gradient and hence variation in surface tension on BG surface, temperature of the hills being lower than that of the valleys. As, energy deposition on the sample increased, melted BG flowed from low surface tension region (valley region) to high surface tension region (hill) and further elongated the microstructures, as evident in case of S3. Repetition rate 10 of used femtosecond laser was 3 kHz and focal spot size of the laser beam on BG surface was 90 μm. Samples S2 and S3 were scanned at speed of 35 μm/sec and15 μm/sec, respectively. Hence, on an average S2 received lesser (~ 7,500 pulses/spot) laser pulses in comparison to S3 (~ 18,000 pulses/spot). Therefore, generation of larger temperature gradient on S3 surface can be expected. This caused formation of high hills with deep valleys on surface of S3. In Fig. 3c, Rt for S3 was measured as 43.7 μm which was significantly larger than Rt for S1 (7.50 μm) and S2 (23.5 μm). (a) (b) Figure 4: Photographs of water contact angle measured for BG (a) S1 (73°), and (b) S3 (43°) Figs. 4a and 4b are the photographs of water contact angle measurement of S1 and S3, respectively. The water contact angle of pristine BG was 73o indicating hydrophilic nature of the surface. The contact angle decreased to 34° for S3. Therefore, wettability of the BG samples was found to increase on laser treatment. (a) (b) Figure 5: XPS survey spectra of (a) untreated and (b)fs-laser treated BG (S3) 11 Surface compositions of BG before and after fs-laser treatment were analyzed by X-ray Photoelectron Spectroscopic technique. The XPS survey spectra of pristine BG and laser modified BG are shown in fig.5a and 5b, respectively. Peaks for the possible constituent elements have been identified and marked on these scans in fig.5. A major difference observed between survey spectra taken before and after laser treatment were the prominent peaks of Ca and O for the laser treated BG representing increased presence of calcium and oxygen post laser treatment. Also, less intense peak of carbon was observed on the surface of laser modified BG in comparison to untreated BG. Lower carbon presence on laser treated BG surface could be due to surface sputtering and cleaning on account of laser irradiation. Carbon contamination of pristine BG surface could also be responsible for absence of detectable peak of Si in case of untreated BG samples. Spectral peaks around 1072.5eV, 531.5eV, 346.5eV, 284.4eV and 102.1eV were assigned to the Na1s, O1s, Ca2p, C1s, and Si2p photoelectrons, respectively. The peak at 469.5eV is assigned to Na KLL Auger transition. In order to have a better understanding of the effect of laser treatment on BG O1s, C1s and Ca2p spectra were deconvoluted into their components. The O1s peak provides very useful information about the oxides in glasses. Silica based glasses generally consist of two kinds of bonding with oxygen atoms i.e. bridging oxygen bond and non-bridging oxygen bond due to presence of alkali network modifiers (Na and Ca) in glass composition. The high resolution O1s spectra for both untreated and fs-laser treated BG have been deconvoluted into four components, as shown in fig.6. (a). For fs-laser treated and untreated BG non bridging oxygen peak appears at 528.6eV and 529.5eV, respectively [22]. Bridging oxygen peak was observed as expected at 531eV for untreated BG [22]. Na-Auger photoelectron peak was observed for both laser treated and untreated BG at binding energy values 535.7eV and 535.9eV [23]. Peak at 533.3eV observed in the case of fs-laser treated sample can be attributed to the presence of OH group [24]. Peak at 531.6eV and 531.8eV, is an indication of the presence of C=O, hydrocarbonaceous contamination on untreated and laser treated BG respectively [25]. In Fig.6b are shown the deconvoluted spectra of Ca2p for untreated and fs-laser treated BG samples. For untreated BG the Ca2p spectra was deconvoluted into two components having peak 12 Figure 6: High resolution XPS spectra of (a) O1s, (b) Ca2p, (c) C1s, of BG untreated surface (left) and fs-laser treated surface (right) positions 346.8eV and 351eV, corresponding to 2p states of CaO and CaCO3. Spectra for fs-laser treated BG when deconvoluted gave three components with peak positions at 346.5eV, 347.4eV, and 350.5eV. Peak appearing around 346.5eV can be attributed to 2p1/2 of Ca(OH)2 or 2p3/2 of CaO, peak at 347.4eV can be associated with 2p3/2 of CaO or CaCO3, while peak around 350.5eV can be 13 assigned to 2p1/2 of CaO or Ca(OH)2 [26]. Hence, these three peaks can be attributed to either 2p1/2 or 2p3/2 states of Ca corresponding to presence of Ca(OH)2, CaO or CaCO3. The high resolution XPS spectra for C1s for both pristine and laser modified BG are shown in fig.6. (c). The carbon spectral components observed at binding energy values 284.8eV and 287.7eV, correspond to C-C and C=O, respectively [24]. Si 2p spectra for laser modified BG showed two peaks at 102.1eV and 104.8eV corresponding to non bridging oxygen and bridging oxygen atoms when bonded to Si atom [22]. Surface contamination of untreated BG samples could be the likely reason for Si 2p spectra not being detected in XPS of pristine BG samples i.e. prior to laser treatment. Dominant peak for Na1s for both untreated and fs-laser treated BG at 1072.5eV and 1072.8eV, respectively indicated the presence of Na2O. While, binding energy value for elemental Na1s is 1071.5eV, 1eV shift of binding energy indicates oxide formation [27]. Figure 7: X-ray diffraction pattern of untreated and fs-laser treated BG Fig.7 is shown XRD pattern of untreated and fs-laser treated BG, respectively. The broad peak centered at 32° for untreated BG confirms the amorphous nature of BG (black) whereas sharp peaks 14 observed in XRD pattern on laser treatment (red), suggest surface crystallization of BG post fs-laser irradiation. Most of these sharp peaks obtained after laser treatment matched with JCPDS # 22-1455, 85-1108 and 87-0674 confirming Na2Ca2Si3O9, CaCO3 and Ca(OH)2 crystalline phase formation, respectively. As demonstrated by various studies, BG is prone to crystallization on heat treatment and forms sodium-calcium silicate phase [28]. Heat treatment at high temperature generally shows the presence of Na2Ca2Si3O9 phase, a prominent crystalline phase of 45S5 bioglass. In our process of laser treatment of BG using a femtosecond laser in addition to sodium-calcium silicate phase we have also observed presence of CaCO3 and Ca(OH)2 crystalline phases. Chemical reaction of BG surface with atmospheric environment during laser surface treatment is expected to have resulted in formation of CaCO3 and Ca(OH)2. Our XRD data thus confirms presence of both CaCO3 and Ca(OH)2 on laser treated surface of BG as was also indicated by our XPS investigations. (a) (b) Figure 8: Photograph of Nutrient agar plate showing replica imprint of BG with laser treated zone (a) S2 and (b) S3 after allowing S. aureus to incubate for 24 hrs. Precise imaging of bacterial adhesion and growth on an implant surface is essential for determining the success and effectiveness of the implant material. We have employed a novel way to test the growth of bacteria adhered to the surface of the implant material based on taking an imprint on the solidified growth medium. For this, BG sample with laser treated zone was immersed in S. aureus bacterial solution for 2 hr and imprint of the sample was taken on nutrient agar plate which 15 was incubated for 24 hrs. Figs. 8a and 8b are the photographs of nutrient agar plate showing replica imprint of S2 and S3, respectively. The laser treated region is marked with arrow in both the figures. Adhesion and agglomeration of S. aureus was observed on untreated and laser treated regions in S2, as seen in Fig. 8a. Unlike this, adhesion of S. aureus significantly reduced on S3, as observed by clear zone in Fig. 8b. Only 10% of total laser treated area allowed colonization of S. aureus on S3 surface as evident in the figure. Figure 9: Photograph of Nutrient agar plate showing replica imprint of S3 adherence of (a) S. aureus , (b) P. aeruginosa, and (c) E. coli. Figs. 9a, 9b and 9c are the photographs of nutrient agar plate showing replica imprint of sample S3 incubated with S. aureus, P. aeruginosa, and E. coli, respectively. The laser treated portion of BG successfully obstructs adhesion of P. aeruginosa, and E. coli as observed by a clear zone (marked by c arrows), in Fig. 9b and 9c, respectively. However, all three bacteria adhere and cohere to the untreated region of the BG. 16 (a) (b) (c) (d) (e) (f) Figure 10: SEM images of laser treated area of S3 showing no adherence of (a) S. aureus , (b) P. aeruginosa, and (c) E. coli, and SEM images of untreated BG showing adherence of (d) S. aureus , (e) P. aeruginosa, and (f) E. coli The antibacterial property of the laser treated BG was also confirmed by independent investigation employing sample imaging under scanning electron microscope (SEM). The bacteria were allowed to adhere to the BG discs for 2 h after which the discs were rinsed mildly in sterile water and air dried for 10 min. The surfaces of the discs were gold coated by sputtering and bacterial adhesion was 17 observed under SEM. Physical adherence of the different bacteria to the laser treated vis-a-vis untreated surface was investigated. As seen in the fig. 10, in all the 3 cases, no bacterial cells were seen on the laser treated surface (figures 10a, 10b, and 10c) while bacterial adhesion was clearly observed on the untreated surface (figures 10d, 10e, and 10f). Figure 11. INT407 Cells growing on (a) untreated, (b) laser treated BG surfaces (S3) and (c) animal cell culture Petri dish Successful bio-integration of implants depends on their bio-compatibility. Hence, bio-integration is a property that needs to remain intact post laser surface treatment. Therefore, adhesion and growth of a typical human cell type INT407 was tested on the laser treated BG surface and compared with an untreated surface. It was evident that there existed no difference in the adherence and growth of INT407 cells on the S1 (Fig. 11a) and S3 (Fig. 11b) surfaces. Adherence of human cell is a desirable property as under real clinical condition when the host cells and tissues accept the laser treated implant surface as a scaffold to adhere and grow, bio-integration of implant is successful. We also confirmed that the growth observed on the 45S5 surface was comparable to the growth of the same cells on animal tissue culture dish that has poly-lysine coating which provides an anchoring surface for the cells to grow as shown in Fig. 11c. 4. Discussions : Topography mediated bacterial adhesion has been studied in the light of a comprehensive description of the surface architecture of 45S5 Hench BG substrate. The effect of surface roughness 18 on adherence of S. aureus, P. aeruginosa and E. coli bacteria has been evaluated. S. aureus is a gram positive bacterium spherical in shape with typical size ~ 0.5-1 .0 μm. This b acterium accounts for ~70% of implant infections [29]. P. aeruginosa is a gram negative rod shaped bacterium with size around 2.0 μm. These are notorious for causing nosocomial infections (~ 8 %) that are pertinent to infections associated with implant devices. E. coli is a rod shaped, gram negative bacterium causing gastroenteritis and it is one of the most established model organism for bacterial research. In our present study, S. aureus adhered and colonolized on the BG surfaces with Ra 0.42 μm (S1) and 1.88 μm (S2). However, for the sample (S3) with Ra 6.25 μm, S. aureus nucleated only on 10% of total laser treated area, as seen in Fig. 8b. Additionally, S3 rejected attachment and bonding of P. aeruginosa and E. coli bacteria completely, as observed in Fig. 9b and 9c, respectively. Maximum roughness (Rt) of S3 was measured as 43.7 μm which was much larger than Rt for S1 (7.50 μm) and S2 (23.5 μm). Red color depicted in Fig. 3b and 3c indicates location of peaks on the surface of the laser treated samples. Summit density; that is a measure of the number of peaks per unit area is expected to be much higher on the surface of S3 in comparison to surface of S2 as confirmed from the distribution of red color in Fig. 3c and 3b, respectively. Increased surface roughness for S3 resulted in discontinuous contacts, and thus reduced the bonding opportunity of bacteria [30]. Additionally, comparatively dense features of high peaks associated with deep valley restricted bacteria to reach the BG surface and limited their agglomeration on S3 surface. In comparison, maximum roughness in case of S2 was relatively lower and summit density also significantly less. Hence, for this sample bacteria could adhere on the BG surface and could agglomerate easily. Kathryn et. al. reported retention of S. aureus bacteria on biomaterial with Ra ~ 2 μm [ 31]. Similarly, in present case Ra for S2 was comparable and it supported attachment of S. aureus. Taylor et al reported increase in P. aeruginosa adhesion on the biomaterial surface with average roughness ~ 0.04 to 1.24 μm, although in the same report it was observed that bacterial adhesion decreased notably when Ra was ~1.86 to 7.89 μm [30]. This is in agreement with our observation in the current investigation where S3 sample with Ra ~ 6.25 μm inhibited bacterial adhesion completely. Bagherifard et. al. reported adhesion and number of colony formation of S. aureus decreases as surface roughness increases [20]. Also, that report clearly mentions that, density of S. aureus on the 19 biomaterial surface is inversely proportional to Rq. The roughness parameters for the sample which stunted adhesion of S. aureus in that work were measured as Ra ~ 8.14 μm, Rq ~ 10.10 μm and Rt ~ 63.43 μm. In our present work, measured roughness parameters for S3 were quite similar and the sample S3 rejected adhesion of all three bacteria completely. In addition to surface topography antibacterial behavior of BG has also been correlated to a high pH value environment. Probable mechanisms responsible for death of bacterial cells when exposed to a strong base have also been proposed. Calcium hydroxide is a strong alkaline substance having a pH of 12.5. In an aqueous environment it dissociates into calcium and hydroxyl ions. Hydroxyl ions show extreme reactivity and have been reported to have lethal effects on bacterial cells [32]. Reported studies on antibacterial behavior of BGs have correlated it to high aqueous pH values, as an alkaline environment has been shown to be detrimental for most of the bacteria [15]. Our XPS results reveal that after fs-laser surface treatment calcium hydroxide formed on the surface of BG. whereas no signature of calcium hydroxide was detected on untreated BG surface. Our XRD results also confirmed presence of calcium hydroxide post laser surface treatment of BG surface. Calcium hydroxide has been extensively used by endodontics because of its antibacterial properties [33]. Therefore, antibacterial effect of fs-laser modified BG sample surface as observed in our study could have also been enhanced owing to the presence of calcium hydroxide on BG surface as a consequence of laser treatment. Our results suggest that surface roughness, as well as, presence of calcium hydroxide on BG surface post fs-laser treatment could collectively be responsible for the observed antibacterial behavior. Thus, both surface topography and chemical composition play a crucial role resulting in antibacterial effects of BG surfaces. Higher wettability with adequate antibacterial property is desirable for an ideal biomaterial, as better wetness helps in tissue attachment and integration [34, 35]. As per our observation, untreated BG surface was hydrophilic in nature and wettability of the sample improved post laser treatment. While, some reports have correlated high bacterial rejection with surfaces having high hydrophobicity [36], recently Prabu et. al. reported bacterial rejection also by hydrophilic surface [35]. 20 5. Conclusion : Antibacterial effect of rough surface and its potential ability to reduce the risk of biofilm formation is of utmost importance. Our results demonstrated that modification of surface texture along with chemical composition holds great potential in controlling microbial attachment. Enhanced antibacterial property with intact biocompatibility of femtosecond laser treated BG, was observed. Surface roughness and wettability of the BG increased after femtosecond laser treatment. Both XPS and XRD results have confirmed presence of calcium hydroxide on the fs-laser modified BG surface. Both surface roughness and presence of calcium hydroxide on laser treated BG surface enhanced its antibacterial behavior. Complete rejection of three types of nosocomial bacteria was seen on most rough BG surface with Ra ~ 6.54 μm and Rt ~ 43.7 μm. Our observations also confirmed that attachment and growth of INT407 human cells on laser treated BG samples remained unaltered when compared to pristine untreated samples. Adherence and growth of human cells on such BG determines its biocompability enabling rapid bio- integration of these implants. Thus, surface modification of BG by femtosecond laser is an attractive means of mitigating bacterial attachment while retaining biocompatibility as revealed by the growth of INT 407 human cells. References 1. Bozic, K. J.; Kurtz, S. M.; Lau, E.; Ong, K.; Chiu, V.; Vail, T. P.; Rubash, H. E.; Berry, D. J. The Epidemiology of Revision Total Knee Arthroplasty in the United States. Clinical Orthopaedics and Related Research® 2010, 468 (1), 45-51. 21 2. Bozic, K. J.; Kurtz, S. M.; Lau, E.; Ong, K.; Vail, T. P.; Berry, D. J. The epidemiology of revision total hip arthroplasty in the United States. J Bone Joint Surg Am 2009, 91 (1), 128-33. 3. Hermansson, M. The DLVO theory in microbial adhesion. Colloids and Surfaces B: Biointerfaces 1999, 14 (1–4), 105-119. 4. Crawford, R. J.; Webb, H. K.; Truong, V. K.; Hasan, J.; Ivanova, E. P. Surface topographical factors influencing bacterial attachment. Adv Colloid Interface Sci 2012, 179-182, 142-9. 5. Jones, J. R. Review of bioactive glass: from Hench to hybrids. Acta Biomater 2013, 9 (1), 4457-86. 6. Raphel, J.; Holodniy, M.; Goodman, S. B.; Heilshorn, S. C. Multifunctional coatings to simultaneously promote osseointegration and prevent infection of orthopaedic implants. Biomaterials 2016, 84, 301-14. 7. Desrousseaux, C.; Sautou, V.; Descamps, S.; Traore, O. Modification of the surfaces of medical devices to prevent microbial adhesion and biofilm formation. J Hosp Infect 2013, 85 (2), 87- 93. 8. Mitik-Dineva, N.; Wang, J.; Truong, V. K.; Stoddart, P.; Malherbe, F.; Crawford, R. J.; Ivanova, E. P. Escherichia coli, Pseudomonas aeruginosa, and Staphylococcus aureus attachment patterns on glass surfaces with nanoscale roughness. Curr Microbiol 2009, 58 (3), 268-73. 9. Bazaka, K.; Jacob, M. V.; Crawford, R. J.; Ivanova, E. P. Plasma-assisted surface modification of organic biopolymers to prevent bacterial attachment. Acta Biomater 2011, 7 (5), 2015-28. 10. Liu, C.; Zhao, Q.; Liu, Y.; Wang, S.; Abel, E. W. Reduction of bacterial adhesion on modified DLC coatings. Colloids Surf B Biointerfaces 2008, 61 (2), 182-7. 11. Qi, H.; Cai, J.; Zhang, L.; Kuga, S. Properties of films composed of cellulose nanowhiskers and a cellulose matrix regenerated from alkali/urea solution. Biomacromolecules 2009, 10 (6), 1597- 602. 12. Hu, S.; Chang, J.; Liu, M.; Ning, C. Study on antibacterial effect of 45S5 Bioglass®. Journal of Materials Science: Materials in Medicine 2009, 20 (1), 281-286. 22 13. Bellantne, M.; Williams, H. D.; Hench, L. Broad-spectrum bacterial activity of Ag2O-doped bioactive glass, Antimicrobial Agents and Chemotherapy 2002, 46, 1940-1945. 14. Gorriti, M. F.; Lopez, J. M. P.; Boccaccini, A. R.; Audisio, C.; Gorustovich, A. A. In vitro study of the antibacterial activity of bioactive glass-ceramc scaffolds, Advance Engineering Materials 2009, 11, B67-B70. 15. Begum, S.; Johnson, W. E.; Worthington, T.; Martin, R. A. The influence of pH and fluid dynamics on the antibacterial efficacy of 45S5 Bioglass. Biomed Mater 2016, 11 (1), 015006. 16. Prabhakar, A. R.; Kumar, S. C. H. Antibacterial effect of bioactive glass in combination with powdered enamel and dentin, Indian Journal of Dental Research 2010, 21, 30-34. 17. Kietzig, A. M.; Hatzikiriakos, S. G.; Englezos, P. Patterned superhydrophobic metallic surfaces. Langmuir 2009, 25 (8), 4821-7. 18. Shaikh, S.;Kedia, S.;Singh, A.;Sharma, K.;Sinha, S. Surface Treatment of 45S5 Bio-glass using Femtosecond Laser to Achieve Superior Growth of Hydroxyapatite. Journal of laser applications 2017. 19. Sharma, K.; Kedia, S.; Singh, A.K.; Chauhan, A.K.; Basu, S.; Sinha, S. Morphology and structural studies of laser treated 45S5 bioactive glass. Journal of Non-Crystalline Solids 2016, 440, 43–48. 20. Sara, B.; Daniel, J. H.; Alba C. de L.; Vera N. M.; Athina, E. M.; Mario, G.; Thomas J. W. The influence of nanostructured features on bacterial adhesion and bone cell functions on severely shot peened 316L stainless steel. Biomaterial 73, 2015, 185-197. 21. Singh, A. K.; Shinde, D.; More, M. A.; Sinha, S. Enhanced field emission from nanosecond laser based surface micro-structured stainless steel. Applied Surface Science 2015, 357, 1313-1318. 22. Dziadek, M; Zagrajczuk, B; Jelen,P; Olejniczak,Z; Cholewa-Kowalska,K. Structural variations of bioactive glasses obtained by different synthesis routes. Ceramic international 2016, 42, 14700-14709 23. Yadegari, H.; Yongliang Li.; Banis M.N.; Xifei Li. ; Wang, B Sun, Q.; Ruying Li .; Sham, T.K.; Cui, X.; Sun, X .On rechargeability and reaction kinetics of sodium–air batteries. Energy Environ. Sci. 2014, 7, 3747–3757. 23 24. Cerruti, M.; Claudia, L.B.; Bonino, F.; Damin, A.; Perardi, A.; Morterra, C. Surface modification of bioglass immersed in TRIS-Buffered solution. A multitechnical spectroscopic study. J.Phys.Chem. B, 2005, 109(30), 14496-14505. 25. Durgalakshmi, D.; Balakumar. S. Analysis of solvent induced porous PMMA–Bioglass monoliths by the phase separation method – mechanical and in vitro biocompatible studies. phys.chem.chem.phys, 2015, 17, 1247-1256. 26. Lillard, R.S.; Enos, D.G.; Scully, J.R. Calcium Hydroxide as a Promoter of Hydrogen Absorption in 99.5% Fe and a Fully Pearlitic 0.8% C Steel During Electrochemical Reduction of Water. Corrosion, 2000, 56(11), 1119-1132. 27. Barrie, A.; Street, F.J. An Auger and X-ray photoelectron spectroscopic study of sodium metal and sodium oxide.J.Electron spectrosc.Relat.Phenom. 1977, 1, 7. 28. Bellucci, D.; Cannillo, V.; Sola, A. An overview of the effects of thermal processing on bioactive glasses. science of sintering, 2010, 42, 307-320. 29. Campoccia, D.; Montanaro, L.; Arciola, C. R. The significance of infection related to orthopedic devices and issues of antibiotic resistance. Biomaterials 2006, 27 (11), 2331-9. 30. Taylor, R. L.; Verran, J.; Lees, G. C.; Ward, A. J. The influence of substratum topography on bacterial adhesion to polymethyl methacrylate. J Mater Sci Mater Med 1998, 9 (1), 17-22. 31. Siqueira Jr, J.F.; Lopes, H.P. Mechanisms of antimicrobial activity of calcium hydroxide: a critical review. International Endodontic Journal, 1999, 32, 361-369. 32. Foreman, P.C.; Barnes, I.E. A review of calcium hydroxide.international endodontic journal , 1990, 23, 283-97. 33. Kathryn, A.W.; John, C.; Joanna, V. Retention of microbial cells in substratum surface features of micrometer and sub-micrometer dimensions. Colloids and Surfaces B: Biointerfaces 2005, 41, 129–138. 34. Emerson, R. J. t.; Bergstrom, T. S.; Liu, Y.; Soto, E. R.; Brown, C. A.; McGimpsey, W. G.; Camesano, T. A. Microscale correlation between surface chemistry, texture, and the adhesive strength of Staphylococcus epidermidis. Langmuir 2006, 22 (26), 11311-21. 24 35. Prabu, V.; Karthick, P.; Rajendran, A.; Natarajan, D.; Kiran, M. S.; Pattanayak, D. K. Bioactive Ti alloy with hydrophilicity, antibacterial activity and cytocompatibility. RSC Advances 2015, 5 (63), 50767-50777. 36. Fadeeva, E.; Truong, V.K.; Stiesch, M.; Chichkov, B.N.; Russell, J.; Crawford,; James, W.; Elena, P.I. Bacterial Retention on Superhydrophobic Titanium Surfaces Fabricated by Femtosecond Laser Ablation. Langmuir 2011, 27 (6), 3012–3019. 25
1805.07744	1	1805	2018-05-20T09:31:26	Optimal morphometric factors responsible for enhanced gas exchange in fish gills	[ "physics.bio-ph" ]	Fish gills are one of the most primitive gas/solute exchange organs, having the highest ventilation volume, present in nature. Such performance is attributed to a functional unit of gill - secondary lamella - that can extract oxygen from an ambience even at a very low partial pressure. For centuries, gills have stood as one of the simplest but an elegant gas/solute exchange organs. Although the role of various morphometric factors of fish gills on gas/solute exchange capabilities have been reported, there has been limited understanding on what makes fish gills as an excellent gas/solute exchange system. Therefore, in the current study, we have theoretically studied the variation of few structural and parametric ratios, which were known to have role in gas/solute exchange, with respect to the weight of fishes. Thereafter, modelling and simulation of convection-diffusion transport through a two dimensional model of secondary lamella were carried out to study different factors affecting the performance of gills. The results obtained from both the studies (theoretical and computational) were in good agreement with each other. Thus, our study suggested that fish gills have optimized parametric ratios, at multiple length scales, throughout an evolution to arrive at an organ with enhanced mass transport capabilities. Further, our study also highlighted the role of length of primary and secondary lamella, surface area of secondary lamellae and inter-lamellar distance on gas/solute exchange capabilities of fish gills. Thus, these defined morphological parameters and parametric ratios could be exploited in future to design and develop efficient gas/solute exchange microdevices.	physics.bio-ph	physics	Optimal morphometric factors responsible for enhanced gas exchange in fish gills Prasoon Kumara,b,c, Prasanna S Gandhib, Mainak Majumderc a IITB-Monash Research Academy, Powai, Mumbai, Maharashtra-400076, India b Suman Mashruwala Advanced Microengineering Laboratory, Department of Mechanical Engineering, Indian Institute of Technology Bombay, Powai, Mumbai, Maharashtra 400076, India c Nanoscale Science and Engineering Laboratory (NSEL), Department of Mechanical and Aerospace Engineering, Monash University, Clayton, Melbourne, Australia Abstract Fish gills are one of the most primitive gas/solute exchange organs, having the highest ventilation volume, present in nature. Such performance is attributed to a functional unit of gill - secondary lamella - that can extract oxygen from an ambience even at a very low partial pressure. For centuries, gills have stood as one of the simplest but an elegant gas/solute exchange organs. Although the role of various morphometric factors of fish gills on gas/solute exchange capabilities have been reported, there has been limited understanding on what makes fish gills as an excellent gas/solute exchange system. Therefore, in the current study, we have theoretically studied the variation of few structural and parametric ratios, which were known to have role in gas/solute exchange, with respect to the weight of fishes. Thereafter, modelling and simulation of convection-diffusion transport through a two dimensional model of secondary lamella were carried out to study different factors affecting the performance of gills. The results obtained from both the studies (theoretical and computational) were in good agreement with each other. Thus, our study suggested that fish gills have optimized parametric ratios, at multiple length scales, throughout an evolution to arrive at an organ with enhanced mass transport capabilities. Further, our study also highlighted the role of length of primary and secondary lamella, surface area of secondary lamellae and inter-lamellar distance on gas/solute exchange capabilities of fish gills. Thus, these defined morphological parameters and parametric ratios could be exploited in future to design and develop efficient gas/solute exchange microdevices. Keywords: Secondary lamella, convection-diffusion, modelling, simulation, parametric ratios 1. Introduction The separation of gases, solutes and solvents under convective flow condition is a requirement in number of applications like artificial kidney dialyser, extra corporeal membrane separators, water filtration, desalination and others [1, 2]. The biomimcry has been used to solve above engineering problems in the recent years by drawing an inspiration from natural systems like avian lung, human lungs, human kidney, human vasculature etc. [3-5]. However, the challenges of developing compact, efficient gas/solute exchange devices still persist. To achieve the above objective, the design of a device need to be improved by taking inspiration from a simple, yet elegant, gas/solute separating organ system. One of such primitive known gas/solute separating organ systems in nature is fish gills. Gills can be taken as a model design for the development of better gas/solute exchanger due to their ability to extract oxygen from water even at a very low partial pressure, having the highest ventilation volume among all species, having design leading to continuous efflux of oxygen from ambient to the blood vessels and an overall compactness of the organ system [6]. Therefore, fish gill needs to be thoroughly studied at different length scales to determine the reasons for it to be an excellent gas/solute exchange organ in nature for centuries. The knowledge of fish gills' design generated through above study can lead to development of microfluidic devices for better gas/solute separation. The gills from different varieties of fishes have been studied for a long time from their physical, morphological, structural and biochemical point of view[7-11]. However, the knowledge of fish gill architecture and morphometrics in the light of their gas/solute exchange capabilities still need considerable attention. The basics structures of gills and their role in gas/solute exchange have been well elucidated in detail through a series of publications by Hughes et al. Through anatomical studies, they have calculated the dimensions of gills of different fishes at multiple length scales. Further, they have also established the role of above determined morphometric factors like thickness of blood-water barrier, size and arrangement of secondary lamellae, on the gas exchange capacities in different weights of fishes[10, 12]. Their studies suggested that surface area of gills, which is responsible for gas/solute exchange, scales with the weight of fishes. Another prominent finding which they reported was the counter-current flow of blood and water stream in secondary lamella and inter-lamellar space, respectively. Later, theoretical model was developed by J. Piiper and P. Scheid to study the gas diffusion in different organ systems. Their model suggested that the effect of counter current flow on gas diffusion is better than the cross-current flow and uniform pool system, assuming the same ventilation perfusion conductance ratio across an epithelial layer. They further substantiated their results through experimental findings in a fish, a bird and a mammal where above three modes of gas exchange operated. In addition, they also commented that the enhanced gas exchange in counter-current flow occurs in a limited range of medium-to-blood conductance ratio (around 1)[13, 14]. Their model beautifully captured the role of counter-current flow on gas diffusion effectiveness, but the role of dimensions of secondary lamella of fish gills on effectiveness of gas/solute exchange remained elusive, except for an enhanced surface area. Later, studies carried out by K Park et al demonstrated an existence of optimal lamellar distances in fish gills irrespective of their mass for an enhanced gas exchange. Their theory and experiments effectively captured the poor dependence of inter-lamellar distance on weight of fishes. Thus, they established the evolutionary conservation of these inter lamellar distances in fishes[15]. Although, above studies identified the role of different morphometric factors on gas/solute exchange, seldom has the study been done to the best of our knowledge to study the structural organization of gills as an organ and parameterization of these morphometric factors in fishes. Thus, there is a need to study and explore dimensional features of a secondary lamella and its functional relationship with gas/solute transport in fishes. Therefore, in our current work, we have theoretically studied the variation in different morphometric features and parametric ratios of fish gills in different sizes of fishes form data reported in literature. Thereafter, computationally investigation of the convective-diffusion process in a simplified model of secondary lamella was carried out to illustrate the role of parametric ratios on gas/solute exchange in a fish gills. Our data analysis suggested that the values of these parametric ratios remain nearly conserved in fishes with different body mass and find good agreement with the values predicted by our computational study. Thus, the study enabled a better understanding of functioning of fish gills. The work also laid a foundation for the design parameters worth attention, before developing a bio-inspired fish's gill devices for application in heat and mass exchanger. 2. Structure and function of fish gills One of the most efficient gas/solute exchange systems developed in an early evolutionary stage of living beings were that of gills in fishes. The outstanding functioning of fish gills as an excellent gas exchanger is without any trade-offs. The structure of fish gills support the transfer of gases and metabolites efficiently from an aquatic surrounding to a fish's body and vice versa. Moreover, the fish gills are capable of extracting oxygen from the water medium under extreme environment; from brackish sea water to murky lakes where oxygen partial pressure are abysmally low[16, 17]. The uniqueness of a fish gill lies in its hierarchical multi-scale structure and functioning. The fish gills' hierarchical structure is shown in the Figure 1. The fish gills are the first vertebrate gas exchange organ, which comprises parallel array of thin epithelial matrix encasing complex bifurcating vasculature known as secondary lamella. These secondary lamellae are stacked in parallel over a primary lamella which in itself is arranged in a rack like structure as shown in the Figure 1. Such arrangements of functional unit of fish gills dramatically increase the surface area to volume ratio. The thin epithelium layer in secondary lamellae provides a shuttle barrier between an aquatic environment and a fish's blood running in vascular network encased in secondary lamella is shown in the Figure 1. The primary lamella is attached to a stem where two primary blood vessels, namely; efferent artery and vein run in parallel. They are further subdivided into capillaries and sub-capillaries while entering into secondary lamellae. When fishes swim in water, the fishes gulp water through their mouth and direct them to flow through inter-lamellar space. The concentration gradient of gases/solute across an aquatic environment and fish's blood drives an exchange of gas and metabolites through high surface area offered by thin epithelial barrier of secondary lamellae. Meanwhile, they also pump blood through blood capillaries encased in secondary lamella with a velocity enough to allow maximum diffusion of gas/solute from water flowing in inter lamellar spaces. Thus, the presence of fractal design of micro/nano capillaries encased in thin secondary lamella supports efficient flow of blood. They also execute majority of functions like respiration, excretion single headedly, which are carried out by pulmonary and renal system in mammals [7]. The exchange of gases/solutes is primarily driven by convection-diffusion phenomena that occur at a site of each secondary lamella of fish gills [9]. Figure 1 Schematic of multi scale architecture of fish gill demonstrating exchange of oxygen from ambient water to fish body[21] The water flows through inter lamellar spaces between secondary lamellae of fish gills. The cross-current flow of water and blood in between secondary lamellae and blood capillaries respectively, and corresponding concentration gradient of gases and metabolites causing highly efficient diffusion pathway, produces high mass transfer ratio. This counter-current flow system always maintain blood with low oxygen concentration as compared to oxygen concentration of water flowing between the secondary lamellae of fish gills; and consequently continual diffusion of oxygen into the blood. The schematic in the Figure 1 explains role of cross current flow system in maximum possible gas and metabolites exchange between external aquatic environment and internal blood vessels [11]. This clearly illustrates the role of fluid flow direction in fish gills to maintain sustained concentration gradient and continuous diffusion flux of gases/solutes. 3. Materials and Methods 3.1 Theoretical study of fish gills The morphometric data of fish gills were obtained from the work of Hughes et.al[11] shown in the Table 1. Table 1 Morphometric features of fish gills Wt of fish (gm) Inter- lamellar dist (mm) Avg. PL length (mm) Number of SL/mm Length of SL (mm) Hgt. of SL (mm) Trachurus trachurus Clupea harengus Gadus merlangus Onos mustela Crenilabrus melops Salmo trutta sp. Tinca tinca Chaenocephalus sp. Lophius piscatorius Plmronectes platessa Zeus faber Trigla gurnardus Cottus bubalis Callionymus lyra 12 40 125 135 85 51 20 80 65 175 140 750 790 1550 86 300 18 40 52 64 46 24 0.017 0.017 0.017 0.02 0.018 0.03 0.023 0.03 0.04 0.023 0.025 0.053 0.057 0.07 0.04 0.04 0.02 0.04 0.04 0.05 0.03 0.047 2.37 4.18 6.16 6.14 4.46 3.18 2.39 4.66 2.96 5.44 8 11.96 9.84 13.31 5.08 4.97 1.84 2.86 3 2.86 2.56 1.77 38 39 38 39 33 21 26 20 21 21 25 12 12 11 20 15 22 16 16 15 16 17 0.18 0.07 0.25 0.7 0.75 0.6 0.6 0.45 0.55 0.85 0.7 0.86 1.6 1.5 1.1 0.1 0.2 0.15 0.17 0.2 0.15 0.27 0.3 0.2 0.1 0.5 0.55 0.4 0.75 0.2 1.1 0.35 0.8 0.8 0.8 0.7 0.6 0.11 0.06 0.3 0.25 0.3 0.25 0.15 Thereafter, the dimensions of different features of fish gills were analyzed for their variation with respect to the weight of different fishes and reported as graphs after considering below dimensionless parameters. a) Thickness_mem/width_water - the ratio width of epithelium and water channels (H2/H1) b) Width_bld/width_water - the ratio width of blood and water channels (H3/H1) c) Blood_parameter - the ratio of the diffusivity and speed in the blood layer (D3xL)/ (V3xH1^2) d) Water_parameter - the ratio of the diffusivity and speed in the water layer (D1xL)/ (V1xH1^2). For the purpose of the calculation of these dimensional less parameters for different fishes, following assumptions were considered. 1) The diameter of a single micro capillary is the maximum width of blood channel spanning across the width of the secondary lamella. 2) The flow calculations were done assuming the rectangular cross-section of inter lamellar space and secondary lamella. 3) The tortuosity and branching of blood channel and its diameter variation along the length are neglected for simplifying the model. 4) The enzymatic reaction based capture of oxygen by hemoglobin in the blood was neglected and only diffusion was considered. 3.2. Computational study of convective- diffusion phenomenon in fish gills To model and simulate the convection diffusion phenomenon in a secondary lamellae, we have studied the phenomenon on a two dimensional model of a secondary lamella as shown in the Figure 2. Figure 2 A) A schematic of secondary lamellae stacked in parallel on primary lamella. The blue arrow indicates the direction of water and red arrow represents direction of blood B) A 2D model of secondary lamella used in computational simulation We have made certain assumptions while simulating the diffusion phenomenon in secondary lamella. These assumptions are 1) A two dimensional model of gas exchange in a single secondary lamella is considered 2) Blood channel represents the width of blood vessel inside the tissue matrix of secondary lamella 3) Water channel width is the width of the inter lamellar space 4) The gray colored domain between blood channel and water channel represents the thin epithelial matrix, a barrier separating blood and water. 5) Steady state convective-diffusion phenomenon is solved in above model 6) The blood and water flow velocity is assumed to be average velocity rather than a velocity with a parabolic profile and flowing in counter-current direction 7) Diffusion of oxygen is considered from water channel to blood channel by crossing thin epithelial matrix. Under above assumptions convective- diffusion phenomenon was modeled in COMSOL Multiphysics 4.3b. The governing equations in each domain are 𝑈𝑤 𝜕𝐶 𝜕𝑥 = 𝐷𝑤( 𝜕2𝐶 𝜕𝑥2 + 𝜕2𝐶 𝜕𝑦2)……. (Water domain) 𝑈𝑏 𝜕𝐶 𝜕𝑥 = 𝐷𝑏( 𝜕2𝐶 𝜕𝑥2 + 𝜕2𝐶 𝜕𝑦2)……. (Blood domain) 𝐷𝑒 𝜕2𝐶 𝜕𝑦2 = 0……. (Epithelial barrier) 𝑁 = −𝐷 𝜕𝐶 𝜕𝑥 + 𝑈𝑥 𝐶 Where, Uw is average velocity in water channel, Ub is average velocity in blood domain, Dw is the diffusion coefficient of oxygen in water, Db is the diffusion coefficient of oxygen in Blood, De is the diffusion coefficient of oxygen in epithelial matrix, C is the concentration of oxygen and N is the flux Initial conditions : C (x,y) = 0 Boundary condition: inlet at water channel  C = Co Outlet at water channel  n.N = 0 Inlet at blood channel  C = 0 Outlet at Blood channel  n.N = 0 Table 2 Parameters used for modelling of convective diffusion phenomenon in secondary lamella Parameters Value Width of water channel 35µm Width of blood channel 8.8µm Thickness of epithelial layer Length of channels 2.8µm 760µm Parameters Diffusion coefficient of oxygen in Blood Diffusion coefficient of oxygen in water Diffusion coefficient of oxygen in epithelial layer Density and viscosity of water Avg. velocity of water 0.014m/s Density and viscosity of blood Avg. velocity of blood 0.0005m/s Conc. Of Oxygen in water stream Value 2.18e-9m2/sec 3.4e-9m2/sec 1.1e-9m2/sec 1000kg/m3 0.001kg/ms 1050kg/m3 0.0035/kg/ms 0.25moles/m3 Thereafter, considering the initial and boundary condition with the values of parameters in the Table 2 [11, 18, 19], we have studied the concentration of oxygen at the exit of blood channel by varying the different parametric ratios under steady state condition. 4. Results and discussion In order to explore the role of primary and secondary lamellae, we analyzed the length of primary and secondary lamella with respect to fish's body masses. We observed that length of secondary lamellae and primary lamellae followed a logarithmic relation with fish body weight Figure 3. However, the length dependence of primary lamellae on a fish body mass was more pronounced as compared to secondary lamella. This might be primarily because secondary lamella being a site of solute/gas exchange, its length is perhaps guided by diffusion length of solute/gas during a convective flow. Due to evolutionary conservation of the inter-lamellar distance across different weight of fishes [15], the time scale of diffusion of gases/solutes across the inter-lamellar space should be less than the residence time of water flowing in the inter- lamellar channel. However, the difference between the two time scales should not be too high and residence time of water in a convective flow should be only marginally higher than diffusion time scale. Moreover, considering the pressure drop along the length is confined by pumping capacity of fishes, the variation in the flow velocity in an lamellar space is minimal and remain laminar [11, 15, 20]. Therefore, length should be just sufficient enough to enable ratio of residence time to diffusion time scale to be equal or greater than 1. Thus, we observed that the length of secondary lamella showed weak dependence on weight of fishes. However, the primary lamella being a structural element that acts as a rack for parallel stacking of secondary lamellae demonstrated a strong dependence on the weight of the fishes. This may be to maximize the number of accommodating secondary lamella which is the respiratory functional unit of fish gills. An increase in the mass of fishes is related to their energy requirements which can be fulfilled by proportionate increase in the surface area of fish gills. The increase in number of secondary lamella eventually increases the surface area available for gas/solute exchange. Therefore, we observed that surface area of gills and number of secondary lamella increases with an increase in fish body mass (Figure 3B). The surface area of fish gills scale linearly with weight of fishes. Thus, the length of the primary lamella demonstrates strong dependence on weight of fishes (Figure 3A). Therefore, the arrangement of primary and secondary lamellae may play a vital role in a head of fish body with different body masses. Figure 3 Graph showing the variation of A) length of primary and secondary lamella with weight of fishes on logarithmic scale b) surface area of secondary lamellae and cross-section area of inter-lamellar space with weight of fishes on logarithmic scale. Further, the gill's primary and secondary lamella are arranged in a compact manner to enable large surface area per unit volume occupied by gills. The fractal dimensions calculated for secondary lamella shows that it is greater than 2 for most of the fishes (Figure 4). It is reported in literature that higher fractal dimensions indicates greater compactness of the structure. These fractal dimensions allows the gills to be compact and efficient increase surface area for effective gas exchange. In the absence of data for primary lamella of fish gills, we could not calculate their fractal dimensions. However, we believe that to ensure their compactness in the fish body, fractal dimensions for primary lamella of fish gills must be above 2. This arrangement of primary and secondary lamella ensures compactness of gills in fishes. Figure 4 shows the variation of fractal dimensions of secondary lamella of fish gills with weight of different fishes In order to develop a device for gas/solute exchange, mimicking the multi-scale architecture of fishes will help in developing micro/nanofluidic devices for mass transfer applications. However, to develop an understanding that can facilitate in translating the concept of mass transfer in fishes for developing scalable micro/nanofluidic devices, independent of the size of fishes will be highly desirable. Further, considering the multiple parameters of fish gills affecting the mass transfer, it is challenging to borrow dimensions from a fish to develop mass transfer device as per our requirement when the above parameter are closely coupled with other parameters. Therefore, having a few dimensionless parameters and knowing their degree of significance in mass transfer will assist in easy and efficient design of user defined size of mass transfer devices. Hence, few dimensional less parameters (parametric ratios) have been defined in this work. The rationale behind selecting these dimensionless parameters is to facilitate easy understanding of the effects of different parameters in their ability to affect the gas exchange process through convection-diffusion phenomenon. These dimensionless parameters summarizes majority of the structural and functional parameters in a four independent dimensionless numbers which governs the gas exchange through convection diffusion phenomena. Thus, the study of these parameters will help us in characterizing gas/solute exchange capabilities of any device where gas/solute exchange happens in two flowing fluid system separated by a membrane. Although, the spatial-arrangement of primary and secondary lamella defines the architecture of gills and forms a basis for design and fabrication of compact biomimetic fish gill structure, they hardly gave any idea about the correlated factors affecting the solute/gas exchange at the site of secondary lamella. Therefore, the parametric ratios for gills of 24 different fishes were evaluated from their morphometric properties and compared. All the parametric ratios plotted against the weight of fishes on a logarithmic scale. It was observed that these parametric ratios remain nearly conserved in all 24 fishes taken for evaluation although their body mass varied from 12g to 1550g as shown in Figure 5. The thickness of membrane to water channel width ratio should be as small as possible to minimize the diffusion barrier length for solute/gases. The calculated data indeed suggested that this ratio for 24 different fishes lied in the range of 0.135 to 0.848 with an average of 0.372 ± 0.189, much less than 1. Further, the width of blood channel should neither be too less leading to high blood pumping pressure due to an increased blood flow resistance nor too high to leading to the condition of washout without proper oxygenation. Therefore, the ratio of width of blood channel to water channel lied between 0.071 and 0.449 with mean of 0.206 ± 0.095 in all 24 fishes. The width of the blood channel appear to be less in comparison with the width of the membrane Figure 5 Graph showing the variation of different dimension less quantities (Parametric ratios) with weight of fishes on logarithmic scale. Further, the blood and water parametric ratio defines the mixing parameter in blood and water channels, respectively. The mixing parameter is a dimensionless number which defines dominance of diffusion phenomenon over convection phenomenon in a laminar flow. The blood parametric ratio existed between 0.535 and 5.778 with mean value 2.11±1.41. The mean value is above 1 indicating the dominance of convective flow over diffusive flow. Such behavior leads to the continuous wash out of oxygen which enters in the blood stream, thereby maintaining a continuous concentration gradient between the blood and the water stream. The kinetics of capture of oxygen by hemoglobin of red blood cells defines the rate of oxygen capture and its removal by convective flow creates oxygen deficient center in the flowing blood stream. The minimum of blood parameter ratio obtained from the above data is 0.535. The large variation in the blood parameter is due to the assumptions on the dimensions of blood channels. When water parametric ratio was considered, it was found that this parameter in different fishes lie between 0.016 and 1.88 with mean of 0.323 ± 0.459. This clearly illustrates the dominance of diffusion phenomena over the convective transport phenomena in the water channel. Such behavior is desirable for fishes because it indicate that maximum of oxygen from given volume of water diffuses out of water stream and enters in blood stream before the water leave the inter-lamellar space. This condition ensures minimum washout of oxygen from the water stream while passing through inter-lamellar space. Although the mean value of water parametric ratio is below 1, the values greater than 1 just indicate the underutilization of gills for oxygen transport. From Figure 5, it was observed that the data of 3 out of 4 different parametric ratios for different fishes showed minimum variation as compared to the weight of fishes. Majority of ratios lie within the window of 1±1. The blood parameter was found to be greater than 1, indicating the dominance of convective process over the diffusion process. This take care of the continuous maintenance of oxygen deficient region in the blood for continued diffusion of oxygen from interlamellar space to blood channel. The value of these parametric ratios do not vary much with the weight of fishes indicating their evolutionary conservation in fishes to maximize it efficiency of gas/solute exchange. Further, the architectural design of primary and secondary lamella is also conserved across different fishes to aid in enhance mass transport. Therefore, the above parametric ratios and arrangement of primary and secondary lamella can be used as engineering design parameters to develop micro/nanodevices for gas and solute exchange applications. However, these data do not speak about the nature of variation of different parametric ratios with the efficiency of oxygen exchange. Therefore, we have studied the gas exchange in a secondary lamella during convective–diffusion phenomena to evaluate the nature of these parametric ratios through computational method. The concentration profile results of oxygen in different layers a) water channel b) thin epithelial layer and c) blood channel was shown in Figure 6B. From the concentration profile, it was evident that there always exist a concentration gradient at an entry of blood channel and exit of water channel, even at steady state condition, providing a continuous diffusion of oxygen from water channel to blood channel via thin epithelial barrier. Further, when oxygen concentration at an exit of blood stream was evaluated against each parametric ratio defined previously, each parameter plays a distinctive role in oxygen concentration flux at the exit of blood channel (Figure 6A). It was observed that thickness of epithelial barrier should be as thin as possible as compared to the water channel to maximize the exchange of gas across the epithelial barrier. Moreover, the width of the blood channel also needs to be small as compared to water channel. The oxygen concentration at the exit of blood channel decreases rapidly with increase in the ratios of width of thin epithelial membrane to the width of the water channel and width of the blood channel to the width of the water channel, respectively (Figure 6A). However, the width of the blood channel need to be of an optimal size to avoid high pumping pressure to overcome increased blood flow resistance and assist in maximum diffusion of oxygen in the blood stream before oxygenated blood comes out of blood channel. Further, the blood parametric ratio suggests that above a critical ratio of 0.33, the effect of increase in ratio hardly affect the oxygen concentration at the exit of blood channel. Below the critical ratio, the diffusion phenomenon dominates the convection phenomena, thereby leading to low oxygen concentration in the blood at the exit. However, at a higher ratio, blood saturated with oxygen washes out of secondary lamella. This suggests that higher ratios are preferable for better gas exchange. The water parametric ratio has negligible effect on the oxygenation of blood within the range of ratios considered in the study. However, at a much higher ratio, the decrease in concentration of oxygen in blood clearly indicated the case of washout. It was observed that all the curves pertaining to different parametric ratios intersect at ratio of 0.33. This ratio marks the optimal ratio of different parameters to enable maximum exchange of gas from water stream to blood stream and is shown as a yellow star in the Figure 6A . Figure 6 A) Graph showing the effect of different dimensional less parameters (ratios) on % of oxygen concentration at an exit of blood channel a) Concentration profile of oxygen in different in blood and water channels. On comparison of results obtained from theoretical and computational model, it was found that overall results show good agreement with each other (Figure 7). Although, the three parametric ratios (ratio of thickness of membrane to width of water channel, ratio of width of blood channel to the width of water channel and water parametric ratios) predicted by computational model is in congruence with theoretical results as shown in figure, the blood parametric ratio calculated through computational model demonstrated significant difference from theoretical results. Figure 7 Graph showing the comparison of parametric ratios evaluated from computational method and theoretical analysis from data available in literature This might be due to fixed velocity assumed in blood channel for all the fishes in the theoretical analysis. The velocity of blood in the secondary lamella of fish varies with species of fishes, activity of fishes and water medium in which they thrive. Therefore, fishes manipulate the blood flow velocity in accordance with the width of inter lamellar space to maximize the sustained diffusion of oxygen from water to blood of fishes. Eventually, our study suggested few parametric ratios and structural parameters that are responsible for higher performance efficiency of gills in mass transfer. 5. Conclusion Fish gills are one of the excellent gas/solute exchange structure available in nature. Such performances are attributed to their extraordinary multi-scale hierarchical structures. Therefore, bio-mimicking these structures at different length scales to form a gas/solute exchange device require detailed study of their structural properties and how they guide the transport properties in fish gills. Further, there is also a need to formulate few dimensionless parameters which can be used as design parameters for development of gas/solute separation device. Hence, we have established four parametric ratios which affect the convection-diffusion phenomenon in fishes and studied their role in gas exchange through modelling and simulation. It was observed that these parametric ratios tend to have an optimal value around 0.33 for the maximization of gas exchange cross secondary lamella. Further, our theoretical analysis on 24 different fishes supported our argument about very limited variability of above parametric ratios with weight of fishes. Thus, structural design of biomimetic gas/solute exchange devices should incorporate these parametric ratios as design parameter. Apart from these parametric ratios, the role of the factors like dimensions of primary and secondary lamella, surface area of fish gills and inter- lamellar distance have been studied in detail. These studies can form a backbone for the biomimetic development of devices for gas/solute exchange. Acknowledgement This research effort was supported partially by Suman Mashruwala Microengineering Laboratory (www.me.iitb.ac.in/~mems) sponsored by IIT Bombay alumnus Mr. Raj Mashruwala and IITB- Monash Research Academy ( A joint venture between IIT Bombay, India and Monash University, Clayton, Melbourne, Australia). Few results in this piece of work has been published as a conference proceedings in international conference, SPIE/NDE, (Las Vegas, USA), 9797-40, 2016. References 1. 2. 3. 4. 5. 6. 7. 8. 9. Li, J.-L. and B.-H. Chen, Review of CO2 absorption using chemical solvents in hollow fiber membrane contactors. Separation and Purification Technology, 2005. 41(2): p. 109- 122. Gabelman, A. and S.-T. Hwang, Hollow fiber membrane contactors. Journal of Membrane Science, 1999. 159(1–2): p. 61-106. Potkay, J.A., The promise of microfluidic artificial lungs. Lab on a Chip, 2014. 14(21): p. 4122-4138. Steven Kim, B.F., Rishi Kant, Benjamin Chui, Ken Goldman, Jaehyun Park, Willieford Moses, Charles Blaha, Zohora Iqbal, Clarence Chow, Nathan Wright, William H. Fissell, Andrew Zydney, Shuvo Roy, Diffusive Silicon Nanopore Membranes for Hemodialysis Applications. PLoS ONE 2016. 11(7): p. e0159526. Nguyen, D.T., Y.T. Leho, and A.P. Esser-Kahn, A three-dimensional microvascular gas exchange unit for carbon dioxide capture. Lab on a Chip, 2012. 12(7): p. 1246-1250. Randall, D.J., 7 Gas Exchange in Fish, in Fish Physiology, W.S. Hoar and D.J. Randall, Editors. 1970, Academic Press. p. 253-292. Evans, D.H., P.M. Piermarini, and K.P. Choe, The Multifunctional Fish Gill: Dominant Site of Gas Exchange, Osmoregulation, Acid-Base Regulation, and Excretion of Nitrogenous Waste. Physiological Reviews, 2004. 85(1): p. 97. Wilson, J.M. and P. Laurent, Fish gill morphology: inside out. JOURNAL OF EXPERIMENTAL ZOOLOGY, 2002. 293(3): p. 192-213. Hughes, G.M. and M. Morgan, THE STRUCTURE OF FISH GILLS IN RELATION TO THEIR RESPIRATORY FUNCTION. Biological Reviews, 1973. 48(3): p. 419-475. Hughes, G.M., Morphometrics of fish gills. Respiration Physiology, 1972. 14(1): p. 1-25. Hughes, G.M., The Dimensions of Fish Gills in Relation to Their Function. Journal of Experimental Biology, 1966. 45(1): p. 177. G. M. HUGHES, M.M., The structure of fish gills in relation to their respiratory function. Biological Reviews, 1973. 48(3): p. 419-475. JOHANNES PIIPER, P.S., Gas transport efficacy of gills, lungs and skin: theory and experimental data. Respiration Physiology, 1975. 23: p. 209-221. JOHANNES PIIPER, P.S., Maximum gas transfer efficacy of models for fish gills, avian lungs and mammalian lungs. Respiration Physiology, 1972. 14: p. 115-124. Park, K., W. Kim, and H.-Y. Kim, Optimal lamellar arrangement in fish gills. Proceedings of the National Academy of Sciences, 2014. 111(22): p. 8067-8070. Reebs, S.G. Oxygen and fish behavior. 2009 [cited 2016 2nd Nov]; Available from: www.howfishbehave.ca. JJ Cech Jr., C.B., Gas exchange: Respiration: An Introduction, in Encyclopedia-of-Fish- Physiology. 2011. p. 791-795. Golding, H.M., Hierarchical structures and transport processes in "fish gill" systems, in Final Year Report. 2010, Monash University: Clayton. Goldstick, T.K., V.T. Ciuryla, and L. Zuckerman, Diffusion of Oxygen in Plasma and Blood, in Oxygen Transport to Tissue - II, J. Grote, D. Reneau, and G. Thews, Editors. 1976, Springer US: Boston, MA. p. 183-190. H, M., Pressure/flow relations in the interlamellar space of fish gills: theory and application in the rainbow trout. Respiratory Physiology, 1989. 78(2): p. 229-41. Jane B. Reece, M.R.T., Eric J. Simon, Jean L. Dickey, Kelly A. Hogan, Campbell Biology: Concepts & Connections. Vol. 8. 2015: Pearson Education. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.
1804.10807	3	1804	2018-11-29T09:51:43	Sojourn-time distribution of virus capsid in interchromatin corrals of a cell nucleus	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.SC" ]	Virus capsids in interchromatin corrals of a cell nucleus are experimentally known to exhibit anomalous diffusion as well as normal diffusion, leading to the Gaussian distribution of the diffusion-exponent fluctuations over the corrals. Here, the sojourn-time distribution of the virus capsid in local areas of the corral, i.e., probability distribution of the sojourn time characterizing diffusion in the local areas, is examined. Such an area is regarded as a virtual cubic block, the diffusion property in which is normal or anomalous. The distribution, in which the Gaussian fluctuation is incorporated, is shown to tend to slowly decay. Then, the block-size dependence of average sojourn time is discussed. A comment is also made on (non-)Markovianity of the process of moving through the blocks.	physics.bio-ph	physics	Sojourn-time distribution of virus capsid in interchromatin corrals of a cell nucleus YUICHI ITTO a and JENS B. BOSSE b a Science Division, Center for General Education, Aichi Institute of Technology, b Heinrich Pette Institute, Leibniz Institute for Experimental Virology, Aichi 470-0392, Japan Hamburg, Germany Abstract Virus capsids in interchromatin corrals of a cell nucleus are experimentally known to exhibit anomalous diffusion as well as normal diffusion, leading to the Gaussian distribution of the diffusion-exponent fluctuations over the corrals. Here, the sojourn-time distribution of the virus capsid in local areas of the corral, i.e., probability distribution of the sojourn time characterizing diffusion in the local areas, is examined. Such an area is regarded as a virtual cubic block, the diffusion property in which is normal or anomalous. The distribution, in which the Gaussian fluctuation is incorporated, is shown to tend to slowly decay. Then, the block-size dependence of average sojourn time is discussed. A comment is also made on (non-)Markovianity of the process of moving through the blocks. 1 1. Introduction Viruses exhibit rich phenomena, which are highly attractive from the viewpoint of physics [1-3]. In particular, a recent experimental study [4] (see also Ref. [5]) has reported a remarkable finding for the diffusion phenomenon of herpesviruses in nuclei of PtK2 cells. There, the cells were infected with pseudorabies virus (i.e., suid herpesvirus 1) or herpes simplex virus 1. Capsid, which is a protein shell surrounding viral DNA, has been labeled with a fluorescent protein, and tracks of such virus capsids have been observed in the nucleus by the single particle tracking. The experiment has shown that the virus capsids diffuse in interchromatin compartments, which are formed by chromatin (i.e., chromosomal substance). Such a compartment is called corral. It has also been found that, during virus infection, chromatin structure becomes more porous, which makes the corral size increase. As a result, the virus capsid moves through the corrals in order to reach nuclear membrane. The diffusion property has been characterized by the mean square displacement of the virus capsid based on each track: x ~2 αt , (1) where t is elapsed time and α is termed the diffusion exponent. Then, the diffusion exponent fluctuates depending on the corrals in a wide range of α [4,5]. More precisely, the exponent takes not only ,1=α i.e., normal diffusion, but also ,1≠α i.e., anomalous diffusion: subdiffusion (superdiffusion) in the case with 0 <<α 1 ( 1>α ). Then, α obeys the following Gaussian distribution: 2 ( αf ~) exp −     2 ) ( αα − 0 2 2 σ     , (2) where 0 =α 85.0 and 24.0=σ are, respectively, the mean value and the standard deviation of .α Thus, the diffusion observed offers an outstanding feature in anomalous diffusion [6-8] under vital investigation in the literature. Regarding the Gaussian fluctuation in Eq. (2), a fact to be emphasized is its robustness in the sense [4,5] that it takes the same form for both types of the virus (i.e., the pseudorabies virus and the herpes simplex virus 1). Thus, this is seen to manifest the existence of universality of the fluctuations over the corrals. We point out that the Gaussian distribution in Eq. (2) has theoretically been derived in a consistent manner in a recent work [9]. In such a situation, a central issue is to understand fundamental dynamics of the virus capsid in the corrals. In fact, to clarify its spatial property, the distribution of the spatial displacement of the capsid in the corrals has been analyzed [4]. The analysis is seen to suggest that this distribution may obey an exponential law, which means that large displacements are not significant. In the random walk scheme [10], this may imply [11] that the spatial property is trivial for the origin of subdiffusion as well as normal diffusion but is not in the case of superdiffusion, where the presence of such displacements is necessary like, e.g., Lévy flights [12], showing how exotic the dynamics of the virus capsid is. In the present work, we focus our attention on the temporal property of diffusion of the virus capsid in the corrals. In particular, we examine the sojourn-time distribution, which is meant as probability distribution of the sojourn time characterizing diffusion in 3 local areas of the corral. Such local areas are treated as virtual cubic blocks, in which the virus capsid exhibits normal or anomalous diffusion. We show that the distribution, in which the Gaussian fluctuation in Eq. (2) is incorporated, tends to slowly decay. Then, we analyze the dependence of average sojourn time on the block size based on the experimental data. In addition, we present a brief discussion about (non-)Markovianity of the process of moving from one block to another in connection with a relation to be satisfied by a class of Markovian processes. This may offer a possible way of examining if there exists a long-term memory in the process or not. The present work is expected to contribute to deeper understanding diffusion of the virus capsid in the nucleus. Throughout the present work, variation of the fluctuations is considered to be very slow on a long time scale, on which the Gaussian fluctuation in Eq. (2) is realized. In other words, α is supposed to be approximately constant on such a time scale. Then, the minimum value of α is not zero and (αf 0~) for such a value, which is supported by the raw data of the experiment [4]. 2. Sojourn-time distribution Consider diffusion of the virus capsid over the region of the corrals in the nucleus. Like in recent works [13,14], we regard this region as a medium for diffusion of the capsid at both viral types, which is composed of many virtual cubic blocks. The virus capsid exhibits normal or anomalous diffusion, depending on these local blocks. It should be noticed [4] that 2x in Eq. (1) has been analyzed for the elapsed time smaller than that for determination of the corral size. Accordingly, the block is identified with a local area of the corral. To evaluate the block size, we express Eq. (1) more precisely as 4 x 2 = 6 αtD [4]. D is a generalized diffusion coefficient denoted here as D ∆= /2 αs , where ∆ and s are positive characteristic constants describing a spatial displacement of the virus capsid and time being required for the displacement, respectively. Then, let l ≡ 2x be the length of the side of the cubic block. Thus, the block size, ,l is determined in this way. As can be seen from the above discussion, l depends on the elapsed time, given a value of .α This, in turn, means that the elapsed time depends on ,α given a value of ,l which is given by /2 α t =     l 6 ∆     . s (3) This offers the sojourn time characterizing diffusion of the virus capsid in a given local block. Then, what we are interested in here is its statistical property, in which the Gaussian fluctuation in Eq. (2) is taken into account. This shows how long/short the virus capsid stays in a given local block at the statistical level. The virus capsid moves through the local blocks with different exponents over the medium and therefore the block size can have different values. However, in order to clarify the statistical property as simple as possible, in the present paper, we discuss the case when l is set as a certain value for all of the blocks in the medium. In other words, the medium is divided into the blocks with an equal block size. [Note that α fluctuates according to the distribution in Eq. (2).] Thus, in such a case, t in Eq. (3) becomes a random variable, t = :) (αt the randomness comes from the diffusion exponent, .α It is noticed that we 5 are interested in the case when t > ,s which requires the block size to fulfill >l .6 ∆ We denote the probability of finding a certain value of t in the interval ,[ ] τττ d+ by P . ( ττ d ) This describes the sojourn-time distribution, which can be formally given as follows: P ( ) τ = ( ) αατδ − ) t ( (4) with α• being the average over the Gaussian distribution in Eq. (2). Therefore, in the case when ,s>τ substitution of Eq. (3) into Eq. (4) leads to the following form of the sojourn-time distribution: P ( τ ~) 1 / ττ ln( [ exp 2 s ])     −    ln[2 ])6(/ l ln( / τ ∆ ) s − α 0 2    2(/ 2 σ ) (5)   ,   showing that the distribution tends to slowly decay, since it is logarithmically related to .τ 3. Block-size dependence As mentioned earlier, it has been observed [4] that the corral size increases, since chromatin structure becomes more porous during virus infection. This seems to correspond to the situation that the block size increases. So, it may be of interest to evaluate the block-size dependence of average sojourn time, which is given, from Eq. (4), by the average of ) (αt with respect to :) (αf t 6 ( α = ) α t ) ( α ( α l ). Below, we wish to discuss this issue based on the experimental data. Here, the characteristic constants and the range of the block size are taken as follows. According to the experiment [4], to determine the positions of the virus capsids at both viral types, the cells have been imaged at a frame rate of 36 frames per second, for which the capsids are present in each consecutive frame. This seems allow us to choose the value of s as the inverse of the rate, .0=s 028 s, for both viral types. Then, it has also been shown that =D .0 035 μm 2 αs/ with .0=α 961 for the pseudorabies virus, whereas =D .0 023 μm 2 αs/ with .0=α 918 for the herpes simplex virus 1. Although these values are obtained from average over not a given track but all of the tracks, we employ them as the representative values of both D and .α Therefore, the values of ∆ are estimated as follows: .0=∆ 034 μm for the pseudorabies virus, whereas .0=∆ 029 μm for the herpes simplex virus 1. Regarding the value of ,l it should be larger than .0 125 μm, which is the diameter of the virus capsid as a sphere. Then, we suppose that the block size has its largest value given by l = 2x with the above representative values at s,36.0=t since the Gaussian distribution in Eq. (2) has been observed at such an elapsed time [4]. Such largest values are estimated as μm28.0 and μm23.0 for the pseudorabies virus and the herpes simplex virus 1, respectively. In Fig. 1, we present the plots of αα) (t for both viral types, in which the raw data of α obtained in the experiment [4] has been employed. There, we see that αα) (t monotonically increases with respect to l at both types. In addition, the value of αα) (t for the pseudorabies virus is seen to be smaller than that for the herpes simplex virus 1 at each value of ,l indicating that the capsid of the former diffuses faster than that of the latter. 7 4. Comment on (non-)Markovian nature The virus capsid moves from one cubic block to another. In other words, the virus capsid passes through the boundary of a given cubic block. We treat this as an event. In this section, we make a comment on (non-)Markovianity of the process of such events over the medium. For it, we examine a relation to be satisfied by a class of Markovian processes, which has been discussed in the context of laser cooling of atoms in Ref. [15]. Suppose a sequence of the events occurred in the time interval ,0[ t ], where the same symbol t as that in Eq. (1) is used for time (i.e., the conventional time) but it will not cause confusion. In this situation, ) (τP in Eq. (4) seems to play a role of the distribution of the time interval between two successive events. Let us denote the mean density of events at time t by (tS ). If the process is Markovian, then the following equation holds [15]: )( tS = )( tP + t ∫ 0 ( tPtd ′ ′− ( ) tSt ′ ). (6) Since this includes a convolution integral, the Laplace transformation of Eq. (6) yields L (][ uS ) = L − [ ) ]( uP ]( [ uP L ) 1 , (7) where L (][ ug ) ∞∫≡ 0 ut− etd ( tg ). We are particularly interested in the long time behavior 8 of (tS ). Using Eq. (4), [ (] uPL ) is therefore calculated, up to the first order of u (i.e., small-u behavior), to be leading to [ L (] uP 1~) αα− ) t ( u , (8) L (][ uS ~) 1 ( ) αα t .1 u (9) Thus, as the long time behavior, we obtain tS ~)( 1 ) ( ααt . (10) Accordingly, the mean density of events behaves as a certain constant at the block size under consideration, which is equal to the inverse of average sojourn time. In this respect, the analysis performed on αα) (t in the previous section tells us about the dependence of the mean density on the block size. The above observation means that, if the process is Markovian, then the virus capsid moves through the blocks at the rate of /1 t αα) ( for large time. Although such a feature is quite general for distributions with finite first moment, the point is in clarifying if the mean density to be observed deviates from the above rate or not. In other words, the violation of Eq. (10) implies the presence of a long-term memory in the 9 process, indicating that the events are not temporally separable. Then, this may be the case, as we shall see below. In Fig. 5 (B) in Ref. [4], the probability distribution of the spatial displacement of the capsid over s5.1 has been presented in the case of the pseudorabies virus. So, we here consider that a typical scale of the displacement, ,∗∆ is given by the spatial extension of the distribution such as its half-width: .μm05.0~∗∆ It seems therefore that the number of events in a certain time subinterval is proportional to ∗∆ 1.5(/ l ) [1/s]. Now, as can be seen from Fig. 1, this quantity deviates from /1 t ( αα) at each value of l in the same subinterval. For example, the ratio of the former to the latter at μm28.0=l is about .67.0 Thus, these facts may imply a possible violation of Eq. (10) and accordingly may support non-Markovianity of the process. We note, however, the following points. In the present case, the sojourn time in the local block is regarded as the time interval between two successive events. Therefore, further studies based on the time interval taken from a set of the time series data of the process seem to be needed for examining (non-)Markovianity of the process. 5. Concluding remarks We have examined the sojourn-time distribution of virus capsid diffusing in interchromatin corrals over nucleus of PtK2 cell for pseudorabies virus and herpes simplex virus 1. We have regarded the region of the corrals as a medium consisting of virtual cubic blocks with equal size. Taking the Gaussian fluctuation of the diffusion exponent into account, we have shown that the distribution tends to slowly decay. Combined with the raw data of experiment, we have performed the analysis of the block-size dependence of average sojourn time. We have also made a comment on 10 (non-)Markovian nature of the process of moving through the blocks. We point out the following. As mentioned in the Introduction, the dynamics of the virus capsid in the corrals seems to be exotic. A recent experimental work in Ref. [16] may provide further information about this point, in which simulation analysis based on the random walk scheme has been performed for herpesvirus capsids. In addition, it may be of interest to examine application of a theoretical framework in Ref. [17] (see also Ref. [18]) based on the fluctuation distribution for describing diffusion of the virus capsid over the corrals. Acknowledgements We thank Lynn Enquist (Princeton University) for support. Y.I. would like to acknowledge the Heinrich Pette Institute, Leibniz Institute for Experimental Virology for the nice working atmosphere, which enabled the present work. He also thanks S. Abe for informative discussions. He was supported by the foreign research project of the Aichi Institute of Technology. The Heinrich Pette Institute, Leibniz Institute for Experimental Virology is supported by the Freie und Hansestadt Hamburg and the Bundesministerium für Gesundheit (BMG). References [1] P.G. Stockley, R. Twarock (Eds.), Emerging Topics in Physical Virology, Imperial College Press, London, 2010. 11 [2] W.H. Roos, R. Bruinsma, G.J.L. Wuite, Nat. Phys. 6, 733 (2010). [3] M.G. Mateu (Ed.), Structure and Physics of Viruses, Springer, Dordrecht, 2013. [4] J.B. Bosse, I.B. Hogue, M. Feric, S.Y. Thiberge, B. Sodeik, C.P. Brangwynne, L.W. Enquist, Proc. Natl. Acad. Sci. USA 112, E5725 (2015). [5] J.B. Bosse, L.W. Enquist, Nucleus 7, 13 (2016). [6] J.-P. Bouchaud, A. Georges, Phys. Rep. 195, 127 (1990). [7] F. Höfling, T. Franosch, Rep. Prog. Phys. 76, 046602 (2013). [8] R. Metzler, J.-H. Jeon, A.G. Cherstvy, E. Barkai, Phys. Chem. Chem. Phys. 16, 24128 (2014). [9] Y. Itto, Phys. Lett. A 382, 1238 (2018). [10] E.W. Montroll, G.H. Weiss, J. Math. Phys. 6, 167 (1965). [11] R. Metzler, J. Klafter, Phys. Rep. 339, 1 (2000). 12 [12] M.F. Shlesinger, G.M. Zaslavsky, U. Frisch (Eds.), Lévy Flights and Related Topics in Physics, Springer, Heidelberg, 1995. [13] Y. Itto, J. Biol. Phys. 38, 673 (2012). [14] Y. Itto, Physica A 462, 522 (2016). [15] F. Bardou, J.-P. Bouchaud, A. Aspect, C. Cohen-Tannoudji, Lévy Statistics and Laser Cooling, Cambridge University Press, Cambridge, 2002. [16] V. Aho, et al., Sci. Rep. 7, 3692 (2017). [17] Y. Itto, Phys. Lett. A 378, 3037 (2014). [18] Y. Itto, in: Atta-ur-Rahman, M. Iqbal Choudhary (Eds.), Frontiers in Anti-Infective Drug Discovery, Vol. 5, Bentham Science Publishers, Sharjah, 2017. 13 Fig. 1 The block-size dependence of average sojourn time, , ( ααt ) of the virus capsids. The filled circles and open squares are for the pseudorabies virus and the herpes simplex virus 1, respectively. 14
1312.3673	1	1312	2013-12-12T23:41:35	Lag, lock, sync, slip: the many "phases" of coupled flagella	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	In a multitude of life's processes, cilia and flagella are found indispensable. Recently, the biflagellated chlorophyte alga Chlamydomonas has become a model organism for the study of ciliary coordination and synchronization. Here, we use high-speed imaging of single pipette-held cells to quantify the rich dynamics exhibited by their flagella. Underlying this variability in behaviour, are biological dissimilarities between the two flagella - termed cis and trans, with respect to a unique eyespot. With emphasis on the wildtype, we use digital tracking with sub-beat-cycle resolution to obtain limit cycles and phases for self-sustained flagellar oscillations. Characterizing the phase-synchrony of a coupled pair, we find that during the canonical swimming breaststroke the cis flagellum is consistently phase-lagged relative to, whilst remaining robustly phase-locked with, the trans flagellum. Transient loss of synchrony, or phase-slippage, may be triggered stochastically, in which the trans flagellum transitions to a second mode of beating with attenuated beat-envelope and increased frequency. Further, exploiting this alga's ability for flagellar regeneration, we mechanically induced removal of one or the other flagellum of the same cell to reveal a striking disparity between the beating of the cis vs trans flagellum, in isolation. This raises further questions regarding the synchronization mechanism of Chlamydomonas.	physics.bio-ph	physics	Lag, lock, sync, slip: the many "phases" of coupled flagella Kirsty Y. Wan, Kyriacos C. Leptos, Raymond E. Goldstein Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences, University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, UK In a multitude of life's processes, cilia and flagella are found indispensable. Recently, the biflagellated chlorophyte alga Chlamydomonas has become a model organism for the study of ciliary coordination and synchronization. Here, we use high- speed imaging of single pipette-held cells to quantify the rich dynamics exhibited by their flagella. Underlying this variabiltiy in behaviour, are biological dissimilarities between the two flagella -- termed cis and trans, with respect to a unique eyespot. With emphasis on the wildtype, we use digital tracking with sub-beat-cycle resolution to obtain limit cycles and phases for self-sustained flagellar oscillations. Characterizing the phase-synchrony of a coupled pair, we find that during the canonical swimming breaststroke the cis flagellum is consistently phase-lagged relative to, whilst remaining robustly phase-locked with, the trans flagellum. Transient loss of synchrony, or phase-slippage, may be triggered stochastically, in which the trans flagellum transitions to a second mode of beating with attenuated beat-envelope and increased frequency. Further, exploiting this alga's ability for flagellar regeneration, we mechanically induced removal of one or the other flagellum of the same cell to reveal a striking disparity between the beating of the cis vs trans flagellum, in isolation. This raises further questions regarding the synchronization mechanism of Chlamydomonas flagella. 1 Introduction Periodicity permeates Nature and its myriad lifeforms. Oscillatory motions lie at the heart of many important biological and physiological processes, spanning a vast dynamic range of spatial and temporal scales. These oscillations seldom occur in isolation; from the pumping of the human heart, to the pulsating electrical signals in our nervous systems, from the locomotive gaits of a quadruped, to cell-cycles and circadian clocks, these different oscillators couple to, entrain, or are entrained by each other and by their surroundings. Uncovering the mechanisms and consequences of these entrainments provides vital insight into biological function. Often, it is to the aid of quantitative mathematical tools that we must turn for revealing analyses of these intricate physiological interactions. following dictate reinhardtii The striking, periodic flagellar beats of one particular discussion: organism shall alga Chlamydomonas whose twin flagella undergo bilateral beating to elicit breaststroke swimming. For these micron-sized cells, their motile appendages, termed flagella, are active filaments that are actuated by internal molecular motor proteins. Each full beat cycle comprises a power stroke which generates forward propulsion, and a recovery stroke in which the flagella undergo greater curvature excursions, thereby overcoming reversibility of Stokes flows [1]. A single eyespot breaks cell bilateral symmetry, distinguishing the cis flagellum (closer to the eyespot), from the trans flagellum (figure 1). Subject to internal modulation by the cell, and biochemical fluctuations in the environs, the two flagella undergo a rich variety of the is a unicellular 1 -- 12; September 12, 2018 tactic behaviours. For its ease of cultivation and well- studied genotype, Chlamydomonas has become a model organism for biological studies of flagella/cilia-related mutations. For its simplistic cell-flagella configuration, Chlamydomonas has also emerged as an idealised system onto which more physical models of flagellar dynamics and synchronization can be mapped [2 -- 5]. With this versatility in mind, the present article has two goals. First, we proffer a detailed exposition of Chlamydomonas flagella motion as captured experimentally by high-speed imaging of live cells; second, we develop a quantitative framework for interpreting these complex nonlinear motions. In the light of previous work, we have found the motion of Chlamydomonas flagella to be sufficiently regular to warrant a low-dimensional phase-reduced description [2, 3, 6, 7]. Single flagellum limit cycles are derived from real timeseries, and are associated with a phase (§3.b). For each cell, dynamics of the flagella pair can thus be formulated in terms of mutually coupled phase oscillators (§3.c), whose pairwise interactions can be determined to sub-beat-cycle resolution. Just as marching soldiers and Olympic hurdlers alike can have preferential footedness, we find that Chlamydomonas is of no exception; resolving within each cycle of its characteristic breaststroke gait we see that one flagellum is consistently phase-lagged with respect to the other. These transient episodes, previously termed slips [2], are to be identified with phase slips that occur when noisy fluctuations degrade the phase- locked synchronization of two weakly coupled oscillators of differing intrinsic frequencies [8]. For each cell, sampled here over thousands of breaststroke cycles, and supported by multi-cell statistics, we clarify the non-constancy of synchrony observed over a typical cycle. In particular, the two flagella are found to be most robustly synchronized in the power stroke, and least synchronized at the transition 3 1 0 2 c e D 2 1 ] h p - o i b . s c i s y h p [ 1 v 3 7 6 3 . 2 1 3 1 : v i X r a 2 (BS-mode), trans- uniflagellated cells can instead sustain the faster mode of beating associated with the phase slip (aBS-mode) (figure 10a). Yet this cis-trans specialization is lost once the cell is allowed to regrow the lost flagellum to full-length, by which time both flagella have recovered the BS-mode. Flagellar tracking has enabled us to acquire true spatial localization of the flagellum throughout its dynamic rhythmicity, complementing recent efforts aiming in this direction [4, 10] The need to know precise waveforms has long been an ambition of historic works, in which manual light-table tracings of Chlamydomonas flagella were used to elucidate behaviour of the wildtype [11, 12], and latterly also of flagellar mutants [13]. We hope that the findings and methodologies herein presented shall be of broad interest to physicists and biologists alike. 2 Background a The enigmatic flagellum beat At a more fundamental level, how is beating of a single flagellum or cilium generated, and moreover, how can multi-ciliary arrays spontaneously synchronize? For each of us, or at least, for the hair-like appendages lining the epithelial cells of our respiratory tracts, this is indeed an important question. Beating periodically, synchronously, and moreover metachronously, multitudes of these cilia drive extracellular fluid flows which mediate mucociliary clearance. These motile cilia, and their non-motile counterparts, are regulated by complex biochemical networks to perform highly-specific functions [14, 15]. Mutations and defects in these organelles have been increasingly implicated in many human disorders including blindness, obesity, cystic kidney and liver disease, hydrocephalus, as well as laterality defects such as situs inversus totalis [16, 17]. Mice experiments in which nodal flows are artificially disrupted directly link mechanical flows to positioning of morphogens, which in turn trigger laterality-signalling cascades [18]. the slender, propulsion-generating appendages possess a remarkably conserved ultrastructure In recent decades, causality from structure to function within eukaryotic ciliary/flagellar established using tools. For the Chlamydomonas in particular, rapid freezing of specimens has made possible the capture of axonemal components in near-physiological states, at molecular- level resolution [20]. Chlamydomonas flagella have a well-characterized 9 + 2 structure of microtubule doublets (MTD), along which are rows of the molecular motor dynein (figure 1). These directional motors generate forces parallel to filamentous MTD tracks, which slide neighbouring MTDs past each other. Anchored to the A-tubule of each MTD by their tail domains, these dyneins detach and reattach onto eukaryotic phylogeny these Across [19]. axonemes has sophisticated molecular been genetics Figure 1. (a) Asymmetric cytoskeletal organization that underlies beating differences between cis and trans flagella. During development, eyespot positioning delineates between the two flagella; that closer to the eyespot is the cis flagellum, and the farther one is the trans flagellum. (b) Inside the axoneme: a peripheral arrangement of microtubule doublets encircles a central pair, and specialized dynein motors initiate interdoublet sliding and beat generation. to the succeeding recovery stroke. This trend appears to be universal to all cells of the wildtype strain. However, the tendency for the two flagella of a given cell to experience phase slips exhibits much greater variation across the population (figure 6). This we take as further indication that Chlamydomonas cells are highly sensitive to fluctuating biochemical cues. Sampled across multiple cells, phase slip excursions between synchronized states can be visualized by a dimension-reduced Poincaré return-map of interflagellar phase difference, showing the synchronized state itself to be globally stable. Examining each flagellar phase slip in detail we show further that the trans flagellum reproducibly transitions to a well-defined transient state with higher beat frequency and attenuated waveform. This evidences an alternative mode of beating - which we conjecture to exist as one of a discrete spectrum of modes on the eukaryotic flagellum. This second mode can also be sustained for longer durations, and in both flagella of a particular phototaxis mutant of Chlamydomonas, as detailed elsewhere [7]. Taken together, figure 3 encapsulates in a single diagram the three possible biflagellate "gaits", their differences and similarities, highlighting the need for a quantitative formulation such as that we present in this article. Intrinsic differences between the two Chlamydomonas flagella, their construction and actuation, underlie this rich assemblage of biflagellate phenomenology. Past experiments have shown such differences to exist, for example in reactivated cell models of Chlamydomonas [9], in which the trans flagellum has a faster beat frequency than the cis. In contrast, we consider here the in vivo case (§3.d); by mechanically inducing deflagellation of either trans or cis flagellum, we render live wildtype cells uniflagellate. This allowed us to compare the intrinsic beating dynamics of cis vs trans flagella. We found that whilst cis- uniflagellated cells tend to beat with the canonical breastroke-like mode (a)(b)C1C2ABradialspokeouterdyneininner dyneinB2cisB1transM4M2D2D4the next B-tubule, consuming ATP in the stepping process. Different dynein species coexist within the flagellar axoneme, with force-generation principally provided by outer dyneins, and modulation of flagellar waveform by the inner dyneins. The central pair, is thought to transduce signals via the radial spokes [21]. Approximately every 96 nm this precise arrangement of dyneins, radial spokes, and linkers repeats itself [22]. Periodic elastic linkages between neighbouring MTDs called nexins provide the coupling by which dynein-driven filament sliding produces macroscopic bending of the flagellum, which in turn propels the cell through the fluid. Treatments of axonemes which disrupt dynein domains have shown these nexin linkages to function in a non-Hookean manner to provide the elastic restoring force that resists internal filament shear [23]. More recently, detailed 3D-tomographic analysis revealed that nexins are in fact one and the same with the dynein regulatory complex, collectively termed the NDRC [24]. The importance of the NDRC within the functioning axoneme, from those of algae to that of humans, has been highlighted in a recent review [25]. With regard to the flagellum-cycling mechanism, consensus has been lacking. Timing-control appears to be absent, yet much experimental evidence points to selective, periodic, dynein-activation [26]. Both power and recovery strokes of the beat cycle involve active force generation by differentially actuated dyneins. Rhythmic beating of the flagellum may arise from dynamical instabilities of dynein oscillations [27, 28], and may be closely coupled to the intrinsic geometry of the axoneme [29, 30]. With these unanswered questions in mind, there is thus much incentive to analyze the flagellum beat in vivo, as the present study seeks to demonstrate. b Molecular origins of cis-trans difference As with many species of flagellated algae, motility is essential not just for swimming, but also for cell taxis. For Chlamydomonas, its two equal-length flagella, each 10 − 15 µm long, emerge from an ellipsoidal cell body ∼ 5 µm in radius. Cell bilateral symmetry is broken by the presence of an eyespot, which functions as a primitive photosensor. Perceived directional light is then converted downstream via secondary messenger ions into differential flagellar response and hence a turn of the cell [22, 31, 32]. The two anterior basal bodies from which the two flagella protrude are connected to each other via distal striated fibres [33]. A system of four acetylated microtubule rootlet-bundles lies beneath the cell membrane, and extends towards the cell posterior [32]. The eyespot is assembled de novo after each cell division, breaking cell bilateral symmetry by its association with the daughter four-membered rootlet (figure 1a, D4). The trans flagellum, nucleated by the mother basal body (B1), has been shown in reactivated cell models to beat at an intrinsic frequency that is ∼ 30 − 40% higher than that of the cis [9]. This frequency mismatch is discernible in vivo in wildtype cells that we 3 rendered uniflagellated through mechanical deflagellation (§3.d), additionally with a discrepancy in beating waveform. Differential phosphorylation of an axonemal docking complex has been suggested to underlie the distinctive cis-trans beat [34]. In particular, differential sensitivity to submicromolar Ca2+ in cell cis-trans models and in isolated axonemes [9, 36] is consistent with the opposing flagellar modulation necessary for cells to perform phototactic turns [13]. Yet, despite these intrinsic differences, the two flagella maintain good synchrony during free swimming [4, 10, 11], as well as when held stationary by a micropipette (here). Interflagellar coupling may be provided by the motion of the fluid medium [35, 37], by rocking of the cell-body [5], or further modulated internally via elastic components through physical connections in the basal region [33]. However, stochastically-induced flagellar phase slips can appear in otherwise synchronized beating flagella in a distinctive, reproducible manner (figure 3). We find that the propensity to undergo these transient slips [13] can vary significantly even between cells of a clonal population (figure 6). 3 Results their native breaststroke In a Three gaits of biflagellate locomotion habitat, The Chlamydomonas cells swim in water (kinematic viscosity ν = 10−6 m2 s−1), at speeds on the order of 100 µm/s, up to a maximum of 200 µm/s depending on strain and culture growth conditions [38]. Oscillatory flows set up by these flagella have a frequency f = ω/2π (cid:39) 60 Hz during the breaststroke. Stroke lengths of L = 10 µm thus produce a tip velocity scale of U = Lω ∼ 4 mm/s. An (oscillatory) Reynolds number Re = ωL2/ν gauges the viscous and inertial force contributions to the resulting flow. Here, Re ≈ 0.001, and cell propulsion thus relies on viscous resistance to shearing of the fluid by the flagellar motion. To overcome the reversibility of such flows, a breaking of spatial symmetry is essential during the swimming breaststroke. The rhythmic sweeping motion of each flagellum can be partitioned into distinct power and recovery strokes: during the power stroke, the flagella are extended to maximize interaction with the fluid, but undergo greater bending excursions during the recovery stroke (figure 3a). Net swimming progress results from the drag anisotropy of slender filaments and the folding of flagella much closer to the cell body during the recovery stroke. Interestingly, a qualitatively similar bilateral stroke can emerge from a theoretical optimisation performed on swimming gaits of biflagellate microorganisms [39] and on single flagella near surfaces [40]. Early microscopic analyses of The phase slip Chlamydomonas flagella suggested that the normal 4 Figure 2. Three Chlamydomonas breaststroke swimming gaits recorded at 3000 fps and shown at intervals of 5 frames (i.e. 1.7 ms intervals). Orange dot marks cell eyespot location, red curves: cis flagellum, blue curves trans flagellum. Shown in order, in-phase synchronized breaststroke (both flagella in BS-mode), a phase slip in the same cell (trans flagellum in aBS-mode), and antiphase synchronization in the phototaxis mutant ptx1 (both flagella in aBS-mode). Scale bar is 5µm. Figure 3. Overlaid sequences of tracked flagella showing (a) normal breaststroke (BS) (5 consecutive beats), (b) a stochastic slip event (4 consecutive slips), in which the trans flagellum transiently enters a different mode (aBS), and (c) in which both flagella of ptx1 sustain the aBS mode (5 consecutive beats). For each flagellum, progression through the beat cycle can be tracked using an angle θ defined relative to the cell bilateral axis (d); sample timeseries for θ are shown for each gait. The aBS-mode can clearly be seen to have an attenuated beat amplitude and a faster beat frequency. (e)-(f): Differences between cis and trans limit cycles are clarified in phase-space coordinates (θ,H(θ)), where H(.) denotes the Hilbert transform (see §3.b), and the color-intensity is obtained by logarithmically scaling the probability of recurrence. slipbreaststrokeantiphasecisbreastrokecistranstransciscistransbreaststroke synchrony "may be disturbed for brief periods" [33]. These interruptions to 'normal' beating, subsequently detailed in manual waveform tracings by Rüffer and Nultsch, were shown to occur in the absence of obvious stimulation, in both free-swimming cells [11] and cells affixed to micropipettes [12]. Crucially, these transient asynchronies do not significantly alter the trajectory of swimming (unpublished observation); instead, during each such episode the cell body is seen to rock back and forth slightly from frame to frame without altering its prior course. These asynchronies are termed slips [2, 3] by analogy with an identical phenomenon in weakly-coupled phase oscillators. Physically, phase slips are manifest in these coupled flagella in a strikingly reproducible manner. Under our experimental conditions (detailed in §5), beating of the trans flagellum transitions during a slip to a distinct waveform, concurrently with a ∼ 30% higher frequency [7], whilst at the same time the cis-flagellum maintains its original mode of beating throughout, apparently unaffected (figure 3b). We find also that the faster, attenuated breaststroke mode (aBS) is sustained by the trans flagellum for an integer number of full beat cycles, after which normal synchronized breaststroke (BS) resumes. The antiphase The aBS waveform assumed by the trans-flagellum during a slip turns out to be markedly similar to that identified in an anti-synchronous gait displayed by a particular phototaxis mutant of Chlamydomonas called ptx1 [41]. In recent, related work, we make these comparisons more concrete, and show that this gait (figure 3c) involves actuation of both flagella in aBS-mode, and in precise antiphase with each other [7]. Although the precise nature of the ptx1 mutation remains unclear, it is thought that emergence of this novel gait in the mutant is closely associated with loss of calcium- mediated flagellar dominance. b Phase-dynamics of a single flagellum Many biological oscillators are spatially extended and are therefore fundamentally high-dimensional dynamical objects, but adopting a phase-reduction approach facilitates quantitative analyses. In such cases, stable self- sustained oscillations can be represented by dynamics on a limit cycle, for which monotonically increasing candidate phases ϕ may be extracted. The natural or interpolated phase of an oscillator is defined to increment linearly between successive crossings of a Poincaré section, and as such is strongly dependent on the precise choice of the section. For discrete marker events {tn}, and tn ≤ t ≤ tn+1, (cid:18) (cid:19) t − tn tn+1 − tn ϕP (t) = 2π n + . (3.1) This method was used in earlier work [2, 3], to extract flagellar phases φcis,trans, by sampling pixel intensity variations over pre-defined regions of interest, 5 Figure 4. (a) Hilbert embedding for the cis-flagellum of a sample cell, recorded over thousands of beat cycles, and coloured according to the equivalent transformed phase φ (via equation 3.6). Points of equal phase lie on isochrones, highlighted here at equi-phase intervals. The rate of phase rotation varies systematically throughout the beat cycle, as indicated by the variable inter-isochrone spacing. Inset: successive zero-crossings of φ − (cid:104)φ(cid:105) map out one Poincaré- section. (b) Snapshots 1 − 5 show typical positions of the flagellum at phases corresponding to five representative isochrones. (c) Phase-velocity Γ(ϕ) = dϕH /dt is approximated by a truncated Fourier series. Shaded regions show one standard deviation of fluctuations in the raw data. on individual frames of recorded video. Specifically, phase values are interpolated between successive peaks in intensity. However as beating of the flagellum corresponds to smooth dynamics, by Poincaré sectioning the dynamics in this way, sub-beat-cycle information is lost. In the present work, we take a more continuous approach by incorporating enhanced resolution to elucidate within-beat-cycle dynamics, and make use of flagellum tracking to define 2D-projections of flagellum oscillations. For this we choose an embedding via the Hilbert transform. This technique derives from the analytic signal concept [42], and is used to unambiguously define an instantaneous phase (and amplitude) from scalar signals with slowly-varying frequency. From a periodic scalar timeseries x(t) we construct its complex extension (c)600200π2π012345Poincaré section0.51.01.5-0.5 0 0.50π2π(a)12345(b)6 Figure 5. Phase-deviation over long-timescales obtained by subtracting from unwrapped phases the linear component that scales with oscillator frequency. General trends are preserved by the phase transformation (equation 3.4), but within-cycle fluctuations are removed (cid:90)∞ −∞ ζ(t) = x(t) + ix(t), where x(t) is given by H(x) = x(t) = PV 1 π x(τ ) t − τ dτ . (3.2) Here, the integral to be taken in the sense of the Cauchy principal value. Polar angle rotation in the x−x phase- plane gives the Hilbert phase, ϕH (t) = tan−1 (cid:18) x − x0 (cid:19) (3.3) , x − x0 which, when unwrapped, serves as a monotonically- increasing candidate phase on our limit cycle projection. The origin (x0, x0) is chosen to be strictly interior of the limit cycle; here, x0 = (cid:104)x(cid:105), x0 = (cid:104)x(cid:105). Candidate phases such as the Hilbert phase do not uniformly rotate and are sensitive to cycle geometry and nonlinearity. That is, ϕ = Γ(ϕ) is in general some non-constant but 2π-periodic function. For any given limit cycle, there is however a unique true phase φ for which ω = dφ/dt is constant, the rate of rotation of φ is equal to the autonomous oscillator frequency. The desired transformation ϕ → φ is (cid:90) φ = ω ϕ (Γ(ϕ))−1 dϕ . (3.4) Whilst φ is unique, ambiguity remains in the choice of limit cycle that best characterizes the original phase- space. Furthermore for noisy timeseries, limit cycle trajectories do not repeat themselves exactly, so that the phase transformation (3.4) can only be performed in a statistical sense. Accuracy of phase estimation is improved with longer observation time. To derive an approximation for dϕ/dt as a function of ϕ, we first sort data pairs (cid:104)ϕ, dϕ(cid:105) and then average over all ensemble realizations of ϕ (figure 4). Direct numerical approximations for dt/dϕ are sensitive to noise, due to the heavy-tailed nature of ratio distributions, to remedy this, we follow the approach of Revzen et al [43] and begin by finding an N-th order truncated Fourier series approximation Γ = FN [Γ], to Γ. Next, we find a similar Fourier approximation to 1/Γ, (ϕ) ≈ FN [1/Γ] = dt dϕ fkeikϕ . (3.5) dϕ = ω−1 where The zeroth coefficient f0 = 1 2π ω is the intrinsic frequency of the oscillator, so that to lowest order, ϕ and φ coincide. Substituting (3.5) into (3.4) gives T 0 dϕ N(cid:88) (cid:16) dt k=−N (cid:82) (cid:17) N(cid:88) k=1 (cid:18) fk k (cid:19) φ = ϕ + 2ω Im (eikϕ − 1) . (3.6) For flagellum oscillations, we choose scalar timeseries x = θcis/trans (Fig. 3). For each flagellum, pairs of values x = (x, x) define a noisy limit cycle in phase space. Using equation. 3.6, we can associate a phase at each x, consistent with the notion of asymptotic phase defined in the attracting region around a limit cycle. For different flagella, the function Γ(ϕ) takes a characteristic form (4c). Points of equal phase lie on isochrones, which foliate the attracting domain (figure 4a). Perturbations that are 2π periodic functions of ϕ are eliminated by the transformation 3.6. Within- period oscillations are averaged out, whilst long-timescale dynamics are preserved, by virtue of the invertibility of the transformation. This can be seen over long-time recordings, in which the two measures of phase deviation: Dφ = φ − ωt and Dϕ = ϕH − ωt coincide (figure 5), but the periodicity of ∆ϕ has been smoothed out (inset). c Phase-dynamics of coupled flagella pair For microorganisms that rely on multiple flagella for swimming motility, precision of coordination is essential to elicit high swimming efficacy. The bilateral geometric disposition of the two Chlamydomonas flagella facilitates extraction of phases for an individual flagellum's oscillations, and in turn, derivation of phase synchrony 01020time (s)-10103011.912.012.112.2phase deviation (rad)247 Figure 6. (a) Lag synchronization in bivariate timeseries of flagella beats, shown for a typical cell. Inset: cis flagellum begins its recovery stroke fractionally ahead of trans. (b) Multi-cell similarity functions show a similar trend. (c) Noisy flagellar dynamics within a population of 60 cells, as represented by the tendency of each cell to undergo slips. Most cells remain synchronized for more than 20 s, whilst some exhibit frequent asynchronies. Red to blue: from high to low probability of recurrence. relations between coupled pairs of flagella. However, a transformation function similar to equation 3.4 that is bivariate in the two phases cannot be derived from observations of the synchronized state alone; therefore in the following, we make use of the Hilbert phase (equation 3.3). the phase difference ϕtrans − ϕcis Phase difference derivation To monitor biflagellar synchrony, is of particular interest. More generically, for coupled noisy phase oscillators i and j, the dynamics of each is perturbed by the motion of the other, as well as by stochastic contributions: where again denotes the Hilbert transform. synchronized behaviour We measured ∆ for a large population of cells (Fig. 6). Phase-slip asynchronies are associated with rapid changes in interflagellar phase difference, and appear as step-like transitions that punctuate (sometimes lengthy) epochs of for which phase difference is constant. We see that over a comparable period of observation time (figure 6c:i-iii), pairs of flagella can experience either perfect synchrony, few slips, or many slips 1. For the population as a whole, the circular representation of figure 6 facilitates simultaneous visualization of general trends in interflagellar phase- synchrony. (i = 1, 2) , ϕi = f (ϕi) + εi g(ϕi, ϕj) + ξi(t) , (3.7) where g(ϕi, ϕj) is the coupling function, and ξ is a noise term. When ε1,2 = 0 we recover the intrinsic motion of a single flagellum. The two oscillators are said to be n : m phase-locked when their cyclic relative phase given by ∆n,m = nϕ1 − mϕ2, satisfies ∆n,m = Const.. For noisy or chaotic systems this condition may be relaxed to ∆n,m < Const. Here, we define the trans-cis phase difference from the respective angle signals θcis(t) and θtrans(t) by Lag synchronization Careful examination of a synchronized epoch shows that ∆ is not strictly constant, but rather fluctuates periodically about a constant value. During execution of breaststroke swimming, Poincaré sectioning of the dynamics has suggested previously that the breaststroke gait is perfectly synchronized [2, 3]. However, plotting θcis against θtrans (figure 6a) we see a consistent lag between the two flagella, which is most pronounced during the recovery stroke. By computing and minimizing the similarity function (cid:115)(cid:104)(θtrans(t + τ ) − θcis(t))2(cid:105) (cid:11)(cid:1)1/2 (cid:11)(cid:10)θ2 (cid:0)(cid:10)θ2 trans cis Λ(τ ) = (3.8) , ∆ = ϕtrans H (t) − ϕcis (cid:32) θtrans(t)θcis(t) − θtrans(t)θcis(t) H (t) (cid:33) = tan−1 θtrans(t)θcis(t) + θtrans(t)θcis(t) 00.40.800.40.8b)c)avg=0.0524lag time (/T)()power recovery recurrentnot recurrent0202a)cistranslagtrans (rad)cis (rad)8 Figure 7. Trends in cis-trans flagellar synchronization in a population of Chlamydomonas cells: each concentric annulus represents data from an individual cell, measured values are plotted on a circular scale 0 → 2π in an anticlockwise sense. (a) Probability of stroboscopically observing φH trans reaches a minimum (i.e. start of new power stroke). (b)-(c) Difference in tracked angles (∆θ = θtrans − θcis) and Hilbert phases (∆φ = φH cis), averaged over thousands of beat cycles. It is seen that ∆θ is greatest during the recovery stroke, and correspondingly ∆φ becomes increasingly negative. cis ∈ [0, 2π] at discrete marker events where θH trans − φH for scalar indicators chosen as lag we find this discrepancy to be indicative of the periodic angle variables synchronization. Here, θcis/trans are the progression of each flagellum through its beat cycle. In particular, the two phases are synchronized with a time lag τmin = minτ Λ(τ ), where Λ(τ ) assumes a global minimum. When the oscillators are perfectly synchronized, τmin = 0. We calculated Λ(τ ) for multiple cells, which displayed a similar profile (figure 6b). With τ normalized by the average inter-beat period T , we see that in every instance the minimum is displaced from 0 (or equivalently 1), with an value of 0.0524 ± 0.01, indicative of persistent directional lag. to subject Stability of fixed points and transients A phenomenological model such as equation 3.7 has a convenient dynamical analogy. Phase difference can be interpreted as particle in a washboard potential V (∆) = −∆δω + cos(∆), overdamped ∆ = −dV (∆)/d∆. Potential minima occur dynamics where ∆ = 0, which requires δω/ < 1. For noise with sufficient magnitude, the particle will have enough energy to overcome the potential barrier, at least transiently. Stochastic jumps between neighbouring potential minima, are manifest in coupled flagella as the slip-mode (figure 3b). In the vicinity of a potential minima, the stationary distribution of ∆ is predicted to be Gaussian (equation 3.7, with white noise). This phase distribution P (∆), can be measured directly from experiment, and assumed to satisfy P (∆) = exp(−U (∆)/kBT ) , (3.9) from which the potential structure U (∆) can be recovered. For each well, the peak location can be used to estimate the phase-lag of the coupled oscillators, while peak width is indicative of strength of noise in the system. We measured P (∆) for 18 cells which did not display slips for the duration of observation (figure 9a). Potential minima have a parabolic profile with a well- defined peak, on average displaced to the left of ∆ = 0, due to the characteristic phase-lag in the direction of the cis-flagellum. For certain cells, this lag is especially pronounced during the recovery-stroke than during the power stroke, resulting in a double-peaked minimum in the fine-structure of the empirical potential. The stability of the synchronized state may be assessed by observing trajectories that deviate from, but eventually return to, this state. Specifically, by measuring ∆ during multiple flagellar phase slips we construct a dimension-reduced return map for the joint system to visualize the potential landscape that extends between neighbouring minima. Slips occur with variable duration (figure 8a), in which the trans flagellum can sustain the faster aBS-mode for a variable but complete number of beat cycles. However for an individual cell, successive slips often exhibit identical dynamics (figure 8b). Figure 9b presents the return map of ∆ associated with > 500 slip events collected from 70 cells, where the discretized phase-difference ∆n is defined to be ∆ evaluated stroboscopically at the position of maximum angular extent of the trans-flagellum, during the nth beat-cycle. We begin by approximating the return map by a polynomial F (∆n). An empirical potential function [44] can be defined by integrating the difference δ∆ = 0.050.010.0020.0004-0.20.20.6-/3-/60/6(a)(b)(c)0.259 Figure 9. A potential analogy for wrapped phase difference, visualized a) within potential minima - ∆ exhibits local fluctuations during the stable synchronized gait; and b) between successive potential minima, via a first return map of cyclic/stroboscopic relative phase. An nth-order polynomial F (∆n) is fit to the multicell return- map statistics, from which an empirical potential function V (∆) can defined. The ability of Chlamydomonas to readily regenerate a lost flagellum has facilitated controlled measurements of flagellar coupling strength as a function of flagellum length [6]. Using the single eyespot as identifier, we removed either the cis or trans flagellum from a pipette- captured cell and recorded the beating dynamics of the remaining flagellum. Histograms of beat-frequencies are plotted in figure 10b. On average, cis- uniflagellated cells tend to beat at a lower frequency than trans- uniflagellated cells. A dissociation of beat-frequency of similar magnitude has been observed previously in reactivated cell models [9]. Moreover, we find that in the absence of the cis flagellum the trans flagellum can sustain the faster aBS-mode for thousands of beats. These differences are highlighted in Figure 10a, for a single cell. Interestingly, the aBS-mode that we can now associate with the intrinsic beating waveform of the trans, emerges transiently during a slip of the wildtype (figure 3b), but in both flagella during an antiphase gait of the mutant ptx1 (figure 3c, and [7]). Indeed, for ptx1, its lack of Figure 8. a) Harmonics of a slip, synchrony resumes after a different (but always integer) number of beats of either flagellum. b) each slip results in a step-like transition in phase-difference ∆. For the same cell, successive slips overlap. Inset: forking of unwrapped phases H and ϕtrans ϕcis H . ∆n+1 − ∆n ≈ F (∆n) − ∆n: V (∆n) = − (cid:90) ∆n δ(∆(cid:48)) d∆(cid:48) , (3.10) 0 V (∆n) = V (∆n) − min which we convert into a locally positive-definite function (3.11) which satisfies V (∆) > 0, ∀ ∆ /∈ argmin V (∆). The resulting effective potential profile (figure 9b, inset) represents the reproducible phase dynamics of a typical flagellar phase slip, from which breaststroke synchrony re-emerges. V (∆) , ∆ d Coupling cis and trans flagella Pre-existing, intrinsic differences between the two Chlamydomonas flagella are essential for control of cell reorientation, loss or reduction in cis-trans specialization may give rise to defective phototaxis in certain mutants of Chlamydomonas [7, 45]. Under general experimental conditions, stochastic asynchronies which we call slips can punctuate an otherwise synchronous breaststroke; more drastic loss of interflagellar synchrony can lead to drifts, which over time, can result in a diffusive random walk in the trajectory of an individual cell [2]. In all these instances, we observe the coupled state of two flagella; in contrast, by mechanically deflagellating wildtype cells (see §5) we can now examine the intrinsic behaviour of each oscillator in isolation. (b)011112020202θcis, θtrans (rad)(a)transcis0.5010100200time (ms)Δ/2πone slipavgtimeunwrapped phaseslip2π02πΔnΔn+1-1-0.500.5-log(P(Δ)Δ510slip(a)(b)syncV(Δ)F(Δ) - Δ2ππ01264208effectual cis-trans specialization has led to speculation that the mutation has renders both flagella trans-like [45]. Specific, structural differences known to exist between the cis vs trans axonemes [34] of the wildtype, may effect this segregation of intrinsic beating modes. used to probe defective swimming behaviours of motility mutants. In these cases, macroscopic measurements of population features may not be instructive to understanding or resolving the mutant phenotype, and would benefit from dynamic flagellum waveform tracking and in-depth analysis at the level of an individual cell. 10 4 Discussion For a unicellular flagellate such as Chlamydomonas, synchrony of its two flagella is intimately regulated by the cell's internal biochemistry; however, the exact mechanism by which messenger ions modulate and shape the flagellum beat remains unclear. Our experimental technique captures the motion of beating flagella in vivo, at high resolution, and with respect to a fixed pivot, thereby permitting long-time analysis. Associating each flagellum oscillator with a continuous phase, we formulated a phase-reduced model of the periodic dynamics. From long-time series statistics of bivariate oscillator phases, we used phase difference to track phase synchrony, quantifying flagellar interactions for a single individual, as well as across the sampled population. Exquisitely sensitive to its surroundings, a flagellum can be found to undergo precise, yet dynamic changes when executing its periodic strokes. Waveform tracking has allowed us to assess these changes in great spatio-temporal detail. In particular, we have found the stable phase- locked breaststroke of C. reinhardtii to exhibit a small but persistent cis-trans phase-lag, the magnitude and direction of which was evaluated from statistics of thousands of beat cycles using a similarity measure, and confirmed for multiple cells. However, often it is not the synchronized state itself but rather the emergence or cessation of synchrony that is most insightful for inferring fluctuations in the physiological state of a complex system. Phase slips are transient excursions from synchrony in which, under our experimental conditions, an alteration of beating mode is observed in the trans flagellum only, and that appear to be initiated by a reduced cis-trans phase lag (figure 8b). These reproducible events highlight the importance of cis-trans specialization of Chlamydomonas flagella. Exploring this further, we mechanically removed individual flagella of wildtype cells to obtain uniflagellated cis or trans versions, revealing significant differences between their isolated beating behaviours for a fully-intact cell, despite these inherent differences in beating frequency and in waveform, coupling interactions either hydrodynamically through the surrounding fluid medium, and/or biomechanically through elastic linkages at the base of the flagellar protrusion, appear sufficient on the most part to enslave the beating of the trans mode to that of the cis. (figure 10). Yet Ours is a very versatile technique for quantifying flagellar synchrony not just of the wildtype system, but such a phase analysis can for instance also be 5 Methods and techniques Single algal cells on micropipettes For purposes of flagella visualization we chose two wildtype C. reinhardtii strains, CC124 and CC125 (Chlamydomonas Center). Stock algae maintained on 2% TAP (Tris- Acetate Phosphate) solid agar slants, were remobilized for swimming motility by inoculation into TAP-liquid medium, and cultures used for experimentation were maintained in exponential growth phase (105 − 106 cells/ml) for optimal motility. Culture flasks are placed onto orbital shakers, and maintained at 24 ◦C in growth chambers illuminated on a 14:10 daily light/dark cycle, so as to imitate the indigenous circadian stimuli. Observation of flagellar dynamics was carried out on a Nikon TE2000-U inverted microscope, at constant brightfield illumination. Additional experiments were also performed with a long-pass filter (622 nm) to minimize cell phototaxis2. Individual cells were captured and held on the end of tapered micropipettes (Sutter Instrument Co. P-97), and repositioned with a precision micromanipulator (Scientifica, UK), and imaged at rates of 1, 000-3, 000 fps (Photron Fastcam, SA3). Digital Image and Signal Processing Recorded movies were transferred to disk for post-processing in MATLAB (Version 8.1.0, The Mathworks Inc. 2013.). Flagellar waveforms were extracted from individual frames, where contiguous dark pixels that localize the moving flagellum were fit to splines. Hilbert transforms were perform in MATLAB (Signal Processing Toolbox), and further timeseries analysis performed using custom MATLAB code. to locate the unique eyespot, Mechanical deflagellation of either cis or trans flagella To obtain the results described in §3.d, individual wildtype cells were first examined under white light thereby differentiating its cis flagellum from the trans. One flagellum was then carefully removed with a second micropipette, by exerting just enough shearing force to induce spontaneous deflagellation by self-scission at the basal region. That cells retain the ability for regrowth of flagella ensures basal bodies have not been damaged by our deflagellation treatment. Cells for which the beating of the remaining flagellum became abnormal or intermittent, and also for which a clear cis-trans identification could not be made, were duly discarded. 11 Figure 10. (a) For a single cell, cis and trans flagella were removed and in turn allowed to regrow to full length. Single-flagellum frequencies separate, but once regrown, lock to a common frequency (cis: 57.12 Hz, trans: 80.52 Hz, both: 63.38 Hz). Insets: typical cis and trans waveforms, the trans waveform is reminiscent of the aBS-mode that onsets during a slip. Waveforms are overlaid on an intensity plot of logarithmically-scaled residence times for each flagellum, over O(100) contiguous beats. (b) When beating in isolation, cis and trans flagella have different frequencies: areas - histograms of interbeat frequencies, lines - averaged frequencies. Data Examples of high-speed movies, flagellar time series, and other data referenced in this work can be found at: http://www.damtp.cam.ac.uk/user/gold/ datarequests.html. We thank Marco Polin for discussions. Financial support is acknowledged from the EPSRC, ERC Advanced Investigator Grant 247333, and a Senior Investigator Award from the Wellcome Trust (REG). Notes 1Of the population of cells analysed for figure 6c, most were observed under white light. However a small percentage (10%) were observed under red light, but which for the sake of clarity, have not been explicitly marked out in the figure. Whilst in both cases variability in frequency of flagellar slips is observed, we find that on average slips occur more prevalently in cells illuminated by red than by white light (discussed further elsewhere). 2Whilst cell phototaxis behaviour is minimized under red-light illumination, physiological cell motility cannot be maintained in prolonged absence of light, unless the experimental conditions were accordingly modified [7]. References [1] Purcell EM. 1977 Life at low reynolds-number. American Journal of Physics, 45, 3-11. (doi:10.1119/ 1.10903) [2] Polin M, Tuval I, Drescher K, Gollub JP, Goldstein RE. 2009 Chlamydomonas swims with two "gears" in a eukaryotic version of run-and-tumble locomotion. Science, 325, 487-490. (doi:10.1126/ science.1172667) [3] Goldstein RE, Polin M, Tuval of I. synchronization beating flagella. Phys. Rev. Lett., 103, 168103. PhysRevLett.103.168103) pairs in 2009 Noise and eukaryotic (doi:10.1103/ [4] Geyer VF, Jülicher F, Howard J, Friedrich BM. 2013 Cell-body rocking is a dominant mechanism for flagellar synchronization in a swimming alga. Proc. Natl. Acad. Sci. U.S.A., 110, 18058- 18063. (doi: 10.1073/pnas.1300895110) [5] Friedrich BM, Julicher F. 2012 Flagellar synchronization independent of hydrodynamic interactions. Phys. Rev. Lett., 109, 138102. (doi:10.1103/PhysRevLett.109.138102) [6] Goldstein RE, Polin M, Tuval during of eukaryotic (doi:10.1103/PhysRevLett.107.148103) the flagella. Phys. Rev. Lett., synchronized beating I. 2011 Emergence of 148103. regrowth 107, [7] Leptos KC, Wan KY, Polin M, Tuval I, Pesci AI, Goldstein RE. 2013 Antiphase synchronization in a flagellar-dominance mutant of Chlamydomonas. Phys. Rev. Lett., 111, 158101. (doi:10.1103/PhysRevLett.111.158101) [8] Pikovsky A, Rosenblum M, Kurths J. 2003 Synchronization: in Nonlinear Sciences., Cambridge A Universal Concept University Press. [9] Kamiya R, Witman GB. 1984 Submicromolar levels of calcium control the balance of beating between the two flagella in demembranated models of Chlamydomonas. Cell Motil. Cytoskeleton, 98 97-107. (doi:10.1083/jcb.98.1.97) [10] Guasto JS, Johnson KA, Gollub JP. 2010 Oscillatory in 168102. flows two (doi:10.1103/PhysRevLett.105.168102). by microorganisms Lett., swimming 105, Phys. Rev. dimensions. induced [11] Ruffer U, Nultsch W. 1985 High-speed cinematographic analysis of the movement of Chlamydomonas. Cell Motil. Cytoskeleton, 5, 251-263. [12] Ruffer U, Nultsch W. 1987 Comparison of the beating of cis-flagella and trans-flagella of Chlamydomonas cells held on micropipettes. Cell Motil. Cytoskeleton, 7, 87-93. (doi:10.1002/cm.970070111) [13] Ruffer U,Nultsch W. coordination in Chlamydomonas cells held on micropipettes. Cell Motil. 41, Cytoskeleton, (doi:10.1002/(SICI)1097- 0169(1998)41:4<297) 1998 Flagellar 297-307. 00.250.51030507090frequency (Hz)00.250.5pdftranscis(b)(a)frequency (Hz)trans204060801001202040608010012000.10.20.30.40.50.60.70.80.9cis00.0.20.30.40.50.60.70.80.91cis103050709000.5pdfisolatedcoupledtrans[14] Smith EF, Rohatgi R. 2011 Cilia 2010: The surprise 4, mr1. Signal., Sci. of organelle decade. (doi:10.1126/scisignal.4155mr1) the [15] Marshall WF, Nonaka S. 2006 Cilia: Tuning in to antenna. Curr. Biol., 16, R604-R614. cellâĂŹs the (doi:10.1016/j.cub.2006.07.012) [16] Ibanez-Tallon I, Heintz N, Omran H. 2003 To beat or not to beat: roles of cilia in development and disease. Hum. Mol. Genet., 12, R27-R35. (doi:10.1093/hmg/ddg061) [17] Fliegauf M, Benzing T, Omran H. 2007 Mechanisms of disease - when cilia go bad: cilia defects and ciliopathies. Nat. Rev. Mol. Cell Biol., 8, 880âĂŞ893. (doi:10.1038/nrm2278) [18] Nonaka S, Shiratori H, Saijoh Y, Hamada H. 2002 Determination of the mouse embryo by artificial nodal flow. Nature, 418, 96-99. (doi:10.1038/nature00849) left-right patterning of [19] Silflow CD, Lefebvre and motility of from Chlamydomonas reinhardtii. Plant Physiol., 127, 1500-1507. (doi:10.1104/pp.010807) cilia and flagella. Assembly lessons eukaryotic 2001 PA. [20] Nicastro D, Schwartz C, Pierson J, Gaudette R, Porter ME, McIntosh JR. 2006 The molecular architecture of axonemes revealed by cryoelectron tomography. Science, 313, 944-948. (doi:10.1126/science.1128618) [21] Smith EF. 2002 Regulation of flagellar dynein by the axonemal central apparatus. Cell Motil. Cytoskeleton, 52, 33- 42. (doi:10.1002/cm.10031) [22] Harris, EH. 2009 The Chlamydomonas sourcebook, Vol 1, 2nd edn. Academic Press. [23] Lindemann CB, Macauley LJ, Lesich KA. 2005 The counterbend phenomenon in dynein-disabled rat sperm flagella and what it reveals about the interdoublet elasticity. Biophys. J., 89, 1165-1174. (doi:10.1529/biophysj.105.060681) [24] Heuser T, Raytchev M, Krell J, Porter ME, Nicastro D. 2009 The dynein regulatory complex is the nexin link and a major regulatory node in cilia and flagella. J. Cell Biol., 187, 921- 933. (doi:10.1083/jcb.200908067) [25] Wirschell M, Olbrich H, Werner C, Tritschler D, Bower R, Sale WS, Loges NT, Pennekamp P, Lindberg S. et al. 2013 The nexin-dynein regulatory complex subunit drc1 is essential for motile cilia function in algae and humans. Nat. Genet., 45, 262-8. (doi:10.1038/ng.2533) [26] Nakano I, Kobayashi T, Yoshimura M, Shingyoji C. 2003 Central-pair-linked regulation of microtubule sliding by calcium in flagellar axonemes. J. Cell Sci., 116, 1627-1636. (doi:10.1242/jcs.00336) [27] Riedel-Kruse, IH, Hilfinger A, Howard J, Jülicher F. 2007 How molecular motors shape the flagellar beat. Hfsp Journal, 1, 192-208. (doi:10.2976/1.2773861) [28] Hilfinger A, Chattopadhyay AK, Jülicher F. 2009 Nonlinear dynamics of cilia and flagella. Phy. Rev. E, 79, 051918. (doi:10.1103/PhysRevE.79.051918) 12 [29] Brokaw CJ. 2009 Thinking about flagellar oscillation. Cell Motil. Cytoskeleton, 66, 425-436. (doi:10.1002/cm.20313) [30] Lindemann, CB, Lesich KA. 2010 Flagellar and ciliary beating: the proven and the possible. J. Cell Sci., 123, 519- 528. (doi:10.1242/jcs.051326) [31] Witman GB. 1993 Chlamydomonas phototaxis. Trends Cell Biol., 3, 403-408. (doi:10.1016/0962-8924(93)90091-E) [32] Dieckmann, CL. 2003 Eyespot placement and assembly in the green alga Chlamydomonas. Bioessays, 25, 410-416. (doi:10.1002/bies.10259) [33] Ringo DL. 1967 Flagellar motion and fine structure of flagellar apparatus in Chlamydomonas. J. Cell Biol., 33, 543-&. (doi:10.1083/jcb.33.3.543) [34] Takada S, Kamiya R. 1997 Beat frequency difference between the two flagella of Chlamydomonas depends on the attachment site of outer dynein arms on the outerdoublet microtubules. Cell Motil. Cytoskeleton, 36, 68-75. (doi:10.1002/(SICI)1097- 0169(1997)36:1<68) [35] Niedermayer T, Eckhardt B, Lenz P. 2008 Synchronization, phase locking, and metachronal wave formation in ciliary chains. Chaos, 18, 037128. (doi:10.1063/1.2956984) [36] Bessen M, Fay RB, Witman GB. 1980 Calcium control of waveform in isolated flagellar axonemes of Chlamydomonas. J. Cell Biol., 86, 446-455. (doi: 10.1083/jcb.86.2.446) [37] Uchida N, Golestanian R. 2011 Generic conditions for hydrodynamic synchronization. Phys. Rev. Lett., 106, 058104. (doi:10.1103/PhysRevLett.106.058104) [38] Racey TJ, Hallett R, Nickel B. 1981 A quasi-elastic light-scattering and cinematographic investigation of motile Chlamydomonas-reinhardtii. Biophys. J., 35, 557-571. [39] Tam D, Hosoi AE. 2011 Optimal feeding and swimming gaits of biflagellated organisms. Proc. Natl. Acad. Sci. U.S.A., 108, 1001-1006. (doi:10.1073/pnas.1011185108) [40] Eloy C, Lauga E. efficient (doi:10.1103/PhysRevLett.109.038101) Phys. Rev. cilium. 2012 Kinematics Lett., of 109, the most 038101. [41] Horst CJ, Witman GB. 1993 Ptx1, a nonphototactic mutant of Chlamydomonas, lacks control of flagellar dominance. J. Cell Biol., 120, 733-741. (doi:10.1083/jcb.120.3.733) [42] Gabor D. 1946 Theory of communication. J. Inst. Electr. Eng. (London), 93, 429-457. [43] Revzen S, Guckenheimer JM. 2008 Estimating the phase synchronized oscillators. Phy. Rev. E, 78, 051907. of (doi:10.1103/PhysRevE.78.051907) [44] Aoi S, Katayama D, Fujiki S, Tomita N, Funato T, Yamashita T, Senda K, Tsuchiya K. 2013 A stability-based mechanism for hysteresis in the walk-trot transition in quadruped locomotion. J. R. Soc. Interface 10, 20120908. (doi:10.1098/rsif.2012.0908) [45] Okita N, Isogai N, Hirono M, Kamiya R, Yoshimura K. 2005 Phototactic activity in Chlamydomonas "non-phototactic" mutants deficient in Ca2+ -- dependent control of flagellar dominance or in inner-arm dynein. J. Cell Sci., 118, 529-537. (doi:10.1242/jcs.01633)
1606.00620	1	1606	2016-06-02T11:03:04	What is the 'minimum inhibitory concentration' (MIC) of pexiganan acting on Escherichia coli? - A cautionary case study	[ "physics.bio-ph" ]	We measured the minimum inhibitory concentration (MIC) of the antimicrobial peptide pexiganan acting on Escherichia coli, and report an intrinsic variability in such measurements. These results led to a detailed study of the effect of pexiganan on the growth curve of E. coli, using a plate reader and manual plating (i.e. time-kill curves). The measured growth curves, together with single-cell observations and peptide depletion assays, suggested that addition of a sub-MIC concentration of pexiganan to a population of this bacterium killed a fraction of the cells, reducing peptide activity during the process, while leaving the remaining cells unaffected. This pharmacodynamic hypothesis suggests a considerable inoculum effect, which we quantified. Our results cast doubt on the use of the MIC as 'a measure of the concentration needed for peptide action' and show how 'coarse-grained' studies at the population level give vital information for the correct planning and interpretation of MIC measurements.	physics.bio-ph	physics	What is the 'minimum inhibitory concentration' (MIC) of pexiganan acting on Escherichia coli? – A cautionary case study Alys K Jepson, Jana Schwarz-Linek, Lloyd Ryan, Maxim G Ryadnov and Wilson C K Poon Abstract We measured the minimum inhibitory concentration (MIC) of the antimi- crobial peptide pexiganan acting on Escherichia coli, and report an intrinsic vari- ability in such measurements. These results led to a detailed study of the effect of pexiganan on the growth curve of E. coli, using a plate reader and manual plating (i.e. time-kill curves). The measured growth curves, together with single-cell obser- vations and peptide depletion assays, suggested that addition of a sub-MIC concen- tration of pexiganan to a population of this bacterium killed a fraction of the cells, reducing peptide activity during the process, while leaving the remaining cells unaf- fected. This pharmacodynamic hypothesis suggests a considerable inoculum effect, which we quantified. Our results cast doubt on the use of the MIC as 'a measure of the concentration needed for peptide action' and show how 'coarse-grained' studies at the population level give vital information for the correct planning and interpre- tation of MIC measurements. Keywords: antimicrobial peptide; pexiganan; Escherichia coli; minimum inhibitory concentration; killing curves; inoculum effect 1 Introduction The discovery of the β -lactam antibiotic penicillin by Fleming in 1928 was a mile- stone of 20th-century medicine. Today, the rampant spread of antimicrobial re- sistance (AMR) constitutes a grand challenge facing medical science in the new century (Aminov, 2010), which, if not met, will turn a 'strep throat' back into a life threatening illness. To confront AMR, more effective ways of using of exist- Alys Jepson, Jana Schwarz-Linek and Wilson Poon ([email protected]) SUPA and School of Physics & Astronomy, The University of Edinburgh, Peter Guthrie Tait Road, Edinburgh EH9 3FD, Scotland, United Kingdom. Lloyd Ryan and Max Ryadnov National Physical Laboratory, Hampton Road, Teddington TW11 0LW, United Kingdom. 1 2 Jepson et al. ing agents, including dosing regimes less prone to generating AMR, are urgently needed, as are new agents. In either case, standardised measures of effectiveness allowing science-based comparison between different agents are clearly required. In the global effort to confront AMR, antimicrobial peptides (AMPs) have at- tracted considerable attention (Fjell et al, 2012). These short peptides (≈ 10 to 50 residues) are widely distributed among metazoans, with diverse sequence and struc- ture. Certain motifs recur, e.g. high net positive charge under physiological condi- tions, and/or α-helical conformation in solution or upon binding to membranes; but these motifs are not universal. AMPs are effective against a wide spectrum of bac- teria, viruses and fungi in natura. The hope is that some natural or synthetic AMPs may be suitable therapeutic antimicrobial agents, especially against AMR strains. The 23-residue AMP magainin-2 secreted by the African clawed frog Xenopus laevis and its 'relatives' have attracted particular attention. Pexiganan (or MSI-78), a synthetic 22-residue magainin analogue (Gottler and Ramamoorthy, 2009), was trialed for topical treatment of diabetic foot ulcers, but was denied approval in 1999 because it seemed no more effective than antibiotics already in use for such ulcers (Moore, 2003); however, future clinical approval remains a possibility (Gottler and Ramamoorthy, 2009). Partly due to on-going efforts to secure such approval, pexi- ganan has been well studied. A large biophysical literature exists on AMPs in general, and pexiganan in par- ticular, focussing on the molecular modus operandi. Partly as a result of substantial research into how pexiganan and similar AMPs interact with lipid bilayers in unil- amellar vesicles (Gottler and Ramamoorthy, 2009), it is widely believed that this and other α-helical AMPs lyse bacteria by membrane poration. We consider the other end of the length scale spectrum, and report a study of the modus operandi of pexiganan on E. coli at the population level. Such 'coarse- grained' studies using rather classical methods (albeit in updated, high-throughput, forms) are seldom performed today. Our results show how such work is needed to complement molecular-level studies, preventing the misinterpretation of pharmaco- dynamic measurements performed to judge the concentration required for antimi- crobial action against live bacteria. We start by measuring the minimum inhibitory concentration (MIC), which is the most important 'one-number characterisation' of the effectiveness of an antimi- crobial agent against a target organism. Loosely, it is the minimum concentration of an antimicrobial agent necessary to cause stasis (no growth). We re-determine the MIC of pexiganan on E. coli, but using more repeat experiments than has ever been reported before. A critique of using MIC to characterise potency based on our mea- surements leads us to study the effect of pexiganan on the growth curve of E. coli, which turns out to be strikingly different from the way many classical antibiotics change the growth curve of the same bacterium. Our growth curves, along with single-cell observations, peptide depletion assays and time-kill curves, suggest that adding sub-MIC concentrations of this AMP to a population of E. coli rapidly kills a fraction of the cells, leaving the rest to grow un- affected while at the same time removing active AMP molecules from the medium. As Udekwu et al (2009) have previously suggested, the depletion or deactivation Pexiganan action on E. coli 3 of an antibiotic causes an 'inoculum effect' (dependence of MIC on initial inocu- lum concentration), which we quantify for pexiganan. We end by discussing the implications of our findings on the interpretation of the MIC in mechanistic and pharmacodynamic contexts. 2 Materials and methods 2.1 Bacteria culture We worked with E. coli K-12 derived strain MG1655 (Blattner et al, 1997). Five colonies, grown on an Mueller Hinton Broth (MHB) agar plate, were touched with a sterile loop and introduced to a 5 ml MHB liquid culture, which were grown at 37◦C and 200rpm to OD=0.5 (600nm). We also grew bacteria in the filtered supernatant of cells lysed by sonication. Ten 30 second pulses of sonication applied to 10ml of E. coli culture resting on ice achieved a 99.98% reduction in viability. The supernatant was filtered (0.22µm) to remove the surviving cells. 2.2 Pexiganan Pexiganan (GIGKFLKKAKKFGKAFVKILKK-NH2) was synthesised on a Liberty microwave peptide synthesizer (CEM Corporation) using standard solid phase Fmoc protocols on Rink amide-MBHA resins with HCTU/DIPEA as coupling reagents. Peptides were purified by semi-preparative RP-HPLC on a JASCO HPLC system (model PU-980, Tokyo, Japan) and confirmed by MALDI-ToF mass spectrometry (Bruker Daltonics Ltd, UK) with α-cyano-4-hydroxycinnamic acid as the matrix. at −20◦C and defrosted immediately before using and refreezing. We prepared stock solutions at 2mM in sterile, distilled water, which were stored 2.3 Growth curves and MIC We followed a published protocol for MIC determination using microdilution assays in microtiter plates (Wiegand et al, 2008) that is consistent with the guidelines of the Clinical and Laboratory Standards Institute and the European Committee on An- timicrobial Susceptibility testing. MIC assays were prepared in 96-well polystyrene microtiter plates (Greiner) with 200µl cylindrical wells. Initial inoculum sizes of n0 = 5×105 cell/ml were incubated in and their optical density read (at 600nm) by a FLUOstar Optima (BMG Labtech) plate reader with lid-covered plates that allowed air flow. We checked that our MIC results were the same using either polystyrene 4 Jepson et al. or polypropylene plates. Note that we report pexiganan concentrations in µM to facilitate comparison with cell concentrations in cells/ml, where 0.4µM = 1µg/ml (using the molecular weight of 2477gmol−1). Literature values in µg/ml have been converted to and quoted in µM. 2.4 Single-cell imaging To determine times to first division, we took time-lapsed phase-contrast images of cells using a Nikon TE300 Eclipse inverted microscope with a 100× PH3 oil- immersion objective and a CoolSNAP HG2 CCD camera (Photometrics). MHB agar pads were set into adhesive Gene Frames (Thermo Scientific). Pexiganan at 3µM was added to E. coli at OD=0.5 and the solution was left to incubate for 3 minutes before pipetting 1µl onto the agar pad. The E. coli were spread by tipping the microscope slide and within ∼ 4.5 minutes all liquid had, by eye, disappeared. We then mounted a glass coverslip to the gene frame in contact with the agar pad. The sample was immediately transferred to the microscope, pre-heated to 37◦C in a temperature-controlled box, for observation. 2.5 Time-kill curves We grew E. coli MG1655 following the same protocol as for MIC assays and di- luted to 5× 105 cell/ml. We worked with 3ml of suspension in two tubes (Greiner, polystyrene, 50ml). Pexiganan was added to one of these at time t = 0 min and both incubated at 37◦C and shaken at 200 rpm. A 100µl sample was removed from both tubes 1min after peptide addition, a range of ten-fold dilutions were spread onto MHB agar plates in triplicate. No more than 5 samples were removed, resulting in a 17% volume reduction, which is somewhat above the recommended maximum re- duction in standard protocols. The agar plates were incubated at 30◦C for 16h before the colonies were counted manually and density of cells calculated in cell-forming units (CFU) per ml. 2.6 Pharmacodynamic studies We followed literature procedures (Udekwu et al, 2009) to determine the effect of residual AMP. Two tubes containing 5ml of MHB and 40µM of pexiganan were prepared, one of which was inoculated with 5× 106 cells/ml. After 18 h incubation both suspensions were filtered and the supernatant used to set up two MIC assays in a 96 well plate each, alongside two replicate control MIC assays using the standard Pexiganan action on E. coli 5 Fig. 1 (a) Data from one dilution experiment in a 96-well plate. Numbers are pexiganan concen- trations in µM in the 200µl wells, each inoculated with n0 = 5× 105 cells/ml. MHB = buffer with no cells. Each column is one dilution series. Yellow = wells with an OD ≈ 10× the OD of MHB wells after 24h. (b) Fraction of wells that showed apparent bacterial growth for 20 replicates as a function of pexiganan concentration, with the zero pexiganan point plotted on the left end of the logarithmic horizontal axis. protocol. A further two assays were set up with supernatant from the tube which had contained both cells and peptide, with 20µM of freshly-added pexiganan. 3 Minimum inhibitory concentration Figure 1(a) shows representative raw data for our determination of the MIC of pex- iganan for E. coli MG1655 using microdilution assays in microtiter plates, with growth after 24h detected by a plate reader.1 In our 96-well plate, each of the 12 columns constitutes a separate MIC determination. Thus, e.g., in column 4 of the plate shown, the minimum concentrating showing no growth at 24h is 5µM, which therefore by definition is the MIC from this particular dilution series. In these 12 replicates, then, five return a MIC of 2.5µM (columns 1, 2, 8, 9 and 11), four re- turn a MIC of 5µM (columns 4, 6, 7 and 12), and three columns (3, 5 and 10) show a seemingly 'impossible' pattern of 're-entrant growth': after being inhibited at 2.5µM, growth apparently restarted at 5µM.2 What, then, is the MIC of pexiganan acting on E. coli MG1655? Our data, Fig- ure 1(a), do not allow us to assign a unique value, but if such a value exists, then it lies in the region of 2.5 to 5µM. Previous studies have returned values in the range of 3.2 to 12.8µM against various isolates (Fuchs et al, 1998; Ge et al, 1999), placing our range of 2.5µM to 5µM at the low end of the spectrum.3 1 See http://datashare.is.ed.ac.uk/handle/10283/1885 to access relevant data on which this article is based. 2 Continuing the experiment to 72h did not change the observed growth/no-growth pattern. 3 This is perhaps unsurprising given that MG1655 is a laboratory strain that has been described as 'deceitful delinquents growing old disgracefully' (Hobman et al, 2007). 123456789101112AMHBMHBMHBMHBMHBMHBMHBMHBMHBMHBMHBMHBB101010101010101010101010C555555555555D2.52.52.52.52.52.52.52.52.52.52.52.5E1.251.251.251.251.251.251.251.251.251.251.251.25F0.630.630.630.630.630.630.630.630.630.630.630.63G0.310.310.310.310.310.310.310.310.310.310.310.31H000000000000(a) 1.00.80.60.40.20.0Fraction of wells showing growth34560.12345612345610Concentration of Pexiganan (µM)0\ \(b)6 Jepson et al. However, to compare our results and previous work in terms of a range of values masks a fundamental qualitative difference between our work and almost all previ- ous MIC measurements. We, like the majority of the literature, follow best-practice guidelines (Wiegand et al, 2008; CLSI, 2012). The difference is that we perform multiple (12) replicates of the same MIC assay on a single plate. This is almost never done in the literature. Instead, commercial 'MIC plates' are used with wells pre-loaded to give single dilution series of multiple antimicrobial agents. Alterna- tively, the same dilution series of a single agent is used to test multiple organisms, again with one dilution series per organism.4 The range of literature values quoted above for pexiganan acting on E. coli therefore arises from single-dilution-series measurements on many isolates, whereas our own range of values arises from vari- ability between multiple replicates on a single strain, Figure 1(a). When a single dilution series is used, as is the case in the majority literature, the issue of disagreement between replicates does not arise. On the rare occasion where two replicates are prepared and they disagree, such as our columns 7 and 8 (Figure 1), the disagreement is typically ascribed to 'dilution error'. Literature protocols do occasionally mention 're-entrant growth', but would ascribe this to ac- cidental 'skip' or to 'single well contamination'. In particular, if we follow Hendrik- sen (2010), we should identify our wells D3, D5, D10 as 'skips', and in these cases take the 'true MIC' to be 10µM. 5 Repeated dilution error, skip or contamination at the same point of our multiple dilution series seem highly improbable. Moreover, repeated measurements showed similar patterns of variability seen in Figure 1(a), in- cluding 'reentrant growth'. Thus, the variability revealed by our results is intrinsic, and reporting a single MIC is misleading. A better way to summarise our findings is to plot the fraction of wells showing growth as a function of pexiganan concen- tration. Figure 1(b) shows such a plot for the 12 replicates shown in Figure 1(a) and another 8 replicates performed using the same peptide stock and inoculum. Intrinsic variability is consistent with a previous meta-study (Annis and Craig, 2005), which ascribed half of the variability uncovered in a survey of literature values of the MICs of various antibiotics against E. coli and Staphylococcus aureus to laboratory-to-laboratory differences. Presumably, then, it is possible that at least part of the other half of the variability is attributable to intrinsic causes, although Annis and Craig attributed this 'commonly shown 3-fold dilution range' entirely to environmental factors such as temperature, inoculum size and incubation time. To begin to elucidate the source of the MIC variability, we turn to consider the full growth curves that were collected and used to generate the MIC data. Pexiganan action on E. coli 7 Fig. 2 Sub-MIC growth curves for E. coli with an initial inoculum size of n0 = 5× 105 cells/ml at the same pexiganan concentrations used for the 96-well plate experiment in Figure 1 and stated in the legend. (a) Raw OD data against time. (b) The same data with OD at t = 0 subtracted. (c) Log- linear plot of the data in (b). (d) The time to reach OD = 0.12, τ0.12, plotted against concentration of pexiganan. In parts (b) and (c), OD = 10.12 is shown as dashed lines. 4 Sub-MIC growth curves Figure 2(a) shows a typical set of growth curves of E. coli MG1655 at sub-MIC con- centrations of pexiganan (background subtracted in Figure 2(b) and (c)), from the MIC assays displayed in Figure 1(a). While such curves are collected in every mi- crotiter assay, they are seldom, if ever, presented or interpreted. Exceptions concern the response of E. coli to tetracycline and amoxicillin (Schuurmans et al, 2009) and to cefotaxime (Baraban et al, 2011). At sub-MIC concentrations tetracycline and ce- fotaxime reduce the population growth rate and stationary level whilst amoxicillin does not significantly affect the growth curves until the MIC is reached. Pexiganan does not influence the maximum growth rate (measured to be α = 0.034 ± 0.01min−1) or the stationary level of the population (a background sub- tracted OD of ≈ 2.5), but lengthens the time until growth is detected6 in a con- centration dependent manner. Figure 2(c) plots the time taken to measure an OD 4 E.g., no replicats are suggested in the guidelines for loading a 96-well plate for testing a range of agents on E. coli and Salmonella isolates in a WHO project (Hendriksen, 2010). 5 See Figure 2 in Hendriksen (2010). If, alternatively, we identify our wells C3, C5 and C10, Figure 1, as contaminated, then the 'true MIC' would be taken to be 2.5µM. 6 We detected growth in a well at 2× 107 cells/ml, similar to what was reported previously for multi-well plate readers (Pin and Baranyi, 2006; M´etris et al, 2006). Lengthened detection times due to peptide action recalls the 'virtual colony count' (VCC) approach developed to measure 3.02.52.01.51.00.50.0OD 10 µM 5 µM 2.5 µM 1.25 µM 0.62µM 0.31 µM 0 µM(a)2.52.01.51.00.50.0OD-OD(0)12008004000Time (min)(b)-6-4-20ln (OD-OD(0))12008004000Time (min)(c)12001000800600400200 t0.12 (min)6543210Concentration of Pexiganan (µM)(d)8 Jepson et al. of 0.12 above the OD of MHB (τ0.12), which increases with pexiganan concentra- tion. Note that without pexiganan, the population takes 160±10 min to grow to OD = 0.12 = 1± 0.4× 108 cell/ml. An initial inoculum of 5× 105 cells/ml will reach 1× 108 cell/ml in α−1 ln(2× 102) = 156min with our measured α. The population lag time is therefore immeasurably small. The variation in τ0.12 between wells increases with pexiganan concentration. In particular, the large differences in τ0.12 between replicates at 2.5µM correlates well with the observation, Figure 1, that some wells do not show growth within 24 h at this concentration. No replicate experiments were reported in previous growth curve studies. In the case of cefotaxime (Baraban et al, 2011), the smooth variation of growth curves at closely-spaced antibiotic concentrations allows us to conclude that there should be little variation between replicates. We therefore infer an intriguing difference vis-`a-vis stocasticity between pexiganan and cefotaxime. We cannot draw a similar conclusion from data for tetracycline and amoxicillin (Schuurmans et al, 2009) because only a few concentrations were studied. 5 Single cell observations Next, we probe the cause of the observed effect of pexiganan concentration on τ0.12, the growth curve detection time. There are two generic ways in which the peptide could affect cells at early times to increase τ0.12. Either some of the initial inoculum dies, and the remaining cells take longer to grow to a given density even if their growth rate remains unaltered, or growth of all the cells is retarded, giving the same macroscopic observed effect on τ0.12. To observe directly the early-time effect of sub-MIC concentrations of pexiganan, we imaged cells spread on agar and recorded the time to first division (TTFD) of each cell, thought to be the sum of the lag time and first generation time (Rasch et al, 2007; M´etris et al, 2005). We chose conditions (caption, Figure 3) that resulted in sufficient cell death to demonstrate the effect of pexiganan, whilst leaving enough live cells to collect meaningful data. The TTFD distribution for the pexiganan-free control, Figure 3 (red), is spread over ∼ 50min due to heterogeneities in growth stages, single-cell lag times7 and generation times. On exposure to 3µM of pexiganan for 3 min before being placed on the agar, a fraction of the cells never go on to divide. Some are clearly dead (showing less contrast and do not grow) by the time the first image was taken, but others grow initially and then stop at later times. Some cells lyse suddenly whilst other fade over time. Interestingly, 29% of the cells which grow and divide give rise to a daughter cell that subsequently dies. Importantly, the TTFD distribution of cells that survive and divide to form colonies does not seem to differ greatly to that defensin activity (Ericksen et al, 2005). However, unlike in VCC, bacteria in our case were exposed to peptides in their growth medium rather than grown in a peptide-free medium after exposure. 7 Cells with the shortest single cell lag times dominate the population lag (Baranyi, 1998), which can be crudely calculated to be ∼ 10 min after the transfer procedure from liquid culture to agar. Pexiganan action on E. coli 9 Fig. 3 Normalised histogram of times to first division for E. coli on agar. Unexposed to pexiganan (red) and exposed to 3µM pexiganan for 3 min (black) Inset shows the same plot if the time to second division is counted for all first divisions resulting in the death of a daughter cell. of the control sample, Figure 3 (black and red respectively). Including cells whose daughters subsequently die shifts the distribution only slightly (inset). These data suggest that exposure to pexiganan results either in death or leaves the cells unaffected to go on to grow and divide, at least for cells inoculated onto agar. A number of caveats are, however, in order. First, the concentration of peptides the cells are exposed to is uncontrolled while the inoculum drop is evaporating after be- ing placed on the agar surface. Thereafter, the peptides are be free to diffuse into the agar away from the cells. It is therefore possible that under more prolonged expo- sure than is possible under our experimental conditions, e.g., in liquid medium, the growth of surviving cells would be retarded. Nevertheless, our experiments certainly show that the main effect of pexiganan is to kill a proportion of the cells rapidly. 6 Depletion of active peptide It is important to know whether this rapid initial killing of cells significantly affects the concentration of free peptides. Towards this end, we first determine if the peptide is depleted with time in our assays we performed the MIC experiments detailed in section 2.6. MIC assays using pre-incubated peptide showed that the activity of the peptide degrades by a factor of 2 to 4 times over 24 h in MHB at 37◦C, perhaps due to peptide aggregation, common for magainins, or adhesion to components of the medium. However, this is a small contribution to what was found when bacteria had also been in the solution. An assay containing 40µM of pexiganan and 5× 106 cell/ml, after being incubated and then filtered to remove any survivors, no longer had any measurable affect on the growth of a new inoculum, so that (from Section 3) we know that the pexiganan concentration must be below 2.5µM. In other words, the activity of the peptide was reduced more than 16-fold. When an additional 20µM of peptide was added to this solution and serially diluted with MHB it resulted in a MIC 0.250.200.150.100.050.00Normalised count14012010080604020Time to first division (TTFD)0.250.200.150.100.050.0080604020 0µM 2% dead. Total count = 234 3µM 46% dead. Total count = 21010 Jepson et al. Fig. 4 Platecounted time-kill curves in MHB for pexiganan concentrations shown in the legend for the inoculum size n0 = 5× 105 cells/ml. 4× greater that that of a control. Not only had the original peptide been depleted but the additional peptide showed less activity in this solution than in MHB. Similar experiments conducted without filtering gave the same results, so that the filtering was not responsible for removing active peptide molecules or complexes. Following these findings, we used a suspension containing the filtered remains of sonicated cells grown in MHB to prepare a MIC assay instead of MHB alone. Even at the highest concentration of 20µM pexiganan, there was no measurable effect on the inoculum of 5× 105 cell/ml, suggesting a reduction in the antimicrobial action of the peptide of by 10 to 20 fold. It appears that the presence of lysed bacteria reduces peptide activity. A plausible hypothesis is that positively-charged peptides are depleted by adhesion to negatively-charged DNA and proteins, and/or by the action of proteolytic enzymes from lysed cells. 7 Sub-MIC time-kill curves If we are right that pexiganan kills a fraction of cells at short times upon addition, our growth curves ought to show an initial decrease in numbers. This is not borne out by our plate reader data, Figure 2. We now demonstrate that this is simply because of lack of detection sensitivity. By using plate counts to measure the number of viable E. coli as a function of time after the addition of pexiganan to their medium, i.e. by determining time-kill curves (Section 2.5), it is possible to detect much lower cell densities than is possible using a plate reader. We added pexiganan at t = 0min to concentrations of 1, 2.5, 5 or 10µM and collected plate-count data for the first 200 min, and then once the following day. Figure 4(a) shows time kill curves for an inoculum size of 5× 105 cell/ml, the same as that used for MIC assays in the plate reader. As can be expected from the results shown in Section 3, no viable cells were recorded at any time point at 10µM. At lower concentrations the density of viable cells in the suspension drops 102 104 106 108 nCFU (CFU/ml)250200150100500Time (min)1062.9 0µM 1µM 2.5µM 5µM\ \Pexiganan action on E. coli 11 at early times; the higher the peptide concentration, the larger the drop in initial viability. This is as expected from the only previous publication of pexiganan time- kill curves (Ge et al, 1999). These measurements however were performed at above- MIC concentrations, and only monitored for 120 minutes. We measured at lower concentrations and for longer and therefore observe the viability of E. coli increasing again at later times. The higher the peptide concen- tration, the longer before recovery starts; however, after one day, all samples have recovered to the same level. Regrowth after an initial drop in viability has been reported in the time-kill curves of other AMPs (Matsuzaki et al, 1998; McGrath et al, 2013; Spindler et al, 2011). It has been attributed to 'resistance', with little evidence given in support. To verify that no resistance had developed in our case, we exposed E. coli that had regrown in our experiments to 2.5 and 5µM pexiganan, and found them as susceptible as cells from the parent population. Our results confirm that the lack of an initial phase of negative growth in our plate reader data, Figure 2, is simply due to lack of detection sensitivity, and strengthens the case for our suggestion that pexiganan kills a fraction of the cells at early times, leaving the rest to grow unaffected. The literature has only recently suggested that such non-monotonic time-kill curves can be due to a dynamic state of balanced divi- sion and death rather than the historically assumed sub-population of non-dividing persister cells (Wakamoto et al, 2013). The killing of a fraction of cells in a clonal population leaving the rest of the cells unaffected is striking. Given the large number of peptides per bacterium in the system (∼ 109, much more than is needed to cover each cell in a monolayer however the peptide is oriented) (Wimley, 2010), the observed heterogeneity is unlikely to be due to fluctuations in the number of peptides strongly interacting with each cell. In- stead, the heterogeneity most likely arises from phenotypic variations in a property or properties of single cells. Our evidence shows that the survivors are not a sub- population of non-dividing cells. Perhaps, then, cell surface heterogeneities, e.g. in the structure of the lipopolysaccharides (Lerouge and Vanderleyden, 2002) or in the expression of fimbrae (Abraham et al, 1985), are responsible for our observations. Note that at peptide concentrations approaching the MIC, e.g., (cid:38) 2.5µM for the inoculum size relevant to Figure 4, the number of viable cells drops to rather low levels (the minima at 2.5µM and 5µM in Figure 4 correspond to 2000 and 20 cells per well respectively) before net growth begins. Such low numbers lead to large fluc- tuations; in particular, some replicates could easily have all cells eradicated at these peptide concentrations. This is the source of the observed fluctuations in Figure 1, which make it difficult to determine 'the MIC' based on a single measurement. 8 The inoculum effect If pexiganan kills a certain fraction of cells outright, depletes the number of active peptides, and leaves the remaining cells to grow, then there should be a strong in- 12 Jepson et al. oculum effect (IE): the MIC should be higher for higher inoculum densities. The IE is important because the in vitro population density of pathogens could be sig- nificantly higher than typical inocula cell densities in MIC assays. The magnitude of the IE for a small number of inocula concentrations is reported in the antimicro- bial literature (Udekwu et al, 2009), including for some AMPs (Levison et al, 1993; Jones et al, 1994); but possible causative mechanisms have seldom been discussed, partly because data for MICs over a large range of inocula densities are, with very few exceptions (Udekwu et al, 2009), not reported. We report such data for pexiganan acting on E. coli. Figure 5(a) shows the frac- tion of growing wells against pexiganan concentration for four different inoculum sizes, n0 = 2.5× 107, 5× 105 (our standard inoculum size), 5× 103 and 5 cell/ml.8 Qualitatively, these data are similar to those shown in Figure 2. Quantitatively, the pexiganan concentration at which the fraction of growing wells drops to zero, which we take in this context to be the MIC, increases with inoculum size, n0. Figure 5(b) plots this MIC as a function of n0 from two sets of results: one which collates data from multiple assays done over a long period of time, and a second set where the data were collected in one experiment using the same peptide stock for each inocu- lum size. In both cases, the MIC approaches a constant value at low n0, and increases sub-linearly at higher n0 (the dashed line in Figure 5(b) has unit slope). Perhaps the simplest way to account for an inoculum effect, given our observa- tion of depletion of active peptides by cell lysis, is that each lysed cell inactivates a fixed number of antimicrobial molecules. However, in its simplest form, this mech- anism should lead to a linear dependence of the MIC on n0 at high enough n0. It is possible that our observed sub-linear n0 dependence is en route to such a linear dependence, but we do not reach high enough inoculum concentrations to observe it. On the other hand, a speculative explanation for a sub-linear dependence at all n0 is that cells aggregate at high densities and high peptide concentrations,9 which may (Moiset et al, 2013) allow peptide molecules or complexes to affect two cells simultaneously. 9 Discussion: What really is the MIC? The overwhelming majority of existing MIC measurements rely on a single series of dilution assays performed at one inoculum size (typically 5× 105 cells/ml) to deter- mine a one-off concentration of antimicrobial agent at which no growth is observed after 24 hours. This 'one-off MIC' is frequently assumed to represent, in the words of Melo et al (2009), 'the macroscopically observable threshold for the onset of [antimicrobial] activity'. This implies that sub-MIC concentrations of antimicrobial agents are sub-lethal towards the target cells. In the light of our findings, both the 8 Note that 5 cells/ml ≡ 1 cell/well, so that the fraction of growing wells is < 1 at zero pexiganan. 9 We have seen such aggregation in optical microscopy (data not show). Pexiganan action on E. coli 13 Fig. 5 (a) Fraction of growing wells as a function of pexiganan concentration, for inocula of n0 = 2.5 × 107 (blue), n0 = 5 × 105 (black), n0 = 5 × 103 (green) and n0 = 5 cells/ml (≡ 1cell/well) (red). (b) The MIC, defined as the concentration of pexiganan at which no replicate grows, plotted agains inoculum size, n0, for one set of experiments using the same pexiganan stock (grey) and data collated from many tests (black). The dashed line has unit slope. practice of determining a 'one-off MIC' value and the quoted interpretation of the significance of this value are problematic, at least for pexiganan acting on E. coli. First, we have found, using multiple replicates, that the 'one-off MIC' shows large intrinsic variability, Figure 1(a). This is because at these peptide concentra- tions, cell numbers drop very low before net regrowth starts, Figure 4, and such low numbers give rise to significant stocasticity. Instead of a 'one-off MIC' from a single dilution series, findings from multiple replicates, reported as the fraction of samples in which no growth is observed after 24 hours as a function of the peptide concentration, Figure 1(b), are necessary to give adequate information. Secondly, we have shown that exposure to sub-MIC levels of pexiganan kills a fraction of E. coli in a population quickly, leaving the remaining cells to grow at a rate that is indistinguishable from that of cells never exposed to the AMP, Figure 3. Indeed, the MIC for a standard inoculum size (5 × 105 cell/ml) is ∼ 10× above its single-cell action threshold: compare the black and red data sets in Figure 5(a). Thus, it is not true that the MIC of pexiganan is 'the macroscopically observable threshold for the onset of [antimicrobial] activity'. Instead, if a single value is de- sired to quantify the 'single-cell action threshold' of pexiganan on E. coli MG1655, a reasonable candidate is the concentration for the onset of lengthening detection times, τ0.12: compare Figure 2(d) with the red curve in Figure 5(a). The interpretation that the MIC itself is a 'single-cell action threshold' clearly lies behind a substantial body of literature. Thus, e.g., Ramamoorthy et al (2006) found a MIC of 4µM for pexiganan acting on E. coli using an inoculum of 107 cell/ml, but observed outer membrane perturbation (according to an ANS uptake assay) at the significantly lower concentrations of 0.67µM after only 5 minutes of incubation. Ramamoorthy et al (2006) infer that factors other than membrane disruption must be involved in the bacterial killing process. A later study came to a similar conclusion on similar grounds (Pius et al, 2012). Our findings mean that this inference is neither necessary nor likely correct: a fraction of cells should be killed (presumably by mechanisms involving membrane perturbation) at 0.67µM (cid:46) 0.2×MIC. In general, 1.00.80.60.40.20.0Fraction of growing wells0.1110Concentration of Pexiganan (µM)\ \0(a)110100 MIC (all replicates) (µM)101102103104105106107108Inoculum size n0 (cell/ml)(b)14 Jepson et al. then, understanding what MIC values mean is crucial if pharmacodynamic results are to be correctly compared with mechanistic studies. Thirdly, there is strong evidence that peptides causing cell death are sequestered and unavailable for further bactericidal action. This finding should influence the way in which comparative data between different AMPs are interpreted. Thus, e.g., Fluorogainin-1 is designed to be a 'more stable' analogue of pexiganan. It displays a lower MIC against S. aureus (Gottler et al, 2008). Our findings suggest that this could be because Fluorogainin-1 is not depleted as rapidly. Separately, one could postulate that mutant strains more resistant to pexiganan (Perron et al, 2006) could have evolved more efficient mechanisms for peptide inactivation in lysed cells, thus raising the measured MIC. These scenarios are speculative, but illustrate once more the need to consider carefully the meaning of MIC measurements. coli, Figure 5(b). There is no IE for n0 (cid:46) 103 cells/ml; thereafter, MIC ∝ nα α ≈ 0.4. This finding has significant practical and theoretical implications. Finally, we have quantified the inoculum effect (IE) of pexiganan acting on E. 0 with Practically, while the IE is well known, it is seldom quantified. In lieu of quan- tification, it is presumably natural to assume that the MIC is essentially proportional to n0, especially if the antimicrobial agent is known to be depleted by the process of killing bacteria. Making this assumption in the case of pexiganan acting on E. coli MG1655 would lead to very substantial errors, with knock-on implication for dosage regimes arising from pharmacokinetic studies based on such an assumption. Theoretically, we note that our power law with α ≈ 0.4 also fits the data for 5 out of the 6 antibiotics for which the IE has been quantified (Udekwu et al, 2009) (we find α between 0.3 and 0.5 fitting their data to power laws). Intriguingly, Udekwu et al (2009) found no depletion of the antibiotic for 3 of these 5 antibiotics. It is a challenge for future theoretical modelling to decide whether a deeper generic mech- anism underly such similarity, or whether two or more mechanisms fortuitously lead to similar quantitative IEs. Overall, we suggest that data like those presented in Figure 5(a) should form the minimum basis for interpreting MIC measurements in the context of mechanistic studies, pharmacodynamic modelling, and clinical decision making for all antimi- crobial agents. More specifically, given that many bactericidal AMPs are structurally similar to pexiganan, we surmise that our findings concerning the modus operandi of pexiganan, e.g., that its action is likely crucially linked to phenotypic heterogeneity of the target organism, may generalise to other compounds. For example, prelim- inary data on the MIC of amhelin, a pore-forming peptide designed by Rakowska et al (2013), shows the same growth curve (Figure 2) and IE (Figure 5(b)) as pexi- ganan. Future research should explore the validity of this surmise for a wider range of AMPs, especially those that, like pexiganan, are bactericidal. Acknowledgements AKJ was funded by an EPSRC studentship. JSL was funded by the National Phys- ical Laboratory and SUPA. LR and MGR were funded by the UK Department of Pexiganan action on E. coli 15 Business, Innovation and Skills. WCKP was funded by EPSRC Programme Grant EP/J007404/1. We thank Simon Titmuss for illuminating discussions, Angela Daw- son for assistance in biological lab work and Vincent Martinez for assistance with data analysis. References Abraham JM, Freitag CS, Clements JR, Eisenstein BI (1985) An invertible element of DNA controls phase variation of type 1 fimbriae of Escherichia coli. Proc Natl Acad Sci USA 82(17):5724–7 Aminov RI (2010) A brief history of the antibiotic era: lessons learned and chal- lenges for the future. Front Microbiol 1:art. 134 Annis DH, Craig BA (2005) The effect of interlaboratory variability on antimicro- bial susceptibility determination. Diagn Microbiol Infect Dis 53:61–64 Baraban L, Bertholle F, Salverda MLM, Bremond N, Panizza P, Baudry J, de Visser JAGM, Bibette J (2011) Millifluidic droplet analyser for microbiology. Lab Chip 11(23):4057–62 Baranyi J (1998) Comparison of Stochastic and Deterministic Concepts of Bacterial Lag. J Theor Biol 192(3):403–408 Blattner F, Plunkett G, Bloch C, Perna N, Burland V, Riley M, ColladoVides J, Glasner J, Rode C, Mayhew G, Gregor J, Davis N, Kirkpatrick H, Goeden M, Rose D, Mau B, Shao Y (1997) The complete genome sequence of Escherichia coli K-12. Science 277:1453–1462 CLSI (2012) Methods for Dilution Antimicrobial Susceptibility Tests for Bacteria That Grow Aerobically; Approved Standard, 9th edition. CLSI document M07- A9. Clinical and Laboratory Standards Institute, Wayne, PA, Ericksen B, Wu Z, Lu W, Lehrer RI (2005) Antibacterial Activity and Specificity of the Six Human α-Defensins. Antimicrob Agents Chemother 49:269–275 Fjell CD, J AH, Hancock REW, Schneider G (2012) Designing antimicrobial pep- tides: form follows function. Nature Rev Drug Discovery 11:37–51 Fuchs PC, Barry AL, Brown SD (1998) In vitro antimicrobial activity of MSI-78, a magainin analog. Antimicrob Agents Chemother 42:1213–1216 Ge Y, MacDonald DL, Holroyd KJ, Thornsberry C, Wexler H, Zasloff M (1999) In Vitro Antibacterial Properties of Pexiganan, an Analog of Magainin. Antimicrob Agents Chemother 43:782–788 Gottler LM, Ramamoorthy A (2009) Structure, membrane orientation, mechanism, and function of pexiganan – a highly potent antimicrobial peptide designed from magainin. Biochim Biophys Acta 1788:1680–6 Gottler LM, Lee HY, Shelburne CE, Ramamoorthy A, Marsh ENG (2008) Using fluorous amino acids to modulate the biological activity of an antimicrobial pep- tide. ChemBioChem 9(3):370–3 16 Jepson et al. Hendriksen RS (ed) (2010) Laboratory Protocols: Level 1 Training Course – MIC determination by broth dilution using Sensititre. No publisher given, available online at http://goo.gl/57Pu7F, last accessed June 3, 2016. Hobman JL, Penn CW, Pallen MJ (2007) Laboratory strains of Escherichia coli: model citizens or deceitful delinquents growing old disgracefully? Mol Microbiol 64:881–885 Jones E, Smart A, Bloomberg G, Burgess L, Millar M (1994) Lactoferricin,a new antimicrobial peptide. J Appl Bact 77:208–214 Lerouge I, Vanderleyden J (2002) O-antigen structural variation: mechanisms and possible roles in animal/plantamicrobe interactions. FEMS Microbiol Rev 26(1):17–47 Levison ME, Pitsakis PG, May PL, Johnson CC (1993) The bactericidal activity of magainins against Pseudomonas aeruginosa and Enterococcus faecium. J An- timicrob Chemother 32:577–585 Matsuzaki K, Mitani Y, Akada KY, Murase O, Yoneyama S, Zasloff M, Miyajima K (1998) Mechanism of synergism between antimicrobial peptides magainin 2 and PGLa. Biochem 37(43):15,144–53 McGrath DM, Barbu EM, Driessen WHP, Lasco TM, Tarrand JJ, Okhuysen PC, Kontoyiannis DP, Sidman RL, Pasqualini R, Arap W (2013) Mechanism of ac- tion and initial evaluation of a membrane active all-D-enantiomer antimicrobial peptidomimetic. Proc Natl Acad Sci (USA) 110(9):3477–82 Melo MN, Ferre R, Castanho MARB (2009) Antimicrobial peptides: linking par- tition, activity and high membrane-bound concentrations. Nature Rev Microbiol 7(3):245–50 M´etris A, Le Marc Y, Elfwing A, Ballagi A, Baranyi J (2005) Modelling the vari- ability of lag times and the first generation times of single cells of E. coli. Int J Food Microbiol 100(1):13–19 M´etris A, George SM, Baranyi J (2006) Use of optical density detection times to assess the effect of acetic acid on single-cell kinetics. Appl Environ Microbiol 72(10):6674–9 Moiset G, Cirac AD, Stuart MCA, Marrink SJ, Sengupta D, Poolman B (2013) Dual action of BPC194: a membrane active peptide killing bacterial cells. PLoS one 8(4):e61541 Moore A (2003) The big and small of drug discovery. EMBO Rep 4:114–117 Perron GG, Zasloff M, Bell G (2006) Experimental evolution of resistance to an antimicrobial peptide. Proc Royal Soc B Biol Sci 273(1583):251–6 Pin C, Baranyi J (2006) Kinetics of single cells: observation and modeling of a stochastic process. Appl Environ Microbiol 72:2163–9 Pius J, Morrow MR, Booth V (2012) H solid-state nuclear magnetic resonance in- vestigation of whole Escherichia coli interacting with antimicrobial peptide MSI- 78. Biochem 51(1):118–25 Rakowska PD, Jiang H, Ray S, Pyne A, Lamarre B, Carr M, Judge PJ, Ravi J, Ger- ling UIM, Koksch B, Martyna GJ, Hoogenboom BW, Watts A, Crain J, Grovenor CRM, Ryadnov MG (2013) Nanoscale imaging reveals laterally expanding an- timicrobial pores in lipid bilayers. Proc Natl Acad Sci (USA) 110(22):8918–23 Pexiganan action on E. coli 17 Ramamoorthy A, Thennarasu S, Lee DK, Tan A, Maloy L (2006) Solid-state NMR investigation of the membrane-disrupting mechanism of antimicrobial pep- tides MSI-78 and MSI-594 derived from magainin 2 and melittin. Biophys J 91(1):206–16 Rasch M, M´etris A, Baranyi J, Bjørn Budde B (2007) The effect of reuterin on the lag time of single cells of Listeria innocua grown on a solid agar surface at different pH and NaCl concentrations. Int J Food Microbiol 113(1):35–40 Schuurmans JM, Nuri Hayali AS, Koenders BB, ter Kuile BH (2009) Variations in MIC value caused by differences in experimental protocol. J Microbiol Meth 79:44–7 Spindler EC, Hale JDF, Giddings TH, Hancock REW, Gill RT (2011) Decipher- ing the mode of action of the synthetic antimicrobial peptide Bac8c. Antimicrob Agents Chemother 55(4):1706–16 Udekwu KI, Parrish N, Ankomah P, Baquero F, Levin BR (2009) Functional rela- tionship between bacterial cell density and the efficacy of antibiotics. J Antimi- crob Chemother 63:745–57 Wakamoto Y, Dhar N, Chait R, Schneider K, Signorino-Gelo F, Leibler S, McKin- ney JD (2013) Dynamic persistence of antibiotic-stressed mycobacteria. Science (New York, NY) 339(6115):91–5 Wiegand I, Hilpert K, Hancock REW (2008) Agar and broth dilution methods to de- termine the minimal inhibitory concentration (MIC) of antimicrobial substances. Nature Protocols 3:163–75 Wimley WC (2010) Describing the mechanism of antimicrobial peptide action with the interfacial activity model. ACS Chem Biol 5(10):905–17
1407.2552	2	1407	2015-03-05T18:57:55	Weak ergodicity breaking of receptor motion in living cells stemming from random diffusivity	[ "physics.bio-ph", "cond-mat.dis-nn", "cond-mat.stat-mech" ]	Molecular transport in living systems regulates numerous processes underlying biological function. Although many cellular components exhibit anomalous diffusion, only recently has the subdiffusive motion been associated with nonergodic behavior. These findings have stimulated new questions for their implications in statistical mechanics and cell biology. Is nonergodicity a common strategy shared by living systems? Which physical mechanisms generate it? What are its implications for biological function? Here, we use single particle tracking to demonstrate that the motion of DC-SIGN, a receptor with unique pathogen recognition capabilities, reveals nonergodic subdiffusion on living cell membranes. In contrast to previous studies, this behavior is incompatible with transient immobilization and therefore it can not be interpreted according to continuous time random walk theory. We show that the receptor undergoes changes of diffusivity, consistent with the current view of the cell membrane as a highly dynamic and diverse environment. Simulations based on a model of ordinary random walk in complex media quantitatively reproduce all our observations, pointing toward diffusion heterogeneity as the cause of DC-SIGN behavior. By studying different receptor mutants, we further correlate receptor motion to its molecular structure, thus establishing a strong link between nonergodicity and biological function. These results underscore the role of disorder in cell membranes and its connection with function regulation. Due to its generality, our approach offers a framework to interpret anomalous transport in other complex media where dynamic heterogeneity might play a major role, such as those found, e.g., in soft condensed matter, geology and ecology.	physics.bio-ph	physics	Weak ergodicity breaking of receptor motion in living cells stemming from random diffusivity Carlo Manzoa,1, † Juan A. Torreno-Pina,1, † Pietro Massignan,1 Gerald J. Lapeyre Jr.,1 Maciej Lewenstein,1, 2 and Maria F. Garcia Parajoc 1, 2 1ICFO-Institut de Ci`encies Fot`oniques, Mediterranean Technology Park, 08860 Castelldefels (Barcelona), Spain 2ICREA-Instituci´o Catalana de Recerca i Estudis Avan¸cats, 08010 Barcelona, Spain (Dated: September 12, 2018) 5 1 0 2 r a M 5 ] h p - o i b . s c i s y h p [ 2 v 2 5 5 2 . 7 0 4 1 : v i X r a [email protected] [email protected] 1 Abstract Molecular transport in living systems regulates numerous processes underlying biological func- tion. Although many cellular components exhibit anomalous diffusion, only recently has the sub- diffusive motion been associated with nonergodic behavior. These findings have stimulated new questions for their implications in statistical mechanics and cell biology. Is nonergodicity a com- mon strategy shared by living systems? Which physical mechanisms generate it? What are its implications for biological function? Here, we use single particle tracking to demonstrate that the motion of DC-SIGN, a receptor with unique pathogen recognition capabilities, reveals nonergodic subdiffusion on living cell membranes. In contrast to previous studies, this behavior is incompatible with transient immobilization and therefore it can not be interpreted according to continuous time random walk theory. We show that the receptor undergoes changes of diffusivity, consistent with the current view of the cell membrane as a highly dynamic and diverse environment. Simulations based on a model of ordinary random walk in complex media quantitatively reproduce all our ob- servations, pointing toward diffusion heterogeneity as the cause of DC-SIGN behavior. By studying different receptor mutants, we further correlate receptor motion to its molecular structure, thus establishing a strong link between nonergodicity and biological function. These results underscore the role of disorder in cell membranes and its connection with function regulation. Due to its gen- erality, our approach offers a framework to interpret anomalous transport in other complex media where dynamic heterogeneity might play a major role, such as those found, e.g., in soft condensed matter, geology and ecology. † These two authors contributed equally to this work 2 I. INTRODUCTION Cell function relies heavily on the occurrence of biochemical interactions between spe- cific molecules. Encounters between interacting species are mediated by molecular transport within the cellular environment. A fundamental mode of transport for molecules in living cells is represented by diffusion, a motion characterized by random displacements. The quantitative study of diffusion is thus essential for understanding molecular mechanisms underlying cellular function, including target search [1], kinetics of transport-limited reac- tions [2, 3], trafficking and signaling [4]. These processes take place in complex environments, crowded and compartmentalized by macromolecules and biopolymers. A prototypical ex- ample is the plasma membrane, where the interplay of lipids and proteins with cytosolic (e.g., the actin cytoskeleton) and extracellular (e.g., glycans) components generates a highly dynamic and heterogeneous organization [5]. The diffusion of a single molecule j, whose position xj is sampled at N discrete times ti = i∆t, is often characterized by the time-averaged mean-square displacement (T-MSD): (cid:0)xj (ti + m∆t) − xj (ti)(cid:1)2, (1) N−m(cid:88) i=1 T-MSD(tlag = m∆t) = 1 N − m which for a Brownian particle scales linearly in the time-lag tlag. Application of fluorescence- based techniques to living cells has evidenced striking deviations from Brownian behavior in the nucleus [6], cytoplasm [7 -- 10] and plasma membrane [11, 12]. Indeed, numerous cellular components show anomalous subdiffusion [13], characterized by a power law dependence of the MSD ∼ tβ, with β < 1 [14 -- 16]. Owing to the implications of molecular transport for cellular function and the widespread evidence of subdiffusion in biology, major theoretical efforts have been devoted to understand its physical origin. Subdiffusion is generally un- derstood to be the consequence of molecular crowding [17] and several models have been developed to capture its main features. In general, subdiffusion can be obtained by mod- els of energetic and/or geometric disorder, such as: (i) the continuous-time random walk (CTRW), i.e., a walk with waiting times between steps drawn from a power law distribu- tion [18]; (ii) fractional Brownian motion, i.e., a process with correlated increments [19]; (iii) obstructed diffusion, i.e., a walk on a percolation cluster or a fractal [15]; (iv) diffusion in a spatially and/or temporally heterogeneous medium [20 -- 22]. Some of these models have been associated with relevant biophysical mechanisms such as trapping [23], the viscoelastic 3 properties of the environment [24] or the presence of barriers and obstacles to diffusion [25]. Advances in single particle tracking (SPT) techniques have allowed the recording of long single-molecule trajectories and have revealed very complex diffusion patterns in living cell systems [5, 11]. Recently, it has been shown that some cellular components show subdiffusion associated with weak ergodicity breaking (wEB) [9, 10, 12], with the most obvious signatures being the non-equivalence of the T-MSD and the ensemble-averaged MSD (E-MSD). The experimentally determined ensemble-averaged MSD over a time interval m∆t is defined by: (cid:0)xj (ti + m∆t) − xj (ti)(cid:1)2, J(cid:88) j=1 E-MSD(tlag = m∆t) = 1 J (2) where J is the number of observed single-particle trajectories and ti is the starting time relative to first point in the trajectory. Moreover, ergodicity breaking has been further confirmed by the presence of aging [26, 27], i.e. the dependence of statistical quantities on the observation time. Based on these findings, several stochastic models presenting nonstationary (and thus nonergodic) subdiffusion have been proposed [20, 28 -- 31]. Among these, CTRW has been used to model nonergodic subdiffusion in living cells [9, 10, 12] and has begun to provide theoretical insight into the physical origin of wEB in biological systems [28], associating the nonergodic behavior with the occurrence of particle immobilization with a heavy-tailed distribution of trapping times. At the same time, these intriguing findings have generated new questions: Is nonergodic subdiffusion a strategy shared by other biological systems? Can biophysical mechanisms other than trapping lead to similar behaviors? What is its functional relevance? Elucidating these issues is crucial to unravel the role of nonergodic subdiffusion in cellular function. The main aim of the present work is to explore other forms of transport in biological systems to provide answers to these questions. Here we used SPT to study the diffusion of a prototypical transmembrane protein, the pathogen-recognition receptor DC-SIGN [32] on living cell membranes. Our experiments and data analysis show that DC-SIGN dynamics display clear signatures of wEB and aging. However, in contrast to recent studies reporting nonergodic behavior of other membrane proteins [12], we find that DC-SIGN very rarely shows trapping events so that the observed wEB cannot be described by the CTRW model. Instead, our analysis shows that DC-SIGN displays a heterogeneous dynamics presenting frequent changes of diffusivity. Our numerical simulations, based on a novel theoretical model of Brownian diffusion in complex media [21], 4 quantitatively reproduce DC-SIGN dynamics demonstrating that nonergodic subdiffusion is a consequence of temporal and/or spatial heterogeneity. Furthermore, structurally mutated variants of DC-SIGN, with impaired function, show very different dynamical features. These results allow us to link receptor transport to molecular structure and receptor function, such as the capability to capture and uptake pathogens. II. WEAK ERGODICITY BREAKING AND AGING IN DC-SIGN DYNAMICS In this work, we performed SPT experiments [5] to follow the lateral diffusion of the pathogen-recognition receptor DC-SIGN [32] on living cell membranes. DC-SIGN is a pro- tein exclusively expressed on the surface of cells of the immune system, such as dendritic cells. The receptor is involved in the binding and uptake of a broad range of pathogens such as HIV-1, Ebola virus, hepatitis C virus, Candida albicans and Mycobacterium tuberculo- sis [33]. Previous studies have shown that DC-SIGN expressed on the membrane of Chinese Hamster Ovary (CHO) cells reproduces the essential features of the receptors naturally oc- curring on dendritic cells [34, 35], thus serving as a valid model system. To characterize its dynamics, we performed video microscopy of quantum-dot labeled DC-SIGN stably trans- fected in CHO cells in epi-illumination configuration (Fig. 1A-B, see Appendix A for details on cell culture and labeling procedures). In order to follow the standard biology nomencla- ture and to differentiate it from its mutated forms, in this manuscript we refer to the full receptor as the wild-type DC-SIGN (wtDC-SIGN). We tracked quantum dot positions with nanometer accuracy by means of an automated algorithm [36]. We acquired more than 600 trajectories, all longer than 200 frames with some as long as 2000 frames, at a camera rate of 60 frame · s−1 to allow the evaluation and the comparison of time and ensemble averaged MSD. The T-MSD of individual trajectories displayed a linear behavior (β ∼ 1), consistent with pure Brownian diffusion (Fig. 1C). The fitting of the average T-MSD provided a value β = 0.95± 0.05. In addition, the distribution of the exponents β obtained by nonlinear fitting of the T-MSDs of the individual trajectories (inset of Fig. 1D) showed an average (cid:104)β(cid:105) = 0.98 ± 0.06. Since the T-MSD values corresponding to different trajectories were broadly scattered, for each trajectory we calculated the diffusion coefficient Ds by a linear fit of the T-MSD 5 at time lags < 10% of the trajectory duration [37]. As expected, the resulting values of Ds were found to have a very broad distribution, spanning more than two orders of magnitude (Fig. 1D). However, in marked contrast with the T-MSD, the E-MSD deviated significantly from linearity, showing subdiffusion with an exponent β = 0.84 ± 0.03 (Fig. 1E). The difference between the scalings of T-MSD and E-MSD is a clear signature of wEB [38]. To inquire whether DC-SIGN dynamics also exhibits aging, we computed the time-ensemble-averaged MSD (TE-MSD) by truncating the data at different observation times T : TE-MSD(tlag, T ) = 1 J 1 T ∆t − m ∆t−m(cid:88) T i=1 j=1 J(cid:88) (cid:0)xj (ti + m∆t) − xj (ti)(cid:1)2, (3) and extracting the corresponding diffusion coefficient DTE by linear fitting [37]. In systems with uncorrelated increments, it can be shown under rather general assumptions that DTE ∼ T β−1 [21, 39]. The observed DTE indeed scaled as a power law with an exponent of −0.17± 0.05 (Fig. 1F), yielding a value of β in good agreement with the exponent determined from E-MSD. These results thus demonstrate that wtDC-SIGN dynamics exhibits aging. III. FAILURE OF THE CTRW MODEL The motion of some biological components, including the Kv2.1 potassium channel in the plasma membrane [12], lipid granules in yeast cells [9] and insulin-containing vesicles in Pancreatic β-cells [10], has been reported to exhibit subdiffusion compatible with the coexistence of an ergodic and a nonergodic process. The nonergodic part of the process has been modeled within the framework of the CTRW [28, 38, 39]. CTRW is a random walk in which a particle performs jumps whose lengths have a finite variance, but between jumps the walker remains trapped for random dwell times, distributed with a power-law probability density ∼ t−(1+β), which for β ≤ 1 has an infinite mean. The duration of trapping events is independent of the previous history of the system. The energy landscape of this process is characterized by potential wells with a broad depth distribution. Such energetic disorder yields nonergodicity, since no matter how long one measures, deep traps cause dwell times on the order of the measurement time. Within 6 the biological context, these traps generally have been associated with chemical binding to stationary cellular components (e.g. actin cytoskeleton [12] or microtubuli [10]), with a distribution of dissociation times with varying lifetimes. T-MSDs of molecules performing CTRW show broadly scattered values, but are on average linear in the lag time tlag [39, 40], similar to our observation in Fig. 1C. The subdiffusive behavior of the motion emerges in the E-MSD, scaling with the same exponent β describing the probability density of trapping dwell-times. Since DC-SIGN dynamics also showed nonergodic subdiffusion and aging, we sought to investigate whether DC-SIGN diffusion agrees with the predictions of the CTRW model. To this end, we searched for the occurrence of transient trapping events on individual trajecto- ries. In SPT experiments, the limited localization accuracy for determining the particle po- sition sets a lower limit for the diffusivity value that can experimentally be measured. In our case, this lower threshold lies at Dth = 6 · 10−4µm2s−1. Therefore, a segmentation algo- rithm [41] was applied to the x- and y-displacements of our trajectories in order to detect events with diffusivity lower than Dth. Surprisingly, transient trapping was only detected over less than 5% of the total recording time (Fig. 2A-C). The detected trapping times dis- played an average duration of 330 ± 30 ms (Fig. 2D). An alternative analysis, based on the transient confinement zone algorithm [42], gave comparable results [43]. In order to understand the nature of these trapping events, we attempted to fit their dis- tribution by means of both an exponential and a power law distribution function ∼ t−(1+β), as expected for CTRW [12]. The power law pdf provided a better fit to the data, yield- ing an exponent β = 0.83 ± 0.05 (Fig. 2D), in agreement with the value obtained for the E-MSD. While a power-law distribution of trapping event durations would be compatible with the behavior expected for the CTRW, it is unlikely that these can have a major role in the ergodicity breaking we observe, given their very small probability of occurrence. In addition, we notice here that various other models predict a similar scaling of the trapping times; an example will be discussed in detail in Sec. IV. To quantify to what extent the small percentage of trapping events actually influences the nonergodic behavior, we cal- culated the E-MSD excluding completely the trajectories showing events compatible with immobilization. Interestingly, this analysis provided an exponent β = 0.84 ± 0.04 exactly coinciding with the value obtained for the full set of trajectories (Fig. 2D), thus confirming 7 that trapping alone can not account for the ergodicity breaking we observe in wtDC-SIGN dynamics. In addition, we constructed the distribution of escape times by identifying the duration of the events in which a trajectory remains within a given radius RTH (Fig. 2E). For a CTRW, the long-time dynamics is dominated by anomalous trapping events and, as a result, this quantity is expected to be independent of RTH [12]. In strong contrast to the CTRW model, the escape-time distributions of DC-SIGN trajectories showed a marked dependence on RTH (Fig. 2F). In summary, the rare occurrence of transient trapping events, the dependence of escape- time distributions on RTH and, most importantly, the fact that T-MSD and E-MSD show different scaling even when the few trajectories showing immobilization are removed from the analysis, are all inconsistent with CTRW, indicating that the main features of the DC-SIGN dynamics may not be explained in terms of this model. IV. DC-SIGN DISPLAY CHANGES OF DIFFUSIVITY Recently, diffusion maps of the cell membrane have shown the presence of patches with strongly varying diffusivity [36, 44, 45]. Based on this evidence, we have recently proposed a class of models describing ordinary Brownian motion with a diffusivity that varies randomly, but is constant on time intervals or spatial patches with random size [21]. These models describe anomalous diffusion and wEB in complex and heterogeneous media, such as the cellular environment, without invoking transient trapping. To address whether the observed nonergodic dynamics of DC-SIGN can be described with this theoretical framework, we further analyzed individual trajectories by means of a change-point algorithm to detect variations of diffusivity in time [41]. In brief, the algorithm consists in a likelihood-based approach to quantitatively recover time-dependent changes in diffusivity, based on the calculation of maximum likelihood estimators for the determination of diffusion coefficients and the application of a likelihood ratio test for the localization of the changes. Notably, DC-SIGN trajectories displayed a Brownian motion with relatively constant diffusivity over intervals of varying length, but that changed significantly between these intervals (Fig. 3A-C). Similar features were identified in a large fraction of trajectories, 8 with 63% showing at least one diffusivity change (Fig. 3D), in qualitative agreement with the models of random diffusivity [21]. To obtain a comprehensive understanding of our data, we considered an annealed model in which randomly diffusing particles undergo sudden changes of diffusion coefficient [21]. The distribution of diffusion coefficients D that a particle can experience is assumed to have a power-law behavior ∼ Dσ−1 for small D (with σ > 0) and a fast decay for D → ∞. Given D, the transit time τ (i.e., the time τ a particle moves with a given D) is taken to have a probability distribution with mean ∼ D−γ (with −∞ < γ < ∞). Since the motion during the transit time τ is Brownian, particles explore areas with radius r ∼ √ τ D, and the radius of the region explored with such diffusion coefficient has probability distribution with mean ∼ D regimes [21], namely: (0) for γ < σ, the long-time dynamics is compatible with ordinary 1−γ 2 . Depending on the values of the exponents σ and γ, this model predicts three Brownian motion and yields an E-MSD exponent β = 1; (I) for σ < γ < σ + 1, the average transit time τ diverges and particles undergo nonergodic subdiffusion with β = σ/γ; (II) for γ > σ + 1, both the average transit time τ and the average of the radius squared r2 of the explored area diverge and one obtains nonergodic subdiffusion with β = 1 − 1/γ. On the other hand, the T-MSD predicted by this model remains linear in time for t (cid:28) T , for every choice of σ and γ. We performed in silico experiments of 2D diffusion (Fig. 4A-B), assuming a distribution of diffusion coefficients D given by: PD(D) = Dσ−1e−D/b bσΓ(σ) and a conditional distribution of transit times τ given by: Pτ (τD) = e−τ Dγ /k Dγ k (4) (5) where b and k are dimensional constants and Γ(x) is the Gamma function. The functional forms of the distributions in Eqs. (4) and (5) comply with the requirements of our model, while at the same time ensure the minimal number of free parameters, making them a natural choice for our theoretical analysis. However, we note here that the asymp- totic behavior of the model is actually independent of the specific functional form of the joint distributions. We performed simulations for different values of σ, with γ = σ/β as in regime (I), and β = 0.84, the exponent obtained from the experimental E-MSD. The simulations 9 quantitatively reproduce not only subdiffusion, nonergodicity and aging, but also the het- erogeneous distribution of diffusion coefficients and escape time distributions (Fig. 4C-H). The remarkable agreement between simulations and experimental data strongly supports heterogeneous diffusion as the origin for DC-SIGN nonergodicity. It must be noticed that, in contrast to CTRW, our model does not assume particle immobilization, but a continuous distribution of diffusivity, with PD(D) ∼ Dσ−1 for small D. However, from the experimental point of view, it is not possible to distinguish immobilization from very slow diffusion. In fact, the limited localization accuracy of SPT experiments translates into a lower limit for the diffusivity value Dth that can be detected. Therefore, in our analysis, trajectories, or portion of trajectories, with diffusivity lower than this threshold value (Dth = 6 · 10−4µm2s−1) are identified as immobile, as shown in Fig. 2A-B. From the model described above, the distribution of the duration of these "apparent" immobilizations (cid:90) Dth can be calculated as: Pimm(τ ) = PD(D)Pτ (τD)dD. (6) 0 We neglect here the possibility that the trajectory of an in-silico particle contains two consecutive segments characterized by diffusivities Di and Di+1 which are both smaller than Dth, as this probability is vanishingly small for the parameter regime of our setup. Independently of Dth, the integral in Eq. (6) scales asymptotically as τ−1−β with β = σ γ , providing for the distribution of immobilization events the same behavior predicted by the CTRW [12]. Therefore, the distribution of immobilization times in Fig. 2D is fully compatible with the prediction of our model, further confirming its agreement with the experimental data. V. DYNAMICS OF RECEPTOR MUTANTS From the structural point of view (Fig. 5A), DC-SIGN is a tetrameric transmembrane protein, with each of the four subunits comprising: (i) an extracellular domain that allows binding of the receptor to pathogens, i.e., ligand binding domain; (ii) a long neck region; and (iii) and a transmembrane part followed by a cytoplasmic tail that allows interactions with inner cell components and facilitates the uptake and internalization of pathogens. [46, 47]. Moreover, DC-SIGN contains a single N -glycosylation site mediating interactions with glycan-binding proteins [43]. To gain insight into the molecular mechanisms of DC-SIGN 10 nonergodic diffusion, we generated three mutated forms of the receptor (Fig. 5A). These mutations have been reported to modify the interaction of DC-SIGN with other cellular components, strongly affecting DC-SIGN function [43, 48, 49]. The N80A mutant lacks the N -glycosylation site. This defect hinders interactions of DC-SIGN with components of the extracellular membrane that bind to sugars [43]. The ∆35 mutant lacks a significant part of the cytoplasmic tail, preventing interactions with cytosolic components such as actin [48]. Finally, the ∆Rep mutant lacks part of the neck region, abrogating interactions between different DC-SIGN molecules [49]. We found that each mutation has a very different effect on the dynamics of the receptor. The N80A mutant (Fig. 5C-F) showed nonergodic subdiffusion, with an exponent β similar to the one measured for wtDC-SIGN. However, N80A showed a significantly larger extent of heterogeneity in the diffusion coefficients distribution, with a lower median diffusivity. The ∆35 mutant (Fig. 5G-L) also showed nonergodic subdiffusion. The anomalous exponent and the distribution of the diffusion coefficients were similar to that of wtDC-SIGN, with only a slight reduction in median diffusivity. We accurately reproduced N80A and ∆35 dynamics by simulations performed in regime (I), i.e., nonergodic subdiffusion, using comparable values of γ for wtDC-SIGN and ∆35, and a smaller value of γ for N80A (Fig. 5B). On the other hand, ∆Rep dynamics yielded ergodic Brownian diffusion (Fig. 5M-P) and a narrower distribution of diffusivity with median value significantly higher than for wtDC-SIGN. Consistently, the behavior of ∆Rep was fully captured by in silico experiments in regime (0), i.e., ordinary Brownian motion. Overall, these results demonstrate that the molecular structure of the receptor strongly influences its diffusive behavior on the cell membrane and the occurrence of weak ergodicity breaking. VI. NONERGODICITY AND BIOLOGICAL FUNCTION Together with our previous biophysical studies on DC-SIGN [43, 49], the data and analysis presented in this paper allow us to link the dynamical behavior of DC-SIGN to its functional role in pathogen capture and uptake (known as endocytosis). In terms of steady-state organization, wtDC-SIGN, N80A and ∆35 preferentially form nanoclusters on the cell membrane, which are crucial for regulating pathogen binding [43, 49], 11 whereas removal of the neck region (∆Rep) reduces nanoclustering and binding efficiency to small pathogens, such as viruses [49]. Our results thus show that the diffusive behavior of the receptor is strongly linked to nanoclustering, but not merely due to size-dependent diffusivity and/or time-dependent cluster formation and breakdown. In fact, dual-color SPT experi- ments performed at high labeling density do not reveal correlated motion between nearby DC-SIGN nanoclusters, excluding the occurrence of dynamic nanocluster coalescence [43]. Moreover, although superresolution imaging has revealed that wtDC-SIGN, N80A and ∆35 form nanoclusters with similar distributions of size and stoichiometry [43, 49], our dynamical data evidence significant differences in their diffusion patterns (Figs. 1 and 5). Our data are in fact consistent with the view of the plasma membrane as a highly dy- namic and heterogeneous medium, where wEB stems from the enhanced ability of DC-SIGN nanoclusters to interact with the membrane environment, including components from the outer and inner membrane leaflet. This interaction is inhibited (or strongly reduced) in the case of the ∆Rep mutant since it does not form nanoclusters [49]. As a result, the motion of ∆Rep is Brownian and ergodic, and interestingly this dynamic behavior correlates with its impaired pathogen binding capability [34, 49]. In contrast, we observed that both wtDC-SIGN and N80A, which show a similar degree of nanoclustering [43], exhibit wEB. But, the distribution of diffusivity of N80A is signifi- cantly broader than that of wtDC-SIGN, and is shifted towards lower diffusivity values (Fig. 5C-F). This increased heterogeneity correlates with altered interactions of the N80A with extracellular components, resulting from the removal of the glycosylation site. Indeed, we have recently shown that the N80A mutant has a reduced capability to interact with extra- cellular sugar binding partners [43]. Thus, it appears that the extracellular milieu next to the membrane contributes to the degree of dynamical heterogeneity sensed by the receptor. Remarkably, this correlation also extends to the functional level, as we have recently shown that interactions of DC-SIGN with extracellular sugar-binding proteins influence encounters of DC-SIGN with the main endocytic protein clathrin. In turn, this resulted in reduced clathrin-dependent endocytosis of the receptor and its pathogenic ligands [43]. Finally, the ∆35 mutant exhibits nanoclustering [49] and wEB similar to that of wtDC- SIGN, From the biological point of view, however, this mutant is not able to interact with cytosolic components in close proximity to the inner membrane leaflet, including actin [48]. Therefore, in contrast to the extracellular influence observed for the N80A mutant, the 12 results obtained for the ∆35 mutant indicate that interactions with the actin cytoskeleton, responsible for the CTRW-like behavior of other proteins [12], do not play a major role in DC- SIGN wEB. Nevertheless, it should be mentioned that the reduced endocytic capability of the ∆35 could not be uniquely attributed to its dynamic behavior on the cell membrane but rather to its impaired interaction with downstream partners involved in internalization [48, 49]. VII. CONCLUSIONS AND OUTLOOK We have demonstrated that the DC-SIGN receptor displays subdiffusive dynamics, char- acterized by weak ergodicity breaking and aging. In contrast to other biological systems, receptor trajectories do not show significant evidence of transient immobilization with power- law distributed waiting times. Therefore, its nonergodic behavior cannot be explained in terms of the CTRW model. However, DC-SIGN dynamics is highly heterogeneous, with trajectories often displaying sudden changes of diffusivity. These features are accurately de- scribed by a novel model of ordinary diffusion in complex media, strongly suggesting inhomo- geneous diffusivity as the cause of DC-SIGN nonergodic behavior. Comparative analysis of three mutated forms of DC-SIGN evidences the importance of specific regions of the receptor structure, known to mediate interactions with other molecules, in receptor dynamics. Since the mutations of these regions differently impair receptor function, the experiments allowed us to establish the relevance of nonergodicity for the regulation of functional mechanisms, such as capacity for pathogen recognition and internalization. The evidence that temporal and/or spatial disorder induces subdiffusion and wEB agrees remarkably well with the current view of the plasma membrane as an extremely complex environment. Here, precise tuning of the spatiotemporal organization of membrane compo- nents, in addition to biochemical interactions with molecules in the inner and outer mem- brane leaflet, orchestrate the triggering of cell signaling pathways. A detailed understanding of how these specific interactions occur and affect dynamics is still lacking. Future experi- ments, involving simultaneous tracking of several proteins by means of multicolor SPT [44] 13 might provide a deeper comprehension of these mechanisms at the molecular level. The model used to interpret our data provides a flexible and realistic framework to de- scribe anomalous motion in cell membranes. Although in the present work we have focused our simulations on time-dependent changes of diffusivity, similar conclusions can be obtained assuming a spatial dependence, with constant diffusivity on membrane patches of random size [21]. The current data do not allow discrimination between the two scenarios. The application of techniques that combine dynamic and spatial mapping at high labeling con- ditions [45, 50] would be necessary to verify the occurrence of spatial maps of diffusivity. In addition, numerical simulations of spatial-dependent random diffusivity require the con- struction of 2-dimensional diffusivity maps consistent with the model's probability densities, which is a non-trivial task. While the work presented here focuses on the cell plasma membrane, we point out that these results have much broader implications. In fact, our model and analysis are very general and can be applied to any diffusive system that shows wEB, in order to investigate the role that heterogeneous diffusivity plays in observed anomalies. Fundamental questions about the nature of anomalous and nonergodic diffusion in disordered media arise in many fields, such as life sciences [28], soft condensed matter [51, 52], ultracold gases [53, 54], geology [55] and ecology [56]. Our work provides an alternative conceptual framework and specific tools for answering these questions. ACKNOWLEDGMENTS We thank A. Cambi for providing DC-SIGN transfected CHO cells, O. Esteban for prepar- ing the Fab fragments, P. Symeonidou Besi for preliminary data analysis and B. Castro for recording the N80A trajectories. This work was supported by Fundaci´o Cellex, Generalitat de Catalunya (Grant No. 2009 SGR 597), the European Commission (FP7-ICT-2011-7, Grant No. 288263), the HFSP (Grant No. RGP0027/2012), ERC AdG Osyris, and the Spanish Ministry of Science and Innovation (Grants FOQUS and No. MAT2011-22887). 14 Appendix A: Cell culture and labeling Chinese hamster ovary (CHO) cells stably transfected with DC-SIGN mutants were cul- tured in HAM's F-12 medium (LabClinics) supplemented with 10% fetal calf serum and antibiotic/antimycotic (Gibco). CHO cells were seeded onto 25 mm coverslips 24 hours be- fore imaging. Streptavidin-coated quantum dots (Qdot655, Invitrogen) were added to an equimolar solution of biotinylated anti-DC-SIGN DCN46 Fab fragment (or anti-AU1 mono- valent Ab, in the case of ∆Rep mutant) and a 50x excess of free biotin (Gibco) in order to obtain a 1:1 Fab fragment-quantum dot ratio (or monovalent Ab-quantum dot ratio, in the case of ∆Rep mutant). In these conditions, we estimated a 0.04% probability of hav- ing multiple Fab fragments or Abs bound to the same quantum dot. Cells were incubated for 5 min at RT with 50 pM conjugated quantum dots in cold PBS buffer supplemented with 6% BSA. Extensive washing was performed to remove non-bound conjugated quantum dots before imaging. In parallel with each experiment, we also performed control experi- ments by labeling DC-SIGN-negative CHO cells. The lack of quantum dots binding to the DC-SIGN-negative CHO cells confirmed the absence of non-specific binding. Furthermore, the low-concentration labeling conditions were chosen to minimize the probability of hav- ing more than one quantum dot within the same region of interest. Fab fragments and monovalent Ab were obtained as described in Refs. [43, 49]. Appendix B: Single particle tracking experiments We performed video microscopy using a custom single-molecule sensitive epi-fluorescence microscope. Continuous excitation was provided by the 488-nm line of an Argon-ion laser (Spectra Physics), with power density at the sample plane of ∼0.3 kW/cm2. Fluorescence was collected by means of a 1.2 NA water immersion objective (Olympus) and guided into an intensified EM-CCD camera (Hamamatsu) after suitable filtering. Movies were recorded on the dorsal membrane of CHO cells at 60Hz frame rate. Experiments were performed in a culture dish incubator (DH-35iL, Warner Instrument) equipped with a temperature controller (TC-324B, Warner Instrument) and a digital CO2 controller (DGTCO2BX, Oko- lab) at 37◦C and in 5% CO2 atmosphere. Trajectories were analyzed with custom Matlab code based on the algorithm described in Ref. [36]. In order to avoid artifacts in trajectory 15 reconnection caused by quantum dots blinking dynamics and/or high local density of quan- tum dots, each trajectory was terminated at the first video frame not displaying a clearly identifiable bright spot in the surrounding of the quantum dot localization obtained in the previous frame. Similarly, the trajectory reconstruction was also interrupted if the presence of multiple bright spots did not allow unambiguous identification of the same quantum dots in successive video frames. As a further check for false-positive reconnection, trajectories were overlaid to raw movies and visually inspected. Appendix C: Data analysis Time-, ensemble- and time-ensemble-averaged mean-squared displacements were calcu- lated as described in [12]. Exponents of the E-MSD, average T-MSD and Dt,ens were obtained by linear fitting of the log-log transformed data. Errors were calculated as the 99% con- fidence interval of the fitting parameters. Short-time diffusion coefficients were extracted from the linear fit of the first 10% of the points of T-MSD curves [37]. Measurements of the apparent diffusion of quantum dots on fixed cells and glass coverslips were used to estimate the smallest detectable diffusivity. Short-time diffusion coefficients were obtained as described above for trajectories of immobilized quantum dots and the corresponding probability distribution was calculated. 95% of the immobilized quantum dots trajectories showed values lower than 6 · 10−4µm2/s, which was therefore set as the threshold (Dth) for classifying a trajectory as mobile. Dynamical changes in the motion of DC-SIGN receptors were identified by application of the change-point algorithm described in Ref. [41]. In brief, the trajectories were recursively segmented and a maximum-likelihood-ratio test was applied to the trajectory displacements (∆x, ∆y) in order to identify sudden changes of diffusivity. The critical values for Type I error rates were set to a confidence level of 99%, corresponding to 1% probability of having a false-positive identification of a change-point. For each dynamical region identified by the algorithm, the short-time diffusion coefficient was calculated from a linear fit of the first 10% of the points of the corresponding MSD curves [37]. Regions showing a short-time diffusion coefficient lower than Dth were considered compatible with transient immobilization. 16 Appendix D: Simulations Simulated trajectories (500 per parameter set) were obtained by generating random dif- fusion coefficients D according to the probability distribution given in Eq. (4). For each diffusion coefficient, the corresponding transit time τ was calculated as a random number drawn from the distribution given in Eq. (5). Particle coordinates r = {x, y} were generated as: rt+∆t(cid:48) = rt + √ 2D∆t(cid:48)ξ where ξ = {ξx, ξy} are pairs of random numbers from a Gaussian distribution with zero mean and unitary standard deviation. The time increment was calculated as ∆t(cid:48) = ∆t/n, where ∆t is the camera acquisition rate and n is a integer depending on D and τ , chosen in order to have at least 10 points for each interval. For comparison with SPT data, trajectories were sub-sampled at the camera acquisition rate. Simulated trajectories were generated with duration Tsim ≥ 3 · Texp, where Texp is the duration of the experimental trajectory. The starting point was randomly drawn from a uniform distribution defined within 0 and Tsim − Texp. Trajectories were then cut to have the same duration Texp as the experimental ones. Gaussian noise corresponding to the experimental localization accuracy (σacc = 20 nm) was subsequently added to the trajectories. [1] S. Condamin, O. B´enichou, V. Tejedor, R. Voituriez, and J. Klafter, First-passage times in complex scale-invariant media., Nature 450, 77 (2007). [2] M. A. Lomholt, I. M. Zaid, and R. Metzler, Subdiffusion and Weak Ergodicity Breaking in the Presence of a Reactive Boundary, Phys. Rev. Lett. 98, 200603 (2007). [3] S. Condamin, V. Tejedor, R. Voituriez, O. B´enichou, and J. Klafter, Probing microscopic origins of confined subdiffusion by first-passage observables., Proc. Natl. Acad. Sci. U. S. A. 105, 5675 (2008). [4] D. Choquet and A. Triller, The role of receptor diffusion in the organization of the postsynaptic membrane, Nature Rev. Neurosci. 4, 251 (2003). [5] A. Kusumi, T. a. Tsunoyama, K. M. Hirosawa, R. S. Kasai, and T. K. Fujiwara, Tracking single molecules at work in living cells., Nat. Chem. Biol. 10, 524 (2014). 17 [6] I. Bronstein, Y. Israel, E. Kepten, S. Mai, Y. Shav-Tal, E. Barkai, and Y. Garini, Transient Anomalous Diffusion of Telomeres in the Nucleus of Mammalian Cells, Phys. Rev. Lett. 103, 018102 (2009). [7] M. Weiss, M. Elsner, F. Kartberg, and T. Nilsson, Anomalous subdiffusion is a measure for cytoplasmic crowding in living cells, Biophys. J. 87, 3518 (2004). [8] I. Golding and E. C. Cox, Physical Nature of Bacterial Cytoplasm, Phys. Rev. Lett. 96, 098102 (2006). [9] J.-H. Jeon, V. Tejedor, S. Burov, E. Barkai, C. Selhuber-Unkel, K. Berg-Sø rensen, L. Odder- shede, and R. Metzler, In Vivo Anomalous Diffusion and Weak Ergodicity Breaking of Lipid Granules, Phys. Rev. Lett. 106, 048103 (2011). [10] S. M. A. Tabei, S. Burov, H. Y. Kim, A. Kuznetsov, T. Huynh, J. Jureller, L. H. Philipson, A. R. Dinner, and N. F. Scherer, Intracellular transport of insulin granules is a subordinated random walk., Proc. Natl. Acad. Sci. U. S. A. 110, 4911 (2013). [11] M. J. Saxton and K. Jacobson, Single-particle tracking: applications to membrane dynamics., Annu. Rev. Biophys. Biomol. Struct. 26, 373 (1997). [12] A. V. Weigel, B. Simon, M. M. Tamkun, and D. Krapf, Ergodic and nonergodic processes coexist in the plasma membrane as observed by single-molecule tracking., Proc. Natl. Acad. Sci. U. S. A. 108, 6438 (2011). [13] F. Hofling and T. Franosch, Anomalous transport in the crowded world of biological cells, Rep. Prog. Phys. 76, 046602 (2013). [14] J.-P. Bouchaud and A. Georges, Anomalous diffusion in disordered media: statistical mecha- nisms, models and physical applications, Phys. Rep. 195, 127 (1990). [15] S. Havlin and D. Ben-Avraham, Diffusion in disordered media, Adv. Phys. 51, 187 (2002). [16] J. Klafter and I. M. Sokolov, First Steps in Random Walks (Oxford University Press, Oxford, 2011). [17] M. J. Saxton, Anomalous diffusion due to obstacles: a Monte Carlo study., Biophys. J. 66, 394 (1994). [18] J. Klafter, A. Blumen, and M. F. Shlesinger, Stochastic pathway to anomalous diffusion, Phys. Rev. A 35, 3081 (1987). [19] B. B. Mandelbrot and J. W. Van Ness, Fractional Brownian motions, fractional noises and applications, SIAM Rev. 10, 422 (1968). 18 [20] A. G. Cherstvy, A. V. Chechkin, and R. Metzler, Anomalous diffusion and ergodicity breaking in heterogeneous diffusion processes, New J. Phys. 15, 083039 (2013). [21] P. Massignan, C. Manzo, J. A. Torreno-Pina, M. F. Garcia-Parajo, M. Lewenstein, and G. J. Lapeyre, Nonergodic Subdiffusion from Brownian Motion in an Inhomogeneous Medium, Phys. Rev. Lett. 112, 150603 (2014). [22] M. V. Chubynsky and G. W. Slater, Diffusing Diffusivity: A Model for Anomalous, yet Brow- nian, Diffusion, Phys. Rev. Lett. 113, 098302 (2014). [23] M. J. Saxton, Anomalous diffusion due to binding: a Monte Carlo study., Biophys. J. 70, 1250 (1996). [24] D. Ernst, M. Hellmann, J. Kohler, and M. Weiss, Fractional Brownian motion in crowded fluids, Soft Matter 8, 4886 (2012). [25] A. V. Weigel, S. Ragi, M. L. Reid, E. K. P. Chong, M. M. Tamkun, and D. Krapf, Obstructed diffusion propagator analysis for single-particle tracking, Phys. Rev. E 85, 041924 (2012). [26] A. P. Young, Spin glasses and random fields (World Scientific, 1997). [27] E. Barkai, Aging in Subdiffusion Generated by a Deterministic Dynamical System, Phys. Rev. Lett. 90, 104101 (2003). [28] E. Barkai, Y. Garini, and R. Metzler, Strange kinetics of single molecules in living cells, Phys. Today 65, 29 (2012). [29] R. Metzler, J.-H. Jeon, A. G. Cherstvy, and E. Barkai, Anomalous diffusion models and their properties: non-stationarity, non-ergodicity, and ageing at the centenary of single particle tracking, Phys. Chem. Chem. Phys. 16, 24128 (2014). [30] A. G. Cherstvy, A. V. Chechkin, and R. Metzler, Particle invasion, survival, and non- ergodicity in 2D diffusion processes with space-dependent diffusivity, Soft Matter 10, 1591 (2014). [31] J. Jeon, E. Barkai, and R. Metzler, Noisy continuous time random walks, J. Chem Phys 139, 121916 (2013), arXiv:1305.1721v1. [32] T. B. Geijtenbeek, D. S. Kwon, R. Torensma, S. J. van Vliet, G. C. van Duijnhoven, J. Middel, I. L. Cornelissen, H. S. Nottet, V. N. KewalRamani, D. R. Littman, C. G. Figdor, and Y. van Kooyk, DC-SIGN, a dendritic cell-specific HIV-1-binding protein that enhances trans-infection of T cells., Cell 100, 587 (2000). 19 [33] Y. van Kooyk and T. B. H. Geijtenbeek, DC-SIGN: escape mechanism for pathogens, Nature Rev. Immunol. 3, 697 (2003). [34] A. Cambi, D. S. Lidke, D. J. Arndt-Jovin, C. G. Figdor, and T. M. Jovin, Ligand-Conjugated Quantum Dots Monitor Antigen Uptake and Processing by Dendritic Cells, Nano Lett. 7, 970 (2007). [35] A. Cambi, I. Beeren, B. Joosten, J. A. Fransen, and C. G. Figdor, The C-type lectin DC-SIGN internalizes soluble antigens and HIV-1 virions via a clathrin-dependent mechanism, Eur. J. Immunol. 39, 1923 (2009). [36] A. Serg´e, N. Bertaux, H. Rigneault, and D. Marguet, Dynamic multiple-target tracing to probe spatiotemporal cartography of cell membranes, Nat. Methods 5, 687 (2008). [37] X. Michalet, Mean square displacement analysis of single-particle trajectories with localization error: Brownian motion in an isotropic medium, Phys. Rev. E 82, 041914 (2010). [38] G. Bel and E. Barkai, Weak Ergodicity Breaking in the Continuous-Time Random Walk, Phys. Rev. Lett. 94, 240602 (2005). [39] A. Lubelski, I. M. Sokolov, and J. Klafter, Nonergodicity Mimics Inhomogeneity in Single Particle Tracking, Phys. Rev. Lett. 100, 250602 (2008). [40] Y. He, S. Burov, R. Metzler, and E. Barkai, Random Time-Scale Invariant Diffusion and Transport Coefficients, Phys. Rev. Lett. 101, 058101 (2008). [41] D. Montiel, H. Cang, and H. Yang, Quantitative characterization of changes in dynamical behavior for single-particle tracking studies., J. Phys. Chem. B 110, 19763 (2006). [42] R. Simson, E. D. Sheets, and K. Jacobson, Detection of temporary lateral confinement of membrane proteins using single-particle tracking analysis, Biophys. J. 69, 989 (1995). [43] J. A. Torreno-Pina, B. M. Castro, C. Manzo, S. I. Buschow, A. Cambi, and M. F. Garcia- Parajo, Enhanced receptor-clathrin interactions induced by N-glycan-mediated membrane mi- cropatterning., Proc. Natl. Acad. Sci. U. S. A. 111, 11037 (2014). [44] P. J. Cutler, M. D. Malik, S. Liu, J. M. Byars, D. S. Lidke, and K. A. Lidke, Multi-color quantum dot tracking using a high-speed hyperspectral line-scanning microscope, PLoS One 8, e64320 (2013). [45] J.-B. Masson, P. Dionne, C. Salvatico, M. Renner, C. G. Specht, A. Triller, and M. Dahan, Mapping the energy and diffusion landscapes of membrane proteins at the cell surface using high-density single-molecule imaging and Bayesian Inference: application to the multiscale 20 dynamics of glycine receptors in the neuronal membrane, Biophys. J. 106, 74 (2014). [46] H. Feinberg, D. A. Mitchell, K. Drickamer, and W. I. Weis, Structural Basis for Selective Recognition of Oligosaccharides by DC-SIGN and DC-SIGNR, Science 294, 2163 (2001). [47] D. A. Mitchell, A. J. Fadden, and K. Drickamer, A novel mechanism of carbohydrate recog- nition by the C-type lectins DC-SIGN and DC-SIGNR Subunit organization and binding to multivalent ligands, J. Biol. Chem. 276, 28939 (2001). [48] A. L. Smith, L. Ganesh, K. Leung, J. Jongstra-Bilen, J. Jongstra, and G. J. Nabel, Leukocyte- specific protein 1 interacts with DC-SIGN and mediates transport of HIV to the proteasome in dendritic cells., J. Exp. Med. 204, 421 (2007). [49] C. Manzo, J. A. Torreno-Pina, B. Joosten, I. Reinieren-Beeren, E. J. Gualda, P. Loza-Alvarez, C. G. Figdor, M. F. Garcia-Parajo, and A. Cambi, The neck region of the C-type lectin DC- SIGN regulates its surface spatiotemporal organization and virus-binding capacity on antigen- presenting cells., J. Biol. Chem. 287, 38946 (2012). [50] S. Manley, J. M. Gillette, G. H. Patterson, H. Shroff, H. F. Hess, E. Betzig, and J. Lippincott- Schwartz, High-density mapping of single-molecule trajectories with photoactivated localization microscopy, Nat. Methods 5, 155 (2008). [51] R. A. L. Vall´ee, N. Tomczak, L. Kuipers, G. J. Vancso, and N. F. van Hulst, Single Molecule Lifetime Fluctuations Reveal Segmental Dynamics in Polymers, Phys. Rev. Lett. 91, 038301 (2003). [52] G. Volpe, G. Volpe, and S. Gigan, Brownian Motion in a Speckle Light Field: Tunable Anoma- lous Diffusion and Selective Optical Manipulation, Sci. Rep. 4, 3936 (2014), 10.1038/srep03936, 1304.1433. [53] E. Lucioni, B. Deissler, L. Tanzi, G. Roati, M. Zaccanti, M. Modugno, M. Larcher, F. Dalfovo, M. Inguscio, and G. Modugno, Observation of Subdiffusion in a Disordered Interacting System, Phys. Rev. Lett. 106, 230403 (2011). [54] S. Krinner, D. Stadler, J. Meineke, J.-P. Brantut, and T. Esslinger, Direct Observation of Fragmentation in a Disordered, Strongly Interacting Fermi Gas, ArXiv e-prints (2013), arXiv:1311.5174 [cond-mat.quant-gas]. [55] B. Berkowitz, A. Cortis, M. Dentz, and H. Scher, Modeling non-Fickian transport in geological formations as a continuous time random walk, Rev. Geophys. 44, n/a (2006). 21 [56] J. Bascompte and R. Sol´e, Modeling spatiotemporal dynamics in ecology, Environmental intel- ligence unit (Springer, 1998). 22 Figure 1. DC-SIGN diffusion shows weak ergodicity breaking and aging. (A) Representa- tive video frames of a quantum-dot-labeled wtDC-SIGN molecule diffusing on the dorsal membrane of a CHO cell. The centroid position of the bright spot (+), corresponding to a single quantum-dot, is tracked and reconnected to build up the DC-SIGN trajectory, shown by the cyan line. (B) Rep- resentative trajectories for the same recording time (3.2 s). (C) Log-log plot of the time-averaged MSD for individual trajectories (blue lines). The dashed lines scale linearly in time, showing that T-MSD is compatible with pure Brownian motion (β = 1). The symbols ((cid:13)) correspond to the av- erage T-MSD. Linear fit to the log-log transformed data (black line) provided β = 0.95± 0.05. (D) Distribution of short-time diffusion coefficients as obtained from linear fitting of the time-averaged MSD for all the trajectories. (Inset) cdf of the exponent β obtained from nonlinear fitting of the T-MSD of all the trajectories. (E) Log-log plot of the ensemble-averaged MSD. Power law fit of the data (dashed line) provides an exponent β = 0.84, showing subdiffusion. (F) Log-log plot of the time-ensemble-averaged diffusion coefficient as a function of the observation time T. The dif- fusion coefficients are obtained by linear fitting of the time-ensemble-averaged MSD. A power-law fit (dashed line) provides an exponent β − 1 = −0.17, revealing aging and in good agreement with the value of β found in (E). 23 1 μm1 μmβ = 0.84 ± 0.03E-MSD (μm2)β-1 = -0.17 ± 0.05DTE (μm2·s-1)10-210-1T-MSD (μm2)10-310-210-11001010.10.050.30.030.10.3tlag (s)10-110010-110010-2tlag (s)T (s)ABCEDF~t~tβ~Tβ-100.040.080.120.160.2frequency10-310-210-1100Ds (μm2·s-1)t=0t=1.7 st=3.5 st=5.2 s1 μm01β0.81.11.0cdfFigure 2. DC-SIGN receptor dynamics is inconsistent with the CTRW model. (A) A trajectory of wtDC-SIGN on living cell membranes showing a short-lived transient immobilization event, highlighted by the yellow circular area. (B) Plot of the x- (blue) and y-displacements (red) as a function of time. The occurrence of transient immobilization (yellow region) corresponds to a reduction in the trajectory displacement. (C) Time-averaged MSD for the entire trajectory ((cid:4)) and for the immobilization region only ((cid:13)). (D) Survival function of the duration of immobilization events for wtDC-SIGN trajectories (black line). Red and blue lines correspond to fits to exponen- tial and power-law distribution functions, respectively. Power law fit provided β = 0.83 ± 0.05, in agreement with the exponent obtained for the E-MSD. (E) Schematic representation of the calcu- lation of the escape time probability from circular areas of different radius RTH. (F) Cumulative probability distribution function (cdf ) of trajectory escape time for different radii RTH=20 ((cid:13)), 50 ((cid:53)), 100 ((cid:3)), 200 ((cid:52)), 300 (•), 500 ((cid:72)) and to 1000 nm ((cid:4)). Dashed lines are guides to the eye. The gray shaded region represents times shorter than the acquisition frame rate. 24 500 nm500 nm50 nmABΔx, Δy (μm)t (s)01234−0.2−0.100.10.2T-MSD (μm2)tlag (s)00.10.200.050.10.3CDsurvival functionimmobilization time (s)00.40.800.511.61.2500 nm500 nmEFRTHDs=0.08 μm2/s Ds=2∙10-5 μm2/s cdfescape time (s)10-210-110010100.20.40.60.81~t-(β+1)~e-(t/τ)Figure 3. DC-SIGN motion experiences changes in diffusivity. (A) Representative wtDC- SIGN trajectory displaying changes of diffusivity. Change-point analysis evidenced 5 different regions represented with different colors. (B) Plot of the x- (blue) and y-displacement (red) for the trajectory in (A) as a function of time. The shaded areas indicate the regions of different diffusivity. The lower panel displays the corresponding short-time diffusion coefficient as obtained from a linear fit of the time-averaged MSD for the 5 different regions. Gray areas correspond to the 95% confidence level. (C) Plot of time-averaged MSD versus time lag for the first three regions of the trajectory in (A). (D) Histogram of the number of changes of diffusion per trajectory. Most of the trajectories (63%) display at least one dynamical change, with an average of 2.2 changes per trajectory. 25 00.10.200.02tlag (s)T-MSD (μm2)AB<n>=2.2±0.5frequency00.20.424680number of diffusion changes per trajectoryCD300 nm300 nmt (s)02468−0.2−0.100.10.20.010.050.005Δx, Δy (μm)Ds (μm2·s-1)1234512345Ds=0.03 μm2/s Ds=0.02 μm2/s Ds=0.005 μm2/s 0.01123Figure 4. Annealed model of heterogeneous diffusion quantitatively reproduces DC- SIGN motion. (A) A simulated trajectory composed by five time intervals with different transit time τi and diffusivity Di. (B) Contour plot of the probability distribution of the simulated diffusion coefficient D and transit time τ for the parameters reproducing the dynamics of wtDC- SIGN (σ = 1.16, γ = 1.38, b = 0.12µm2/s, k = 0.10µm2γsγ+1). The white line represents the power law dependence between diffusivity and average transit time with exponent −γ. (C) Log-log plot of the time-averaged MSD for simulated trajectories (black lines). The symbols ((cid:13)) correspond to the average T-MSD. Linear fit to the log-log transformed data provided β = 0.98 ± 0.03. (D) Log- log plot of the ensemble-averaged MSD for the simulated trajectories. The dashed line represents a power law with the theoretical exponent β = σ/γ = 0.84. (E) Log-log plot of the time-ensemble averaged diffusion coefficient as a function of the observation time T. The dashed line represents a power law with the theoretical exponent β − 1 = −0.16. (F) Simulated trajectories for the same recording time (3.2 s). (G) Distribution of short-time diffusion coefficients as obtained from linear fitting of the time-averaged MSD for all the simulated trajectories. (Inset) cdf of the exponent β obtained from nonlinear fitting of the T-MSD of all the trajectories. (H) cdf of trajectory escape time for different radii. Curves from left to right correspond to radii RTH=20, 50, 100, 200, 300, 500 and 1000 nm. 26 0123450123450123456t (s)x (µm)y (µm)τ1,D1τ2,D2τ3,D3τ4,D4τ5,D5ABCDFEH10-210-1100101escape time (s)0cdf0.20.40.60.8100.040.080.120.160.2frequency10-310-210-1100Ds (μm2·s-1)β = 0.84E-MSD (μm2)10-310-210-110010110-110010-2tlag (s)~tβ10-210-1T-MSD (μm2)0.030.10.3tlag (s)β-1 = -0.16DTE (μm2·s-1)0.10.050.310-1100T (s)~Tβ-11 μm1 μmG10-110010-210-310-110010110210310-2D (μm2·s-1)τ (s)-γ = -1.3801β0.81.11.0~tcdfFigure 5. Effect of mutations on the dynamics of DC-SIGN. (A) Schematic representation of wtDC-SIGN and its mutated forms. (B) Contour plot of the probability distribution of the simulated diffusion coefficient (D) and transit time (τ ) for the parameters reproducing the dynamics of N80A (σ = 0.58, γ = 0.68, b = 0.09µm2/s, k = 0.74µm2γsγ+1), ∆35 (σ = 1.04, γ = 1.23, b = 0.08µm2/s, k = 0.07µm2γsγ+1) and ∆Rep (σ = 2.11, γ = 1.91, b = 0.07µm2/s, k = 0.07µm2γsγ+1). The white line represents the power law dependence between diffusivity and average transit time with exponent −γ. (C) Log-log plot of the ensemble-averaged MSD for N80A trajectories (•) and simulated data (+). (D) Log-log plot of the time-ensemble averaged diffusion coefficient for N80A trajectories (•) and simulated data (+) as a function of the observation time T. (E) Distribution of short-time diffusion coefficients as obtained from linear fitting of the time-averaged MSD for the N80A (filled bars) and the simulated trajectories (empty bars). (F) cdf of the escape time for N80A (symbols) and simulated trajectories (lines) for different radii. The meaning of the symbols is the same as in Fig. 2F. (G-L) Dynamical behavior of the ∆35 mutant. (M-P) Dynamical behavior of the ∆Rep mutant. 27 C00.050.100.150.200.25frequency10-310-210-1100Ds (μm2·s-1)Dβ = 1.01 ± 0.02E-MSD (μm2)10-310-210-110010110-110010-2tlag (s)FEβ-1 = 0.00 ± 0.08DTE (μm2·s-1)0.10.050.310-1100T (s)10-310-210-1100Ds (μm2·s-1)β = 0.85 ± 0.03E-MSD (μm2)10-310-210-110010110-110010-2tlag (s)00.040.120.08β-1 = -0.13 ± 0.03DTE (μm2·s-1)0.050.010.110-1100T (s)frequency0.02A00.100.20frequency10-310-210-1100Ds (μm2·s-1)E-MSD (μm2)10-310-210-110010110-110010-2tlag (s)β = 0.82 ± 0.02ΔRepN80AΔ35simulationssimulationssimulationsβ-1 = -0.15 ± 0.03DTE (μm2·s-1)0.050.010.110-1100T (s)0.02neck regioncytoplasmictailwtDC-SIGNN80AΔ35ΔRepligandbindingdomainglycosylationsite10-210-1100101escape time (s)0cdf0.20.40.60.8110-210-1100101escape time (s)0cdf0.20.40.60.8110-210-1100101escape time (s)0cdf0.20.40.60.81B10-110010-210-310-110010110210310-2τ (s)10-110010-210-3D (μm2·s-1)10-110010-210-3N80AΔ35ΔRepGHLIMNPOplasmamembraneextracellularcytosol-γ = -0.68-γ = -1.23-γ = -1.91
1011.1023	1	1011	2010-11-03T22:17:54	Electrostatic fluctuations promote the dynamical transition in proteins	[ "physics.bio-ph" ]	Atomic displacements of hydrated proteins are dominated by phonon vibrations at low temperatures and by dissipative large-amplitude motions at high temperatures. A crossover between the two regimes is known as a dynamical transition. Recent experiments indicate a connection between the dynamical transition and the dielectric response of the hydrated protein. We analyze two mechanisms of the coupling between the protein atomic motions and the protein-water interface. The first mechanism considers viscoelastic changes in the global shape of the protein plasticized by its coupling to the hydration shell. The second mechanism involves modulations of the motions of partial charges inside the protein by electrostatic fluctuations. The model is used to analyze mean square displacements of iron of metmyoglobin reported by Moessbauer spectroscopy. We show that high flexibility of heme iron at physiological temperatures is dominated by electrostatic fluctuations. Two onsets, one arising from the viscoelastic response and the second from electrostatic fluctuations, are seen in the temperature dependence of the mean square displacements when the corresponding relaxation times enter the instrumental resolution window.	physics.bio-ph	physics	Electrostatic fluctuations promote the dynamical transition in proteins Center for Biological Physics, Arizona State University, PO Box 871604, Tempe, AZ 85287-1604 Dmitry V. Matyushov∗ Alexander Y. Morozov Atomic displacements of hydrated proteins are dominated by phonon vibrations at low temper- atures and by dissipative large-amplitude motions at high temperatures. A crossover between the two regimes is known as a dynamical transition. Recent experiments indicate a connection between the dynamical transition and the dielectric response of the hydrated protein. We analyze two mech- anisms of the coupling between the protein atomic motions and the protein-water interface. The first mechanism considers viscoelastic changes in the global shape of the protein plasticized by its coupling to the hydration shell. The second mechanism involves modulations of the motions of partial charges inside the protein by electrostatic fluctuations. The model is used to analyze mean square displacements of iron of metmyoglobin reported by Mossbauer spectroscopy. We show that high flexibility of heme iron at physiological temperatures is dominated by electrostatic fluctuations. Two onsets, one arising from the viscoelastic response and the second from electrostatic fluctuations, are seen in the temperature dependence of the mean square displacements when the corresponding relaxation times enter the instrumental resolution window. PACS numbers: 87.14.E-, 87.15.H-, 87.15.kr, 87.10.Pq Keywords: Dynamical transition, Mossbauer spectroscopy, hydrated protein. I. INTRODUCTION Measurements of the Mossbauer absorption of 57Fe in metmyoglobin crystals revealed that the mean square dis- placement (msd) of this atom starts to grow faster with increasing temperature above TD ≃ 200 K.1 This find- ing was followed by similar observations from neutron scattering,2 which by now have been reported for a large number of proteins and other biopolymers,3 all demon- strating the same phenomenology.4,5 The increase in the slope of the protein msd as a function of temperature was called a "dynamical transition", presently assigned to a rather broad range of onset temperatures TD ≃ 200−240 K. The basic observation is that the high-temperature flexibility of the proteins much exceeds the linear extrap- olation of the low-temperature behavior characteristic of an expanding solid. The low-temperature msd is well characterized by the observed phonon spectrum of the protein,6 while the high-temperature msd excess is linked to dissipative long-wavelength modes with energies below ≃ 3 meV.7 -- 10 Early explanations of the dynamical transition offered scenarios ranging from detrapping of the protein con- formational motions from low-energy states6,11,12 to the glass transition in bulk water.2 Several recent observa- tions have shifted the focus to the protein hydration shell. The disappearance of the dynamical transition in dry protein powders,13 and its separate existence for the hydration water,14,15 point to a strong link between the dynamical transition and the protein hydration shell. Despite somewhat different semantics, the views in the field seem to converge to the notion of the critical role of the hydration shell in driving the dynamical transi- tion. According to Doster:16 "The onset of the dynamical transition depends on the solvent viscosity near the pro- tein surface.. . . The protein-water α-process consists of a concerted librational motion of protein surface residues, coupled to translational jumps of water molecules on the same time scale." The physical mechanism behind the transition is assigned in this view to the caging of the protein by water's hydrogen bonds, stiffening its confor- mational flexibility. As the temperature increases, the population of broken hydrogen bonds grows exponen- tially, resulting in a release of the protein conformational flexibility in a narrow range of temperatures. Although appealing, this concept does not address the key question of how ∼ 0.5 − 2 ps events of hydrogen bond breaking develop into a ∼ 2 µs collective relaxation process at TD. Frauenfelder and co-workers have recently suggested a somewhat different scenario, also focusing on the dynam- ics of the protein hydration shell.17 -- 19 According to their view, protein conformations are coupled to motions of the hydration layer, with a relaxation time following an Arrhenius law. This relaxation mode is therefore con- sidered to be a secondary, β-process according to the common classification adopted in the field of supercooled liquids.20 This secondary relaxation is assessed by dielec- tric spectroscopy of protein samples embedded into solid poly(vinyl)alcohol (PVA), thus eliminating the bulk wa- ter relaxation from the dielectric response.18,19 The use of dielectric absorption of PVA-confined pro- teins yields a surprisingly accurate account of the temper- ature dependence of the Lamb-Mossbauer factor equal to the fraction of recoilless absorption, f = exp[−k2 0h(δx)2i]. Here, h(δx)2i is the msd of the heme iron of metmyo- globin in projection on the wavevector k0 of γ-radiation. The iron msd can be separated into a vibrational, low- temperature component h(δq)2i ∝ T , described by the vibrational density of states (VDOS) of the protein, and a high-temperature component h(δQ)2i appearing at T > TD. Correspondingly, f = fqfQ becomes a product of two components, fq and fQ. It turns out that the temperature variation of the Lamb-Mossbauer factor fQ originating from the high-temperature msd is exception- ally well described by the variance of the sample dipole moment at the same hydration level fQ = exp[−k2 0h(δQ)2i] = h(δM )2i</h(δM )2i. (1) In this equation, h(δM )2i< is defined as the integral of the frequency-dependent variance of the sample dipole moment Mω over the frequencies below (subscript "<") the instrumental frequency ωobs = 1/τobs, τobs = 140 ns.19 The parameter fQ then determines the fraction of the sample dipole that has not had a chance to alter on the life-time of the iron nucleus, thus keeping the nuclei in resonance for Mossbauer absorption. The empirical connection between the dynamical tran- sition and dipolar fluctuations is additionally supported by recent observations of breaks in the dependence of the terahertz dielectric absorption on temperature at typi- cal values of TD.21 All these observations, although ad- vancing the field toward identifying the physical modes responsible for high-temperature flexibility of proteins, pose a significant conceptual challenge. Both Mossbauer and neutron-scattering techniques probe translational atomic motions on their correspond- ing resolution windows. It seems therefore natural to relate the break in the temperature dependence of the msd to changes in the dynamic and/or static properties of atomic translations. This is the conceptual frame- work behind the glass-transition scenario16 which, even in the current form emphasizing the hydration layer, is focused on the caging arrest, i.e., on the primary effect of short-ranged repulsive interactions in the system. On the contrary, the dielectric measurements19,21,22 shift the focus to the long-ranged electrostatic interactions, which is what dielectric spectroscopy is sensitive to at the first place. What is the correct view? Answering this question requires gaining deeper in- sights into the actual physical modes coupled to the pro- tein msd at high temperatures, the problem that has evaded direct experimental inquiry so far. Numerical simulations point to enhanced fluctuations of hydrogen bonds of hydration water at high temperatures,23 but those can be projected on either density or dipolar col- lective modes. Since a strong link between the dynam- ical transition and the hydration shell has been clearly established, the main question posed by recent studies is whether density or orientational collective modes drive the transition.24 They are mostly decoupled by symmetry and can therefore be considered as two distinct mecha- nisms of altering the protein flexibility. The goal of this paper is to present some initial esti- mates of the relative importance of the density and ori- entational fluctuations in the protein dynamical transi- tion. We model the density fluctuations at the protein- water interface by viscoelastic response and the orien- tational dipolar fluctuations in terms of electrostatic re- 2 sponse. The dependence of the onset temperature TD on the observation window is an important ingredient of the observations,16,25 which is introduced into the theory by limiting the range of frequencies over which the response functions are integrated,26 similarly to Eq. (1). We start with formulating the model, followed by the results of calculations. II. MODEL The purpose of our model is to determine the msd of a single atom, heme iron of metmyoglobin, as a func- tion of temperature. The internal motions of the protein can roughly be separated into two modes, the phonon vibrations q and dissipative large-scale motions Q. Even though phonons do not formally exist in proteins, we will use this language to distinguish short-wavelength vibra- tions from motions altering the protein's global shape. Accordingly, we will split the coordinate of the iron atom r = q+ Q into two statistically independent components, q and Q. The former can be expanded in protein's nor- mal modes or directly calculated from the VDOS mea- sured, for instance, by the phonon-assisted Mossbauer scattering.6,27,28 The corresponding msd of the vibra- tional coordinate q is, in the classical limit, a linear func- tion of temperature h(δq)2i = aqT. (2) The proportionality coefficient aq is calculated from the VDOS according to the standard prescriptions. The dissipative motions of the protein are seen in neu- tron scattering spectra as a quasielastic peak with ener- gies below ≃ 4 meV, growing in intensity with increasing temperature.9 With the sound velocity in a protein8,29 of ≃ 1700 m/s, the wavelength of the corresponding vi- brations is about 26 A, which is comparable with the diameter of myoglobin, 2R = 36 A. These modes there- fore alter the global shape of the protein, which is the domain of the viscoelastic response. In order to obtain a first-order estimate of the geom- etry changes involved, we will model the protein mo- tions as radial viscoelastic vibrations of a sphere of ra- dius R immersed in a viscoelastic water continuum. The low-frequency variance of the iron coordinate is then simply related to the radius fluctuations of the sphere as h(δQ2)i = (r/R)2h(δR)2i. The latter can be found by solving the standard equations of viscoelasticity30,31 yielding the response function χR(ω) connecting the change of the sphere's radius R to an oscillatory pres- sure p(t) = p0eiωt applied to the sphere's surface. The result is30 χR(ω) = − 1 1 4πR 3∆Kp(ω) + 4µw(ω) . (3) Here, ∆Kp(ω) = Kp(ω) − Kp,0 is the viscoelastic bulk modulus of the protein minus its bulk modulus Kp,0 at zero frequency. Further, µw(ω) is the shear modulus of water. Applying the fluctuation-dissipation theorem32 to Eq. (3), one gets h(δQω)2i = − 2kBT r2 3ωVp Im 1 3∆Kp(ω) + 4µw(ω) , (4) where Vp is the protein volume. We now proceed to calculating the Lamb-Mossbauer factor33,34 f = h(cid:12)(cid:12)heik0xi(cid:12)(cid:12) 2 ihet, (5) where x = k0 · (q + Q) is the projection of the iron displacement on the direction of photon propagation, k0 = k0/k0. There are two averages in this definition: the inner angular brackets denote an ensemble average over the protein and water modes affecting the position of iron in a single protein, while the outer angular brackets denote the average over the proteins in the sample. This second average carries the subscript "het" to emphasize that it reflects the heterogeneity of the sample, such as for instance variations in the hydration level among dif- ferent proteins in the protein powder. We do not consider the heterogeneous average in our present study and limit ourselves by the inner average only. This approxima- tion amounts, in experimental techniques, to considering the narrow feature of the Mossbauer absorption line and subtracting the broad base-line originating from the sam- ple heterogeneity.6 Accordingly, the experimental points shown in Fig. 1 are obtained from the area f (T ) of the narrow line as − ln f (T )/(k0)2, k0 = 53.2 A−1. The ensemble average over the water/protein statis- tical distribution can be described in terms of the free energy F (x) such that the inner brackets in Eq. (5) be- come heik0xi = Z −1Z eik0x−βF (x)dx, (6) where β = 1/(kBT ) and Z = R exp[−βF (x)]dx. The free energy F (x) is determined by projecting35 the manifold of q and Q coordinates on the single coordinate x 3 where the variance of q is given by Eq. (2) and the vari- ance of Q requires additional explanation. The limited instrumental time τobs affects the observ- ables and, in fact, the dynamical transition itself becomes possible only when the relaxation time of a collective mode of the hydration shell coupled to high-temperature protein's motions enters the experimental observation window.16,19 Therefore, the variance of the slow disper- sive motions of the heme iron is not a thermodynamic variable referring to an infinite observation window, but a property affected by instrumental resolution.26,36 This is reflected by the subscript ">" in Eq. (8) which indi- cates that h(δQ)2i> is calculated by integrating the re- sponse function in Eq. (4) over the frequencies exceeding the observation frequency ωobs = τ −1 obs h(δQ)2i> = Z ∞ ωobs h(δQω)2i(dω/π). (9) The statistical average over the electrostatic fluctua- tions can be simplified by a first-order expansion of the potential φw in x: φw(q, Q) ≃ φw,0 − xEw, where φw,0 is the potential at the equilibrium position of iron and Ew is the electric field projected on k0. Assuming that Ew is a Gaussian variable, one gets a Gaussian form of βF (x) = x2/(2h(δx)2i) where the variance h(δx)2i is h(δx)2i = h(δx)2iel/ME. (10) Here, the elastic msd h(δx)2iel = h(δq)2i + h(δQ)2i> is the sum of two statistically decoupled components. Fur- ther, the correction ME represents softening of atomic vibrations by electrostatic fluctuations of the hydration shell. Similarly to the viscoelastic effect, the electrostatic softening depends on the observation window. Account- ing again for the frequency cutoff introduced by the finite instrumental resolution, it is given in the form ME = 1 − (βz)2h(δx)2ielZ ∞ ωobs CE(ω)dω/(2π), (11) e−βF (x) = Z δ[x − k0 · (q + Q)] where CE(ω) is the Fourier transform of the time auto- correlation function of the field Ew(t) (7) he−βH0(q,Q)−βzφw(q,Q)iw dqdQ. In this equation, H0(q, Q) is the Hamiltonian of classical harmonic vibrations of the protein and φw is the elec- trostatic potential of the surrounding dielectric medium acting on the heme iron carrying charge z. The subscript "w" in Eq. (7) specifies water as the main source of the electrostatic fluctuations. It is not, however, required by the theory, and slow protein motions, not included in the calculation of h(δq)2i, can contribute to the fluctuations of the electrostatic potential as well (see below). The Hamiltonian H0(q, Q) can be given in the Gaus- sian form βH0(q, Q) = δq2 2h(δq)2i + δQ2 2h(δQ)2i> , (8) CE(ω) = Z ∞ −∞ hδEw(t)δEw(0)ieiωtdt. (12) By applying the fluctuation-dissipation theorem32 once again, one can recast Eq. (11) in terms of the response function χE(ω) representing the polar response to an os- cillating dipole probe m(t) = m0 exp(iωt) placed at the position of the iron atom ME = 1 − βz2h(δx)2ielZ ∞ ωobs χ′′ E(ω)dω/(πω). (13) This form of the correction factor accounting for the dipo- lar fluctuations of the water shell is used in the calcula- tions below. ) 2 Å ( 〉 ) 2 x δ ( 〈 0.04 0.02 0 0.1 0.08 0.06 0.04 0.02 0 0 100 200 300 100 200 T (K) 300 FIG. 1. Temperature dependence of the msd of iron in met- myoglobin. Experimental msd6 are shown by circles. Di- amonds refer to the msd calculated from the VDOS mea- sured from phonon-assisted Mossbauer effect at different temperatures,6 whereas the dashed line is the same calcula- tion from the VDOS at 235 K.27 The dash-dotted line refers to the msd combining phonons with viscoelastic oscillations of the protein shape and the solid line refers to the combination of all three effects: protein's phonons, viscoelastic shape oscil- lations, and dipolar fluctuations of the hydration layer. The solid line is obtained with ∆Kp(T ) as explained in the text, while the dotted lines refer to the temperature-independent ∆Kp = 3.2 GPa obtained from the fit of Eq. (17) to the protein intrinsic compressibility at 298 K.37 The inset shows the rise of the msd due to the elastic motions near the glass transition temperature of the protein, Tg ≃ 180 ± 15 K. III. MEAN SQUARE DISPLACEMENT Here we outline the calculations performed using the model developed in this paper. We need to mention that many parameters entering the model are not ex- perimentally available. Some of them can be potentially extracted from numerical simulations. The usefulness of simulations is, however, limited for interpreting the ex- perimental data since reproducing heterogeneous condi- tions of partially hydrated protein powders presents sig- nificant challenges to simulation protocols. Likewise, the viscoelastic model used here should be viewed as only a first step toward a more realistic description of the elastic response of hydrated proteins. However, one of the major conclusions of this paper is the dominance of electrostat- ics in the high-temperature flexibility of proteins and a relatively small effect of viscoelastic motions on the iron msd. This observation puts high priority to the develop- ment of the electrostatic component of the model, and makes the limitations of the modeling of the viscoelastic response less critical. The viscoelastic response functions entering Eq. (3) were taken in the Maxwell form32 ∆Kp(ω) = µw(ω) = ∆Kpiωτp(T ) 1 + iωτp(T ) G∞iωτw(T ) 1 + iωτw(T ) . , (14) In this equation, ∆Kp = Kp,∞ − Kp,0 is the change in the bulk protein modulus between infinite and zero fre- 4 quencies and G∞ is the high-frequency shear modulus of water; τp,w(T ) are the corresponding relaxation times. From the Maxwell equation, one gets the shear viscosity ηw(T ) = G∞(T )τw(T ) which is well tabulated down to the water nucleation temperature.38 The shear relaxation time was obtained from ηw(T ) and G∞(T )/GPa = 1.68− 0.0127(T − 273) taken from Ref. 39. The resulting τw(T ) turns out to be close to the exponential relaxation time of the longitudinal modulus extracted from inelastic x-ray scattering:40 τℓ(T )/s = 0.84 × 10−15 exp(1910 K/T ). The VDOS of metmyoglobin powders is well estab- lished by phonon-assisted Mossbauer measurements,6,8,27 and the temperature slope aq in Eq. (2) was calculated from the experimental VDOS (dashed line in in Fig. 1). Figure 1 also shows a rise in the msd due to the onset of viscoelastic oscillations of the protein (dash-dotted line). These results were obtained by adopting Eq. (14) for the elastic moduli with the protein's relaxation time τp(T ) from the measurements done on dry protein powders.22 The relaxation process in dry proteins is too slow to enter the observation window of the spectrometer and ∆Kp(ω) ≃ ∆Kp in Eq. (14). This notion implies that that the frequency depen- dence of the moduli, e.g., Debye vs dispersive relaxation does not significantly affect the outcome of the calcula- tions and only ∆Kp(T ) matters for the temperature de- pendence of the viscoelastic msd. The latter was adopted to reproduce the experimental intrinsic compressibility41 of myoglobin37 βT = 11.04 Mbar−1 at T = 298 K and the temperature variation of Young's moduli of myoglobin crystals at lower temperatures42,43 (see below). The use of experimental compressibility to parametrize the model also implies that the overall viscoelastic component of atomic displacements is limited by the thermodynamic experimental value and can only be lower for a given in- strumental observation window. The most uncertain part of our analysis is the calcu- lation of the electric field response function χE(ω). It can be calculated by solving the Poisson equation for the heme immersed in the heterogeneous dielectric formed by the protein and its hydration layer. However, the as- signment of the dielectric constants to both the protein and the thin layer of water surrounding it in the powder is subject to significant uncertainties. We will therefore restrict ourselves to rough estimates aimed to establish whether electrostatic fluctuations can produce a signifi- cant effect on the msd under a reasonable set of assump- tions. The results of fitting are additionally supported by Molecular Dynamics simulations of the fully hydrated metmyoglobin.44 The dielectric response to charges immersed in a po- lar medium is dominated by longitudinal modes of dipo- lar polarization45 characterized by the dielectric modulus ǫ(ω)−1, where ǫ(ω) is the complex, frequency-dependent dielectric constant of the protein-water mixture. The dielectric response function χE(ω) establishes the reac- tion field of the dielectric medium in response to a probe dipole placed at the position of the heme iron. It scales as ] ) s ( E τ [ 0 1 g o l - 10 5 0 Ref 49, h = 0.5 Ref 50, h = 0.3 Ref. 24, cryst. Ref 19, h = 0.4 3.5 4 103 K/T 4.5 5 ) 1 - r a b M ( T β 5 10 0.8 0.4 5 0 0 100 150 200 250 300 150 200 T (K) 250 300 FIG. 2. Dielectric longitudinal relaxation time τL(T ) = (ǫ∞/ǫs)τE(T ) reported in the literature19,47,48 with the pro- tein hydration levels indicated in the legend. The results from Ref.19 refer to myoglobin confined in PVA, results from Ref.24 are for metmyoglobin crystals, and Refs.47,48 refer to met- myoglobin powders. The solid line refers to the longitudinal, Debye relaxation time of the reaction field response function χE(ω) [Eq. (16)], τL(T )/s = 3.2 × 10−13 exp[3000 K/T ], used to fit the experimental msd for the heme iron.6 the inverse cube of the characteristic size d of the heme. The loss function χ′′ E(ω) can therefore be written in the form χ′′ E(ω) = 1 d3 ǫ′′(ω) ǫ(ω)2 . (15) Dielectric properties of partially hydrated proteins have not been well characterized since the results are strongly affected by both the sample preparation and the hydration level. Even dielectric relaxation times mea- sured on samples of close hydration level are rather incon- sistent. This point is illustrated in Fig. 2 where results on partially hydrated myoglobin powders of hydration level h = 0.3 − 0.5 (in g of water per g of protein) have been assembled.19,46 -- 48 We also show measurements done on myoglobin crystals24 for the sake of comparison. Multiple relaxation processes are common for such measurements, and the fastest relaxation, commonly attributed to the hydration shell,22,47 is shown in Fig. 2. Even if we could firmly establish the proper relaxation time for the the sample dipole moment, this would not necessarily give us the relaxation time of the reaction field correlation function required for χ′′ E(ω). Because of the linear scaling of the dipole moment variance with the number of dipoles, dielectric measurements empha- size the effect of outer solvation shells, while χ′′ E(ω) is dominated by waters closest to the probe dipole (heme's iron). In view of these uncertainties, we have constructed the temperature-dependent relaxation time τE(T ) that, together with the parameter d in Eq. (15), allows us to fit the experimental msd. Assuming the Debye form for the fast relaxation com- ponent in ǫ(ω),22,48 the response function in Eqs. (13) and (15) gains the form βz2χ′′ E(ω) = 1 δ2 ωτL 1 + (ωτL)2 , (16) FIG. 3. Intrinsic isothermal compressibility βT (T ) of a pro- tein calculated from Eq. (17) with ωobs = 1 (dashed line), 102 (solid line), and 103 MHz (dash-dotted line). The protein and water parameters are those adopted for the calculation of the myoglobin msd, with ∆Kp(T )/GPa = 3.22 − 0.03 × (T − 298)+0.00025×(T −298)2 obtained to fit the protein intrinsic compressibility at 298 K37 and the temperature dependence of Young's moduli (crosses in the inset). The inset shows experimental expansivity of the hydration shell of lysozyme (circles),51 experimental inverse Young's moduli of myoglobin crystals exposed to air of 95 -- 100% (diamonds)43 and 75% hu- midity (crosses).42 The solid line shows βT (T ) calculated from Eq. (17) at ωobs = 102 MHz; all curves are normalized to the corresponding values at T = 298 K. s ∞ − ǫ−1 where the parameter δ, δ−2 = βz2c0/d3 sets up a chara- tecteristic length and c0 = ǫ−1 is the Pekar factor. In addition, τL = (ǫ∞/ǫs)τE is the longitudinal dielectric time and ǫ∞ and ǫs are the high-frequency and static di- electric constants, respectively. The parameters δ = 0.12 A and τL(T )/s = 3.2 × 10−13 exp[3000 K/T ] (solid line in Fig. 2) were used in fitting the experimental msd. The fitting relaxation time τL(T ) is generally consistent with dielectric measurements and, in addition, the Arrhe- nius slope of τL(T ) matches our simulations of the pro- tein Stokes-shift dynamics at elevated temperatures.49 More detailed calculations might require replacing one- relaxation Debye dynamics in Eq. (16) with dispersive dynamics characterized by a distribution of relaxation time, as suggested by the NMR experiment.50 The overall fluctuations of the protein volume are de- termined by the response function in Eq. (3) combining the dynamic elastic moduli of the protein and the hydra- tion shell. This connection can be used to parameterize the model on volumetric properties of hydrated proteins, in particular on protein's intrinsic compressibility.41 For a given instrumental resolution, one obtains from Eq. (3) for the isothermal compressibility βT ∝ h(δVp)2i of the protein βT = −(6/π)Z ∞ ωobs Im [3∆Kp(ω) + 4µw(ω)]−1 (dω/ω). (17) In Fig. 3 we show βT (T ) for the parameters adopted in the calculations of the iron msd and several values of ωobs. The intrinsic compressibility of the protein rises sharply at the point close to protein's glass transition Tg. The latter depends on the observation window, but 250 200 150 100 ) 2 - Å ( ) ω ( ″ E χ 2 z β 50 F 0 0 0.2 P+W P W 0.4 0.6 ω (ns-1) 0.8 1 FIG. 4. The loss function βz2χ′′ E(ω) obtained from the fitting of the experimental msd to Eqs. (10), (13), and (16) (marked as "F") and from direct MD simulations of the electric field acting on the iron of metmyoglobin. The plot shows the result for the overall electric field produced by protein and water ("P+W") and by protein ("P") and water ("W") separately. The simulation trajectories (T = 300 K) were 45 ns long, 35 ns of which were used for collecting the correlation functions.44 is close to reported values Tg ≃ 180 ± 15 K43,51,52 marked by breaks in several observable parameters.52 The rise of compressibility at Tg in our calculations is caused by the water component of the viscoelastic response func- tion when the relaxation time τw becomes smaller than τobs. This result is consistent with the glass transition of the hydration shell expansivity51 shown in the inset in Fig. 3. The inset in Fig. 3 also shows compressibilities obtained from experimentally reported Young's moduli of myoglobin crystals42,43 assuming that their Poisson ratios are independent of temperature. The temperature variation of ∆Kp(T ) in our calculations shown in Fig. 1 was chosen to match these data. The fit of ∆Kp(T ) to crystalline Young's moduli results in an upward increase of the elastic msd at the highest temperatures shown in Fig. 1. This up- ward increase reflects pre-melting of myoglobin crys- tals when their Young's moduli approach zero.43 Since the melting temperature is typically higher in protein powders,43 ∆Kp(T ) obtained from fitting the crystal data might overestimate these effects; the iron msd with a temperature-independence ∆Kp = 3.2 GPa is shown by the dotted line in Fig. 1. The results shown in Fig. 1 indicate that electrostatic fluctuations far outweigh viscoelastic vibrations in the iron msd. We additionally confirm this outcome by com- paring the function βz2χ′′ E(ω) from our fitting to the same function obtained from MD simulation of the fully hydrated metmyoglobin.44 Figure 4 shows the response functions from the electric field fluctuations produced by the protein and water combined and by each component separately. The height of the maximum quantifies the strength of the msd modulation by the corresponding electrostatic component, and it is of main importance for this comparison. 6 cally non-polar, and its hydration shell is the main source of the electrostatic fluctuations. It does not need to be so. Low-frequency motions, not included in the VDOS used to caculate h(δq)2i, can modulate the protein's par- tial charges (dipole moments of α-helices, ionized sur- face residues, etc.) and compete in the elecrostatic noise with the hydration layer. This might be particularly true for partially hydrated protein powders where the fluctuations of the water dipoles are probably reduced to motions of polarized domains around ionized surface residues. Figure 4 in fact shows that the protein component of χ′′ E(ω) exceeds that of water, and its maximum is higher than that of the fitting function from Eq. (16). The elec- trostatic fluctuations of the protein itself are therefore sufficient to produce the observable msd and, in addition, our estimates do not seem to overestimate the effect of the electrostatic fluctuations on the msd. The primary role of water in powders might be reduced to ionizing the surface residues of the protein and plasticizing its motions above Tg (Fig. 3). Water in patches solvating ionized residues is strongly coupled to the protein both electrostatically and by surface hydrogen bonds. The re- laxation times of their electrostatic response functions are therefore close (Fig. 4), resulting in matching on- set temperatures of the dynamical transition for each component.14,15 Further, the overall loss function χ′′ E(ω), which in- cludes cross-correlations between the water and protein electric fields, shows a slower relaxation time than each component separately. The relaxation time of 6.3 ns of the essentially Debye overall function χ′′ E(ω) is close to τL(300 K) ≃ 7 ns adopted in fitting of the experimental msd. It is this loss function, combining the protein and water electrostatics, that is of primary interest for the modeling of the high-temperature flexibility of proteins. IV. DISCUSSION The picture presented here assigns an increase in the protein msd at the dynamical transition to the en- trance of a collective relaxation time of the protein- water interface into the observation window of the spectrometer.19,22,53 We consider two types of interfacial fluctuations, elastic modes changing the global shape of the protein and electrostatic fluctuations. Electrostat- ics turn out to be the main factor affecting the high- temperature portion of the msd. The longitudinal relaxation time of the electric field fluctuations, τL(T ), determines the transition tempera- ture by the condition ωobsτL(TD) ≃ 1. With the Arrhe- nius form for the relaxation time τL(T ), this condition predicts a logarithmic dependence of TD on the observa- tion frequency, TD ∝ ln[ωobsτ0]−1 , (18) We have assumed so far that the protein is electrostati- where τ0 is the preexponent in τL(T ). For instance, with 0.16 0.12 2 V e 0.08 0.04 0 100 150 T (K) 200 250 FIG. 5. Variance of the water's electrostatic potential at the active site, ∆q2h(δφ)2i of the protein plastocyanin from MD simulations (circles).55 Diamonds show the difference of water potentials in equilibrium with the active site carrying charges q1 and q2, β−1∆q(hφi1 − hφi2), ∆q = q2 − q1. This latter quantity is sensitive to high-frequency ballistic modes of the hydration water, but not to collective fluctuations of the shell dipole.56 The two calculations coincide in the linear response approximation, which is valid at low temperatures. Linear response breaks down when the collective mode of water's dipolar polarization enters the observation window fixed by the length of the simulation trajectory. The spike at ≃ 220 K in the potential variance carries signatures of a weak first- order transition, but its origin is currently unclear. the observation window of neutron scattering of ≃ 500 ps and of Mossbauer spectroscopy of 140 ns, the above equation yields 1.4 for the ratio of TD values measured by neutron and Mossbauer techniques (τ0 = 10−13 s). This estimate assumes equal electrostatic relaxation times for (mostly surface) protons and heme iron, which is likely not true. The actual picture is also more complex as several slope changes contribute to the overall tempera- ture dependence of the msd.54 It is also the case with the present model producing two different onsets arising from viscoelastic and electrostatic fluctuations. An increase in TD was also reported for proteins solvated in glycerol and in concentrated sucrose-water solutions.16 Although an increase in viscosity does shift TD in the right direc- tion according to Eq. (18), the alteration of the effective polarity of the hydration layer and the surface charge distribution of the protein might be other factors con- tributing to the shift. Generally, the the present model predicts a decrease in the protein atomic displacements for hydration in solvents of lower polarity. The main physical question looming behind the phe- nomenon of the dynamical transition is what are the mechanisms and physical modes allowing high flexibil- ity of proteins at physiological temperatures. We em- phasize here electrostatic fluctuations as the primary ori- gin of the increase in the protein's atomic displacements. This mechanism connects the translational manifold of the protein's interior to the dipolar orientational mani- fold of the hydration layer. While this connection was es- tablished empirically by experiment [Eq. (1)], numerical 7 simulations directly show the same basic phenomenology for the electrostatic fluctuations and the atomic msd. Figure 5 shows the results of numerical simulations for the variance of the electrostatic potential produced by the water hydration shell at the active site of the protein plastocyanin.55 A break in the temperature dependence at TD refers to the time-scale of ≃ 10 ns fixed by the length of the simulation trajectory. The difference of the first moments of the potential in the two redox states of the protein (diamonds in Fig. 5) gives the component of the same property produced by the ballistic dynam- ics of the hydration shell and not sensitive to its collec- tive relaxation.56 In a sense, the diamonds in Fig. 5 are analogs to the diamonds in Fig. 1 referring to the vibra- tional component of the msd.6 There is a clear qualitative similarity between laboratory and numerical results pre- sented in Figs. 1 and 5. Because the response function of the water's electric field scales as d−3 with the distance d from the sur- face inside the protein, dipolar fluctuations of the hy- dration shell will mostly affect protein's surface residues. The vibrations of the surface protons will therefore be softer than of interior protons, and they will stronger contribute to the observable msd. There is also a pos- sibility of "surface melting" when ME = 0 in Eq. (10) is reached with rising temperature for a group of atoms. The low-temperature conformation of the corresponding residues will become unstable, with the instability re- leased through a conformational transition. The present model predicts higher flexibility for atoms carrying higher partial charges. In case of heme iron this implies higher flexibility of the protein oxidized state compared to the reduced state. While this prediction qualitatively agrees with experiment,57,58 more detailed studies are required to distinguish the effect of electro- static fluctuations from the alteration of the VDOS also occurring upon changing the redox state. V. CONCLUSIONS The model proposed here treats high-temperature atomic displacements of the protein as a combination of viscoelastic deformation of the global protein shape and electrostatic fluctuations coupled to the atomic charge. We suggest that electrostatic fluctuations dominate the high-temperature flexibility of proteins. ACKNOWLEDGMENTS This research was supported by the National Science Foundation (CHE-0910905). We are grateful to Alexei Sokolov, Jan Swenson, and Guo Chen for communicat- ing results of their dielectric measurements to us. DVM has greatly benefited from useful discussions with Robert Young and Alexei Sokolov. 8 ∗ [email protected] 1 F. Parak and H. Formanek, Acta Crystallogr. A 27, 573 Phys. 132, 085103 (2010). 30 L. D. Landau and E. M. Lifshits, Theory of elasticity (El- (1971). sevier, Amsterdam, 1986) p. 18. 2 W. Doster, S. Cusack, and W. Petry, Nature 337, 754 31 R. M. Christensen, Theory of Viscoelasticity (Dover Pub- (1989). 3 G. Caliskan, R. M. Briber, D. Thirumalai, V. Garcia-Sakai, S. A. Woodson, and A. P. Sokolov, J. Am. Chem. Soc. 128, 32 (2006). 4 F. Gabel, D. Bicout, U. Lehnert, M. Tehei, M. Weik, and G. Zaccai, Quat. Rev. Biophys. 35, 327 (2002). 5 F. G. Parak, Rep. Prog. Phys. 66, 103 (2003). 6 F. G. Parak and K. Achterhold, J. Phys. Chem. Solids 66, 2257 (2005). 7 M. Diehl, W. Doster, W. Petry, and H. Schober, Biophys. J. 73, 2726 (1997). 8 K. Achterhold and F. G. Parak, J. Phys.: Condens. Matter lications, Inc., Mineola, N. Y., 2003). 32 J. P. Boon and S. Yip, Molecular Hydrodynamics (Dover Publications, Inc., New York, 1991). 33 K. S. Singwi and A. Sjolander, Phys. Rev. 120, 1093 (1960). 34 T. E. Cranshaw, B. W. Dale, G. O. Longworth, and C. E. Johnson, Mossbauer spectroscopy and its applications (Cambridge University Press, Cambridge, 1985). 35 P. M. Chaikin and T. C. Lubensky, Principles of condensed matter physics (Cambridge University Press, Cambridge, 1995). 36 E. W. Knapp, S. F. Fischer, and F. Parak, J. Chem. Phys. 15, S1683 (2003). 78, 4701 (1983). 9 M. Marconi, E. Cornicchi, G. Onori, and A. Paciaroni, 37 K. Mori, Y. Seki, Y. Yamada, and H. M. K. Soda, J. Chem. Chem. Phys. 345, 224 (2008). 10 K. Hinsen and G. R. Kneller, Proteins 70, 1235 (2008). 11 H. Frauenfelder, S. G. Sligar, and P. G. Wolynes, Science 254, 1598 (1991). 12 G. Zaccai, Science 288, 1604 (2000). 13 V. Kurkal, R. M. Daniel, J. L. Finney, M. Tehei, R. V. Dunn, and J. C. Smith, Chem. Phys. 317, 267 (2005). 14 K. Wood, A. Frolich, A. Paciaroni, M. Moulin, M. Hartlein, G. Zaccai, D. J. Tobias, and M. Weik, J. Am. Chem. Soc. 130, 4586 (2008). 15 X.-Q. Chu, A. Faraone, C. Kim, E. Fratini, P. Baglioni, J. B. Leao, and S.-H. Chen, J. Phys. Chem. B 113, 5001 (2009). 16 W. Doster, Biochim. Biophys. Acta 1804, 3 (2010). 17 P. W. Fenimore, H. Frauenfelder, B. H. McMahon, and Phys. 125, 054903 (2006). 38 K. R. Harris and L. A. Woolf, J. Chem. Eng. Data 49, 1064 (2004). 39 W. M. Slie, J. A. R. Donfor, and T. A. Litovitz, J. Chem. Phys. 44, 3712 (1966). 40 G. Monaco, A. Cunsolo, G. Ruocco, and F. Sette, Phys. Rev. E 60, 5505 (1999). 41 D. P. Kharakoz, Biophys. J. 79, 511 (2000). 42 V. N. Morozov and S. G. Gevorkian, Biopolymers 24, 1785 (1985). 43 V. N. Morozov and Y. Y. Morozova, J. Biomol. Struct. Dyn. 11, 459 (1993). 44 D. V. Matyushov, unpublished. 45 D. V. Matyushov, J. Chem. Phys. 120, 1375 (2004). 46 J. Swenson, H. Jansson, and R. Bergman, Phys. Rev. Lett. R. D. Young, Proc. Natl. Acad. Sci. 101, 14408 (2004). 96, 247802 (2006). 18 G. Chen, P. W. Fenimore, H. Frauenfelder, F. Mezei, 47 G. Schir`o, A. Cupane, E. Vitrano, and F. Bruni, J. Phys. J. Swenson, and R. D. Young, Phil. Mag. 88, 33 (2008). 19 H. Frauenfelder, G. Chen, J. Berendzen, P. W. Fenimore, H. Jansson, B. H. McMahon, I. R. Stroe, J. Swenson, and R. D. Young, Proc. Natl. Acad. Sci. 106, 5129 (2009). 20 M. D. Ediger, C. A. Angell, and S. R. Nagel, J. Phys. Chem. 100, 13200 (1996). Chem. B 113, 9606 (2009). 48 M. Bonura, G. Schir`o, and A. Cupane, Spectroscopy 24, 143 (2010). 49 D. N. LeBard, V. Kapko, and D. V. Matyushov, J. Phys. Chem. B 112, 10322 (2008). 50 S. A. Lusceac, M. R. Vogel, and C. R. Herbers, Biochim. 21 Y. He, P. I. Ku, J. R. Knab, J. Y. Chen, and A. G. Markelz, Biophys. Acta 1804, 41 (2010). Phys. Rev. Lett. 101, 178103 (2008). 22 S. Khodadadi, S. Pawlus, J. H. Roh, V. G. Sakai, E. Ma- montov, and A. P. Sokolov, J. Chem. Phys. 128, 195106 (2008). 23 M. Tarek and D. J. Tobias, Phys. Rev. Lett. 88, 138101 (2002). 51 W. Doster, S. Busch, A. M. Gaspar, M.-S. Appavou, J. Wuttke, and H. Scheer, Phys. Rev. Lett. 104, 098101 (2010). 52 S. Khodadadi, A. Malkovskiy, A. Kisliuk, and A. P. Sokolov, Biochim. Biophys. Acta 1804, 15 (2010). 53 R. M. Daniel, J. L. Finney, and J. C. Smith, Faraday Dis- 24 G. P. Singh, F. Parak, S. Hunklinger, and K. Dransfeld, cuss. 122, 163 (2002). Phys. Rev. Lett. 47, 685 (1981). 54 M. Krishnan, V. Kurkal-Siebert, and J. Smith, J. Phys. 25 S. Magaz`u, F. Migliardo, and A. Benedetto, J. Phys. Chem. B 112, 5522 (2008). Chem. B 114, 9268 (2010). 26 D. V. Matyushov, J. Chem. Phys. 130, 164522 (2009). 27 K. Achterhold, C. Keppler, A. Ostermann, U. van Burck, W. Sturhahn, E. E. Alp, and F. G. Parak, Phys. Rev. E 65, 051916 (2002). 28 B. M. Leu, Y. Zhang, L. Bu, J. E. Straub, J. Zhao, W. Sturhahn, E. E. Alp, and J. T. Sage, Biophys. J. 95, 5874 (2008). 29 B. M. Leu, A. Alatas, H. Sinn, E. E. Alp, A. H. Said, H. Yavas, J. Zhao, J. T. Sage, and W. Sturhahn, J. Chem. 55 D. N. LeBard and D. V. Matyushov, Phys. Rev. E 78, 061901 (2008). 56 D. N. LeBard and D. V. Matyushov, J. Phys. Chem. B 114, 9246 (2010). 57 E. N. Frolov, R. Gvosdev, V. I. Goldanskii, and F. G. Parak, J. Biol. Inorg. Chem. 2, 710 (1997). 58 A. M. Jorgensøn, F. Parak, and H. E. M. Christensen, Phys. Chem. Chem. Phys. 7, 3472 (2005).
1203.4666	1	1203	2012-03-21T07:44:34	Active contractility in actomyosin networks	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.CB" ]	Contractile forces are essential for many developmental processes involving cell shape change and tissue deformation. Recent experiments on reconstituted actomyosin networks, the major component of the contractile machinery, have shown that active contractility occurs above a threshold motor concentration and within a window of crosslink concentration. We present a microscopic dynamic model that incorporates two essential aspects of actomyosin self-organization: the asymmetric load response of individual actin filaments and the correlated motor-driven events mimicking myosin-induced filament sliding. Using computer simulations we examine how the concentration and susceptibility of motors contribute to their collective behavior and interplay with the network connectivity to regulate macroscopic contractility. Our model is shown to capture the formation and dynamics of contractile structures and agree with the observed dependence of active contractility on microscopic parameters including the contractility onset. Cooperative action of load-resisting motors in a force-percolating structure integrates local contraction/buckling events into a global contractile state via an active coarsening process, in contrast to the flow transition driven by uncorrelated kicks of susceptible motors.	physics.bio-ph	physics	Active contractility in actomyosin networks Shenshen Wang1, 2 and Peter G. Wolynes1, 2 1Department of Physics, Department of Chemistry and Biochemistry, and Center for Theoretical Biological Physics, University of California, San Diego, La Jolla, CA 92093, USA 2Department of Chemistry, and Center for Theoretical Biological Physics, Rice University, Houston, TX 77005, USA (Dated: November 1, 2018) Abstract Contractile forces are essential for many developmental processes involving cell shape change and tissue deformation. Recent experiments on reconstituted actomyosin networks, the major compo- nent of the contractile machinery, have shown that active contractility occurs above a threshold motor concentration and within a window of crosslink concentration. We present a microscopic dynamic model that incorporates two essential aspects of actomyosin self-organization: the asym- metric load response of individual actin filaments and the correlated motor-driven events mimicking myosin-induced filament sliding. Using computer simulations we examine how the concentration and susceptibility of motors contribute to their collective behavior and interplay with the network connectivity to regulate macroscopic contractility. Our model is shown to capture the forma- tion and dynamics of contractile structures and agree with the observed dependence of active contractility on microscopic parameters including the contractility onset. Cooperative action of load-resisting motors in a force-percolating structure integrates local contraction/buckling events into a global contractile state via an active coarsening process, in contrast to the flow transition driven by uncorrelated kicks of susceptible motors. 2 1 0 2 r a M 1 2 ] h p - o i b . s c i s y h p [ 1 v 6 6 6 4 . 3 0 2 1 : v i X r a 1 I. INTRODUCTION Contractile forces are essential for many processes vital to development, ranging from cytokinesis and cell motility1 to wound healing and gastrulation2. Networks of filamentous actin (F-actin) and the molecular motor, type II myosin have been identified as the major components of the contractile machinery. The actin network provides a structural scaffold on which the myosin motors move, powered by ATP hydrolysis. Actomyosin networks generate contractile forces through the activity of myosin motors, which themselves assemble into bipolar minifilaments that generate sustained sliding of neighboring actin filaments relative to each other in order to reorganize F-actin networks and generate tension.3 When coupled to the cell substrate or using cell-cell adhesions, contractile actomyosin networks transmit forces to their environment. In addition to the microtubule-kinesin system, another important filament-motor as- sembly in cells that forms well-focused mitotic spindle poles driven by polarity sorting mechanism4,5 to accomplish high-accuracy segregation of duplicated chromosomes, acto- myosin condensates appear in diverse tissues and organisms as transient structures that coalesce into still larger arrays that exert contractile forces. Examples include the contrac- tile rings driving cytokinesis and wound healing, and the contractile networks that deform epithelial cell layers in developing embryos and drive polarizing cortical flows6,7. Some recent theoretical efforts have modeled the contractile actin cortex as an active po- lar gel and have derived effective continuum theories within a hydrodynamic framework.8 -- 13 These macroscopic approaches based on generic symmetry considerations predict the for- mation of diverse patterns in acto-myosin gels such as asters and ring-like structures, which have been observed in studies in vitro.14 Recently Bendix and coworkers have reconstituted contractility in a simplified system of F-actin, muscle myosin II motors, and α-actinin crosslinks.15 The well-controlled na- ture of this in vitro system allows a systematic study of the dependence of contractility on microscopic parameters, such as the number and activity of myosin motors, crosslink density and actin network connectivity. It has been shown that contractility occurs above a threshold motor concentration and within a window of crosslink concentrations. Whereas earlier experiments on purified actomyosin solutions have established that contraction of F-actin networks by myosin II motors at physiological ATP level requires the presence of F- 2 actin crosslinks,16,17 as has been confirmed by recent theoretical work,18 this newly observed non-monotonic dependence of contraction tendency on crosslinking strength still calls for explanation. There are two important aspects in actomyosin self-organization: (1) Actin filaments have a highly asymmetric response to axial loading: They strongly resist tensile forces but easily buckle under compressive loads of several piconewtons. The ability to sustain large tension allows the motor-induced stresses to propagate significant distances through the network, whereas the buckling instability promotes formation of local actomyosin aggregates that coalesce into larger arrays exerting contractile forces (as exemplified in the spontaneous for- mation of myosin foci starting from a uniform distribution in an isotropic actin network19,20). (2) Motor-induced movements come in correlated pairs: the motor acts on a pair of parallel filaments to slide them past one another, inducing a pair of equal and oppositely directed moves at the crosslinks which are thus pulled together. Despite the success of hydrody- namic theories in predicting diverse patterns, a more microscopic model that could capture the nonlinear buckling behavior and the correlated motor-driven events, both of which are crucial for actomyosin self-organization and active contractility, is needed. Here we provide a microscopic dynamic model for active contractility that combines the motor-driven stochastic processes (modeled as correlated kicks on motor-bonded crosslinks) with the asymmetric load response of individual actin filaments. This minimal model will be shown to exhibit the experimentally observed dependence of macroscopic contraction on motor concentration and actin network connectivity. The model also highlights the key role, in structural development, of motor susceptibility, a parameter characterizing how sensitively the motors respond to imposed forces. By performing dynamic Monte Carlo simulations, we investigate the formation and dynamics of nonequilibrium structures in an actomyosin network modeled as a "cat's cradle"21,22 consisting of crosslinked nonlinear-elastic filaments subject to anti-correlated kicks on motor-bonded crosslinks (see Fig. 1 for a schematic illustration). We first study how the concentration and susceptibility of motors determine the collective behavior gen- erating diverse patterns. We further construct a phase diagram for active contractility as a function of motor concentration and network connectivity at a given motor susceptibil- ity. This diagram identifies the threshold motor concentrations and contains a window of network connectivity for active macroscopic contraction, consistent with observation.15 We 3 also find that at high connectivity contraction can still occur (at intermediate motor con- centrations) but only if the excluded volume effect is negligibly small. We finally compare the structures that develop for systems with correlated kicks to those when there are only uncorrelated kicks as in our earlier work23. In particular, under uphill-prone motor kicks, the formation of "asters" is replaced by the formation of disordered condensates resulting from an active multistage aggregation process due to buckling of connected actin structures induced by cooperative anti-correlated kicks. This prediction receives support from recent in vitro experimental studies on how the collective action of myosin motors organizes actin filaments into contractile structures.20 Multiple factors that cause dynamic remodeling in cells are clearly absent in the re- constituted assay,15 such as the disassembly of contractile structures and the transience of physiological actin cross-linking proteins. In our numerical model, we likewise assume that network connectivity and motor distribution are quenched once initially assigned, in line with the fact that in vitro structures are irreversibly assembled because many in vivo fac- tors allowing fast pattern renewals are left out. Therefore the nonequilibrium dynamics and structures exhibited in our numerical study arise solely from the intrinsic activity of motors firmly built into the actin network driving correlated movements stochastically. This con- trasts with the two-fluid model that treats the cytoskeletal network as an elastic continuum where motor (un)binding kinetics leads to enhanced low-frequency stress fluctuations.10 The discreteness of the power strokes and thus the kick steps in our model bears relevance to the pulsed contraction observed in actomyosin networks in vivo.24 4 II. MODEL A. Model system and dynamic rule We model the actomyosin network as a cat's cradle21,22 consisting of nonlinear elastic filaments built on a three dimensional random lattice of volumeless crosslinks (Fig. 1). These nonlinear filaments stretch elastically with effective stiffness βγ when their con- tour length r exceeds the relaxed length Le but buckle and become floppy upon short- ening, as described by the pair interaction potential between bonded neighboring crosslinks βU(r) = Θ(r − Le)βγ(r − Le)2/2 where β = 1/kBT and Θ(·) is the Heaviside step func- tion. The assumed weakness of the excluded volume effect allows large-scale structural rearrangements. We assign two mean field parameters to characterize the architecture of this filament-motor assembly: (1) the network connectivity, Pc, which denotes the fraction of nearest-neighbor pairs of crosslinks bonded by filaments; (2) the motor concentration, Pa, which indicates the fraction of active bonds, i.e. those attached by motors which induce equal and oppositely directed kicks on the connected crosslink pair. Network connectivity and motor concentration determine the fraction of active nodes against passive nodes: active nodes have motor-attached bonds and are subject to anti-correlated kicks with their motor- bonded neighbors whereas passive nodes have no motor-attached bonds and only undergo Brownian motion. To mimic the motor-driven filament sliding in actomyosin networks we describe the mo- tors as generating anti-correlated kicks on pairs of crosslinks along their lines of centers. Assuming a fixed kick step size l, consistent with the nearly periodic structure of the actin filaments, an anti-correlated kick pair acting on nodes i and j can be represented by (~lij,~lj i) = l(rij, −rij), where rij is a unit vector pointing from node i to node j. These anti-correlated kick pairs with equal size automatically satisfy momentum conservation on the macroscopic scale. Yet if we include explicitly the aqueous environment in which the ac- tomyosin network is immersed, hydrodynamic interactions between the nodes via the solvent should be taken into account. These interactions might modify the current simplified pic- ture and counteract any motor-induced force imbalance on individual nodes, thus validating momentum conservation on the microscopic scale. Dynamical evolution of the many-particle configuration {~ri} due to these motor-driven 5 nonequilibrium processes can be described by a master equation ∂Ψ/∂t = LNEΨ for the configurational probability density Ψ({~ri}; t) with LNEΨ({~r}, t) = R Πid~r′ i[K({~r′} → {~r})Ψ({~r′}, t) − K({~r} → {~r′})Ψ({~r}, t)] where the integral kernel K({~r′ i} → {~ri}) encodes the probability of transitions between different crosslink/node configurations. Our earlier description23,25 of the motor kicking rate, k, still applies to the current case of correlated kicks, i.e., k = κ[Θ(∆U) exp(−suβ∆U) + Θ(−∆U) exp(−sdβ∆U)], (1) where κ is the basal kicking rate and su(sd) denotes motor susceptibility to energetically uphill (downhill) moves, except that the free energy change ∆U now arises from pairs of displacements. Explicitly we write LNEΨ({~ri}; t) = ×nδ(~ri − ~r′ −δ(~ri − ~r′ 1 2 κXi Xj i − ~lij)δ(~rj − ~r′ i + ~lij)δ(~rj − ~r′ j iZ d~r′ Cij Z d~r′ j + ~lij)whU(· · · , ~r′ j − ~lij)whU(· · · , ~ri, · · · , ~rj, · · · ) − U(· · · , ~r′ i, · · · , ~r′ j, · · · ) − U(· · · , ~ri, · · · , ~rj, · · · )i Ψ({~r′ i}; t) i, · · · , ~r′ j, · · · )i Ψ({~ri}; t)o. The factor 1/2 avoids double counting in the summation over all pairs. The quantity Cij, much like an element of a contact map in description of protein structures, defines whether the node pair (i, j) is connected by an active bond and thus subject to anti-correlated displacements (~lij, −~lij): Cij = Cji = 1 for motor-bonded pairs while Cij = Cji = 0 for non- bonded pairs. Our description of the rates gives w[Ui − Uf ] = Θ(Uf − Ui) exp[−suβ(Uf − Ui)] + Θ(Ui − Uf ) exp[−sdβ(Uf − Ui)]. Assuming symmetric motor susceptibility, i.e. su = sd = s, one finds more simply LNEΨ({~ri}; t) = 1 2 κXi Xj Cij ×ne−sβ[U (~ri,~rj)−U (~ri−~lij ,~rj+~lij )]Ψ({· · · , ~r′ i = ~ri − ~lij, · · · , ~r′ −e−sβ[U (~ri+~lij ,~rj −~lij )−U (~ri,~rj)]Ψ({· · · , ~ri, · · · , ~rj, · · · }; t)o. j = ~rj + ~lij, · · · }; t) (2) We assume that kicks on different node pairs at any time are uncorrelated. The rates of possible kicking events depend on the instantaneous node configuration reflecting an assumed Markovian character of the dynamics. There is no angular average due to the definiteness of kicking directions for a given configuration. 6 Note that the motor power strokes and thus kick steps are discrete occurring in a stochas- tic fashion. The correlated motions pull in slack locally while pulling taut neighboring fil- aments until a global balance is reached or a macroscopic collapse occurs, depending on whether the motors are downhill-prone (with a large positive s) or load-resisting (with a small or negative s), respectively. The latter may be relevant to the contractile rachet-like behavior24 that operates to incrementally drive cell shape change and deform tissues. B. Numerical translation: dynamic Monte Carlo simulation To realize the finite-jump Markov process described by the (chemical) master equation (Eq. 2) we performed dynamic Monte Carlo26 simulations on the model actomyosin network (Fig. 1). In these simulations we generated initially a three-dimensional random lattice of volumeless nodes (mimicking the crosslinking proteins) and connected the nearest-neighbor nodes (defined by the first shell of the pair distribution function) with nonlinear elastic bonds22 (mimicking the actin filaments) at a given probability Pc. We then distributed the myosin motors uniformly to the bonds at a given probability Pa and obtained an active bond map. Considering the anti-correlated kicks along individual active bonds as chemical reac- tion channels, we adopted a stochastic simulation algorithm26 to execute the moves following the stochastic process defined by the model motor kicking noise (Eq. 1). For a sufficiently large system, kicking events on different node pairs are effectively decoupled (consistent with summing over independent reaction channels in the master equation). Intermediate thermal moves between successive chemical moves obey Brownian dynamics27 implemented via the position Langevin equation. The bond properties are given by the elasticity onset or relaxed length of actin filaments Le = 1.2 and the effective stretch modulus βγ = 2. Since the relaxed length is larger than the mean node separation (set as the length unit), the initial homogeneous (but amorphous) network has a considerable fraction of floppy bonds. We assumed a relatively high basal kicking rate (κ = 0.01) and a large kick step size (l = 0.2) such that the dimensionless motor activity (defined as κl2/D0 where D0 is the thermal diffusion constant) is close to 1 and thus the strength of chemical noise is at least comparable to that of the thermal noise. The system size is N = 256 and periodic boundary conditions are applied. The relevant biophysical parameters to vary include network connectivity (Pc) as well as motor 7 concentration (Pa) and susceptibility (s). III. MAIN FINDINGS A. Role of motor susceptibility (s) and concentration (Pa) in contractile behavior We first study how the concentration (Pa) and susceptibility (s) of motors contribute to the collective behavior. As illustrated in Fig. 2, depending on the specific combinations of Pa and s, distinct nonequilibrium structures emerge. For a force-percolating network (i.e. one with connectivity beyond the percolation threshold) when partially motorized (Pa < 1) un- der susceptible (large s) anti-correlated kicks, the active nodes (those with motor-attached bonds; shown as red spheres) begin to aggregate and tend to separate from the passive nodes (those with no active bonds; shown as blue spheres). The left panel of Fig. 2a shows snapshots of both the initial and later node configurations. The corresponding developed network structure (Fig. 2a right) exhibits clumps of floppy bonds (concentrated short green lines) connected by tense bonds (long red lines). The overall rigidity (i.e. homogeneity on large scales) of the structure is protected by susceptible motors which tune the balance between local bond contraction and neighboring bond stretching such that energetically un- favorable tense states are avoided. At a low concentration of adamant motors (small s), however, active nodes and their aggregates tend to "glue" together progressively the pas- sive nodes and their condensates (Fig. 2b). A finite spatial extent of the condensate and a non-vanishing fraction of taut bonds remain due to the insufficient cooperativity (low Pa) between local aggregation events. More dramatically, significant spatial heterogeneity forms when the system is driven by a large number of uphill-prone motors with negative suscepti- bility: the cooperative action of load-resisting motors induces a multistage aggregation and coarsening of the nodes (Fig. 2c upper row) finally leading to a macroscopic contraction of an initially homogeneous network into a dense clump of buckled filaments (Fig. 2c lower row). This multistage coarsening process involves three steps: (1) local bond contraction and node aggregation giving floppy clumps connected by tense filaments; (2) coarsening of the aggregates leading to filament alignment and formation of tense bundles; (3) coalescence of the larger aggregates into a single condensate accompanies collapse of the tense bundles into a floppy clump. The aligned tense bundles formed before the eventual collapse constitute a 8 taut state that can generate contractile forces. We will investigate the scenario of motor-driven aggregation and pursue its analogy to arrested phase separation later. In the present work we will focus on the regime for macro- scopic contraction. To identify the required motor properties for active contractility, we performed many simulations to obtain the evolution (in Monte Carlo time tMC) of the statistical character- istics for a series of motor susceptibilities (Fig. S1a) and concentrations (Fig. S1b). These measures consistently indicate the existence of a threshold motor concentration ((Pa)th) and a threshold susceptibility (sth) for the onset of macroscopic contraction within the simulation time window. When s is less than sth at an intermediate Pa value (Fig. S1a: s ≤ 0.02, Pa = 0.5) or for Pa > (Pa)th at a small s value (Fig. S1b: Pa ≥ 0.3, s = 0), the fraction of taut bonds drops to essentially zero, indicating that an initially homogeneous percolating network collapses into a floppy clump. More interestingly, the total energy first rapidly rises and reaches a maximum before falling to zero with the fraction of taut bonds. This nonmonotonic behavior suggests that the system first works against an energy barrier due to the formation of a transient tense state (having highly stretched bundles induced by adamant or uphill-prone motor kicks), and then cooperative action of sufficiently load- resisting motors drives the system over the barrier to allow energetically downhill moves via subsequent coarsening (as shown in Fig. 2c). The rapid increase and saturation of the mean squared displacement (MSD) mirrors the evolution of the fraction of taut bonds and results from the formation of a single isolated floppy clump. Larger s and/or higher Pa yields a lower barrier to the collapsed state (as indicated by arrows in Fig. S1). Moreover, lower Pa necessitates having a smaller s to induce macroscopic contraction. In other words, weaker motor cooperativity requires a stronger load-resisting tendency to trigger contractile instability. When Pa < (Pa)th and/or s > sth the structure remains homogeneous except for modest local node aggregation and network deformation. B. State diagram for active contractility: Interplay of network connectivity and motor cooperativity We now examine the interplay of network connectivity (Pc) and motor concentration (Pa) in forming contractile structures. The equivalence between the crosslink concentration and 9 the fraction of bonded neighboring crosslinks Pc (both are proportional to the number of inter-crosslink segments and define the number of bond constraints), and between the motor concentration and the fraction of active bonds Pa (both are proportional to the number of crosslinks subject to motor kicks and determine the spatial cooperativity between motors) allows us to compare our state diagram constructed on the Pa-Pc plane with the experimental result15 shown for the parameter space of concentration ratios [myosin]/[actin] versus [α- actinin]/[actin]. We present our state diagram showing the dependence of macroscopic contractility on the network connectivity and motor concentration at a small s value (s = 0.01) in Fig. 3. Red crosses denote contractile networks while blue circles denote non-contractile networks. By "contractile" we mean a complete collapse of an initially homogeneous network into a floppy clump within 107 MC steps, monitored by the vanishing of total potential energy and the fraction of taut bonds. On top of the figure, we display the initial network structures for several typical values of network connectivity Pc with the average number of bonded neighbors z. As observed experimentally,15 we identify a threshold motor concentration and a window of network connectivity for active contractility, which define a parameter region as marked by the purple open frame. The new feature is a small window of motor concen- tration for contraction at high connectivity (marked by a green closed frame). These two aspects vividly demonstrate the interplay of network connectivity and motor concentration for global contraction: at any connectivity beyond the percolation threshold (Pc ≥ 0.2), a sufficiently high motor concentration is required to achieve cooperativity among local con- traction events; on the other hand, since the bond constraints are strong at high connectivity, motor concentration cannot be too high since force asymmetry (imbalanced tug-of-war) is necessary to trigger local contraction. In the in vitro experiments,15 no macroscopic contrac- tion was observed on the hour time scale at high crosslink concentrations. This observation is not incompatible with our results since inclusion of excluded volume effects should give a dramatic slow-down of the contractile dynamics owing to jamming and/or glass transition which would account for the absence of observable contraction on laboratory time scales. Outside the framed regions, at low motor concentration and/or low connectivity as well as at high motor concentration and high connectivity, there is no macroscopic contraction. We illustrate the failure of contractility at low connectivity (Fig. S2a) or low motor concentration (Fig. S2b). For Pc as low as 0.1, the average number of bonded neighbors is no greater than 10 1, lack of tension percolation thus prevents global contraction. This becomes more obvious as we increase the motor concentration; an increasing trend of local collapse with rising Pa value is apparent (Fig. S2a), leading to more compact aggregates and disconnected floppy clumps (several typical spots have been circled for the highest Pa case). At very low motor concentration Pa = 0.1 (Fig. S2b), formation of sparse and small active-node foci did not dramatically reshape the network, since rare and separate contraction events are insufficient to trigger global contractile instability. A quantitative demonstration of the interplay between Pc and Pa is given in Fig. S3. Macroscopic contraction occurs either for high motor concentration (upper row, Pa = 0.7) at intermediate connectivity (Pc = 0.3, 0.5) or for high connectivity (lower row, Pc = 0.7) at intermediate motor concentrations (Pa = 0.3, 0.5). Lack of percolation at Pc = 0.1 is signalled by the diffusive behavior of the mean square node displacement MSD linearly in- creasing with tMC (red line indicated by arrow in Fig. S3a right panel) since the disconnected aggregates merely undergo thermal motion. On the other hand, a small steady value for the MSD at Pa = 0.1 (red line indicated by arrow in Fig. S3b right panel) reflects the low cooperativity which causes no more than modest local distortions. A balanced tug-of-war at high Pc and high Pa disfavors even local deformations yielding a lower fraction of taut bonds, lower energy and smaller MSD as Pa increases. This can be seen by comparing the measures for Pa = 0.1 and Pa = 0.7 at Pc = 0.7 in Fig. S3b. C. Contrast with the case for uncorrelated kicks Our earlier study of a model cytoskeleton with uncorrelated isotropic kicks acting on in- dividual nodes23 revealed that force percolation and mechanochemical coupling due to sus- ceptible motors can conspire to maintain a spontaneous flow, whereas adamant motor kicks promote fluidization (characterized by a vanishing localization strength of the nodes and formation of a disordered tense network). Under anti-correlated kicks acting along the lines of centers of motor-bonded node pairs, however, no vectorial flow transition is found even for a very large kick step size (l ≃ 0.5). This is probably because the restrictiveness of (local) kicking directions for a given configuration impedes a global concerted net movement of the whole lattice. Instead, balanced local contraction and neighboring stretch in the presence of force percolation results in a network of floppy spots connected by tense filaments which 11 remains homogeneous on large scales and bears some resemblance to arrested phase separa- tion occurring in low-packing-fraction physical gels with strong short-range attractions28. At intermediate connectivity, force imbalance sensed by susceptible motors still induces phase separation into large floppy clumps connected by taut inter-clump bonds yet oscillations are no longer found. Fluidization is replaced by global contraction at above-threshold motor concentration. When the uncorrelated-kicking system was driven by uphill-prone motors with negative susceptibility, aster-like patterns formed (Fig. 4 left). Instead, now, multistage coarsening and eventual macroscopic collapse occur (Fig. 4 right), after surmounting a high energy bar- rier caused by transient tense states with stretched bundles. Comparing the bond tension patterns for these two cases in Fig. 4, we see that global contraction requires correlated move- ments that locally buckle a filament yet impose strain on the bonds at both ends. Without this correlation the aster pattern cannot collapse. We display in Fig. S4 the snapshots in the course of aster formation which clearly demonstrate that node aggregation and coarsening leads to progressively more tight filament bundling and bundle alignment, a mechanism that operates for both the correlated- and uncorrelated-kick cases. IV. CONCLUSION AND DISCUSSION We have simulated a microscopic model for the actomyosin cytoskeleton as a motorized cat's cradle that combines the asymmetric load response of individual actin filaments with correlated motor-driven events. This model reproduces the dependence of active contractility on microscopic parameters observed in reconstituted actomyosin networks. The simulations allow us to identify several necessary conditions for active contractility: highly asymmetric load responses of the filaments are needed for local contractile behavior under anti-correlated kicks, and a minimal network structure beyond percolation threshold is required to prop- agate local contraction. Sufficiently cooperative load-resisting motors manage to drive the system through energy-costly intermediate states and incorporate local buckling events into macroscopic contraction via a multistage coarsening process. This numerical study provides an explanation for the formation and contractile dynamics of disordered condensed state of actomyosin in vivo. The fact that such a simplified model is able to mimic cellular self-organization states and their contractile dynamics suggests that 12 purely physical interactions contribute to the regulation of cell and tissue morphogenesis. Nevertheless specific biochemical signaling events29 certainly contribute to the localized as- sembly and activation of myosin foci such as occurs in cleavage furrow during cytokinesis and in wound borders. It is clearly necessary in the future to take into account the excluded volume effect which is expected to dramatically slow the contractile dynamics at high crosslink concentrations, thus accounting for the failure to see contractility on laboratory time scales. Long-range hydrodynamic interactions between the node pairs may also change the qualitative physics. Hydrodynamic correlations should facilitate propagation of nearby kicking events and break the continuous symmetry so as to allow spontaneous directed motion with motor-driven hydrodynamics-mediated pulsed contractions. Finally, in vivo one must incorporate motor attachment and detachment and crosslink binding and unbinding for complete realism. Support from the Center for Theoretical Biological Physics sponsored by the National Science Foundation (Grant PHY-0822283) is gratefully acknowledged. 1 T. D. Pollard and G. G. Borisy. Cellular motility driven by assembly and disassembly of actin filaments. Cell 112, 453-465 (2003). 2 J. Holtfreter. A study of the mechanics of gastrulation (part 1). J. Exp. Zool. 94, 261-318 (1943). 3 A. C. Martin. Pulsation and stabilization: contractile forces that underlie morphogenesis. Dev. Biol. 341, 114-125 (2010). 4 T. Surrey, F. N´ed´elec, S. Leibler, and E. Karsenti. Physical properties determining self- organization of motors and microtubules. Science 292, 1167-1171 (2001). 5 F. N´ed´elec, T. Surrey, A. C. Maggs, and S. Leibler. Self-organization of microtubules and motors. Nature 389, 305-308 (1997). 6 S. Hird and J. White. Cortical and cytoplasmic flow polarity in early embryonic cells of Caenorhabditis elegans. J. Cell Biol. 121, 1343-1355 (1993). 7 M. Mayer, M. Depken, J. S. Bois, F. Julicher, and S. W. Grill. Anisotropies in cortical tension reveal the physical basis of polarizing cortical flows. Nature 467, 617-621 (2010). 8 K. Kruse, J. F. Joanny, F. Julicher, J. Prost, and K. Sekimoto. Asters, vortices, and rotating spirals in active gels of polar filaments. Phys. Rev. Lett. 92, 078101 (2004). 13 9 R. Voituriez, J. F. Joanny, and J. Prost. Generic phase diagram of active polar films. Phys. Rev. Lett. 96, 028102 (2006). 10 F. C. MacKintosh and A. J. Levine. Nonequilibrium mechanics and dynamics of motor-activated gels. Phys. Rev. Lett. 100, 018104 (2008). A. J. Levine and F. C. MacKintosh. The mechanics and fluctuation spectrum of active gels. J. Phys. Chem. B 113, 3820-3830 (2009). 11 K. Kruse and F. Julicher. Self-organization and mechanical properties of active filament bundles. Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 67, 051913 (2003). 12 A. Lau, B. Hoffmann, A. Davies, J. Crocker, and T. Lubensky. Microrheology, stress fluctua- tions, and active behavior of living cells. Phys. Rev. Lett. 91, 198101-198104 (2003). 13 T. Liverpool and M. Marchetti. Instabilities of isotropic solutions of active polar filaments. Phys. Rev. Lett. 90, 138102 (2003). 14 F. Backouche, L. Haviv, D. Groswasser, and A. Bernheim-Groswasser. Active gels: Dynamics of patterning and self-organization. Phys. Biol. 3, 264-273 (2006). 15 P. M. Bendix, G. H. Koenderink, D. Cuvelier, Z. Dogic, B. N. Koeleman, W. M. Briehera, C. M. Fielda, L. Mahadevan, and D. A. Weitz. A quantitative analysis of contractility in active cytoskeletal protein networks. Biophys. J. 94, 3126-3136 (2008). 16 L. W. Janson, J. Kolega, and D. L. Taylor. Modulation of contraction by gelation/solation in a reconstituted motile model. J. Cell Biol. 114, 1005-1015 (1991). 17 R. E. Kane. Interconversion of structural and contractile actin gels by insertion of myosin during assembly. J. Cell Biol. 97, 1745-1752 (1983). 18 A. E. Carlsson. Contractile stress generated by actomyosin gels. Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 74, 051912 (2006). 19 G. H. Koenderink, Z. Dogic, F. Nakamura, P. M. Bendix, F. C. MacKintosh, J. H. Hartwig, T. P. Stossel, and D. A. Weitz. An active biopolymer network controlled by molecular motors. Proc. Natl. Acad. Sci. USA 106, 15192-15197 (2009). 20 M. S. Silva, M. Depken, B. Stuhrmann, M. Korsten, F. C. MacKintosh, and G. H. Koenderink. Active multistage coarsening of actin networks driven by myosin motors. Proc. Natl. Acad. Sci. USA 108, 9408-9413 (2011). 21 T. Shen and P. G. Wolynes. Statistical mechanics of a cat's cradle. New J. Phys. 8, 273 (2006). 22 S. Wang, T. Shen, and P. G. Wolynes. The interplay of nonlinearity and architecture in equi- librium cytoskeletal mechanics. J. Chem. Phys. 134, 014510 (2011). 14 23 S. Wang and P. G. Wolynes. On the spontaneous collective motion of active matter. Proc. Natl. Acad. Sci. U.S.A. 108, 15184-15189 (2011). 24 A. C. Martin, M. Kaschube, and E. F. Wieschaus. Pulsed contractions of an actin-myosin network drive apical constriction. Nature 457, 495-500 (2009). F. M. Mason and A. C. Martin. Tuning cell shape change with contractile ratchets. Curr. Opin. Genet. Dev. 21, 1-9 (2011). 25 S. Wang and P. G. Wolynes. Effective temperature and glassy dynamics of active matter. J. Chem. Phys. 135, 051101 (2011). 26 D. T. Gillespie. A general method for numerically simulating the stochastic time evolution of coupled chemical reactions. J. Comp. Phys. 22, 403-434 (1976). 27 D. L. Ermak and J. A. McCammon. Brownian dynamics with hydrodynamic interactions. J. Chem. Phys. 69, 1352-1360 (1978). 28 E. Zaccarelli. Colloidal gels: equilibrium and non-equilibrium routes. J. Phys.: Condens. Matter 19, 323101 (2007). 29 J. L. Tan, S. Ravid, and J. A. Spudich. Control of nonmuscle myosins by phosphorylation. Annu. Rev. Biochem. 61, 721-759 (1992). 15 F-actin Myosin-II active node passive node FIG. 1: Schematic of the model system: A cat's cradle composed of a three dimensional amor- phous network of crosslinked nonlinear-elastic bonds (purple lines) where motor-driven anti- correlated kicks induce pulsed local bond contraction and node aggregation. Spheres represent nodes/crosslinks and the yellow fuzzy objects stand for myosin II motor proteins. The size of the nodes and motors is exaggerated; excluded volume interaction is not implemented in current simulations. Red nodes are active nodes having motor-attached bonds. These are subject to motor kicks. Blue nodes are passive nodes only undergoing Brownian motion. An enlarged view of a unit of local contraction (circled) shows a pair of motor-bonded crosslinks undergoing anti-correlated kicks with fixed step size l along the line connecting their centers where rij is a unit vector pointing from node i to node j. 16 A Motor-driven aggregation B Active "glue" initial later initial later C Multistage coarsening/aggregation FIG. 2: Illustration of how motor concentration Pa and susceptibility s contribute to their col- lective behavior. (a) Left: motor-driven aggregation of active nodes (red spheres) among the passive nodes (blue spheres); right: formation of dense floppy clumps (concentrated short green lines) inter-connected by tense bonds (long red lines). Circles mark the corresponding regions of node aggregation and bond collapse. Pc = 0.2, Pa = 0.5, s = 1. (b) Active nodes tend to "glue" together passive nodes and their aggregates. Pc = 0.5, Pa = 0.2, s = 0. (c) Multistage coarsening/aggregation of active condensates driven by high-concentration (Pa = 1) uphill-prone (s = −0.5) motors. Pc = 0.5. Shown are the temporal evolution of the node configuration (upper row) and of the corresponding network structure (lower row). 17 Pc=0.1 (z=1) Pc=0.3 Pc=0.5 Pc=0.8 (z=3) (z=6) (z=9) a P 1 0.8 0.6 0.4 0.2 0 Contractile networks Non-contractile networks Pa threshold and Pc window Present only if excluded volume is negligible 0 0.2 0.4 0.6 0.8 1 Pc FIG. 3: State diagram showing the dependence of active contractility on motor concentration Pa and network connectivity Pc. s = 0.01. Red crosses denote contractile networks and blue circles denote non-contractile networks. Top row displays the initial homogeneous network structures with increasing connectivity Pc and average number of bonded neighbors z. Macroscopic contraction occurs in the two framed regions: (1) purple open frame: intermediate Pc and above-threshold Pa; (2) green closed frame: high Pc and intermediate Pa. 18 Uncorrelated kicks Correlated kicks FIG. 4: Aster versus condensate. Pc = 0.5, Pa = 1, s = −0.5. Left: Under uncorrelated kicks, an initially disordered and homogeneous network self-organizes into highly tense and ordered "asters", composed of tense bundles radiating from the junctions where floppy bonds concentrate (see the zoom-in image). Right: Driven by anti-correlated kicks, an initially homogeneous force- percolating network first develops transient tense states consisting of highly-stretched bundles, but then abruptly collapses into a single floppy clump. 19 A s d n o b t u a t f o n o i t c a r F B s d n o b t u a t f o n o i t c a r F Threshold motor susceptibility (s) for contractility y g r e n E s-- s-- D S M s-- Threshold motor concentration (Pa) for contractility y g r e n E D S M Pa++ Pa++ 100000 Fig. S1 Thresholds for active contractility. (a) Evolution of the statistical measures for a series of motor susceptibilities (s = 0, 0.01, 0.02, 0.05 and 0.1). Pc = 0.5, Pa = 0.5. When s ≤ 0.02 the fraction of taut bonds and the total energy drop to essentially zero after surmounting an energy barrier due to tense intermediates. Mean square displacement (MSD) mounts to a plateau as a consequence of the formation of a single floppy clump. Smaller s yields a higher barrier and faster collapse (indicated by arrows). (b) Evolution of the measures for a series of motor concentrations (Pa = 0.1, 0.2, 0.3, 0.4, 0.5 and 1). Pc = 0.5, s = 0. When Pa ≥ 0.3 global contraction occurs. Larger Pa leads to lower barrier and faster collapse (indicated by arrows). 20 A Low network connectivity B Low motor concentration Increasing Pa initial later Fig. S2 Illustrations of the failure of contractility at low network connectivity or low motor concentration. s = 0. (a) Pc = 0.1. Lack of tension percolation due to low degree of bonding prevents global contraction. As Pa increases (left to right: Pa = 0.1, 0.5, 0.8 and 1) system exhibits increasing trend of local aggregation (lower row) resulting in an increased fraction of floppy bonds (upper row). When driven by high-concentration (Pa = 1) adamant motors, the initially homogeneous network exhibits local collapses into disconnected clusters of buckled filaments (circles mark the typical regions). (b) Pa = 0.1, Pc = 0.5. Sparse and modest local network distortion (upper) and node aggregation (lower) are insufficient to trigger global contractile instability. 21 Interplay of Pc and Pa in macroscopic contraction A High motor concentration (Pa=0.7) y g r e n E B High network connectivity (Pc=0.7) y g r e n E s d n o b t u a t f o n o i t c a r F s d n o b t u a t f o n o i t c a r F D S M D S M Fig. S3 Interplay of network connectivity and motor concentration in macro- scopic contraction. s = 0.01. (a) Pa = 0.7, Pc = 0.1, 0.3, 0.5, 0.7: contraction occurs for intermediate Pc. Larger Pc yields higher "barrier" to collapsed state since more bond con- straints results in more intense tug-of-war. (b) Pc = 0.7, Pa = 0.1, 0.3, 0.5, 0.7: contraction occurs for intermediate Pa. Larger Pa lowers the barrier since cooperativity between local contraction events is enhanced. 22 Aster formation Fig. S4 Snapshots in the course of aster formation. Upper row shows the bond structure: filament bundling evolves from loose to tight. Lower row presents the corre- sponding node configuration: aggregation toward the corners of the simulation box becomes progressively more compact. Pc = 0.3, Pa = 1, s = −0.5. 23
0910.2509	2	0910	2010-01-26T15:58:49	Efficient equilibrium sampling of all-atom peptides using library-based Monte Carlo	[ "physics.bio-ph" ]	We applied our previously developed library-based Monte Carlo (LBMC) to equilibrium sampling of several implicitly solvated all-atom peptides. LBMC can perform equilibrium sampling of molecules using the pre-calculated statistical libraries of molecular-fragment configurations and energies. For this study, we employed residue-based fragments distributed according to the Boltzmann factor of the OPLS-AA forcefield describing the individual fragments. Two solvent models were employed: a simple uniform dielectric and the Generalized Born/Surface Area (GBSA) model. The efficiency of LBMC was compared to standard Langevin dynamics (LD) using three different statistical tools. The statistical analyses indicate that LBMC is more than 100 times faster than LD not only for the simple solvent model but also for GBSA.	physics.bio-ph	physics	Efficient equilibrium sampling of all-atom peptides using library-based Monte Carlo Ying Ding, Artem B. Mamonov and Daniel M. Zuckerman∗ Department of Computational Biology, School of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania 15260 Abstract We applied our previously developed library-based Monte Carlo (LBMC) to equilibrium sampling of several implicitly solvated all-atom peptides. LBMC can perform equilibrium sampling of molecules using the pre-calculated statistical libraries of molecular-fragment configurations and energies. For this study, we employed residue-based fragments dis- tributed according to the Boltzmann factor of the OPLS-AA forcefield describing the individual fragments. Two solvent models were employed: a simple uniform dielectric and the Generalized Born/Surface Area (GBSA) model. The efficiency of LBMC was compared to standard Langevin dynamics (LD) using three different statistical tools. The statistical analyses indicate that LBMC is more than 100 times faster than LD not only for the simple solvent model but also for GBSA. 1 Introduction Conformational fluctuations in proteins are well appreciated to be essential to bio- chemistry, in roles accompanying binding, catalysis and locomotion [1]. In recent years, the importance of fluctuations has been further underscored by recognition ∗Electronic mail: [email protected] 1 1 Introduction 2 of the widespread presence of disordered regions in proteins [2, 3]. Structural ex- periments, however, are fairly limited in their power to characterize such fluctua- tions from a true ensemble perspective. For a given protein, X-ray crystallography generates one or a very small number of configurations, typically excluding the most flexible regions [4, 5]. NMR studies yield highly approximate structure sets based on simplified forcefields and non-canonical algorithms [6]. Cryo-EM can characterize large structural fluctuations but at low resolution [7]. In principle, computations are ideal for characterizing fluctuations in biomolecules, but sampling power is typically inadequate except for small systems. The ba- sic reason is the well-known gap in timescales between simulation and biolog- ical behavior [8]. To bridge this gap, much effort has been undertaken in the field to develop new efficient sampling techniques. Many of these techniques are based on "generalized ensembles" including replica exchange [9, 10, 11, 12]. Other techniques use modified energy surfaces [13, 14, 15] or modified dynamics [16, 17, 18, 19, 20]. The "resolution exchange" (ResEx) algorithm uses a ladder of different resolution models with occasional exchanges between levels [21]. Both replica-exchange and ResEx can be implemented in the serial top-down scheme [22, 23]. Another strategy for speeding calculations is to exploit computer memory to store frequently used information. In particular, libraries of molecular fragment configurations can be stored and re-used. Libraries were previous used by Rosetta [24], but not for true canonical sampling. In our previous work we introduced the statistical rigorous library based Monte Carlo (LBMC) and used it to incorporate atomic details into a coarse-grained protein model at a small computational cost [25]. The semi-atomistic model was applied to equilibrium sampling of several proteins containing up to 309 residues. LBMC was also applied to equilibrium sampling of several peptides described by OPLS-AA forcefield [26] with a simple uniform dielectric model to model the solvent [25]. A large efficiency gain of LBMC compared to standard Langevin dynamics was observed. Inspired by the results of our previous study , here we further investigate the application of LBMC to equilibrium sampling of all-atom peptides. In this study we apply LBMC to several implicitly solvated peptides described by OPLS-AA forcefield with two different implicit solvent models: a simple uni- form dielectric of 60 and the Generalized Born/Surface Area (GBSA) model [27]. The efficiency of LBMC was quantified by comparison to Langevin Dynamics using three different statistical tools. The first tool is based on the autocorrelation behavior of the end-to-end distance, the second uses our previously developed "decorrelation time" analysis [28] , and the third employs a block averaging anal- 2 Methods 3 ysis [29] of the end-to-end distances. All the analyses point to efficiency gains of two to three orders of magnitude in the three peptides studied - tetraalanine, octaalanine and Met-enkephalin. 2 Methods 2.1 Library-based Monte Carlo Method (LBMC) The library-based Monte Carlo (LBMC) method was recently introduced, along with complete derivations [25]. Here we briefly review the method. LBMC uses the simple idea to divide a molecule into non-overlapping frag- ments, each of which is pre-sampled into a library of Boltzmann-distributed frag- ment configurations. For peptides and proteins, fragments based on amino acids are natural. Trial moves consist of swapping the present configuration of one or more fragments with members of the corresponding libraries. LBMC, which is a rigorous MC scheme , has several noteworthy features. (i) Libraries -- e.g., for each amino acid -- are generated one time and can be re-used in multiple sim- ulations; accordingly, the internal-to-fragment interactions are never calculated during a simulation, saving some CPU cost. (ii) Because fragment configurations are pre-sampled based on all interactions internal to the fragment, those energy terms do not enter the Metropolis acceptance criterion. (iii) Perhaps most impor- tantly, the complex correlations among degrees of freedom internal to a fragment are fully accounted for in the library-generation stage -- i.e. , the "price" for the internal timescales is paid in advance. A Metropolis-Hastings criterion for an LBMC trial move is derived in the usual way based on the detailed-balance condition [25]. In outline, the derivation is accomplished by first separating the full set of degrees of freedom ~r into M fragments,−→r = {~r1, ..., ~rM } . Similarly, the total energy of a forcefield U tot, which could include implicit solvent terms, is decomposed into components: the terms internal to each fragment i, denoted U frag , and all the "rest," which are cross-fragment interaction terms lumped into U rest. Thus we have i U tot(~r1, ..., ~rM ) = X U frag i (~ri) + U rest(~r1, ..., ~rM ) (1) In the present study, U tot represent the OPLS-AA forcefield plus an implicit solvent model, as described below. 2 Methods 4 In our previous work, we derived Metropolis criteria for two types of library- swap trial moves [25]. The first is a simple swap move in which configurations from one or more fragments are swapped with configurations chosen uniformly from the corresponding libraries (Each library is already Boltzmann distributed according to U frag , as described below.) In the simple swap move, the generating probability for the trial/new configuration (n) is completely independent of the old configuration (o). This leads to significant cancellation of terms, and one finds the acceptance probability to be [25] i pacc(o → n) = min(cid:2)1, exp(−β∆U rest)(cid:3) We will also employ a second type of swap move based on "neighbor lists." In the context of LBMC a neighbor list is, for each configuration, the list of config- urations deemed to be similar by an arbitrary criterion. The trial move of interest, then, is to choose a library configuration for swapping only among the neighbor- ing configurations for a single fragment i. When trial configurations are selected uniformly among neighbors, it can be shown that the acceptance criterion is [25] (2) pacc(o → n) = min(cid:20)1, exp(−β∆U rest) i ko kn i (cid:21) (3) i and kn where ko i are the number of neighbors of configuration o and n respec- tively for fragment i. If neighbor lists are constructed to have the same number of neighbors for all configurations in a given library, then the acceptance criterion of Eq. 3 reduces to Eq. 2. Below we will explain our procedures for generating libraries and neighbor lists. 2.2 Practical library generation The fragment used in this study correspond to individual amino acids, which are the natural building blocks of peptides. In previous LBMC work [25], we used both peptide planes and amino acids as fragments in separate simulations. How- ever, amino acid fragments have the advantage of including detailed "Ramachan- dran correlations" among φ and ψ angles, as well as the other degrees of freedom. In practical terms, this means that the timescales and correlations associated with Ramachandran effects are pre-sampled within the libraries. Fragment configurations in the libraries were generated according to the Boltz- mann factor of OPLSAA forcefield [26], plus the appropriate implicit solvent 2 Methods 5 model, for all interactions internal to the fragment. Fragment libraries must in- clude not only atomic coordinates, but also the six degrees of freedom necessary for connecting one fragment to the next. Full details of the degrees of freedom for amino acid libraries were given in [30], our previous work. In brief, we used dummy atoms "borrowed" from the next fragment to facilitate sampling the coor- dinates necessary for connecting fragments. Interactions with dummy atoms were fully accounted for to yield the correct ensemble of the whole molecule -- as can be seen in our results below. Although library ensembles, in principle, can be generated using any canonical sampling scheme, we found it most convenient to use internal-coordinate Monte Carlo (ICMC). ICMC readily permits fixing degrees associated with the dummy atoms which we did not wish to vary. Our use of ICMC properly accounted for internal-coordinate Jacobians, which ensure that the distribution obtained agrees with that from using the (natural) Cartesian coordinates. The standard Jacobians were employed -- i.e., r2 for bond lengths r and sin θ for bond angles θ. For each amino acid fragment, ICMC was run for 109 steps to produce libraries of 105 configurations. See Figure 1. The library configurations may not be fully statistically independent, but we do carefully assess the statistical quality of the ultimate ensembles of the full molecule -- as shown below. 2.3 Neighbor-list construction In LBMC, "neighbor lists" of library configurations similar to each library mem- ber provide a conveninent way to attempt relatively local moves in configuration space. As explained in our initial derivation [25], neighbors can be defined in an arbitrary way. Natural choices include criteria based on a pairwise "distance" similarity metric such as the root mean square deviation (RMSD) or the sum of absolute differences over all bond and dihedral angles in a given fragment as was used in our previous work [25]. When constructing neighbor lists, if configura- tion i contains configuration j in its neighbor list then j must have i, to satisfy microscopic reversibility. In the present work the neighbor lists were constructed to generate groups of n = 10 similar configurations. To minimally perturb the molecule's overall struc- ture, similarity between two configs was quantified by the RMSD of six atoms, three at each end of the fragment. To construct neighbor lists, the following algo- rithm was used. A first "reference" structure is selected randomly from the whole library and the nearest n − 1 configurations in the RMSD space are chosen for the first neighbor list. The next reference structure is randomly selected from the 2 Methods 6 remaining configurations and again the closest n − 1 configurations are chosen for this neighbor list. This process is repeated until the whole library is partitioned into equi-sized neighbor lists. In this study, each library has 105 configurations and is partitioned into 104 neighbor groups of size n = 10. In general, it is not necessary to make equal size clusters, nor is it necessary to strictly partition the whole library into "disconnected" neighbor lists. If there is a strict partitioning (as in the present study), then non-neighbor trial moves are required to ensure the possibility of ergodicity. By adjusting the fraction of local to global moves, the acceptance rate can be tuned. In the future, it will be worthwhile to construct and test overlapping (non-isolated) neighbor lists. 2.4 Efficiency analysis It is critical to quantify the sampling quality of any new method, in comparison to a standard technique. In this study we assess the convergence of LBMC sim- ulations and compare its efficiency relative to standard Langevin dynamics using three different statistical tools. One of these methods is semi-qualitative and the other two are quantitative. Because there are no true physical timescales in our Monte Carlo simulations, our primary focus is to compare sampling efficiency in terms of single-processor wall-clock time. We recognize that different Langevin implementations (i.e., in different software packages) will vary in speed. However, we anticipate such differences will be small compared to the orders of magnitude efficiency we report below for LBMC. Furthermore, our reference Langevin simulations employ a low friction constant, which is recognized to improve sampling speed compared to a more physical water-like value [?]. The semi-qualitative tool we use to analyze sampling is the standard autocor- relation function of some slowly changing variable. The autocorrelation function is given, as usual, by Cx(τ ) = hx(t) x(t + τ )i − hxi2 hx2i − hxi2 (4) where hxi is the average value of x(t), and τ is the time interval or number of MC steps between configurations in the trajectory. Because all correlations in an LBMC "trajectory" are sequential, a "time" correlation description is valid. A number of useful slow coordinates can be defined [31], and we choose the end- to-end distance of a peptide as a simple geometric measure which illustrates the 2 Methods 7 key timescales. However, the auto-correlation behavior is not used to quantify efficiency in our study, but only to depict it graphically. We measured time in units of wall-clock minutes to facilitate comparison between LBMC and standard Langevin simulation. The second statistical tool is based on our previously developed "structural de-correlation time" analysis which determines how much simulation time must elapse between configurations in the trajectory in order for them to exhibit statis- tical independence [28]. The ratio of the overall trajectory length to the decorrela- tion time provides an objective estimate of the effective sample size (ESS) -- i.e., the number of independent configurations. Importantly, because all correlations are sequential in the LBMC Markov chain, the ESS for LBMC can be calculated from the same ratio of trajectory length to decorrelation time. We therefore define the first efficiency factor as the gain in the sampling speed of LBMC over Langevin dynamics based on the ratio of CPU cost per independent configuration: γ1 = ESSLBMC/tLBMC ESSLangevin/tLangevin (5) where tLBMC and tLangevin are the total wallclock times of LBMC and Langevin simulation respectively. The last statistical method is based on the more traditional block averaging analysis [29, 32] of some slowly changing variable. In this approach, a trajectory is divided into "blocks" of different size. The mean value of the variable along with the standard error of the mean (SE) is calculated for different size blocks. As the block size increases, so does the standard error because blocks become more independent from each other. At some block size the standard error levels off, indicating that the blocks have become effectively independent from each other. This plateau is the true value of SE. We use block-averaging of the end-to- end distance, as a representative slow coordinate. We therefore define the second efficiency factor based on the ratio of CPU cost per "unit of precision": γ2 = tLangevinSE2 tLBMCSE2 Langevin LBMC (6) where SELangevin and SELBMC are the standard errors for Langevin and LBMC simulation, respectively, estimated from block averaging. Note that SE2 is ex- pected to vary linearly with inverse simulation time [32]. 2 Methods 8 2.5 System and simulation details We applied LBMC to three implicitly solvated peptides including Ace-(Ala)4- Nme, Ace-(Ala)8-Nme, and Met-enkephalin described by OPLS-AA forcefield [26]. No cutoffs were used in these relatively small systems. We chose these peptides because they have been extensively studied experimentally and compu- tationally [33, 34, 35]. Two different implicit solvent models were employed: a uniform dielectric constant of ε = 60 and the more standard GBSA model [27]. The constituent atomic radii for GBSA are taken from the OPLS-AA force field and the nonpolar solvation is calculated via the ACE approximation [36]. The Born radii used in GBSA are recomputed for every MC step. For LBMC simulations of poly-alanine systems, three libraries were employed corresponding to Ace, Ala and Nme fragments. For Met-enkephalin (Tyr-Gly- Gly-Phe-Met), six libraries were used corresponding to Ace, Gly, Phe, Tyr, Met and Nme residues. Different libraries were used depending on the solvent models. For LBMC with uniform dielectric solvent we used fragment libraries sampled according to the uniform dielectric model, whereas for GBSA simulations we used libraries sampled according to GBSA solvent. For all LBMC simulations reported here, the trial move was a single fragment swap with the corresponding library. For our systems, this was found to be the most efficient based on simulations with different number of fragments swapped per MC step. All system were sampled at 298 K. For LBMC simulations of both solvent models Ace-(Ala)4-Nme was run for 105 MC steps with configurations saved every 10 MC steps resulting into 104 frames. Ace-(Ala)8-Nme was run for 107 MC steps with frames saved every 100 MC steps resulting in 105 structures. Met-enkephalin was run for 106 MC steps with frames saved every 10 MC steps resulting into 105 frames. To compare LBMC with Langevin dynamics we ran LD simulations for the same three systems and both solvent models. Specifically, all three peptides were run for 100 ns with frames saved every picosecond resulting into 105 structures. All Langevin simulations were run at the temperature of 298 K and the friction constant of 5 psec-1, as implemented in the Tinker software package [37]. 3 Results 3 Results 3.1 Ensemble Quality 9 We first verified that LBMC samples the correct equilibrium distributions. For this purpose we prepared "structural bins" which are randomly selected regions of configuration space which cover the whole space, and can sensitively quantify sampling [38]. The bins were constructed using a Voronoi procedure as described in [30]. For all three systems we compared the fractional populations of the struc- tural bins obtained from LBMC and Langevin simulations. The results for the uniform dielectric solvent model are shown in Figure 2 and for GBSA in Figure 3, indicating good agreement between two methods. To examine the ensembles from a more traditional perspective, we also cal- culated hydrogen bond population and the helical content of octa-alanine. Based on the hydrogen-bond definition given in [19], we find the average number of hydrogen bonds in tetraalanine, octaalanine and Met-enkephalin to be (2.400 ± 0.006 , 5.00 ± 0.02, 3.20 ± 0.01) in the simple solvent model and (2.84 ± 0.08 , 6.63 ± 0.06, 3.98 ± 0.07) in GBSA from LBMC simulation. For comparison, we found (2.408 ± 0.002 , 5.03 ± 0.02, 3.21 ± 0.01) in simple solvent model and (2.75 ± 0.08 , 6.51 ± 0.06, 4.06 ± 0.09) in GBSA from Langevin simulation. Here the uncertainty is quantified as the standard error of mean of the number of hydro- gen bonds in the specific ensemble. Helical content was defined to be the fraction of residues in the system whose (φ, ψ) dihedrals were within ±25◦ of the ideal angles of approximately (−60◦, −40◦). We found helical population for octaala- nine is 11.7%(±0.4%) in the simple solvent model and 30%(±2%) in GBSA from LBMC as compared 11.1%(±0.3%) in simple solvent model and 31%(±2%) in GBSA from Langevin simulation, respectively. These structural measures further verify our results. 3.2 Efficiency Analysis The efficiency of LBMC to sample the equilibrium distributions was compared to Langevin using the three statistical tools discussed in Sec. 2.4. The first tool is the autocorrelation function (ACF) of the end-to-end distance for each peptide. For all systems, the end-to-end distances was calculated based on coordinates of the methyl carbon atom of the Ace group and the methyl carbon of the Nme group. The autocorrelation function was calculated according to Eq. 4. As shown in Figures 4 and 5 , we calculated ACFs for all systems in the two solvent models 3 Results 10 and using two time measures. Most importantly, we depict the ACF vs. wallclock time, which suggests the high efficiency of LBMC compared to Langevin in these systems. For reference, we also computed each ACF as a function of the number of simulation steps. The second statistical method is the "de-correlation time" analysis which we used to calculate the number of statistically independent configurations (i.e., the effective sample size (ESS)) in the trajectory [28]. The ESS results for LBMC and Langevin simulations, along with the efficiency factors γ1 calculated using Eq. 5, are reported for the uniform dielectric model in Table 1 and for GBSA in Table 2. From Table 1 it follows that for the uniform dielectric model LBMC is more than three orders of magnitude faster than Langevin for Ace-(Ala)4-Nme, more than two orders of magnitude faster for Ace-(Ala)8-Nme and almost three orders of magnitude faster for Met-enkephalin. Table 2 indicates that for GBSA solvent LBMC is more than three orders of magnitude faster than Langevin for Ace-(Ala)4-Nme, over two order of magnitude faster for Ace-(Ala)8-Nme and over two orders of magnitude faster for Met-enkephalin. For Langevin simula- tions, the decorrelation time is also a physical timescale [28] as tabulated in Table 1 and Table 2: it is about 1nsec or less in the three peptides. The third statistical tool is based on the more traditional block averaging anal- ysis and we used it to confirm the efficiency results obtained from the previous method. We again employed the end-to-end distance which is a slowly changing variable. The standard errors (SE) of the mean for end-to-end distances from the block averaging, along with the efficiency factors γ2 calculated using Eq. 6, are reported for the uniform dielectric model in Table 1 and for GBSA in Table 2. Comparison of γ1 with γ2 indicates that the block averaging technique esti- mates similar efficiency factors to the de-correlation analysis, demonstrating the robustness of our analysis. 3.3 Regarding GBSA GBSA affects both simulation cost (per timestep) and the ability to sample (by changing the landscape's roughness). Both these factors, in turn, affect efficiency. The cost, however, is implementation specific. We now briefly address GBSA effi- ciency and implementation. For GBSA, the efficiency factors are slightly smaller than for the uniform dielectric model. When using GBSA solvent the ESS de- creased by the factor of ca. 2.5 for both Langevin and LBMC. For LBMC the acceptance rate decreased as well. This indicates that sampling becomes more difficult for both methods in the more complicated energy landscape provided by 4 Discussion 11 GBSA. Note that all other parameters, such as the number of atoms and the num- ber of steps, was the same for each method. We can also compare solely the wallclock cost to run the same number of MC or Langevin steps for simple solvent model and GBSA. When GBSA solvent was employed, the simulation time increased by a factor of 4 for Langevin and by a factor of 6 for LBMC. The larger increase of the wallclock time for LBMC can be attributed to a relatively inefficient implementation of GBSA in our algorithm compared to the Tinker program. We therefore believe that the decrease of effi- ciency factors, γ1 and γ2, for GBSA simulations can be attributed to our inefficient implementation of GBSA rather than the difficulty of LBMC to sample complex energy landscapes. See further comments on GBSA in Sec. 4. 3.4 Neighbor-based trial moves The use of neighbor swap moves in LBMC (Secs 2.1 and 2.3) is suggested by Tables 1 and 2, because the acceptance rate significantly dropped for LBMC sim- ulations with GBSA solvent compared to the uniform dielectric model. To test the ability of neighbor-list trial moves to increase the acceptance rate for LBMC simulation of all-atom peptides, we employed two sets of trial moves: one with 30% and the other with 70% of local (neighbor-list) moves. Both sets helped to increase the acceptance rate. For example, for Ace-(Ala)8-Nme using 30% local moves increased the acceptance rate from 0.18 to 0.27 and 70% local moves led to 0.41. For Met-enkephalin 30% local moves increased the acceptance rate from 0.17 to 0.31 and 70% local moves led to 0.38. However, the efficiency analysis in- dicated that for simulations with local trial moves the efficiency factors γ1 and γ2 turned out to be smaller compared to simulations with regular (100% global) trial moves. As discussed in Sec 4, additional exploration of neighbor-list construction is needed. 4 Discussion The initial results for sampling all-atom peptides in implicit solvent via LBMC are very encouraging. We wish to make some observations and also point out several avenues for improvement as well as some limitations. First of all, LBMC is not simply internal-coordinate MC (ICMC) with "win- dow dressing". That is, using libraries is not merely a way to employ large trial moves, such as drastic changes to φ and ψ dihedrals. Indeed if large dihedral 4 Discussion 12 moves are used in ICMC, the acceptance rate is over an order of magnitude smaller than for global swap moves in LBMC. The key to LBMC success (in the systems studied) is the correlated nature of trial moves: large φ and ψ changes are ac- companied, by construction, with correlated changes of other coordinates in the fragment (i.e., residue). There appear to be two principal limitations for application of LBMC to all- atom sampling with standard forcefields. These limitations stem, in a sense, from the strength of LBMC for small flexible systems: extremely large trial moves (not feasible or physical in dynamics) lead to rapid sampling. For instance, LBMC will occasionally jump from one region of the Ramachandran plane to a completely different part. Such large moves immediately suggest that, first of all, LBMC with large trial moves will not be suitable for explicit solvent. Secondly, even in implicit solvent, once a single molecule becomes large and "dense" -- such as a full protein -- large-scale trial moves will again prove nearly impossible to accept. In an explicit solvent simulation of a peptide, for example, trial moves for the peptides could be governed by LBMC and solvent moves via "ordinary" MC. In the LBMC acceptance criterion, U rest would include solvent interactions. Whether such an approach proves practical will depend sensitively on the construction of suitably local LBMC trial moves -- i.e., crankshaft-like, see below. It is also possible that future methodologies will permit the conversion of implicit solvent ensembles to explicit solvent [39]. We note that, in principle, LBMC is not limited to implicit solvent. LBMC can however employ more "local" trial moves, for instance based on "neighboring" library configurations as described above. As noted in the Sec. 3, such local moves actually decrease sampling efficiency (despite increasing the acceptance ratio) in the small systems studied here. In larger systems, however, we expect neighbor-based moves to be helpful. Local moves should also prove important in sampling loops with LBMC. In the future, more sophisticated con- structions of neighbor lists should be possible, as compared to our fairly simple approach described in Sec. 2. Further improvements to GBSA-based sampling via LBMC appear to be pos- sible, given our inefficient implementation of GBSA as discussed in Sec. 3. In- deed, generally speaking, GBSA is not well-suited to Monte Carlo simulation, as previously noted [40], because the non-pairwise energy terms depend on the en- tire molecule even when only part of it is changed as in typical MC trial moves. Therefore, other solvent models or approximations to GBSA [40, 41, 42] should improve LBMC efficiency further in terms of wall-clock time. Like almost any canonical sampling method, LBMC can be employed in more 5 Summary and Conclusions 13 sophisticated sampling strategies, such as replica, Hamiltonian, or resolution ex- change [9, 10, 43, 21], as well as the related dual-chain MC approaches [13, 14, 15]. However, LBMC would appear to have a particular advantage for multi- resolution approaches: the positions of all atoms can be stored at essentially zero run-time cost, even if a "coarse grained" forcefield is employed. That is, because all degrees of freedom are maintained, LBMC provides a natural means for cast- ing resolution-exchange simulation in terms of "simple" Hamiltonian exchange. We believe this idea warrants further investigation. As noted in our earlier paper [25] and echoing the above discussion, LBMC provides a natural platform for coarse and mixed resolution models. Most sim- ply, all atoms can be accounted for, with only a subset used as interaction sites in a "coarse-grained" model -- allowing flexibility to tune the coarse interactions. In a "mixed modeling" approach, atoms in critical regions such as binding sites can retain their full interactions, while distant residues are coarse-grained. Such a strategy might be useful in binding-affinity or ensemble-based docking calcula- tions. 5 Summary and Conclusions Multiple statistical efficiency analyses show that library-based Monte Carlo (LBMC) can obtain remarkable efficiency for peptide systems. LBMC employs pre-calculated libraries of equilibrium configurations of molecular fragments -- in this case, amino acid fragments. We applied LBMC to three peptides (4, 5, and 8 residues) de- scribed by a standard all-atom forcefield, OPLS-AA, with a simple dielectric "solvent" as well as the common GBSA implicit solvent. In every case, two in- dependent methods of quantifying efficiency indicate that LBMC out-performed Langevin dynamics by two orders of magnitude. The success of LBMC in flexible peptides derives from several factors. First, large trial moves -- with significant φ and ψ changes -- are frequently attempted. Second, because the libraries are pre-sampled to include all interactions and cor- relations internal to each residue, only long-range interactions present an obstacle to accepting a trial move. Finally, once the libraries have been generated they can be re-used repeatedly without the need to re-calculate the tabulated energies internal to fragments. LBMC thus appears to be promising for loop-sized peptides (~10 residues), particularly if trial moves to neighboring library configurations can be better ex- ploited. In addition, LBMC can readily be combined with "advanced" techniques 5 Summary and Conclusions 14 such as those based on the exchange idea [9, 10, 43, 21, 23, 13, 14, 15]. The ability of LBMC simulation to store all atomic coordinates at no run-time cost suggests it will provide an ideal platform for flexible coarse-graining approaches based on using a subset of interaction sites. Acknowledgement The authors thank Xin Zhang, Bin Zhang, and Divesh Bhatt for helpful dis- cussions. Funding was provided by the NIH through Grants GM070987 and GM076569, as well as by the NSF through Grant MCB-0643456. References [1] Jeremy M. Berg. Biochemistry. W.H.Freeman, 2006. [2] A. Keith Dunker, J. David Lawson, Celeste J. Brown, Ryan M. Williams, Pedro Romero, Jeong S. Oh, Christopher J. Oldfield, Andrew M. Campen, Catherine M. Ratliff, Kerry W. Hipps, Juan Ausio, Mark S. Nissen, Ray- mond Reeves, ChulHee Kang, Charles R. Kissinger, Robert W. Bailey, Michael D. Griswold, Wah Chiu, Ethan C. Garner, and Zoran Obradovic. Intrinsically disordered protein. J. Mol. Graph. Model., 19:26 -- 59, 2001. [3] Dunker,A.K.; Brown,C.J.; Lawson, J.D; Iakoucheva,L.M.; Obradovic, Z. In- trinsic Disorder and Protein Function. Biochemistry, 41:6573 -- 6582, 2002. [4] Mark A DePristo, Paul I.W de Bakker, and Tom L Blundell. Heterogeneity and inaccuracy in protein structures solved by x-ray crystallography. Struc- ture, 12(5):831 -- 838, May 2004. [5] Eran Eyal, Sergey Gerzon, Vladimir Potapov, Marvin Edelman, and Vladimir Sobolev. The Limit of Accuracy of Protein Modeling: Influence of Crystal Packing on Protein Structure. J. Mol. Biol., 351(2):431 -- 442, August 2005. [6] Spronk, C. A. E. M.; Nabuurs, S. B.; Bonvin, A. M. J. J.; Krieger, E.; Vuister, G. W.; Vriend, G. The precision of NMR structure ensembles revisited. J. Biomol. NMR, 25:225 -- 234, 2003. [7] Helen R. Saibil. Conformational changes studied by cryo-electron mi- croscopy. Nat. Struct. Mol. Biol., 7(9):711 -- 714, September 2000. 5 Summary and Conclusions 15 [8] Peter L. Freddolino, Feng Liu, Martin Gruebele, and Klaus Schulten. Ten- Microsecond Molecular Dynamics Simulation of a Fast-Folding WW Do- main. Biophys. J., 94(10):L75 -- L77, May 2008. [9] Robert H. Swendsen and Jian-Sheng Wang. Replica Monte Carlo Simulation of Spin-Glasses. Phys. Rev. Lett., 57(21):2607 -- 2609, November 1986. [10] C.J. Geyer. Markov chain Monte Carlo maximum likelihood. Proceedings of the 23rd Symposium on the Interface Foundation, page 156, 1991. [11] B. A. Berg and T. Neuhaus. Multicanonical Ensemble: A New Approach to Simulate First-Order Phase Transitions. Phys. Rev. Lett., 68:9 -- 12, 1992. [12] Yuko Okamoto. Generalized-ensemble algorithms: enhanced sampling tech- niques for Monte Carlo and molecular dynamics simulations. J. Mol. Graph. Model., 22(5):425 -- 439, May 2004. [13] Radu Iftimie, Dennis Salahub, Dongqing Wei, and Jeremy Schofield. Using a classical potential as an efficient importance function for sampling from an ab initio potential. J. Chem. Phys., 113(12):4852 -- 4862, 2000. [14] Lev D. Gelb. Monte Carlo simulations using sampling from an approximate potential. J. Chem. Phys., 118(17):7747 -- 7750, 2003. [15] Balazs Hetenyi, Katarzyna Bernacki, and B. J. Berne. Multiple "time step" Monte Carlo. J. Chem. Phys., 117(18):8203 -- 8207, 2002. [16] Zhongwei Zhu and Mark E. Tuckerman. Ab Initio Molecular Dynamics Investigation of the Concentration Dependence of Charged Defect Transport in Basic Solutions via Calculation of the Infrared Spectrum. J. Phys. Chem. B, 106(33):8009 -- 8018, August 2002. [17] Lula Rosso, Peter Minary, Zhongwei Zhu, and Mark E. Tuckerman. On the use of the adiabatic molecular dynamics technique in the calculation of free energy profiles. J. Chem. Phys., 116(11):4389 -- 4402, 2002. [18] Peter Minary, Mark E. Tuckerman, and Glenn J. Martyna. Dynamical Spatial Warping: A Novel Method for the Conformational Sampling of Biophysical Structure. SIAM J. Sci. Comput., 30(4):2055 -- 2083, 2008. 5 Summary and Conclusions 16 [19] Jerry B. Abrams and Mark E. Tuckerman. Efficient and Direct Generation of Multidimensional Free Energy Surfaces via Adiabatic Dynamics without Coordinate Transformations. J. Phys. Chem. B, 112(49):15742 -- 15757, De- cember 2008. [20] Luca Maragliano and Eric Vanden-Eijnden. A temperature accelerated method for sampling free energy and determining reaction pathways in rare events simulations. Chem. Phys. Lett., 426(1-3):168 -- 175, July 2006. [21] Edward Lyman, F. Marty Ytreberg, and Daniel M. Zuckerman. Resolution Exchange Simulation. Phys. Rev. Lett., 96(2):028105, 2006. [22] D. D. Frantz, D. L. Freeman, and J. D. Doll. Reducing quasi-ergodic be- havior in Monte Carlo simulations by J-walking: Applications to atomic clusters. J. Chem. Phys., 93(4):2769 -- 2784, August 1990. [23] Edward Lyman and Daniel M. Zuckerman. Resolution Exchange Simula- tion with Incremental Coarsening. J. Chem. Theory Comput., 2(3):656 -- 666, 2006. [24] Carol A. Rohl, Charlie E. M. Strauss, Kira M. S. Misura, and David Baker. Protein Structure Prediction Using Rosetta. Methods Enzymol., 383:66 -- 93, 2004. [25] Artem B. Mamonov, Divesh Bhatt, Derek J. Cashman, Ying Ding, and Daniel M. Zuckerman. General Library-Based Monte Carlo Technique En- ables Equilibrium Sampling of Semi-atomistic Protein Models. J. Phys. Chem. B, 113(31):10891 -- 10904, August 2009. [26] William L. Jorgensen, David S. Maxwell, and Julian Tirado-Rives. De- velopment and Testing of the OPLS All-Atom Force Field on Conforma- tional Energetics and Properties of Organic Liquids. J. Am. Chem. Soc., 118(45):11225 -- 11236, 1996. [27] D Qiu, PS Shenkin, FP Hollinger, and WC Still. The GB/SA Continuum Model for Solvation. A Fast Analytical Method for the Calculation of Ap- proximate Born Radii. J. Phys. Chem. A, 101(16):3005 -- 3014, April 1997. [28] Edward Lyman and Daniel M. Zuckerman. On the Structural Convergence of Biomolecular Simulations by Determination of the Effective Sample Size. J. Phys. Chem. B, 111(44):12876 -- 12882, 2007. 5 Summary and Conclusions 17 [29] H. Flyvbjerg and H. G. Petersen. Error estimates on averages of correlated data. J. Chem. Phys., 91:461 -- 466, 1989. [30] Xin Zhang, Artem B. Mamonov, and Daniel M. Zuckerman. Absolute free energies estimated by combining precalculated molecular fragment libraries. J. Comput. Chem., 30(11):1680 -- 1691, 2009. [31] Andreas Vitalis, Xiaoling Wang, and Rohit V. Pappu. Quantitative Char- acterization of Intrinsic Disorder in Polyglutamine: Insights from Analysis Based on Polymer Theories. Biophys. J., 93(6):1923 -- 1937, September 2007. [32] A.. Grossfield and D. M. Zuckerman. Quantifying uncertainty and sampling quality in biomolecular simulations. Annu. Rep. Comput. Chem., 5:23 -- 48, 2009. [33] Ulrich H. E. Hansmann. Parallel tempering algorithm for conformational studies of biological molecules. Chem. Phys. Lett., 281(1-3):140 -- 150, De- cember 1997. [34] Robert R. Hudgins and Martin F. Jarrold. Helix formation in unsolvated alanine-based peptides: Helical monomers and helical dimers. J. Am. Chem. Soc., 121(14):3494 -- 3501, April 1999. [35] Reinhard Schweitzer-Stenner, Fatma Eker, Kai Griebenow, Xiaolin Cao, and Laurence A. Nafie. The Conformation of Tetraalanine in Water Determined by Polarized Raman, FT-IR, and VCD Spectroscopy. J. Am. Chem. Soc., 126(9):2768 -- 2776, March 2004. [36] Michael Schaefer and Martin Karplus. A Comprehensive Analytical Treat- ment of Continuum Electrostatics. J. Phys. Chem., 100(5):1578 -- 1599, Jan- uary 1996. [37] Jay W. Ponder and Frederic M. Richards. An efficient newton-like method for molecular mechanics energy minimization of large molecules. J. Com- put. Chem., 8(7):1016 -- 1024, 1987. [38] Edward Lyman and Daniel M. Zuckerman. Ensemble-Based Convergence Analysis of Biomolecular Trajectories. Biophys. J., 91(1):164 -- 172, July 2006. 5 Summary and Conclusions 18 [39] Divesh Bhatt and Daniel M. Zuckerman. Absolute free energies and equi- librium ensembles of dense fluids computed from a nondynamic growth method. J. Chem. Phys., 131(21):214110 -- 10, December 2009. [40] Julien Michel, Richard D. Taylor, and Jonathan W. Essex. Efficient General- ized Born Models for Monte Carlo Simulations. J. Chem. Theory Comput., 2(3):732 -- 739, 2006. [41] John Mongan, Carlos Simmerling, J. Andrew McCammon, David A. Case, and Alexey Onufriev. Generalized Born Model with a Simple, Robust Molecular Volume Correction. J. Chem. Theory Comput., 3(1):156 -- 169, January 2007. [42] Andreas Vitalis and Rohit V. Pappu. ABSINTH: A new continuum solva- tion model for simulations of polypeptides in aqueous solutions. J. Comput. Chem., 30(5):673 -- 699, 2009. [43] Yuji Sugita, Akio Kitao, and Yuko Okamoto. replica-exchange method for free-energy calculations. 113(15):6042 -- 6051, 2000. Multidimensional J. Chem. Phys., 5 Summary and Conclusions 19 Figures Fig. 1: The residue-based fragment libraries employed for library-based Monte Carlo (LBMC) are illustrated for all 20 amino acids. We note that the number of configurations shown in the graph does not represent the actual library size. 5 Summary and Conclusions 20 A B C Fig. 2: Confirmation of correct equilibrium sampling in simple solvent. Frac- tional population of structural bins obtained from LBMC (red dashed line) and Langevin simulations (black solid line) are shown for three peptides: (A) Ace-(Ala)4-Nme, (B) Ace-(Ala)8-Nme and (C) Met-enkephalin. The peptides were sampled according to the OPLS-AA forcefield with the uni- form dielectric of 60 to model the solvent. Error bars represent one stan- dard deviation for each bin, calculated from 10 independent simulations for both LBMC and Langevin. 5 Summary and Conclusions 21 A B C Fig. 3: Confirmation of correct equilibrium sampling in GBSA. Fractional pop- ulation of structural bins obtained from LBMC (red dashed line) and Langevin simulations (black solid line) are shown for three peptides: (A) Ace-(Ala)4-Nme, (B) Ace-(Ala)8-Nme (C) Met-enkephalin. The peptides were sampled according to the OPLS-AA forcefield with GBSA solvent. Error bars represent one standard deviation for each bin, calculated from 10 independent simulations of both LBMC and Langevin. 5 Summary and Conclusions 22 A B C Fig. 4: Comparison of autocorrelation functions in simple solvent for three pep- tides based on LBMC (green) and Langevin simulations (blue). The left column shows the autocorrelation function (ACF) of the end-to-end dis- tance vs wallclock time and the right column shows the ACF vs timestep: (A) Ace-(Ala)4-Nme, (B) Ace-(Ala)8-Nme, (C) Met-enkephalin. The pep- tides were sampled according to the OPLS-AA forcefield with a uniform dielectric of 60 to model the solvent. 5 Summary and Conclusions 23 A B C Fig. 5: Comparison of autocorrelation functions in GBSA for three peptides be- tween LBMC (green) and Langevin simulations (blue). The left column shows the autocorrelation function (ACF) of the end-to-end distance vs wallclock time and the right column shows the ACF vs timestep: (A) Ace-(Ala)4-Nme, (B) Ace-(Ala)8-Nme, (C) Met-enkephalin. The peptides were sampled according to the OPLS-AA forcefield with the GBSA im- plicit solvent model. 5 Summary and Conclusions 24 Tables Tab. 1: Efficiency in simple solvent. The results of the "de-correlation" and block averaging analyses of LBMC and Langevin simulations are reported for three peptides: Ace-(Ala)4-Nme, Ace-(Ala)8-Nme, and Met-enkephalin. The peptides were sampled according to the OPLS-AA forcefield with the uniform dielectric of 60 to model the solvent. M is the number of atoms, t is the total wallclock time, tdecorr is the decorrelation time of Langevin simulation in physical units, Acc is the average acceptance rate of LBMC simulation, ESS is the effective sample size, and SE is the standard error of the mean end-to-end distance. The factors γ1 and γ2 represent the efficiency gain of LBMC relative to Langevin dynamics and are defined in Eq. 5 and Eq. 6 respectively. System Ace-(Ala)4-Nme Ace-(Ala)8-Nme Met-enkephalin M 52 92 84 t 16h 43.5h 48h Langevin tdecorr ESS 1333 0.08ns 0.5ns 200 142 0.7ns t LBMC Acc ESS SE 454 0.07 0.69 0.22 20min 0.28 200 142 0.30 0.26 10sec 3min γ1 γ2 SE 0.10 1961 2822 0.10 632 839 0.27 145 924 5 Summary and Conclusions 25 Tab. 2: Efficiency in GBSA implicit solvent. The results of the "de-correlation" and the block averaging analyses of LBMC and Langevin simulations are reported for three peptides: Ace-(Ala)4-Nme, Ace-(Ala)8-Nme and Met-enkephalin. The peptides were sampled according to the OPLS-AA forcefield with GBSA solvent. M is the number of atoms, t is the total wallclock time, tdecorr is the decorrelation time of Langevin simulation in physical units, Acc is the average acceptance rate of LBMC simulation, ESS is the effective sample size, and SE is the standard error of the mean end-to-end distance. The factors γ1 and γ2 represent the efficiency gain of LBMC relative to Langevin dynamics and are defined in Eq. 5 and Eq. 6 respectively. System Ace-(Ala)4-Nme Ace-(Ala)8-Nme Met-enkephalin M 52 92 84 t 58h 147h 120h Langevin tdecorr ESS 0.2ns 500 111 0.9ns 2ns 50 LBMC Acc ESS SE 0.44 200 0.11 200 0.19 0.18 0.22 11min 0.17 50 t 58s 2h γ1 γ2 SE 0.17 1438 1507 119 0.15 0.30 367 133 524 5 Summary and Conclusions Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . Introduction . Methods . . . . 1 . 2 2.1 Library-based Monte Carlo Method (LBMC) . 2.2 Practical library generation . . 2.3 Neighbor-list construction . 2.4 Efficiency analysis . . . 2.5 System and simulation details . . . 3 . . . 3.1 Ensemble Quality . . . . . 3.2 Efficiency Analysis . . 3.3 Regarding GBSA . . . . 3.4 Neighbor-based trial moves . Discussion . . 4 . . . Summary and Conclusions . 5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 1 3 3 4 5 6 8 9 9 9 10 11 11 13 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1509.08924	1	1509	2015-08-21T12:22:54	Balance of interactions determines optimal survival in multi-species communities	[ "physics.bio-ph", "physics.soc-ph", "q-bio.PE" ]	We consider a multi-species community modelled as a complex network of populations, where the links are given by a random asymmetric matrix J, with fraction 1-C of zero entries, where C reflects the over-all connectivity of the system. The non-zero elements of J are drawn from a gaussian distribution with mean 'mu' and standard deviation . The signs of the elements J reflect the nature of density-dependent interactions, such as predatory-prey, mutualism or competition, and their magnitudes reflect the strength of the interaction. In this study we try to uncover the broad features of the interspecies interactions that determine the global robustness of this network, as indicated by the average number of active nodes (i.e. non-extinct species) in the network, and the total population, reflecting the biomass yield. We find that the network transitions from a completely extinct system to one where all nodes are active, as the mean interaction strength goes from negative to positive. We also find that the total population, displays distinct non-monotonic scaling behaviour with respect to the product 'mu'C, implying that survival is dependent not merely on the number of links, but rather on the combination of the sparseness of the connectivity matrix and the net interaction strength. Moreover, the total population levels are optimal when the network has intermediate net positive connection strengths. At the local level we observe marked qualitative changes in dynamical patterns, ranging from anti-phase clusters of period 2 cycles and chaotic bands, to fixed points, under the variation of mean 'mu' of the interaction strengths. Lastly, we propose an effective low dimensional map to capture the behavior of the entire network, and this provides a broad understanding of the interplay of the local dynamical patterns and the global robustness trends in the network.	physics.bio-ph	physics	Balance of interactions determines optimal survival in multi-species communities Anshul Choudhary and Sudeshna Sinha Department of Physical Sciences, Indian Institute of Science Education and Research (IISER) Mohali, Knowledge City, SAS Nagar, Sector 81, Manauli PO 140 306, Punjab, India. * E-mail of corresponding author: [email protected] Abstract We consider a multi-species community modelled as a complex network of populations, where the links are given by a random asymmetric connectivity matrix J, with fraction 1 − C of zero entries, where C reflects the over-all connectivity of the system. The non-zero elements of J are drawn from a gaussian distribution with mean µ and standard deviation σ. The signs of the elements Jij reflect the nature of density-dependent interactions, such as predatory-prey, mutualism or competition, and their magnitudes reflect the strength of the interaction. In this study we try to uncover the broad features of the interspecies interactions that determine the global robustness of this network, as indicated by the average number of active nodes (i.e. non-extinct species) in the network, and the total population, reflecting the biomass yield. We find that the network transitions from a completely extinct system to one where all nodes are active, as the mean interaction strength goes from negative to positive, with the transition getting sharper for increasing C and decreasing σ. We also find that the total population, displays distinct non-monotonic scaling behaviour with respect to the product µC, implying that survival is dependent not merely on the number of links, but rather on the combination of the sparseness of the connectivity matrix and the net interaction strength. Interestingly, in an intermediate window of positive µC, the total population is maximal, indicating that too little or too much positive interactions is detrimental to survival. Rather, the total population levels are optimal when the network has intermediate net positive connection strengths. At the local level we observe marked qualitative changes in dynamical patterns, ranging from anti-phase clusters of period 2 cycles and chaotic bands, to fixed points, under the variation of mean µ of the interaction strengths. We also study the correlation between synchronization and survival, and find that synchronization does not necessarily lead to extinction. Lastly, we propose an effective low dimensional map to capture the behavior of the entire network, and this provides a broad understanding of the interplay of the local dynamical patterns and the global robustness trends in the network. Introduction Complex networks provides a common framework to address a vast range of phenomena in large interactive systems [1]. The use of network theory in studying the stability and dynamics of model ecosystems started with the landmark paper of Robert May [2] and the success and effectiveness of such inquiry can be gauged from the fact that even today most studies in theoretical ecology heavily rely on the network 1 framework [3]. Our current understanding of the stability of an ecological network hinges around two key aspects: interaction topology and nature of interactions. Now, there are wide ranging situations where one either does not have sufficient information on the exact underlying topology, or one finds that the web of interactions essentially appears to be a random network. In such cases, the interactions are modelled by random connectivity matrices, and the broad nature of interactions is our only guiding principle in analyzing the dynamics and survival properties of the complex system. In this study, we are going to explore the effect of the balance of different kinds of interactions in a multi-scpecies community on the collective dynamical behaviour of the network. Our focus will be on the global robustness of the system, as exemplified by the total population of all species [4]. It is evident that the total population relects the global state of the network effectively, while being sensitive to the underlying dynamics at the local species level as well. So here we will explore how diversity in interactions influence the emergent dynamics, and the relation of these dynamical patterns to survival of populations, lending yet another perspective to the stability-diversity debate. Recently, Mougi & Kondoh [5] studied the interesting effects of diversity in interaction types on the stability of an ecological community and they found that diversity is a key element in determining stability and biodiversity. However their results are based on linear stability analysis for small perturbations about local equilibria, and they do not give the relationship between survival and the emergent dynamical patterns. In this context, our study provides a complementary exploration of the global survival features of such systems [6] and also relates it to dynamical behavior of the constituent populations. Model The model we consider here is inspired by the earlier theoretical studies conducted by Robert May [2, 7]. However, we would like to mention here that unlike most studies regarding stability [5, 8] which assumes the existence of time-independent population densities when system reaches steady state, we consider the more general condition where the attracting state can have complex temporal behavior, rather than a fixed point solution [6]. The principal motivation for this approach to the question of stability is its wider relevance and broader applicability. Further, rather than local stability about an equilibrium, we will focus on a different set of global quantifiers of robustness and survival in the complex network. Specifically, in this work we consider a prototypical map, the Ricker (Exponential) Map, modelling population growth of species with non-overlapping generations: f (x) = x er(1−x) with r interpreted as an intrinsic growth rate. We then consider the evolution of N such interacting populations given as: xi(n + 1) = f (xi(n)) + Jij xj(n) 1 N Xj (1) (2) where i = 1, . . . N , and the connectivity or community matrix, J represents how species are mutually interacting. Further we consider that xi(n + 1) = 0 if xi(n + 1) < xthreshold, where xthreshold is the minimum population density below which population cannot sustain on their own and therefore becomes extinct, namely the Allee Effect [9]. So we have a system (Eqn. 2) with well mixed populations at the nodes that display chaotic dynamics, and are extinction prone due to the population 2 threshold when uncoupled. So clearly, a persistent community can only be sustained through suitable interactions among the species. The connectivity matrix J is a random asymmetric matrix, with fraction 1 − C of zero entries, where C reflects the over-all connectivity of the system. The non-zero elements are drawn from a gaussian distribution, standard deviation σ. 1 √2πσ2 e −(x−µ)2 2σ2 , with mean µ and The signs of the elements Jij of the connectivity matrix J reflects the nature of density-dependent interactions. In general, neutral interactions are reflected by zero matrix elements. Interactions that reduce the population at a node, for instance through parasitism, grazing, and predation, will be reflected by a negative sign, while interactions that benefit a species, for instance by provding refuge from physical stress, predation or competition, will bear a positive sign. So then, when we have mutualism or symbiosm, both Jij and Jji are positive. When we have competition or antagonistic interactions, both Jij and Jji are negative. When the effect of one species on the other is positive, but neutral the other way round, we have commensalism, reflected by Jij > 0 and Jji = 0. Similarly, amensalism is reflected by Jij < 0 and Jji = 0. General predator-prey interactions are captured by Jij and Jji having different signs. In general, positive interactions or facilitative interactions between species that benefit the growth of the species, give rise to more positive mean µ. Namely, high positive µ implies the dominance of mutualism in the ecosystem where as µ ∼ 0 would imply balance of different kinds of interactions in the system. The standard deviations, σ on the other hand controls the degree of variability in the strength of these interactions. Finally, connectedness, C is another important factor that tell us how many species are interacting with each other. Namely, it reflects the fraction of neutral interactions in the network. The main aim of this study is to understand the relation between broad features of the interaction matrix and the collective dynamics of the system, and then go on to link this to the local and global survival in the system. Methods At the outset, we present our tools and describe the measures for analyzing the survival properties of the system. To gauge the robustness of the system, we first calculate the number of active nodes, namely the number of non-extinct species with non-zero population, after transience. This quantity is then averaged over a period of time T and further averaged over Nic different initial conditions. We denote this averaged number of non-extinct nodes by hNactivei, and it is defined as: hNactivei = φi(t) =(cid:26) 1 0 1 NicT XNic XT ( N Xi=1 φi(t)) where, if xi(t) > xthreshold, otherwise, (3) (4) The next important measure of global survival is the total population ΣN i=1xi, of the system, reflecting the biomass yield in a multi-species community. This quantity, averaged over a period of time T and over Nic different initial realizations, is denoted by hxtotali, and mathematically expressed as: hxtotali = 1 NicT XNic XT ( N Xi=1 xi(t)) (5) 3 Synchronization Order Parameter : In order to probe collective patterns in the network, we studied the level of synchronization that emerges in the system. To quantify the degree of synchronization we have employed two different order parameters. (i) Zsync: Here we measure synchronization error as the mean square deviation of the local state of the nodes, averaged over time T (after transience) and over Nic different realizations [23 -- 25], mathematically expressed as : Zsync = 1 NicT XNic XT ( 1 N [xi(t) − hx(t)i]2) N Xi=1 (6) When it goes to zero, this measure reflects complete synchronization in the system. (ii) Zphase: This is a phase order parameter that reflects the degree of variation in the phases of the local dynamics at the nodes. Specifically, it is a measure of the fraction of nodes in the largest phase cluster, averaged over time T and over different network realizations Nic. When Zphase = 1, it implies that the entire system is phase synchronized (though not necessarily in complete synchronization). For the specific case of the local dynamics being a 2-cycle or being in a period 2 chaotic band (which is observed in this system over a large parameter range) Zphase then is the supremum of the quantity 1 NicT XNicXT ( 1 N ϕi(t)) N Xi=1 where ϕi(t) is 1 if xi(t) lies in the specified band, and 0 otherwise. So these measures provide complementary information about the synchrony in phase and amplitude of the dynamics of the local constituents of the network. System Parameters: In this work parameter r = 4 in Eq. 1, namely the local map is in the chaotic regime, and the threshold value xthreshold = 0.0001. All results reported here are robust with respect to small variations around these values. The survival and synchronization measures were calculated by averaging over 100 random initial conditions, i.e. Nic = 100 in the equations above. The system sizes ranged from 100 ≤ N ≤ 800, with connectedness 0 ≤ C ≤ 1 and standard deviation 0.1 ≤ σ ≤ 0.5 in the connectivity matrix J. Further, to explore the effect of the mean interaction strength µ on the dynamics, which is a focus of our work here, we investigated the range: −1 ≤ µ ≤ 1. Note that several earlier studies have been confined to the balanced situation, where µ = 0. In the sections below, we present the principal observations from our extensive simulations over this wide-ranging window of parameters. Results and Analysis Survival in the Network We first calculate the average number of active nodes (namely the average number of non-extinct species) hNactivei, as a function of the mean interaction strength µ of the connectivity matrix J. As evident from the results displayed in Fig. 1, the average number of active nodes hNactivei in the network rises sharply as a function of mean interaction strength µ around µ ∼ 0. When the mean interaction strength is quite negative, the number of active nodes goes to zero, i.e. the entire system is driven to 4 extinction. For positive µ all nodes in the network are non-zero, i.e. no species goes extinct. The connectedness C and the variability of interaction strengths σ then does not affect the number of active nodes in the network when the network is far from balanced, namely considerably positive mean interaction strengths yield hNactivei = N , while considerably negative interactions results in hNactivei = 0, irrespective of C and σ. The transition from complete extinction to a completely active network is sharper for systems with low variability in interaction strengths (i.e. low σ), and for systems with higher connectedness (i.e. high C). To gain further quantitative understanding of the nature of this transition, we explore the scaling behaviour near the transition, and discover that the average number of active sites scales with respect to C as: hNactivei ∼ Θ((µ − µc)C α) Further the number of active sites scales with respect to σ as: hNactivei ∼ Ω((µ − µc)σβ ) (7) (8) Here, α and β are appropriate critical exponents for scaling functions Θ and Ω respectively. A good data collapse, shown in Fig. 1 (insets), is obtained for µc = 0, with α = 0.45 ± 0.02 and β = 1. These scaling relations suggest that a transition from complete extinction to a fully active ecosystem occurs around µ = 0, namely around the state of balanced interaction strengths or completely neutral interactions [10]. We also performed finite size scaling with respect to system size N , and found the simple scaling: hNactivei ∼ N f (σ, µ, C), implying that the active fraction hNactivei/N is independent of N . The next important measure of global survival is the average total population hxtotali of the system, reflecting the biomass yield in a multi-species community. The variation of hxtotali, as a function of mean µ of the interaction strengths and connectedness C of the interaction matrix J, is shown in Figs. 2-3. It is clear that for C = 0, i.e, when there are no interactions, the local extinctions accumulate, eventually leading to mass extinction. When interactions are present, different global scenarios emerge with respect to varying mean µ and connectedness C of matrix J. For fixed C, the total population increases sharply with increasing µ, namely with increasing net positive interactions, around µ ∼ 0. So we find that for networks close to the balanced situation, we have enhanced population densities indicating greater survival, for increasing net positive interactions. Further, there exists an interval of mean µ, around µ ∼ 0, where the average total population always increases with the increase in the number of interactions among species, namely increasing C. This would imply that connectivity always enhances survival of the system here. However, when mean µ is smaller, or larger, than the above interval, one finds that at intermediate values of connectedness, the population is the largest(cf. Fig.2). Namely, a mix of neutral interactions along-side other interactions is most conducive to enhanced population yeild. There is a critical negative mean, µextinction c (where µextinction c is a function of C) for which the local species experience severe loss of population leading to global extinction. There is also a critical positive mean, µexplosion such that for mean µ > µexplosion growth, destabilizing the whole network. We consider µ < µexplosion the nodes experience unbounded and explosive in our study. Interestingly, we uncovered a scaling pattern between the total population and c c , where µexplosion C ∼ 1, c c characteristics of the connectivity matrix. The data collapse of the population onto a single non-monotonic curve in Fig. 3B reveals a scaling relation between total 5 population and the product of the mean interaction strength, µ and connectedness, C of the network. This implies that the most relevant quantity in understanding the global behaviour of the network is µC, rather than µ and C alone. So clearly, survival is dependent not merely on the number of links, but on the combination of the sparseness of the connectivity matrix and the net interaction strength. For instance, fewer interactions (i.e. low C) tends to decrease the population, but this effect may be compensated by more positive interactions, i.e. higher µ. More importantly, the existence of an intermediate window of positive µC where the total population is maximum indicates that too little or too much positive interactions is detrimental to survival. Infact survival is optimal when the network has intermediate net positive connection strengths. So counter intuitively, if positive interactions such as mutualistic or symbiotic links dominate other kinds of interactions too much, its effect ceases to be beneficial, causing the total population to reduce. Local Dynamics Now we attempt to correlate these global features to local species-level dynamics. Namely, we attempt to correlate the survival and global stability of the ecological network to dynamical patterns emerging in the network as a result of interactions. From the bifurcation diagrams displayed in Fig. 4, one can clearly discern the presence of coherent collective dynamics in the system. This coherence breaks down as one approches µ = 0, as evident in Fig. 4, with the network displaying unsynchronized spatio-temporal chaos. In order to gauge the degree of synchronization among the nodes quantitatively we calculate the synchronization order parameter Zsync. Our attempt now will be to find the correlation between synchronization and survival. This is an important question, as synchronization has often been seen as increasing risks of extinction. Fig. 5 exhibits this synchronization error, along side the number of active patches (i.e. nodes with non-zero population), the total population and collective dynamics of the whole network. It is clear that synchronization does not necessarily lead to extinction. In fact for positive mean interactive strengths, even when the entire system is completely synchronized, all nodes are active. The rationale for the above observation is that synchronization is a threat only when the synchronized dynamics covers a large range of population densities, such as in synchronized chaos, which typically is ergodic over state space. Here on the other hand, the synchronized dynamics is confined to the "safe zone" and the attractor trajectory does not enter the extinction region [11]. So the synchronized patches survive. We further investigate the nature of the time series of the local patches to discern cluster formation, and the phase relation between the clusters. We find that when the mean of the interaction strengths has a low positive value, the populations are attracted to a period 2 cycle, and the sytem divides into two anti-phase clusters (cf. Fig. 6). Namely, alternately in time, one set of nodes in the network have low population densities, while the other set has high population densities. This behaviour is reminiscent of the field experiment conducted by Scheffer et al [12] which showed the existence of self-perpetuating stable states alternating between blue-green alage and green algae. We also studied the phase clusters emerging in the system, by calculating a phase order parameter Zphase, which gives the fraction of species whose dynamics are in-phase in the network. This quantity is ∼ 0.5 in a large range of positive mean interaction strengths (µ ∼ 0.1 − 0.6), indicating that here the network always splits into two approximately equal clusters. In each cluster the nodes are in-phase with respect to each one another, and anti-phase with respect to nodes in the other cluster. However note that the degree of complete synchronization, which depends on both 6 phase and amplitude, will be dependent on µ (as evident from Fig. 5). So changing the mean interaction strength changes the nature of the dynamics without destroying this phase relationship and two phase synchronized clusters of varying amplitudes emerge in a large range of µ. Effective map for nodal dynamics To gain further understanding of the dynamical patterns, we construct an effective map to mimick the essential dynamics of the nodes. Our approach is to split the interactive part in Eqn. 2 into an average part and a term capturing the fluctuations. Here the mean interaction strength, which is the dominant contribution, is µC, as there are a fraction C of non-zero interaction strengths drawn from a distribution with mean µ. So as first approximation, neglecting fluctuations, we can model the local dynamics as: X(n + 1) = f (X(n)) + µC X(n) (9) when X(n + 1) > xthreshold, and X(n + 1) = 0 otherwise. Such an effective map is an accurate representation of the population dynamics when there is a high degree of coherence in the system. To further gauge if the bifurcation diagram obtained numerically in Fig. 4(d) can be understood qualitatively using this effective map picture, we argue that the deviations from the effective dynamics can be modelled by random fluctuations about the mean interaction strength µC. So we study the dynamics given by Eqn. 9 under the influence of multiplicative noise as well. The results are shown in Fig. 7. It can be clearly seen that the effective dynamical map, under random noise, qualitatively captures the collective dynamics of the multi-species communities. Analysis: We can also straight-forwardly analyse the effective map dynamics given by Eqn. 9 to find the windows where the positive steady state is stable. Note that this non-trivial fixed point, which is a function of µC, can be obtained as a solution of: X∗ = 1 1 − µC f (X∗) (10) and its stability is determined by the condition f ′(X∗) + µC < 1. Clearly as µC → 1, the fixed point becomes unboundedly large. This also explains the presence of the critical positive mean µexplosion C ∼ 1, in the system. From Fig. 8A it is clear that the parameter yielding fixed populations in the system is very close that found in the effective map (cf. Fig. 7). , with µexplosion c c Further we employed another, more accurate, approach to gauge the stability of the synchronized steady state. In this extension, the stability analysis takes into account the entire network by considering the extremal eigenvalues of the connectivity matrix. This yields stability conditions on the fixed point (which is a solution of Eqn. 10) given by: f ′(X∗) + λmax < 1, f ′(X∗) + λmin > −1 (11) where, λmax and λmin are the average maximum and minimum eigenvalues of the random gaussian matrix, J. From Fig. 8B, one observes that the region where synchronized steady state is stable, namely where f ′(X∗) lies within the two bounds, corresponds quite closely to µ ∼ 0.8. This matches the value observed in simulations very closely (cf. Fig.4), and thus provides an accurate description of the effective collective behavior of the system. Lastly, consider the scenario that leads the dynamics of the nodes to extinction, namely to X∗ = 0. This will happen if X(n + 1) < xthreshold, which then will map to 7 X = 0. Notice that for populations very close to zero, f (X) ∼ 0. So from Eqn. 9, X(n + 1) ∼ µC. This implies that the subsequent iterate can become negative if and only if µ is negative, as C is non-negative and f (X) > 0 if X > 0. This suggests why extinctions are seen to arise for µ < 0. Effect of the degree of variability in inter-species interactions Having gained understanding of the collective dynamics of the system in terms of the dynamics of a single effective stochastic map, we now try to understand the effect of the standard deviation σ of the connectivity matrix J on the dynamics of the multi-species community. It seems reasonable to argue that the strength of the noise term in the effect dynamical map is directly related to σ, namely we can associate the stochasticity in the effective map to variability in the inter-species interaction strength across the network. Thus we investigate the changes in the bifurcation structure for two different values of σ, which represents different degrees of spatial variability in the network. From Fig. 9, one can observe that with increasing σ, the bifurcation diagram gets more noisy. This indicates that one can incorporate the spatial variability in interaction structure easily in the effective map. Also, as discussed earlier, for populations close to zero, Eqn. 9 effectively gives X(n + 1) ∼ µC, and these are driven to extinction if X(n + 1) < xthreshold. This implies that the sign of the subsequent iterate for a system close to zero is very sensitive to large fluctuations in the distribution of interaction strengths. Namely, large variability around the mean value µC in Eqn. 9 can push the system into the extinction zone, or out of it, when µ, C and X are close to zero. This accounts for the spread in hNactivei values around µ ∼ 0, in the presence of large σ in Fig. 1. Discussions In summary, we have analyzed the survival properties in ecological networks. In particular, we considered a complex network of populations where the links are given by a random asymmetric connectivity matrix J, with fraction 1 − C of zero entries, where C reflects the over-all connectivity of the system. The non-zero elements are drawn from a gaussian distribution with mean µ and standard deviation σ. The signs of the elements Jij of the connectivity matrix J reflect the nature of density-dependent interactions, such as predatory-prey, mutualism or competition, and their magnitude reflect the strength of the interaction. Unlike many earlier studies, we investigate the full range of mean interaction strengths, and do not confine ourselves to the balanced situation which assumes µ = 0. Also note that one can potentially draw a parallel between our model and the system of metapopulations with density dependent dispersal [13]. Namely, our system can also be interpreted as a network of metapopulation patches [14], or "a population of populations" [15]. In particular, it can describe a system comprising many spatially discrete sub-populations connected by migrations where inter-patch dispersal is both high enough to ensure demographic connectivity among patches, yet low enough to maintain some degree of independence in local population dynamics. The connectivity matrix in this scenario reflects density dependent dispersal and migration, as is commonly seen in vertebrate and invertebrate populations [16 -- 21]. A problem of vital importance here is understanding how broad features, such as the connectedness and net positive interaction strength, modulates the emergent dynamics in such a network. First, in order to gauge the global stability of the system, we calculate the average number of active nodes, namely number of non-extinct species, in the network. We find that the network transitions from a completely 8 extinct system to one where all nodes are active, as the mean interaction strength goes from negative to positive. This transition, marked distinctly by scaling relations, gets sharper with increasing C and decreasing σ. This result has much relevance, as realistic ecosystems are unlikely to have a perfect balance of interactions. So understanding the implications of imbalance in interaction types and strengths in the network (namely µ 6= 0) is important. Another significant observation is that the total population, reflecting the biomass production in a multi-species community, displays distinct non-monotonic scaling behaviour with respect to the product µC, implying that survival is dependent not merely on the number of links, but rather on the combination of the sparseness of the connectivity matrix and the net interaction strength. Interestingly, in an intermediate window of positive µC, the total population is maximal, indicating that too little or too much positive interactions is detrimental to survival. Infact survival is optimal when the network has intermediate net positive connection strengths. Counter-intuitively then, if positive interactions such as mutualistic or symbiotic links are too dominant, its effect ceases to be beneficial and in fact results in reduction of the total population. At the local level we observed marked qualitative changes in dynamical patterns, ranging from fixed points to spatioteporal chaos, under variation of mean µ of the interaction strengths. Specifically we found anti-phase clusters of period 2 cycles and the presence of period-2 chaotic bands, in certain windows of mean µ. This behaviour is reminiscent of the field experiment conducted by Scheffer et al [12] which showed the existence of self-perpetuating stable states alternating between blue-green alage and green algae. We also studied the correlation between synchronization and survival, and find that synchronization does not necessarily lead to extinction. Lastly, we proposed an effective low dimensional map to capture the behavior of the entire network, and this provides a broad understanding of the interplay of the local dynamical patterns and global stability of the network. References 1. M. Newman, Networks: An Introduction (Oxford University Press, Oxford, UK, 2010). 2. R. May, Nature, 238, 413-414(1972). 3. See representative work: B Blasius, A Huppert, L Stone, Nature 399 (1999), 354; T Gross, B Blasius, J. of R. Soc. Interface 5 (2008), 259; M. Baurmann, T. Gross, U. Feudel, J. of Theor. Bio. 245 (2007) 220. 4. A. R. Ives & S. R. Carpenter, Science, 317 58-62(2007). 5. A. Mougi & M. Kondoh, Science, 337, 349-351(2012). 6. S. Sinha and S. Sinha, Phys. Rev. E, 71, (2005) 020902; S. Sinha and S. Sinha, Phys. Rev. E, 74, (2006) 066117. 7. R. May, Ecology, 54, 638-641(1973). 8. S. Allesina & S.Tang, Nature, 483, 205-208(2012).. 9. PA Stephens, WJ Sutherland & RP Freckleton, Oikos(JSTOR), 87, 185 (1999) 10. N.K. Kamal and S. Sinha, Chaos, Solitons and Fractals 44 (2011) pp. 71-78 11. A. Choudhary, V. Kohar, S. Sinha, Eur. Phys. J. B 87 (2014), 202 9 12. M. Scheffer, S. Rinaldi, A. Gragnani, L. R. Mur, & E. H. van Nes Ecology, 78, 272-282(1997). 13. I Hanski, Nature, 396, 41 (1998). 14. R Levins, Extinction. In Some mathematical questions in biology (ed. M. Gerstenhaber), pp. 77 -- 107. Providence, RI: The American Mathematical Society (1970). 15. R Levins, Bulletin of the ESA, 15, 237 (1969). 16. B Kerr, MA Riley, MW Feldman & BJM Bohannan, Nature, 418, 171 (2002). 17. B Kerr, C Neuhauser, BJM Bohannan & AM Dean, Nature, 442, 75 (2006). 18. S Dey & A Joshi, Science, 312, 434 (2006). 19. JH Brown & A Kodric-Brown, Ecology, 58, 445 (1977). 20. T Reichenbach, M Mobilia & E Frey, Nature, 448, 1046 (2007) 21. B Cazelles, S Bottani & L Stone, Proc. R. Soc., B, 268, 2595 (2001) 22. Stachowicz J.J., Mutualism, facilitation, and the structure of ecological communities, Bioscience (2001) 51 235 23. Amritkar, R. E., & Rangarajan, G. (2006) Phys. Rev. Letts., 96, 258102. 24. Nishikawa, T., & Motter, A. E. (2010). PNAS 107, 10342. 25. Sorrentino, F., & Ott, E. (2008) Phys. Rev. Letts., 100, 114101. 10 σ = 0.1 σ = 0.2 σ = 0.3 σ = 0.4 σ = 0.5 (A) -0.40 -0.2 0.2 0 µ-µ c -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 C = 0.0 C = 0.2 C = 0.3 C = 0.5 C = 0.7 C = 0.8 C = 1.0 (B) 100 80 60 40 20 0 -0.5 100 80 60 40 20 s e d o N e v i t c A f o . o N . g v A s e d o N e v i t c A f o . o N . g v A 0 -0.25 -0.15 -0.40 -0.2 0.2 0 µ-µ c 0 µ 0.15 0.25 Figure 1. Average number of active nodes, hNactivei, as a function of mean µ of the interaction strengths of the connectivity matrix J, for different values of connectivity C and different variabilities in interaction strengths σ. Here system size N = 100, and so hNactivei = N = 100 in the figures reflect a network where no species becomes extinct. The insets show the collapse of the scaled hNactivei, with respect to µ − µc, where critical µc = 0. 11 s s e n d e t c e n n o C 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 300 250 200 150 100 50 0 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 Mean Figure 2. Average total population hxtotali of a system of size N = 100 (represented by the color scale), as a function of mean µ of the interaction strengths and the connectedness C (giving the number of non-zero entries in connectivity matrix J). Here standard deviation σ of the non-zero matrix elements is 0.5. 12 (A) (B) µ C = 0.1 C = 0.3 C = 0.5 C = 0.8 C = 1.0 C=0.2 C=0.3 C=0.5 C=0.8 C=1.0 300 200 100 0 300 200 100 l n o i t a u p o P l a t o T n o i t a l u p o P l a t o T 0 -0.4 -0.2 0 0.2 µC 0.4 0.6 0.8 0.95 Figure 3. Average total population of hxtotali of a system of size N = 100, as a function of mean µ of the interaction strengths for (A) different values of connectedness parameter C. In (B), the average total population is given as a function of the product µ C, and it is evident that the data collapses onto a single curve implying a functional relation between total population and µ C. Here the average total population is obtained by averaging the total over 100 random realizations, and the standard deviation σ of the connectivity matrix J is 0.1. 13 C = 0.3 C = 1 5 4 3 2 1 ) t ( i X 0 0.2 0.6 0.4 Mean, µ 0.8 0 0.2 0.4 0.6 Mean, µ 0.8 1 Figure 4. Bifurcation Diagram of the population dynamics of the network of 100 species, as a function of mean µ of the interaction strengths, for representative values of connectedness: (a) C = 0.3 and (b) C = 1. Here we show xi(t), for all i = 1, . . . N , over a period of time, after transience. Standard deviation σ of J is 0.1. On careful observation we found that all bifurcation diagrams collapse on to a single pattern when viewed as a function of µC. 14 6 5 4 3 2 1 0 -0.2 0 0.2 0.4 Mean, µ 0.6 0.8 1 Figure 5. Synchronization error (dotted black), total population (red), bifurcation diagram for the emergent behavior of the population densities (cyan) and the average number of active nodes (blue), as a function of mean µ of the interaction strengths. Here the connectivity matrix J has C = 1, σ = 0.1 and the size of network is N = 100. 15 (A) (C) ) t ( i X 6 5 4 3 2 1 0 ) t ( i X 4.5 4 3.5 3 2.5 2 1.5 1 (B) (D) 6 5 4 3 2 1 1.475 1.47 1.465 1.46 1.455 1.45 1.445 9920 9922 9924 9926 9928 9930 9920 9922 9924 9926 9928 9930 Time Time Figure 6. Time series of 10 representative sites in a network of 100 species, for different mean µ of the interaction strengths: (a) µ = 0 (b) µ = 0.12 (c) µ = 0.40 and (d) µ = 0.85. Here the connectivity matrix J has C = 1 and σ = 0.1. 16 (A) (B) 6 5 4 ) t ( i X 3 2 1 0 0.0 0.5 µC 1 0.0 0.5 µC 1 Figure 7. Bifurcation diagrams of (Left) the effective map given by Eqn. 9, as a function of µC and (Right) the same map under fluctuations given as: X(n + 1) = f (X(n)) + (µC + Dξ) X(n) , where ξ is a random variable drawn from a zero-mean gaussian distribution and D governs the strength of fluctuations. Here D = 0.0003. 17 (A) f'(x)+ µC 2 1.5 1 0.5 0 0.4 0.5 0.6 0.7 µC (B) f'(x) Lower Bound Upper Bound 1 0.5 0 -0.5 -1 -1.5 -2 0.8 0.9 1 -0.4 -0.2 0 0.2 Mean, µ 0.4 0.6 0.8 1 1.2 Figure 8. (A) Stability curve for the fixed point obtained from Eqn. 10. The fixed point loses stability when f ′(x∗) + µC > 1, and this occurs for µC ∼ 0.6, which is consistent with the results in Fig. 7(B) Upper bound (1 − λmax), lower bound (−1 − λmin) and f ′(x∗). The region where the f ′(x∗) lies between upper bound and lower bound (cf. Eqn. 11) gives us the region of stability of global synchronized state of the network. 18 (A) (B) 6 5 4 ) t ( i X 3 2 1 0 0.0 0.5 Mean, µ 1 0.0 0.5 Mean, µ 1 Figure 9. Bifurcation diagram of the population dynamics of all the species as mean interaction strength µ is varied, for two different values of standard deviation: (a) σ = 0.05 and (b) σ = 0.5. For this representative case, the network is taken to be fully connected, i.e. C = 1. 19
1302.3797	1	1302	2013-02-15T16:30:11	Traffic jams, gliders, and bands in the quest for collective motion	[ "physics.bio-ph", "cond-mat.stat-mech" ]	We study a simple swarming model on a two-dimensional lattice where the self-propelled particles exhibit a tendency to align ferromagnetically. Volume exclusion effects are present: particles can only hop to a neighboring node if the node is empty. Here we show that such effects lead to a surprisingly rich variety of self-organized spatial patterns. As particles exhibit an increasingly higher tendency to align to neighbors, they first self-segregate into disordered particle aggregates. Aggregates turn into traffic jams. Traffic jams evolve toward gliders, triangular high density regions that migrate in a well-defined direction. Maximum order is achieved by the formation of elongated high density regions - bands - that transverse the entire system. Numerical evidence suggests that below the percolation density the phase transition associated to orientational order is of first-order, while at full occupancy it is of second-order. The model highlights the (pattern formation) importance of a coupling between local density, orientation, and local speed.	physics.bio-ph	physics	Traffic jams, gliders, and bands in the quest for collective motion Fernando Peruani,1 Tobias Klauss,2 Andreas Deutsch,2 and Anja Voss-Boehme2 1Max Planck Institute for the Physics of Complex Systems, Nothnitzer Str. 38, 01187 Dresden, Germany 2Zentrum fur Informationsdienste und Hochleistungsrechnen, Technische Universitat Dresden, Zellescher Weg 12, 01069 Dresden, Germany (Dated: September 6, 2018) We study a simple swarming model on a two-dimensional lattice where the self-propelled particles exhibit a tendency to align ferromagnetically. Volume exclusion effects are present: particles can only hop to a neighboring node if the node is empty. Here we show that such effects lead to a surprisingly rich variety of self-organized spatial patterns. As particles exhibit an increasingly higher tendency to align to neighbors, they first self-segregate into disordered particle aggregates. Aggregates turn into traffic jams. Traffic jams evolve toward gliders, triangular high density regions that migrate in a well-defined direction. Maximum order is achieved by the formation of elongated high density regions - bands - that transverse the entire system. Numerical evidence suggests that below the percolation density the phase transition associated to orientational order is of first-order, while at full occupancy it is of second-order. PACS numbers: 87.18.Gh, 87.10.Hk, 05.65.+b, 05.70.Ln Self-propelled particle (SPP) systems are found at all scales in nature. Examples in biology range from human crowds [1] and animal groups [2, 3], down to insects [4, 5], bacteria [6], and even to the microcellular scale with, e.g., the collective motion of microtubules driven by molecu- lar motors [7]. SPP systems are not restricted to living systems. There are examples in non-living matter, as for instance in driven granular media [8 -- 10]. Interestingly, the statistical properties of the large-scale self-organized patterns emerging in SPP systems depend only on a few microscopic details: the symmetry associated to the self- propulsion mechanism of the particles, which can be ei- ther polar [11 -- 13] or apolar [14], the symmetry of the velocity alignment mechanism, which can be either fer- romagnetic [11, 12] or nematic [13, 14], and very impor- tantly, the presence or absence of volume exclusion ef- fects, as we are going to discuss here. In addition, the nature of the supporting space where the particles move plays also a crucial role: the dimension of the space, and whether this space is continuous [11 -- 14] or discrete [15 -- 18]. In this letter, we focus on polar SPPs moving on a two-dimensional lattice that align their orientation, re- spectively their moving direction, via a local ferromag- netic alignment mechanism. We explicitly model vol- ume exclusion effects: nodes can be occupied at most by one particle. We show that such effects introduce a coupling between particle speed, local density and local alignment that lead to a surprisingly rich variety of self- organized spatial patterns unseen in previous swarming models. As particles exhibit an increasingly higher ten- dency to align to neighbors, the system passes through three distinct phases. For weak alignment strength, the system exhibits orientational disorder, while particles self-segregate, Fig. 1(a). Within this initial phase, there is a transition from a spatially homogeneous to an ag- gregate phase. The onset of orientational (polar) order marks the beginning of the second phase which is charac- terized by the emergence of locally ordered, high density regions: traffic jams, Fig. 1(b). As the tendency to align is enhanced, traffic jams evolve toward triangular high density aggregates that migrate in a well-defined direc- tion. We refer to these dynamical traffic jams as gliders, Fig. 1(c). The third phase emerges when the particles self-organize into highly ordered, elongated, high density regions: bands, Fig. 1(d). In contrast to the traveling bands observed in off-lattice SPP models with ferromag- netic alignment [12], these bands are formed by particles aligned to the long axis of the band and are rather static. We find evidence that the phase transition to ori- entational order is discontinuous below the percolation threshold. When the lattice is fully occupied, the sys- tem reduces to the classical planar Potts model and the phase transition to orientational order is undoubtedly of second-order [24, 25]. Previous lattice swarming mod- els were found to exhibit a continuous phase transition from a homogeneous to a condensed phase in 1D [17, 18], while in 2D, both, first and second-order transitions to orientational order have been claimed. Bussemaker et al. reported a second-order transition in a cellular automa- ton model with 4 moving directions [15], while Csah´ok and Vicsek found, for a lattice-gas model with 6 mov- ing directions, a weakly-first-order transition to collec- tive migration [16]. Here we show that all these phenom- ena occur in a minimal 2D lattice swarming model, but where in contrast to previous models the system dynam- ics is dominated by volume exclusion effects that lead to a completely novel spatial self-organization of particles. Model. Particles move on a two-dimensional lattice with periodic boundary conditions, and have four possible ori- entations: up, down, left, and right, whose associated vectors are v1, v2, v3, and v4, respectively. The state of 2 FIG. 1: (Color online) Example of self-organized spatial patterns: (a) disordered aggregates, (b) traffic jams, (c) gliders, and (d) bands. Particle orientation is indicated by the orientation of the small triangles, and is also color-coded: right (red), left (black), down (green), and up (blue). Parameter values are indicated in Fig. 2. a particle is given by its position on the lattice and its orientation. As it will become clear below, the orienta- tion of a particle fully determines its moving direction. The particles are able to perform two actions: i) they can change their orientation, and ii) they can migrate in the direction given by their orientation. These actions have associated transition rates which specify the aver- age number of events per time unit. Let us start out with the reorientation transition rate TR that a particle at x with orientation v turns its orientation into direction w: TR ((x, v) → (x, w)) = exp(g Xy ∈A(x) hwV(y)i) , (1) where the sum runs over the nearest lattice neighbors of x, represented by A(x). The vector V(y) returns the orientation of the particle placed in node y, if there is any, and the null vector otherwise. The symbol h..i in- dicates the inner product between two vectors, while g is a parameter which controls the alignment sensitivity. For positive g, eq. (1) defines a stochastic ferromagnetic alignment mechanism. As result of this alignment, first nearest neighbor particles tend to be aligned. The migration rate is defined in the following way: TM ((x, v) → (y = x + v, v)) = (cid:26) v0 if node y is empty if node y is occupied. 0 (2) Thus, a particle at position x and pointing in direction v migrates to the neighboring node y = x + v with a transition rate v0 as long as the node at y is empty. If the y-node is occupied, the particle will not jump. The only action that is allowed to the particle in this situation is to change its orientation. At this point, it is important to understand how this continuous-time process is simulated. Let us assume that at time t0 the system is in a given state. Then, we com- pute the time at which the next event will take place in the system, i.e., we calculate t1. Now we have to de- cide which is the event that will take place at time t1. We choose at random one out of all the possible events that could take place, but we weight each of these events according to their associated transition rate, eqs. (1) 1 0.8 L = 49 L = 97 L = 193 > m < 0.6 0.4 (b) (I) strong clustering 0.2 0 0 (a) 0.5 (d) (c) (III) oriented bands traffic jams & gliders (II) g* 1 g 1.5 gb 2 FIG. 2: (Color online) Mean orientation hmi vs. the align- ment sensitivity g for systems with density d = 0.3 and mi- gration rate v0 = 100. Simulations were carried out for 2 107 time steps. The different curves correspond to different sys- tem sizes L. The boundary between phase I and II, g ∗, and the between phase II and III, gb are indicated by the verti- cal dashed lines. (a) to (d) refer to the simulation snapshots shown in Fig. 1. and (2). This procedure is an adaption of the classi- cal Gillespie algorithm [19] to interacting particle sys- tems [20]. m(t) = (1/N )Px Results. We start by fixing the migration rate v0 and particle density d, and use g as control parame- ter. The degree of orientational order in the system at time t is characterized by the (global) orientation V(x), where N is the total number of particles in the system, the sum runs over all lattice sites, and V(x) is defined as above. Fig. 2 shows the behavior of the mean orientation hmi as function of the alignment sensitivity g, with h. . .i a temporal average taken once the system reaches the steady-state after a short transient. The system exhibits a phase transition to orientational order above a critical g∗ for all densities, as long as v0 > 0. Here we focus on 0 < d < dp, where dp refers to the (site) percolation threshold in a 2D (square) lattice, dp ∼ 0.59. The system exhibits three phases with g, labeled I, II, and III by increasing alignment sensitiv- ity g. Phase I corresponds to g < g∗ and is characterized by exhibiting no macroscopic orientational order. Fig. 3 shows that within phase I there is a dynamic phase transition from a spatially homogeneous to an aggregate 3 FIG. 3: (Color online) Phase transition to aggregation (phase I). (a) to (d) correspond to simulation snapshots whose value of g is indicated in (f) (other parameters as in Fig. 2). The average cluster size hki normalized by the total number of particles, d.L2, as function of g is shown in (e) for various system sizes, and in (f) for various densities. (Color online) Phase transition to orientational or- FIG. 4: der. Time series m(t), histogram, and typical spatial configu- rations for phase II, g ∗ < g ≤ gb. Letters in (c) indicate sim- ulation snapshots shown in (d)-(f). Parameters as in Fig. 2. phase as g is increased. The degree of aggregation is characterized by the average cluster size hki, which is computed as the temporal average of hki = Pk k p(k, t), with p(k, t) = k nk(t)/N , where nk(t) is the number of clusters of mass k at time t and N = d L2 is the number of particles in the system. The figure shows that there exists a critical value ga above which a phase transition to aggregation occurs, with ga < g∗. As g approaches g∗, more than 85% of the particles in the system form a large aggregate. The transition point between phase I and II is at g∗, where the curves hmi(g, L) for different system sizes L meet. Interestingly, g∗ seems to be inde- pendent of the density d, as confirmed with simulations for various system sizes with density d = 0.2, 0.3, and 0.4 (data not shown). Moreover, g∗ coincides with the critical point for the full occupancy problem, i.e., d = 1. Fig. 4(a) shows time series of m(t) for values of g close to g∗. The order parameter m(t) exhibits fluctuations between high and low values. Low m-values correspond to the appearance of round traffic jams, while high values correspond to elongated traffic jams where two directions dominate over the other two. Traffic jams results from the jamming of four particle clusters attempting to move to the left, right, upward, and downward, respectively. Fluctuations are due to the competition between these four clusters. Fig. 4(b) shows that the distributions p(m) obtained from the m(t) time series for values of g close to g∗ do not exhibit a Gaussian shape as expected for a second order transition, but rather a bimodal distribu- tion (the arrow indicates the second peak) as expected for first order transitions. The coexistence of several particle configurations (or "phases") exhibiting different degrees of ordering, i.e., values of m, is evident for values of g deep into phase II. Fig. 4(c) shows that m(t) jumps be- tween well-defined values corresponding to different spa- tial patterns. Values of m(t) ∼ 1/√2 are associated to gliders, while higher values of m(t) correspond to bands. Lower values of m(t) are due to the presence of traffic jams. Gliders are dynamical traffic jams moving back- wards with respect to their mean average orientation. Their presence affects the temporal evolution of the cen- ter of mass of the system, xcm(t), which exhibits bal- listic motion whenever there is a glider, while otherwise is Brownian. As result of this, the average speed of the center of mass hVcmi peaks at values of g where gliders are more stable [21], see Fig. 5. Gliders are remarkably different from traffic jams observed in 2D traffic mod- els [22], arguably due to the presence of the alignment mechanism, Eq. (1). How frequently gliders appear and for how long they survive, depends on the value of g and L. For example, for g = 1.4, m(t) displays excursions from low to high values that reflect the fact the system alternates between traffic jams and gliders. For an illus- tration of this dynamics, see [23]. For g > gb, i.e. in (a) > m c v < 0.4 0.3 0.2 0.1 0 (I) (II) (III) (b) 120 80 g = 0.5 g = 1.5 0.5 1 g 1.5 80 120 FIG. 5: (Color online) (a) average speed of the center of mass hVcmi as function of g. (b) Trajectories of the center of mass for two different values of g. phase III, the only stable configurations is a band. contrast to g∗, gb is highly dependent on L. In Discussion. In the limit of full occupancy, d = 1, par- ticles are frozen in their positions and the only action allowed to them is reorientation. The system defined by Eqs. (1) and (2) becomes an equilibrium system in this limit, whose order is again characterized by m(t). Since, by definition, there are only four possible orientations, we can safely claim that the model reduces to a 4-state planar Potts model [24]. It has been shown that this model can be reduced to the standard 2-Potts model [25]. In two-dimensions, the standard q-Potts model exhibits a continuous transition for q ≤ 4, and a discontinuous one for q > 4 [24]. Thus, our model exhibits a second- order transition for d = 1, as confirmed via simulations (data not shown). For v0 > 0 and d < 1, we are in a pure non-equilibrium scenario. The migration rule, Eq. (2), breaks detailed balance and prevents us from writing down a free energy. Nevertheless, it is worth to compare our system with its equilibrium counterpart, the diluted Potts model with annealed vacancies. If we represent the absence of particles in a lattice position with an ex- tra vector direction, we end up with a 5-Potts system with full occupancy, instead of a diluted 4-Potts system. In consequence, according to what was said above, the transition would be first-order. The argument applies to the standard Potts model and assumes vacancies are in thermal equilibrium, which is not true for Eq. (2). Nev- ertheless, it helps to realize that a discontinuous dynamic phase transition is quite possible in our non-equilibrium system. In summary, we have shown through a minimal model that volume exclusion effects, when they are allowed to stop particle motion, can lead to a surprisingly rich va- riety of self-organized patterns. Such effects introduce a coupling between local density, local orientation and par- ticle speed that strongly affects the large-scale behavior of the system, with the jamming of particles playing a dom- inant role. This coupling is present in many real systems as in gliding bacteria, animal groups, etc. Certainly, sev- eral features of the self-organized patterns described here depend on the discrete nature of the model. Nevertheless, 4 we expect similar phenomena to emerge in off-lattice, continuum symmetry systems. For instance, static traf- fic jams are probably a robust property of all systems where stagnation can occur. Here we have also learned that the jamming of self-propelled particles can lead to unexpected self-organized structures in two-dimensions like dynamical traffic jams, e.g., gliders. The presence of an alignment mechanism induces (local) orientational or- der, and provided particles are oriented, density waves of stagnated particles should emerge. The results reported here are a first step toward a deep understanding of the possible phenomena that such a coupling may induce. We thank A. Greven and C.F. Lee for useful comments. [1] D. Helbing, I. Farkas, and T. Vicsek, Nature (London) 407, 487 (2000). [2] A. Cavagna et al., Proc. Natl. Acad. Sci. 107, 11865 (2010). [3] K. Bhattacharya and T. Vicsek, New J. Phys. 12, 093019 (2010). [4] J. Buhl et al., Science 312, 1402 (2006). [5] P. Romanczuk, I.D. Couzin, and L. Schimansky-Geier, Phy. Rev. Lett. 102, 010602 (2009). [6] H.P. Zhang et al., Proc. Natl. Acad. Sci. 107, 13626 (2010). [7] V. Schaller et al., Nature 467, 73 (2010). [8] V. Narayan, S. Ramaswamy, and N. Menon, Science 317, 105 (2007). [9] A. Kudrolli et al., Phys. Rev. Lett. 100, 058001 (2008); A. Kudrolli, Phys. Rev. Lett. 104, 088001 (2010). [10] J. Deseigne, O. Dauchot, and H. Chat´e, Phys. Rev. Lett. 105, 098001 (2010). [11] T. Vicsek et al., Phys. Rev. Lett. 75, 1226 (1995). [12] G. Gr´egoire and H. Chat´e, Phys. Rev. Lett. 92, 025702 (2004). [13] F. Peruani, A. Deutsch, and M. Bar, Eur. Phys. J-Spec. Top.,157, 111 (2008); F. Ginelli et al., Phys. Rev. Lett. 104, 184502 (2010). [14] H. Chat´e, F. Ginelli and R. Montagne, Phys. Rev. Lett. 96, 180602 (2006). [15] H.J. Bussemaker, A. Deutsch, and E. Geigant, Phys. Rev. Lett. 78, 5018 (1997). [16] Z. Csah´ok and T. Vicsek, Phys. Rev. E 52, 5297 (1995). [17] O.J. O'Loan and M.R. Evans, J. Phys. A: Math. Gen. 32, 99 (1999). [18] J.R. Raymond and M.R. Evans, Phys. Rev. E 73, 036112 (2006). [19] D.T. Gillespie, J. Phys. Chem. 81, 2340 (1977). [20] T. Klauss and A. Voss-Boehme, in Mathematical Model- ing of Biological Systems, vol. II, 353 (2008). [21] hVcmi has been computed as hVcmi = hxcm(t + τ ) − xcm(t)/τ it, with τ = 104 and h. . .i a temporal average. [22] O. Biham, A.A. Middleton and D. Levine, Phys. Rev. A 46, R6124 (1992); T. Nagatani, Phys. Rev. E 48, 3290 (1993). [23] See supplementary material at http://link.aps.org/ sup- plemental/XXXXX for a movie. [24] F.Y. Wu, Rev. Mod. Phys. 54, 235 (1982). [25] D.D. Betts, Can. J. Phys. 42, 1564 (1964).
1307.0715	4	1307	2013-11-19T14:47:30	Effect of magnesium ions on dielectric relaxation in semidilute DNA aqueous solutions	[ "physics.bio-ph", "cond-mat.soft", "q-bio.BM" ]	The effect of magnesium ion Mg2+ on the dielectric relaxation of semidilute DNA aqueous solutions has been studied by means of dielectric spectroscopy in the 100 Hz-100 MHz frequency range. De Gennes-Pfeuty-Dobrynin semidilute solution correlation length is the pertinent fundamental length scale for sufficiently low concentration of added salt, describing the collective properties of Mg-DNA solutions. No relaxation fingerprint of the DNA denaturation bubbles, leading to exposed hydrophobic core scaling, was detected at low DNA concentrations, thus indicating an increased stability of the double-stranded conformation in Mg-DNA solutions as compared to the case of Na-DNA solutions. Some changes are detected in the behavior of the fundamental length scale pertaining to the single molecule DNA properties, reflecting modified electrostatic screening effects of the Odijk-Skolnick-Fixman type. All results consistently demonstrate that Mg2+ ions interact with DNA in a similar way as Na1+ ions do, their effect being mostly describable through an enhanced screening.	physics.bio-ph	physics	Effect of magnesium ions on dielectric relaxation in semidilute DNA aqueous solutions D. Grgičin,∗ S. Dolanski Babić,† T. Ivek, and S. Tomić Institut za fiziku, 10000 Zagreb, Croatia R. Podgornik Department of Physics, University of Ljubljana and J. Stefan Institute, 1000 Ljubljana, Slovenia (Dated: February 5, 2018) The effect of magnesium ion Mg2+ on the dielectric relaxation of semidilute DNA aqueous solutions has been studied by means of dielectric spectroscopy in the 100 Hz -- 100 MHz frequency range. de Gennes -- Pfeuty -- Dobrynin semidilute solution correlation length is the pertinent fundamental length scale for sufficiently low concentration of added salt, describing the collective properties of Mg- DNA solutions. No relaxation fingerprint of the DNA denaturation bubbles, leading to exposed hydrophobic core scaling, was detected at low DNA concentrations, thus indicating an increased stability of the double-stranded conformation in Mg-DNA solutions as compared to the case of Na-DNA solutions. Some changes are detected in the behavior of the fundamental length scale pertaining to the single molecule DNA properties, reflecting modified electrostatic screening effects of the Odijk-Skolnick-Fixman type. All results consistently demonstrate that Mg2+ ions interact with DNA in a similar way as Na1+ ions do, their effect being mostly describable through an enhanced screening. PACS numbers: 87.15.H-, 82.39.Pj, 77.22.Gm I. INTRODUCTION The biological functions of the highly negatively charged DNA are intimately coupled to the positive coun- terions that neutralize them (for a review see Refs. 1 and 2). A particularly significant effect ubiquitous in biological environment is the DNA condensation in mul- tivalent salt DNA solutions [3 -- 8]. In this case the pres- ence of a multivalent ion atmosphere creates effective at- tractive interactions between nominally equally charged DNA molecules that can lead to toroidal aggregate for- mation, which seems to be the preferred morphology for high-density packaging of DNA [9]. In fact, most verte- brate sperm cells contain DNA toroidal aggregates, each measuring about 100 nm in outside diameter, condensed by arginine-rich and thus highly charged proteins. Con- densed DNA aggregates seem to be relevant also for gene packing in bacteriophages and for their potential impact in artificial gene delivery [10]. While DNA condensation is also at least in part due to electrostatic interactions, it can not be explained within the mean-field Poisson- Boltzmann imagery [11]. A radical reformulation of the theory of electrostatic interactions is needed, based on the concept of "strong coupling" between the multiva- lent salt counterions and the charges on the DNA back- bone, in order to understand the counterintuitive change in sign of electrostatic interactions between nominally equally charged bodies [12]. While the general frame- work of the strong-coupling effects is well understood, there are different ways of its exact implementation that ∗ http://real-science.ifs.hr/ † Permanent address: Department of physics and biophysics, Med- ical School, University of Zagreb, 10000 Zagreb, Croatia. accentuate different facets of the interaction between the DNA backbone and the mobile multivalent ions in the bathing solution [2, 13]. In order to elucidate and eventu- ally differentiate between different theoretical approaches more detailed experiments on the effects of multivalent ions are needed. While the investigation of equilibrium properties of multivalent counterion DNA solutions are vigorously pursued [3 -- 8], the concomitant elucidation of their dynamical properties leaves a lot to be desired. Dynamics of multivalent counterions would be in par- ticular interesting to probe in order to reveal otherwise inaccessible features of the strong-coupling attractive in- teractions between DNA molecules [11]. However, not all multivalent ions in DNA solutions act the same and they are clearly differentiated in the way they affect the ds- DNA. Early light-scattering studies indicated that DNA collapse into compact structures occurs when ≈ 90% of the DNA phosphate charges are neutralized by con- densed counterions [14]. It also became clear that in fact very few cations induce ds-DNA condensation. Uni- valent and divalent cations, excluding transition-metal ions such as Mn2+, Ni2+, and Cu2+ [15], do not con- dense ds-DNA even when present at very high con- centrations, while almost all divalent cations condense the single-stranded DNA but not the ds-DNA. On the other hand, alkaline-earth divalent cations, such as Mg2+, Ba2+ and Ca2+, do condense triple stranded DNA with a more highly charged helix than the ds-DNA form, but not the ds-DNA [16]. Furthermore, counterions Mn2+, Cd2+, Co(NH3)3+, polyamines such as spermidine3+, spermine4+, polylysine+, etc. do condense ds-DNA at finite concentrations. While electrostatics without doubt plays an important role in the DNA condensation mech- anism [2], it cannot be the sole factor affecting it as, e.g., Co(NH3)3+ is more efficient in condensing DNA than spermidine3+, both being trivalent counterions. The best condensing agents appear to be those that bind into one of the DNA grooves [17]. Based on these varied proper- ties of multivalent counterions, it is thus clear that their dynamical behavior could provide an additional clue to the puzzles posed by DNA collapse. In what follows, we thus plan to investigate the dynam- ics of multivalent counterions by using the dielectric spec- troscopy (DS) to study DNA solutions at various DNA densities and salt content. DS [18] has been successfully used to probe the dynamics of aqueous polyelectrolyte so- lutions [19, 20]. The higher-frequency dielectric response located in high MHz-GHz region (near 17 GHz) is usu- ally associated with the macromolecular hydration and the bulk dielectric response of water [21]. On the other hand, the lower-frequency modes (between 100 Hz and 100 MHz) are dominated by the counterion atmosphere around each polyelectrolyte and this fluctuating charge cloud responds to the applied electric field [20]. We re- cently resolved the counterion response to external fields into two fundamental dissipative peaks that appear as fingerprints of two relaxation modes arising from the dif- fusive motion of counterions [22 -- 25]: the dynamics of free diffuse counterions detected in the MHz frequency range probes collective properties of the polyelectrolyte solution, whereas the dynamics of condensed counteri- ons in the kHz range probes the single-chain polyelec- trolyte properties. It is noteworthy that on the latter time scale the polarization of counterions happen along an essentially stationary DNA chain. We have shown that the details of the counterion dielectric response de- pend on various parameters characterizing the polyelec- trolyte chain and the polyelectrolyte solution as a whole, such as the length of the polyelectrolyte chain, its charge density and flexibility, and the concentration of polyelec- trolyte chains, as well as the ionic strength of the added salt. DNA has been studied by DS since the early 1960s [20, 26, 27]. In the case of univalent cations such as Na1+, the MHz relaxation in the semidilute regime of DNA so- lutions corresponds to the collective (de Gennes -- Pfeuty -- Dobrynin) correlation length as the defining property. This is true for DNA as well as for other polyelectrolytes, e.g., hyaluronic acid [25]. Since the DS technique can be applied at very low polyelectrolyte concentrations, even below 10 g/L, it can be seen as complementing scatter- ing techniques, such as SAXS and SANS, that are usually applicable only for much higher polyelectrolyte solution concentrations [28]. Recent experiments by Salomon et al. [29] showed that the characteristic length scale probed in SAXS measurements corresponds to the characteristic length scale of the HF dielectric relaxation obtained by DS, thus confirming that both techniques probe the same correlation length of the semidilute DNA solutions. The specific features of DNA, not observed for other poly- electrolytes, are pronounced flexibility of short double- stranded DNA (ds-DNA) fragments [24, 30, 31] and lo- cally fluctuating regions of exposed hydrophobic cores of 2 long DNA [23]. The latter can be associated with the nucleation of DNA denaturation bubbles stemming from broken segments of several consecutive base pairs [32]. The kHz relaxation, on the other hand, reveals the single molecule properties such as the flexibility of a single DNA chain as described by its persistence length. Its vari- ation with the solution electrostatic screening (Debye) length can be rationalized by the Odijk-Skolnick-Fixman (OSF) theory, applicable to rigid and semi flexible chains [33, 34]. In order to shed additional light on the behavior of ds- DNA in solutions of polyvalent salts we thus embark on a systematic study of the dielectric response of DNA so- lutions in multivalent salt bathing solutions. As a point of departure we choose the Mg2+ cation. Our choice is based on the fact, vide supra, that Mg2+ does not in- duce any DNA condensation but nevertheless still affects DNA solution dynamics through enhanced screening, as will become evident as we proceed. This differentiation between the condensation effects and the screening ef- fects is important in order to keep a clear track of the specificities of the multivalent ions. In subsequent stud- ies we plan to move from the predominant screening ef- fects to more complicated condensation effects as present with Mn2+ or Co(NH3)3+ ions, elucidating their interre- lation and prevalence. Consistent with our previously elaborated methodology, we will probe the two dielectric response modes involving counterions in DNA solutions and attempt to rationalize our findings within the con- ceptual framework elaborated for the case of univalent salt DNA solutions and adapted specifically to the case of multivalent counterions. II. MATERIALS AND METHODS Salmon testes lyophilized Na-DNA threads were pur- chased from Sigma-Aldrich. Previous gel electrophoresis measurements showed that the majority of DNA frag- ments were in the 2 -- 20 kbp range, so that the estimated average DNA fragments were 4 µm long [35]. As for the Na-DNA solutions we were always in the semidilute regime [23]. A crude estimate based on the de Gennes arguments gives the chain overlap concentration c∗ of the order of 0.001 g/L, which is one order of magnitude below the lowest concentration found in our experiments and even if we take into account the shortest chains in our DNA solutions we estimate that c∗ ≈ 0.007 g/L, which is still lower than the lowest concentration of DNA solu- tions in our measurements [36, 37]. First we prepared the Mg2+ salt of DNA in pure wa- ter and in solutions with added MgCl2 salt. The reason for this is that we want to study DNA aqueous solutions containing only magnesium cations in order to avoid pos- sible complications due to the presence of sodium cations [38]. Thus Mg-DNA pure water solutions within con- centration range 0.01 g/L ≤ c ≤ 4 g/L were prepared by exhaustive dialysis according to the protocol as de- 3 Figure 1. Normalized absorption (A/l, where l is the light path length) of Mg-DNA pure water solutions (open circles) and of Mg-DNA solutions with added salt MgCl2, Is = 0.3 mM (full circles) as a function of DNA concentration. Inset: extinction coefficient versus Mg-DNA concentration indicat- ing double-stranded conformation in the whole concentration range. scribed in Ref. 39. Then DNA solutions with differ- ent added salt ionic strengths were made. MgCl2 was added to Mg-DNA water solution with a concentration chosen in such a way as to have the added salt ionic strength in the range 0.003 mM ≤ Is ≤ 4 mM [40]. In addition, Mg-DNA solutions with concentrations in the range 0.01 g/L ≤ c ≤ 4 g/L and with the added salt ionic strength Is = 0.3 mM and Is = 3 mM were prepared [41]. Finally, and for the sake of comparison, Na-DNA solu- tions in the presence of a MgCl2 buffer were also made according to protocols as above. UV spectrophotometry measurements of the DNA ab- sorbance intensity at 260 nm were done in order to ver- ify the nominal DNA concentration. The concentration was then determined assuming double-stranded confor- mation, implying the extinction coefficient at 260 nm equal to 20 L/gcm. The measured concentrations for DNA solutions in 1 mM of added salt were consistently smaller by about 20% than the nominal ones, as found before [23]. This difference is due to water content not taken into account by the spectrophotometry approach. Throughout this paper we refer to these measured con- centrations as the DNA concentrations. We have also used UV spectrophotometry in order to verify the sta- bility of the double-stranded conformation of Mg-DNA solutions. The obtained data clearly showed that the ds conformation was fully preserved not only in added salt, but also in the pure water solutions in the whole range of studied concentrations (Fig. 1). Dielectric spectroscopy measurements were performed Figure 2. Imaginary (ε′′) and real (ε′) part of the dielectric function at T = 25◦C of (a),(b) pure water Mg-DNA solutions and (c),(d) Mg-DNA solutions with added salt MgCl2, Is = 0.3 mM for representative a1 -- a3 (4, 0.5, 0.022 g/L) and b1 -- b3 (2, 0.4, 0.08 g/L) DNA concentrations. The full lines are fits to the sum of the two Cole-Cole forms; the dashed lines represent a single CC form. at room temperature (25◦C) using a setup which con- sisted of a home made capacitive chamber with parallel platinum electrodes and a temperature control unit, in conjunction with the Agilent 4294A precision impedance analyzer operating in ν = 40 Hz -- 110 MHz frequency range. The capacitive chamber enables reliable complex admittance measurements with reproducibility of 1.5% of samples in solution with small volume of 100 µL and with conductivities in the range 1.5 -- 2000 µS/cm. The chamber constant value is l/S = 0.1042 ± 0.0008 cm−1, where S = 0.98 cm2 is the effective electrode cross sec- tion corresponding to the sample of 100 µL and l = 0.1021 ± 0.0001 cm is the distance between the elec- trodes. Extrinsic effects, especially those due to free ions and electrode polarization, were removed by using the reference subtraction method. To this end, MgCl2 so- lutions of different molarities were chosen for reference samples and measured in addition to DNA samples. This method gives reliable results up to the concentration of 2 g/L. At higher concentrations the influence of electrode polarization is sufficiently large both due to counterions and added salt ions, so that it cannot be subtracted in a satisfactory way. Because of that we were only able to determine the parameters of the high-frequency relax- ation for concentrations below 2 g/L. Detailed measure- ment procedure and data analysis was given previously [23, 35]. Since then an additional improvement in the data analysis has been implemented. In our previous 4 Figure 3. Dielectric strength (∆ε, upper panel) and mean relaxation time (τ0, lower panel) of Mg-DNA pure water so- lutions as a function of DNA concentration. Open circles and triangles stand for the HF and LF mode, respectively. Figure 4. Dielectric strength (∆ε, upper panel) and mean relaxation time (τ0, lower panel) of Mg-DNA water solutions with added salt MgCl2; Is = 0.3 mM as a function of DNA concentration. Full circles and triangles stand for the HF and LF mode, respectively. method, the most time-consuming and difficult to auto- mate task was finding the appropriate electrolyte solution to be used as a reference: simple algorithmic matching in most cases requires an additional manual adjustment. With this in mind we have written a custom fitting pro- gram in Python using the open source Enthought Tool Suite (Traits, TraitsUI, Chaco). It provides interactive visualization and matching of the background to sample spectra. Model fitting can then be applied immediately to the resulting dielectric spectra. In the case of two over- lapping dielectric modes near the high-frequency limit of our experimental window, we additionally rely on fitting the model to the real part of conductivity. In particular, this improves the accuracy of the extracted mode width. III. RESULTS Figure 2 shows representative spectra for Mg-DNA pure water solutions and for solutions with 0.1 mM added salt. The complex dielectric spectra can be described by the sum of the two Cole-Cole functions is the dielectric strength, τ0 the mean relaxation time, and 1−α the symmetric broadening of the relaxation time distribution function of the low-frequency (LF) and high- frequency (HF) dielectric mode. The broadening param- eter of both modes is typically 1 − α ≈ 0.8. The concen- tration dependencies of the dielectric strengths and mean relaxation times for pure water Mg-DNA solutions and solutions with added salt are shown in Fig. 3 and Fig. 4, respectively. From the measured mean relaxation time τ0 we first ex- tract the characteristic length scales L along which coun- terions diffuse following the applied ac electric field. We use the expression τ0 ∝ L2/Din, where Din is the diffu- sion constant of counterions which is well approximated by the diffusion constant of bulk ions [23, 35, 42]. Since we work with Mg-DNA aqueous solutions, we use the dif- fusion constant of Mg2+ ions Din = 0.706 × 10−9 m2/s [43]. A. HF mode ε(ω) − ε∞ = ∆εLF 1 + (iωτ0,LF)1−αLF + ∆εHF 1 + (iωτ0,HF)1−αHF , (1) where ε∞ is the high-frequency dielectric constant, ∆ε Our first important result concerns the characteristic length of the HF mode which probes the collective prop- erties of DNA solutions. For pure water DNA solutions with Mg2+ counterions, the characteristic length LHF fol- 5 Figure 5. Characteristic length of the HF mode (LHF) for pure water Mg-DNA solutions (open circles) and for Mg-DNA solutions with added MgCl2 salt; Is = 0.3 mM (full triangles) and 3 mM (full circles) as a function of DNA concentration (c). The full line is a fit to the power law LHF ∝ c−0.53±0.03. The dashed lines stand for the theoretically expected Debye screening lengths for the investigated Is of added salt. lows the power law LHF ∝ c−0.53±0.03 in an almost three- decades-wide range of concentration (Fig. 5). Thus, in the semidilute solution regime with divalent Mg2+ coun- terions the de Gennes -- Pfeuty -- Dobrynin (dGPD) correla- tion length or the mesh size ξ ∝ c−0.5 [36, 44] is revealed as the most relevant length scale. This is in distinc- tion to the univalent Na1+ counterions, where the mesh size ξ characterizes the organization of DNA chains only at concentrations larger than 0.5 g/L, whereas at lower concentrations it is replaced by the exposed hydropho- bic core scaling c−0.33, corresponding to incipient denat- uration bubbles [23, 28]. As we suggested previously, this may reflect the appearance of locally fluctuating de- naturation bubbles due to the relatively weak effect of Na1+ counterions that are unable to screen the repulsion between two DNA strands at these low concentrations [22, 23]. An increased stability of the double-stranded conformation detected in the case of Mg-DNA can thus be interpreted with diminished repulsion between two DNA strands. Both effects can be associated with the pure screening action of divalent magnesium counteri- ons. These results thus suggest a much enhanced screen- ing effect of magnesium counterions as compared with sodium, eventually yielding an increased stability of the ds conformation of DNA. A very similar conclusion has been reached by Lyons et al. [38] on the basis of their counterion activity coefficient data and assuming that binding is basically determined by the long-range elec- trostatic interactions, with site-specific binding playing only a marginal role. They have also found that the ac- tivity coefficient of Mg-DNA was five times smaller than the one for Na-DNA which indicates much stronger bind- ing of magnesium cations to DNA. With MgCl2 as added salt, the behavior of LHF re- mains unchanged (Fig. 5), so that again the dGPD cor- Figure 6. Normalized dielectric strength ∆ε/(c ·L2) of the HF mode (open circles) and of the LF mode (open triangles) as a function of DNA concentration c for pure water Mg-DNA solutions. The dashed lines are guides for the eye. relation length is the only relevant length scale. This remains the case as long as the ionic strength of the Mg- DNA is larger than the ionic strength of the added salt. At lower DNA concentrations, the LHF clearly levels off, with a limiting value close to the Debye screening length theoretically expected for this salt ionic strength. Quali- tatively, this result is similar to the one found previously for Na-DNA water solutions with NaCl as the added salt. However, a significant change is found at the quantitative level indicating again an enhanced screening capability of DNA with Mg2+ cations as compared to DNA with Na1+ cations. Previously observed results indicate that NaCl added salt with the ionic strength only twice the ionic strength of Na-DNA prevails in the DNA solution screening and promotes the Debye screening length to the fundamental length scale [22, 23]. Conversely, the re- sults displayed in Fig. 5 show that adding of MgCl2 to an ionic strength of six times the effective value of Mg-DNA is needed to overcome the intrinsic screening of Mg-DNA [45]. Finally, our data enable an estimate of the number f of oscillating counterions which is given by ∆ε/(c · L2) [23]. Increasing DNA concentration leaves the fraction of free counterions participating in the HF relaxation process unchanged (Fig. 6); that is, it features qualitatively sim- ilar, concentration-independent f as found for Na-DNA [23]. This result validates the standard theoretical mod- els which use the Manning-based definition of f as the concentration-independent parameter. B. LF mode The second important result concerns the LF mode which characterizes the single-chain properties. As for Na-DNA, we find the characteristic length scale LLF ∝ c−0.23±0.02 to behave as predicted for a Gaussian chain composed of correlations blobs, that is LLF ∝ c−0.25 (Fig. 1000 Mg-DNA with added MgCl2 6 ) m n ( F L L 100 0.04 g/L Mg-DNA 0.4 g/L Mg-DNA 25oC 0.001 0.01 0.1 Is (mM) 1 10 Figure 8. Characteristic length of the LF mode (LLF) for DNA solutions with varying added salt (Is) for two represen- tative DNA concentrations: c = 0.04 g/L (stars); c = 0.4 g/L (diamonds). The full line is a fit to the expression Lp = L0 + a · I −1 s with L0 = 57 ± 5 nm and a = 4.4 ± 1.2 nm mM. 2 z 2 32.4 nm mM, where α is the dard OSF theory a = α degree of ionisation, z is counterion valency [46]. There could be several reasons for this type of discrepancy, quite probably based on the limitations inherent in the OSF theory itself that completely disregards any ion-specific effects [50]. Ion specificity is a compound effect and has many potential sources including polarizability, London- dispersion effects, and other nonpolar interactions that could play a significant role also in ion-DNA interactions [51]. While these effects can modify the ion-DNA inter- action they are as a rule long range and cannot be viewed as short-range tight binding of ions to specific sites along the DNA. Figure 8 also attests to the respective roles of intrin- sic DNA counterions and ions from the added salt in ionic screening. As in the case of Na-DNA, the OSF model applies as long as the ionic strength of added salt is larger than the ionic strength pertaining to DNA. In the opposite limit the OSF behavior is replaced by the DNA self-screening. Again, comparing with the Na-DNA case, an important quantitative difference is detected in- dicating that ten times stronger ionic strength of added salt is needed for Mg-DNA as compared to Na-DNA in order for the OSF behavior to prevail. The data for c = 0.04 and 0.4 g/L deviate from the OSF behavior for Is < 0.01 mM and for Is < 0.1 mM, respectively. At these low added salt values Is < 0.04IMg-DNA, where IMg-DNA = 4c [45], the intrinsic counterions become dom- inant in determining the behavior of LLF and indeed it attains the same value as in pure water Mg-DNA solu- tions (Fig. 8). On the other hand, Na-DNA experiments [23] show that deviation from the OSF behavior takes place when Is < 0.4INa-DNA, where INa-DNA = 3c. In the case of divalent Mg counterions it thus appears that the screening from DNA and its counterions largely overpow- ers the screening effects of added salt, thus promoting the average size of the chain, as opposed to the OSF persis- Figure 7. Characteristic length of the LF mode (LLF) for pure water Mg-DNA solutions (open triangles) and for DNA solutions with added MgCl2 salt; Is = 0.3 mM (full triangles) as a function of DNA concentration (c). The full line is a fit to the power law LLF ∝ c−0.23±0.02. The dashed line designates theoretically expected value for DNA structural persistence length L0 = 50 nm. 7). However, in Mg-DNA this characteristic length scale is shorter by about 1.5 times than in the Na-DNA [22, 23], an effect which can again be ascribed to the enhanced screening in Mg-DNA. This result is also in accord with the viscosity data which show that Mg-DNA is hydro- dynamically shorter than Na-DNA [38]. The enhanced screening in Mg-DNA is also reflected in an effective num- ber of condensed counterions participating in the LF re- laxation (Fig. 6) which, in contrast to the Na-DNA [23], decreases with an increase in DNA concentration. The latter result indicates that the concentration-independent Manning-based definition for the number of oscillating counterions is not valid in this case. A rather surprising result in comparison to Na-DNA data is then obtained in the presence of added MgCl2 salt with Is = 0.3 mM (Fig. 7). In this case, LLF stays approximately constant in the whole measured concentration range at the level of L0 = 50 nm, which suggests that in this regime LLF is proportional to the structural persistence length of DNA. The dependence of LLF on the added salt ionic strength Is is shown in Fig. 8 for two selected Mg-DNA concen- trations. The observed data follow the OSF behavior Lp = L0 + a · I −1 , where L0 is DNA structural per- sistence length and a is the effective charge density of DNA [33, 34]. From the fit to the OSF expression we get L0 = 57 ± 5 nm and a = 4.4 ± 1.2nm mM [46]. While the value of L0 close to 50 nm is in accordance with standard expectations as well as experimental re- sults for DNA structual persistence length [47, 48], the value of the coefficient a which describes the effective linear charge density is two times smaller than the value found in Na-DNA. A qualitatively similar result has been previously found by a magnetic birifrigence experiment [49]. It is noteworthy that both values of the coefficient a are much smaller than the value expected by the stan- s 7 tence length, to the fundamental length scale. V. CONCLUSION IV. DISCUSSION It is important to note that our experiments were per- formed in the absence of any other kind of cation except Mg2+. This presents an important point and allows for a straightforward interpretation when comparing our data with other published qualitatively similar results on DNA with magnesium cations (see [38] and references therein). Specifically, a shorter statistical end-to-end distance was found by Foerster resonance energy transfer (FRET) ex- periments for the single-stranded DNA [52]. In addi- tion, the FRET measurements also showed that Mg2+ is even 20 -- 40 times more efficient in screening than Na1+ in the case of single-stranded DNA. In a previous DS in- vestigation, Na-DNA solutions with concentration 0.25 g/L were studied in the presence of different amounts of MgCl2 so that the effects of increasing ratio Mg2+/base pair were tracked [53]. Owing to the restricted frequency range in that work, only the dielectric relaxation in MHz range (that is HF) was detected. The data showed an increase of the correlation length ξ and a decrease of the number of oscillating counterions ∆εHF/(c · ξ2) with increasing ratio Mg2+/base pair. The authors of that study suggested that their results might be understood as a consequence of a strong site-binding of Mg2+ so that they could not contribute to the dielectric relaxation [54]. However, these DS data showing larger values of the cor- relation length for larger amounts of Mg2+ ions can be easily understood within a framework where both sodium and magnesium cations can respond equally to applied ac electric fields. Namely, at the low Na-DNA concen- tration used in that work, the correlation length is short if compared with pure Mg-DNA due to the prevailing hydrophobic scaling with exponent 0.33 instead of the dGPD mesh size scaling with exponent 0.5. Evidently the increasing Mg2+ content leads to a larger screening thus yielding larger values of the correlation length. In- deed, we have verified this scenario in our experiments on Na-DNA solutions with MgCl2 added salt. Site-specific binding to DNA was also excluded by Raman data which indicated relatively weak interactions of Mg2+ as well as of Na1+ with both base and phosphate sites [55]. Thus we are lead to conclude that the magnesium counterions in- teract with DNA prevalently via long-range electrostatic forces and that they respond to applied ac fields in the same way as the sodium cations. In conclusion, we have analyzed the behavior of mag- nesium counterions in ds-DNA semidilute solutions by means of dielectric spectroscopy measurements. Our re- sults demonstrate that divalent magnesium cations sig- nificantly contribute to the DNA self-screening of electro- static interactions. Their screening efficiency is in fact anomalously large when compared to univalent sodium cation standard. However, with magnesium counterions there is as yet no net attraction between DNA segments that would be capable of inducing the full DNA collapse. Other types of multivalent counterions will be studied in order to probe these effects. As follows from our DS experiments, the effect of mag- nesium cations on the properties of DNA solutions can be characterized by a shortening of the statistical end-to- end distance of the DNA chains and a smaller effective linear charge density than for the Na-DNA, a clear indi- cation of an ion-specific effect. The magnesium cations also assure the stability of the double-stranded conforma- tion even in the limit of low DNA and added salt concen- tration by making the dGPD solution correlation length the most relevant fundamental length scale that describes the structural organization of DNA chains in solution. We also find that specific-site short-range strong counte- rion binding, sometimes assumed to be relevant for DNA counterion interaction, appears to not be particularly rel- evant for the interaction of magnesium cations with DNA in the studied concentration range of DNA and added salt. On the contrary, we confirm that the counterion- DNA interaction is primarily due to long-range electro- static forces acting in a similar fashion as in the case of sodium counterions. Further characterization of a dena- tured state of DNA in the presence of magnesium cations and other multivalent counterions is currently on the way. ACKNOWLEDGMENTS ICP-AES was performed by I. Ladan at the Croatian National Institute of Public Health. Discussions with T. Vuletić and D. Rau in the early stage of this work, as well as with M. Tomić are acknowledged. This work was supported by the Croatian Ministry of Science, Ed- ucation and Sports under Grant No. 035-0000000-2836. R.P. acknowledges the financial support from the Slovene Agency for Research and Development (Grants No. P1- 0055 and No. J1-4297). [1] A. A. Kornyshev, D. J. Lee, S. Leikin, A. Wynveen, Rev. [5] J. Pelta, D. Durand, J. Doucet, and F. Livolant, Biophys. Mod. Phys. 79, 943 (2007). J. 71, 48 (1996). [2] A. G. Cherstvy, Phys. Chem. Chem. Phys. 13, 9942 [6] J. Pelta, F. Livolant and J.-L. Sikorav, J. Biol. Chem. (2011). [3] D. C. Rau, V. A. Parsegian, Biophys. J. 61, 246 (1992). [4] D. C. Rau, V. A. Parsegian, Biophys. J. 61, 260 (1992). 271, 5656 (1996). [7] G. E. Plum and V. A. Bloomfield, Biopolymers 27, 1045 (1988). 8 [8] E. Raspaud, I. Chaperon, A. Leforestier and F. Livolant. Biophys. J. 77, 1547 (1999). [9] N. V. Hud and I. D. Vilfan, Annu. Rev. Biophys. Biomol. Struct. 34, 295 (2005). [10] R. Podgornik, D. Harries, J. DeRouchey, H. H. Strey and V. A. Parsegian, in Gene Therapy: Therapeutic Mecha- nisms and Strategies, 3rd ed., edited by N. Smyth Tem- pleton (Marcel Dekker, New York, 2008), pp. 443 -- 484. [11] A. Naji, M. Kanduč, R. R. Netz, and R. Podgornik, in Understanding Soft Condensed Matter via Modeling and Computations, edited by W.-B. Hu and A.-C. Shi (World Scientific, Singapore, 2010), p. 265. [12] A.Naji, M. Kanduč, J. Forsmann, and R. Podgornik, J. Chem. Phys. 139, 150901 (2013). [13] V.B. Teif and K. Bohinc, Prog. Biophys. Mol. Biol., 105, 208 (2011). [37] c∗ is given by the concentration where there is only one polymer molecule in the volume of a polymer globule c∗ = molecule mass/Vc, where molecule mass is N · mbp, c = N 3 · a3; mbp is a mass of a base pair and Vc ≈ L3 ≈ 10−18 mg, Lc = N · a; a = 0.34 nm; a is the monomer size. [38] J. W. Lyons and L. Kotin, J. Am. Chem. Soc. 86, 3634 (1964) and references there-in. [39] Stock solution was prepared by dissolving dry DNA in 25 mM of MgCl2 for 48 h at 4◦C. Next, dialysis start- ing from the stock DNA solution was performed in three steps: against 10 mM MgCl2, against 1 mM MgCl2, and finally against pure water. Each dialysis with two ex- changes with ratio 1:500 was carried out during 48 h at 4◦C. Such a procedure was sufficient to remove all sodium ions as determined by ion chromatography. [14] R. W. Wilson and V. A. Bloomfield, Biochemistry 18, [40] Note that the ionic strength and molar concentration for 2192 (1979). [15] N. V. Hud and M. Plavec, Curr. Opin. Struct. Biol. 11, 293 (2001). [16] X. Qiu, V. A. Parsegian, and D. C. Rau, Proc. Natl. Acad. Sci. U.S.S. 107, 21482 (2010). [17] A.A. Kornyshev and S. Leikin, Phys. Rev. Lett. 82, 4138 (1999); J. DeRouchey, V. A. Parsegian and D. C. Rau, Biophys. J. 99, 2608 (2010). [18] R. Buchner and G. Hefter, Phys. Chem. Chem. Phys. 11, 8984 (2009). [19] N. Nandi, K. Bhattacharyya and B. Bagchi, Chem. Rev. 100, 2013 (2000). [20] F. Bordi, C. Cametti, T. Gili and R. H. Colby, J. Phys.: Condens. Matter 16, R1423 (2004). [21] J. Hasted, Aqueous Dielectrics, (Chapman and Hall, Lon- don, 1973). [22] S. Tomić, T. Vuletić, S. Dolanski Babić, S. Krča, D. Ivanković, L. Griparić, R. Podgornik, Phys. Rev. Lett. 97, 098303 (2006). [23] S. Tomić, S. Dolanski Babić, T. Vuletić, S. Krča, D. Ivanković, L. Griparić, R. Podgornik, Phys. Rev. E 75, 021905 (2007). [24] S. Tomić, S. Dolanski Babić, T. Ivek, T. Vuletić, S. Krča, F. Livolant and R. Podgornik, EPL 81, 68003 (2008). [25] T. Vuletić, S. Dolanski Babić, T. Ivek, D. Grgičin, S. Tomić, R. Podgornik, Phys. Rev. E 82, 011922 (2010). [26] M. Sakamoto, H. Kanda, R. Hayakawa and Y. Wada, Biopolymers 15, 879 (1976). [27] M. Mandel, Ann. NY Acad. Sci. 303, 74 (1977). [28] S. Tomić, D. Grgičin, T. Ivek S. Dolanski Babić, T. Vuletić, G. Pabst, R. Podgornik, Macromol. Symp. E 305, 43 (2011). MgCl2 are related as Is = 3c. [41] The procedure starting from the stock solution was ap- plied with the final dialysis against 0.1 mM and 1 mM of MgCl2, respectively. [42] T. E. Angelini, R. Golestanian, R. H. Coridan, J. C. But- ler, A. Beraud, M. Krisch, H. Sinn, K. S. Schweizer and G. C. L. Wong, Proc. Natl. Acad. Sci. USA 103, 7962 (2006). [43] P. Vanysek in CRC Handbook of Chemistry and Physics, edited by D. R. Lide (CRC Press Taylor & Francis Group, Boca Raton, FL, 2008) pp. 5 -- 76. [44] A. V. Dobrynin and M. Rubinstein, Prog. Polym. Sci. 30, 1049 (2005); A. V. Dobrynin, R. H. Colby and M. Rubinstein, Macromolecules 28, 1859 (1995). pcp + z2 [45] Ionic strength of Na-DNA and Mg-DNA can be written as: IDNA = (1/2)[z2 i ci] where zpcp = 2c, zpcp = zici and zici = 2c. z is valency, c is the concentration of base pairs, while ci is the counterion concentration which for Na-DNA reads ci = 2c, while for Mg-DNA: ci = c. Thus it follows that the ionic strength of Na-DNA and Mg-DNA reads IDNA = 3c and I = 4c, respectively. [46] OSF expression reads Lp = L0 + Lel, where Lel = α2lB/4κ2b2 and α = 1. Taking into account the OM con- densation η = zlB/b = 1, one gets Lel = α2/z24κ2lB. Using κ−2[nm2] = 96.3I −1 S [mM] one gets Lel = (α2/z2)× 32.4I −1 [mM] . and Lel = aI −1 s s [47] V. A. Bloomfield, D. M. Crothers and I. Tinocco, Jr., Nu- cleic Acids (University Science Books, Sausalito, 2000). [48] C. G. Baumann, S. B. Smith, V. A. Bloomfield and C. Bustamante, Proc. Natl. Acad. Sci. USA 94, 6185 (1997). [49] G. Maret and G. Weill, Biopolymers 22, 2727 (1983). [50] D. Ben-Yaakov, D. Andelman, R. Podgornik, D. Harries, [29] K. Salamon, D. Aumiler, G. Pabst and T. Vuletić, Macro- Curr. Opin. in Colloid & Interface Sci. 16, 542 (2011). molecules 46, 1107 (2013). [51] V. K. Misra and B. Honig, Proc. Natl. Acad. Sci. USA [30] C. Yuan, H. Chen, X. W. Lou, and L. A. Archer, Phys. 92, 4691 (1995). Rev. Lett. 100, 018102 (2008). [31] O.-c. Lee, J. H. Jeon, W. Sung, Phys. Rev. E 81, 021906 (2010). [32] D. Poland and H. A. Scheraga, Theory of Helix-Coil Transitions in Biopolymers (Academic Press, New York, 1970). [33] T. Odijk, J. Polym. Sci.: Polym. Phys. 15, 477 (1977). [34] J. Skolnick and M. Fixman, Macromolecules 10, 944 (1977). [35] S. Tomić, D. Grgičin, T. Ivek, T. Vuletić, S. Dolanski Babić, and R. Podgornik, Physica B 407, 1958 (2012). [36] P. G. de Gennes, P. Pincus, R. M. Velasco and F. Brochard, J. Phys. (Paris) 37, 1461 (1976). [52] H. Chen, S. P. Meisburger, S. A. Pabit, J. L. Sutton, W. W. Webb and L. Pollack, Proc. Natl. Acad. Sci. USA 109, 799 (2012). [53] A. Bonincontro, E. Bultrini, G. Onori and G. Risuleo, J. of Non-Crystall. Solids USA 307-310, 863 (2002). [54] R. M. Clement, J. Sturm and M. P. Daune, Biopolymers, 12, 405 (1973). [55] J. Duguid, V. A. Bloomfield, J. Benevides and G. J. Thomas, Jr., Biophys. J. 65, 1916 (1993).
1704.03267	1	1704	2017-04-11T12:42:35	Kinetic model of selectivity and conductivity of the KcsA filter	[ "physics.bio-ph" ]	We introduce a self-consistent multi-species kinetic theory based on the structure of the narrow voltage-gated potassium channel. Transition rates depend on a complete energy spectrum with contributions including the dehydration amongst species, interaction with the dipolar charge of the filter and, bulk solution properties. It displays high selectivity between species coexisting with fast conductivity, and Coulomb blockade phenomena, and it fits well to data.	physics.bio-ph	physics	Kinetic model of selectivity and conductivity of the KcsA filter W.A.T. Gibby Department of Physics Lancaster University Lancaster LA1 4YB, UK Email: [email protected] D.G. Luchinsky SGT Inc., Greenbelt MD, 20770, USA I.Kh. Kaufman A. Ward and P.V.E. McClintock Department of Physics Lancaster University 7 1 0 2 r p A 1 1 ] h p - o i b . s c i s y h p [ 1 v 7 6 2 3 0 . 4 0 7 1 : v i X r a Abstract -- We introduce a self-consistent multi-species kinetic theory based on the structure of the narrow voltage-gated potassium channel. Transition rates depend on a complete energy spectrum with contributions including the dehydration amongst species, interaction with the dipolar charge of the filter and, bulk solution properties. It displays high selectivity between species coexisting with fast conductivity, and Coulomb blockade phenomena, and it fits well to data. I. INTRODUCTION Narrow biological ion channels allow the permeation of ions at near to the diffusion rate, whilst being highly selective. Paradoxically, potassium (K+) channels allow conduction at close to 108s−1, coexisting with high selectivity amongst mono-valent species [1]. The general permeation properties through the filter are preserved among all voltage-gated K+ channels due to the conserved amino acid structure of the filter [2]. Experimentally it is known that conduction occurs via the single-file concerted motions of ions separated by a water molecule, such that we have transitions between 2 and 3 ions [3], [4]. The filter charge Qf = nf q plays an important role in the permeation process. It is attributed to the dipolar interactions with the oxygen atoms in the permeation pathway, and these 20 atoms form 5 binding sites within the filter S1-S4 and 1 site in the entrance to the extracellular bulk S0. Kinetic theory has been widely studied and applied to channels [5] -- [8], with some success in comparing predictions with experimental data of K+ channels [9], [10] and explaining properties such as rectification. Yet kinetic models are often criticised for inconsistencies in the definition of the transition rates and the choice of energy barriers [11]. Here we introduce a novel self-consistent kinetic model of the KcsA filter that resolves its selectivity vs conductivity paradox. The model is based on a first principles statistical analysis of the energy of the filter that incorporates the effects of site-binding, dehydration, excluded volume etc by including the excess chemical potential difference between the bulk and the channel ∆¯µb i [12]. The self-consistent transition rates are included into the model using Grand Canonical Monte Carlo theory [13]. We validate the theory by comparing model predictions with experimental data, including mixed species solutions and mutants [9], [14]. i − ¯µc i = ¯µb Lancaster LA1 4YB, UK Fig. 1. Left: reduced image of closed conformation high-K+ KcsA (pdb 1K4C), rendered with VMD software [15]. The yellow ribbons represent two of the amino acid chains, whilst the narrow selectivity filter has its residues highlighted. The 4/5 binding sites are displayed with a S0S2S4 occupation. Right: Schematic of the transition S2S4↔S0S2S4 (note that we may also have an additional two ion state S1S3). In the work that follows, with SI units: q, k, N0, T , 0 and w, respectively represent the proton charge, Boltzmann's and Avogadro's constants, system temperature, and water and vacuum dielectric permittivities (w = 80) and constants. We K = 1.96 × use the following bulk diffusion coefficients: Db N a = 1.33× 10−9 m2s−1, and ionic radii: 10−9 m2s−1 and Db RK = 1.4 A and RN a = 1 A. The filter has the following geometry, radius 1.5 A, length 12 A, and 5 binding sites. II. MIXED-SPECIES KINETIC THEORY We consider a filter diffusively and thermally coupled to intra-cellular (L), and extra-cellular (R) bulk reservoirs, between which the membrane potential (φm) acts from left to right. Anions cannot enter the filter and so we only account for their presence via their contributions to the energy spectrum. We introduce indistinguishable sites, with the following reduced state space describing the number and species of occupying ions {nj}, {K +K +},{K +K +K +},{N a+K +K +}. (1) To simplify notation the subscript 0 will denote the ground state, whilst either excited state will be distinguished by its species: K+ or Na+, denoted by i. Transitions are described via a set neighbouring states master-equations,  = −ΓK  P0 PK PN a 01 01 − ΓN a ΓK 01 ΓN a 01  ·  ,  P0 PK PN a ΓK −ΓK 10 0 10 ΓN a 10 0 −ΓN a 10 (2) where we have introduced the notation: Γi = ΓL,i + ΓR,i. The general solution can be found using standard linear algebra methods when we consider probability conservation, taking a simplified form by considering the binding factor B [13] defined via the ratio of probabilities to the ground state, P ({nj}) = B({nj}) (cid:80){nj} B({nj}) , B({nj}) = P ({nj}) P0 . (3) To proceed in analysing the theory we first need to derive our rates, and we follow [13], [16], [17], and use our statistical theory. This approach starts by considering transitions between the mouth, which is in quasi-equilibrium with the bulk solution and the channel. We establish the local balance relation between each bulk b and the channel, and normalise such that rates sum to 1, leaving sigmoidal rates as functions of the full transition energy barrier: ∆G({nj}; nf ; b). Each rate is a function of the energy barrier from each bulk and so we allow the condition of equilibrium between both bulks to be broken. This ensures we have a non-equilibrium steady-state with non-zero current and failure to establish the Boltzmann ratio. i /L2 10 = τi × Db Γb,i 01 = τi × Db Γb,i exp[(∆G({nj}; nf ; b))/kT ] 1 + exp[(∆G({nj}; nf ; b))/kT ] i /L2 1 + exp[(∆G({nj}; nf ; b))/kT ] The prefactor τi × Db i /L2, is added under the condition that 0 < τi ≤ 1. These rates converge to a Kramer's exponential with a large energy barrier, and 1/2τi × Db i /L2 at the barrier- less condition (∆G ≈ 0) because the net flow is zero [18]. (4) 1 , The free energy barrier was derived and defined earlier in [12], and here we present only a brief discussion. ∆G({nj}; nf ; b) = ∆E({nj}; Qf ) − ∆φb − kT ln(xb i ) + kT ln[(ni + 1)/nw] − ∆¯µb i . (5) It represents the energy required for a transition between the bulk and channel to occur. It consists of an electrostatic component ∆E({nj}; Qf ) between ions and Qf [12], [19], bulk parameters: mole fraction ∼ cb w (where c is the con- centration of either species or water molecule) and influence from the membrane potential ∆φb in transition to a site at η. We include a permutation factor due to binding statistics where ni and nw are the initial numbers of ions and water i /cb molecules in the filter, and the excess chemical potential difference between bulk and channel ∆¯µb i. This final term provides a distinction between species be- cause it depends strongly on the ionic concentration, charge and radii. These contribute via a large number of terms including: dehydration at either filter entrance, site bonding, ion-ion interactions in the bulk and volume exclusion [12], [20], [21]. Our term takes the following form, (cid:115) N0q2(cid:80) (cid:17) (cid:17) · ∆¯µb i− i = ∆µb q2κ 8πw0(1 + κRi) , κ = i 2z2 i cb i (6) kT 0w where the term: ∆µb i includes dehydration, site-bonding and volume exclusion, and will be used as a fitting parameter. The final term is calculated explicitly and given by the Debye- Huckel ion-ion interaction, screened by a solvent and includes the presence of anions. The difference of ∆¯µb i between two species gives the thermodynamic selectivity ∆∆µb, and for K+ and Na+ this is around ∼ 8kT [22], [23], strongly disfavouring Na+. In the next section we introduce equations for K+ and Na+ current and consider selectivity. III. SELECTIVITY vs CONDUCTIVITY OF THE KCSA FILTER The current can be computed in the standard way as the to i . It is given below for balance of probability fluxes over each barrier, subject Kirchoff's law where Ii = I L the left barrier, i = I R 01 · P0 − ΓL,i ΓL,i 10 · Pi . (7) After inserting the solutions to the master equations (2), it can be simplified to, IK = q · Dc 01 − ΓR,K ΓL,K 01 IN a = q · Dc N a L2 01 − ΓR,N a ΓL,N a 01 1 K L2 + ΓK 10 2 Dc (cid:17) · ΓN a 01 ΓN a 10 1 2 Dc N a L2 + ΓN a 10 (8) . (9) ΓK 01 ΓK 10 This system is easily reducible back to two states describing K+, if we take the Na+ concentration to be zero. The occupancy of the filter can be calculated from the ensemble average, (cid:104)n(cid:105) = 2 × P0 + 3 × (PK + PN a) (10) In Fig: 2 we display current and occupancy of the filter for both species vs Qf . K+ is energetically favoured and so its conductivity and contribution to the occupancy is much larger. Hence the resonant transition occurs at the midpoint of the step which corresponds to a degeneracy in the energy spectrum i.e. the barrier-less knock-on condition ∆GK = 0, because here the ground and K+ excited state probabilities are non-zero and equal to each other. The Na+ peak is shifted because it is trying to enter a stable fully-occupied filter and hence requires Ii = q ·(cid:16) L2 ·(cid:16) ·(cid:16) K a K+ ion to exit; hence it peaks at the minimum in energy to remove a K+ and add a Na+. This phenomena is known as Coulomb blockade (CB) [19], [24] because outside of the K + peak the filter is in a blockaded state with minimal conduction. find a non-negligible Na+ current which continues to increase with even larger φm until it reaches its diffusion limit. This non-negligible current is observed experimentally and known as punch-through [25]. IV. EXPERIMENTAL COMPARISONS There is a vast experiential literature about the behaviour of K+ channels and in this paper we will compare the theoretical current with current-voltage (I − V ) curves under differing concentrations, mixed solutions and the effect of filter mutations, from the Shaker channel [9], KcsA [14]. Unless stated, only K+ is present in the solution, and hence the state space reduces to: {K +K +},{K +K +K +}, and we take Qf = −2.5q. In Fig: 3 we display the results of comparisons with the Shaker channel, noting that the diamonds indicate symmet- rical solutions and crosses the asymmetrical solutions, with concentrations given in the legend. The asymmetrical solutions consisted of K+ in the extracellular solution with concentra- tion 0.095M, and either 0.105M K+ or Na+ in the intracellular solution. Fig. 2. Top: IK and IN a× 102 vs Qf , with: τi = 0.5, φm = 50mV, η = 0.5, ∆µb K = 4kT and xi = 0.1/55. Each species provides a peak although Na+ is shifted and << IK. Bottom Selectivity between species decreases vs ∆∆µb with parameters: τi = 0.5, Qf =-2.5q, η = 0.5, ∆µb K = 4kT and xi = 0.1/55. For ∆∆µb (cid:46) 1kT the filter disfavours K+ and this is due to the binding statistics, for values (cid:38) 1.1kT we see high selectivity. The inset plot shows selectivity decreasing vs φm at ∆∆µb = 8kT , culminating with S ∼ 160 at φm = 0.2V. Fig. 3. Theoretical I − V curves are compared with the experimental values for the Shaker K+ channel [9]. Diamonds indicating symmetrical solutions, are fitted with the paramaters: τK = 0.0486, η = 0.617 and ∆µb K = 3.92kT . The assymetrical solutions denoted by crosses are fitted with: τi = 0.102, η = 0.629, and ∆µb i = 4.14kT . Selectivity is defined via the ratio: S = IK/IN a, and is plotted in the bottom figure of Fig: 2 and its inset. As expected selectivity strongly depends on the difference ∆∆µb, but decreases with membrane potential. We also observe that, due to the binding factor when ∆∆µb (cid:46) 1.1kT , it results in selectivity favouring Na+. The membrane voltage provides the driving force governing conduction in symmetrical solutions, and we observe later that at large voltages the current saturates. Thus when φm ∼ 0.2V K+ conduction has, or is approaching, saturation at its diffusion limit, and it becomes independent of voltage; meanwhile the energy barrier for Na+ conduction is lowered allowing conduction. Thus at these large voltages we The fitting parameters were very similar, although we note that the diffusion rate doubled when we had asymmetrical solutions. Importantly, however it was much less than the bulk value in agreement with simulations [26]. The remaining fitting parameters are consistent with both experiments for, η ∼ 0.6 and ∆µb K ∼ 4kT . This indicates a slight asymmetry in the channel and a low energy barrier for entry. The theory fits better at low concentrations possibly due to the break-down of the Debye-Huckel interaction term, which works best for dilute solutions. In the final comparison we compare our theory to wild-type (WT) KcsA and its mutant (MUT), created by replacing the Qf [q]-3-2.75-2.5-2.25-2IK, INa [pA]01020h n i 22.22.42.62.83IKINa × 102h n i∆∆ µb [kT]10-1100101S 10-2100102104φm=0.01Vφm=0.2Vφm [V]00.050.10.150.2S × 10300.20.40.6φm [V]-0.1-0.0500.050.10.15I [pA]-6-4-20240.043M0.073M0.206M0.325M0.605M1.15MCNaL 0.105MCKL 0.105M[3] A. L. Hodgkin and R. D. Keynes, "The potassium permeability of a giant nerve fibre," J. Physiol., vol. 128, no. 1, pp. 61 -- 88, 1955. [4] J. H. H. Morais-Cabral, Y. Zhou, and R. MacKinnon, "Energetic optimization of ion conduction rate by the K+ selectivity filter," Nature, vol. 414, no. 6859, pp. 37 -- 42, 2001. [5] P. Lauger, "Ion transport through pores: A rate-theory analysis," Biochim. Biophys. Acta (BBA) -- Biomembranes, vol. 311, no. 3, pp. 423 -- 441, 1973. [6] E. von Kitzing, "A novel model for saturation of ion conductivity in transmembrane channels," in Membrane Proteins: Structures, Inter- actions and Models: Proc. 25th Jerusalem Symposium on Quantum Chemistry and Biochemistry, Jerusalem, May 18-21, 1992, A. Pullman, J. Jortner, and B. Pullman, Eds. Dordrecht: Kluwer, 1992, pp. 297 -- 314. [7] I. S. Tolokh, S. Goldman, and C. Gray, "Unified modeling of conduc- tance kinetics for low-and high-conductance potassium ion channels," Phys. Rev. E, vol. 74, no. 1, p. 011902, 2006. [8] P. H. Nelson, "A permeation theory for single-file ion channels: One- and two-step models," J. Chem. Phys., vol. 134, pp. 165 102 -- 165 114, 2011. [9] L. Heginbotham and R. MacKinnon, "Conduction properties of the cloned shaker k+ channel," Biophys. J., vol. 65, no. 5, pp. 2089 -- 2096, 1993. [10] D. Meuser, H. Splitt, R. Wagner, and H. Schrempf, "Exploring the open pore of the potassium channel from streptomyces lividans," FEBS letters, vol. 462, no. 3, pp. 447 -- 452, 1999. [11] K. Cooper, P. Gates, and R. Eisenberg, "Diffusion theory and discrete rate constants in ion permeation," J. Membr. Biol., vol. 106, no. 2, pp. 95 -- 105, 1988. [12] D. G. Luchinsky, W. A. T. Gibby, I. Kaufman, D. A. Timucin, and P. V. E. McClintock, "Statistical theory of selectivity and conductivity in biological channels," arXiv[physics.bio-ph], p. 1604.05758v3, 2016. [13] B. Roux, "Statistical Mechanical Equilibrium Theory of Selective Ion Channels," Biophys. J., vol. 77, no. 1, pp. 139 -- 153, 1999. [14] M. Zhou and R. MacKinnon, "A mutant kcsa k+ channel with altered conduction properties and selectivity filter ion distribution," J. mol. biol., vol. 338, no. 4, pp. 839 -- 846, 2004. [15] W. Humphrey, A. Dalke, and K. Schulten, "VMD -- Visual Molecular Dynamics," J. Mol. Graph., vol. 14, pp. 33 -- 38, 1996. [16] W. Im, S. Seefeld, and B. Roux, "A grand canonical Monte CarloBrown- ian dynamics algorithm for simulating ion channels," Biophys. J., vol. 79, no. 2, pp. 788 -- 801, 2000. [17] B. Roux, T. Allen, S. Berneche, and W. Im, "Theoretical and computa- tional models of biological ion channels," Quart. Rev. Biophys., vol. 37, no. 1, pp. 15 -- 103, 2004. [18] C. W. Gardiner, Handbook of Stochastic Methods: for Physics, Chem- istry and the Natural Sciences. Berlin: Springer, 2002. [19] I. K. Kaufman, P. V. E. McClintock, and R. S. Eisenberg, "Coulomb blockade model of permeation and selectivity in biological ion chan- nels," New J. Phys., vol. 17, no. 8, p. 083021, 2015. [20] D. A. McQuarrie, Statistical Mechanics, 1st ed. University Science Books, Jun. 1976. [21] D. Krauss, B. Eisenberg, and D. Gillespie, "Selectivity sequences in a model calcium channel: role of electrostatic field strength," Eur. Biophys. J., vol. 40, pp. 775 -- 782, 2011. [22] S. Y. Noskov and B. Roux, "Importance of hydration and dynamics on the selectivity of the KcsA and NaK channels," J. Gen. Physiol., vol. 129, no. 2, pp. 135 -- 143, 2007. [23] S. W. Lockless, "Determinants of cation transport selectivity: Equilib- rium binding and transport kinetics," J. Gen. Physiol., vol. 146, no. 1, pp. 3 -- 13, 2015. [24] C. W. J. Beenakker, "Theory of Coulomb-blockade oscillations in the conductance of a quantum dot," Phys. Rev. B, vol. 44, no. 4, pp. 1646 -- 1656, 1991. [25] C. M. Nimigean and C. Miller, "Na+ block and permeation in a k+ channel of known structure," J. Gen. Physiol, vol. 120, no. 3, pp. 323 -- 335, 2002. [26] D. P. Tieleman, P. C. Biggin, G. R. Smith, and M. S. P. Sansom, "Simulation approaches to ion channel structure-function relationships," Quart. Rev. Biophys., vol. 34, no. 4, pp. 473 -- 561, 2001. Fig. 4. Theoretical I − V curves are compared with the experimental values for WT (blue) and MUT (orange) [14]. Resulting in: τK = 0.161, ∆µb W T = 4.1kT , ∆µb M U T = 0.979kT , ηW T = 0.175 and ηM U T = 0.446. amino-acid threonine with cysteine at the S4 site (for full mu- tagenisis details refer to [14]), strongly affecting conduction. To model this mutation we allow for differences in symmetry and binding, and hence vary ∆µb K and η, between the WT and MUT. In Fig: 4 we display the results, and there is a strong K . In WT KcsA the energy difference in the symmetry and ∆µb K was much barrier for entry was much smaller because ∆µb larger than the mutant, and tuned to produce faster conduction. The difference in the η parameters were large suggesting the site mutation caused an asymmetry from the original WT. V. CONCLUSION We have introduced a self-consistent kinetic model de- scribing transitions of state in the selectivity filter of narrow channels. It demonstrates large selectivity amongst mono- valent species coexisting with high conductivity, and CB effects. Predictions of the model are in good agreement with experimental data and we have considered the effect of a filter mutation. It can be extended by considering site-specific binding energies to increase the number of states, and with further comparison to experimental data. We fully expect the theory to be applicable to other narrow channels such as voltage-gated sodium channels e.g. NaCh- Bac, and to artificial nano-pores. ACKNOWLEDGEMENTS We are grateful to Bob Eisenberg, Miroslav Barabash, Alek- sandra Pidde and Aneta Stefanovska for helpful discussions. The research was supported by the Engineering and Physical Sciences Research Council UK (grant No. EP/M015831/1). REFERENCES [1] B. Hille, Ion Channels Of Excitable Membranes, 3rd ed. Sunderland, MA: Sinauer Associates, 2001. [2] Q. Kuang, P. Purhonen, and H. Hebert, "Structure of potassium chan- nels," Cell. Mol. Life Sci., vol. 72, no. 19, pp. 3677 -- 3693, 2015. φm [V]-0.2-0.100.10.2I [pA]-1001020WTMUT
1606.02977	1	1606	2016-06-09T14:33:24	Multiphasic profiles for voltage-dependent K+ channels: Reanalysis of data of MacKinnon and coworkers	[ "physics.bio-ph" ]	In a study of the role that voltage-dependent K+ channels may have in the mechanosensation of living cells (Schmidt et al. Proc Soc Natl Acad Sci USA 109: 10352-10357. 2012), the data were as conventionally done fitted by a Boltzmann function. However, as also found for other data for ion channels, this interpretation must be rejected in favor of a multiphasic profile, a series of straight lines separated by discontinuous transitions, quite often in the form of noncontiguities (jumps). The data points in the present study are often very unevenly distributed around the curvilinear profiles. Thus, for 43 of the 75 profiles, the probability is less than 5% that the uneven distribution is due to chance, for 26 the probability is less than 1%, and for 12 the probability is less than 0.1%, giving a vanishingly low overall probability for all profiles. Especially at low voltages, the differences between the fits to curvilinear and multiphasic profiles may be huge. In the multiphasic profiles, adjacent lines are quite often parallel or nearly so, in which case the transitions are, necessarily, in the form of jumps. The lines in the multiphasic profiles appear to be perfectly straight, with no indication of any curvilinearity. The r values are for the most part high, quite often exceedingly high. In addition to activation of ion channels, a wide variety of biological as well as non-biological processes and phenomena involving binding, pH, folding/unfolding and effect of chain length can be well represented by multiphasic profiles (Nissen 2015a,b, 2016. Posted on arXiv.org with Paper ID arXiv: 1511.06601, 1512.02561 and 1603.05144). Introduction	physics.bio-ph	physics	Multiphasic profiles for voltage-dependent K+ channels: Reanalysis of data of MacKinnon and coworkers Per Nissen Norwegian University of Life Sciences Department of Ecology and Natural Resource Management P. O. Box 5003, NO-1432 Ås, Norway [email protected] 2016 2 Abstract In a study of the role that voltage-dependent K+ channels may have in the mechanosensation of living cells (Schmidt et al. Proc Soc Natl Acad Sci USA 109: 10352-10357. 2012), the data were as conventionally done fitted by a Boltzmann function. However, as also found for other data for ion channels, this interpretation must be rejected in favor of a multiphasic profile, a series of straight lines separated by discontinuous transitions, quite often in the form of noncontiguities (jumps). The data points in the present study are often very unevenly distributed around the curvilinear profiles. Thus, for 43 of the 75 profiles, the probability is less than 5% that the uneven distribution is due to chance, for 26 the probability is less than 1%, and for 12 the probability is less than 0.1%, giving a vanishingly low overall probability for all profiles. Especially at low voltages, the differences between the fits to curvilinear and multiphasic profiles may be huge. In the multiphasic profiles, adjacent lines are quite often parallel or nearly so, in which case the transitions are, necessarily, in the form of jumps. The lines in the multiphasic profiles appear to be perfectly straight, with no indication of any curvilinearity. The r values are for the most part high, quite often exceedingly high. In addition to activation of ion channels, a wide variety of biological as well as non-biological processes and phenomena involving binding, pH, folding/unfolding and effect of chain length can be well represented by multiphasic profiles (Nissen 2015a,b, 2016. Posted on arXiv.org with Paper ID arXiv: 1511.06601, 1512.02561 and 1603.05144). Introduction In addition to multiphasic profiles for ion uptake in plants (Nissen 1971, 1974, 1991, 1996), such profiles have been recently (Nissen 2015a,b, Nissen 2016) reported for many other processes and phenomena. In the present paper, data (Schmidt et al. 2012) for voltage-dependent Kv channels will be reanalyzed to compare the fits to curvilinear profiles with the fits to multiphasic profiles. Original data have been kindly provided by Daniel Schmidt. In addition to the r values, slopes ± SE (or only slopes) are given on the plots. Except for the data for Figs 11-13 and 27-29 (authors' Figs 1L and S2C), all slopes have been multiplied by 1000. The Runs test (Wald and Wolfowitz 1940) gives the probability for the uneven distribution of points around the curvilinear profile being due to chance. Independent probabilities have been combined by the method of Fisher (Fisher 1954). I Reanalysis 3 Fig. 1. Authors' Fig. 1D, H and L. Plot D: Paddle chimera in the same outside-out patch with 0 mm Hg (triangles), 5 mm Hg (circles) and 15 mm Hg (squares) of transient pressure applied. Plot H: Shaker Kv in the same outside-out patch at different time points after patch excision: 0 min (squares), 4 min (triangles) and 8 min (diamonds). Plot L: Kv2.1 channels in the same on-cell patch with 0 mm Hg (squares), 5 mm Hg (triangles) and 15 mm Hg (diamonds) of transient pressure applied. See also the original legend. Fig. 2. Reanalysis of data in Fig. 1. The profile for 0 mm Hg can be well represented as pentaphasic, with the transitions at -42.6, -24.3 and 12.2, and between 40 and 50 mV (jump). The r values for lines III and IV are very high and lines IV and V are precisely parallel, indicating that the data are highly precise. The 6-point line I has a low r value. This could, most likely, be because of its very shallow slope (slight errors will cause marked decreases in the r values for such lines). However, there could also be several phases in this range (the line for -90 and -80 mV and the line for -60 and -50 mV have about the same slope, 1.0 and 1.1, respectively). Fig. 3. The enlargement of Fig. 2 at high negative voltages shows the huge differences in the fits to the curvilinear and the multiphasic profile. In contrast to the precisely multiphasic profiles, the fits to the curvilinear profile are poor. Furthermore, the points are very unevenly distributed around this profile. Points 1-5 and 13-17 are above the line, points 6-12 are below the line. The probability of this uneven distribution, or an even more unlikely one, occurring by chance is only 0.09% (by the Runs test). 4 Fig. 4 (left above). Reanalysis of data in Fig. 1. The profile for 5 mm Hg can be well represented by 8 phases, with the transitions between -100 and -90, at -62.2, -45.0, -20 and 2.3, between 20 and 30 (jump), and between 50 and 60 mV. The r values for lines II, IV, V and VII are very high. The multiphasic profile is precise, and the curvilinear profile is imprecise. Fig. 5 (right above). Although not evident from Fig. 4, this enlargement shows that the fits to the curvilinear profile are very poor also at the lowest voltages. (A single line in the range of phases I and II will have an r value of only 0.547.) The fits to the curvilinear profile are poor when compared to the multiphasic profile. Furthermore, the distribution around the curvilinear profile is exceedingly uneven. The first 10 points are all above the line, the remaining 7 are all below the line. The probability of this happening by chance is only 0.01%. Fig. 6 (left above). Reanalysis of data in Fig. 1. The profile for 15 mm Hg can be well represented as heptaphasic, with the transitions at -73.3 and -51.3, between -30 and -20, between -20 and -10, at 13.4, and between 30 and 40 mV (jump). The data are insufficiently detailed in the range of phase IV for the line to be resolved. The r values are high for lines I and V, and very to exceedingly high for lines III and VII. Fig. 7 (right above). As also for lower Hg (Figs 3 and 5), the data can be well represented as multiphasic, but not as curvilinear, also at low y values (right), where the curvilinear profile gives, for the most part, exceedingly poor fits. The distribution of points around the curvilinear profile is somewhat uneven. Points 1-3, 8-9, 13 and 15-17 are above the profile, points 4-7, 10-12, and 14 are below the curvilinear line. The probability of this or a more uneven distribution being due to chance is 15.7%. The probability that the uneven distribution around the three profiles in Fig. 1D is due to chance is less than 0.001% (by Fisher's method for combining independent probabilities). 5 Fig. 8. Reanalysis of data in Fig. 1H. The profile for 0 min can be well represented by 8 phases, with the transitions between -90 and -80 (jump), at -53.4, between -40 and -30 (jump), between -20 and -10 (jump), at 10, between 30 and 40, and between 40 and 50 mV. The data are insufficiently detailed in the range of phase VII for the line to be resolved. Lines I and II are parallel. The r values for lines V and VI are very to exceedingly high. Runs test: P = 31.9%. Fig. 9. The profile for 4 min can be well represented as pentaphasic, with the transitions at -52.2 and -16.0, between 10 and 20 (jump), and between 40 and 50 mV (jump). Lines I and III are parallel, lines IV and V are about parallel. Runs test: P = 6.9%. Fig. 10. The profile for 8 min can be well represented as heptaphasic, with the transitions at -61.0, between -50 and -40 (jump), between -30 and -20, between -20 and -10, between 20 and 30, and between 30 and 40 mV. The data are insufficiently detailed in the range of phases IV and VI for the lines to be resolved. The r value for line VII is high. Runs test: P = 8.0%. The probability that the uneven distribution around the three profiles in Fig. 1H is due to chance is less than 5%. Fig. 11. Reanalysis of data in Fig. 1L. The profile for 0 mm Hg can be well represented by 8 phases, with the transitions between -80 and -70 (jump), between -60 and -50 (jump), at -31.9 mV, between -10 and 0 (jump), between 10 and 20 (jump), between 30 and 40 (jump), and between 60 and 70 mV. A single line in the range of lines I-III will have an r value of only 0.941. Lines VI and VII are about parallel. Runs test: P = 23.8%. 6 Fig. 12. The profile for 5 mm Hg can be well represented by 9 phases, with the transitions between -100 and -90, at -70, -50, -30 and -10, between 10 and 20 (jump), between 30 and 40 (jump), and between 50 and 60 mV (jump). Line II is essentially horizontal. A single line in the range of lines VII-IX will have an r value of 0.979, which is low compared to the r values for lines III-VI. Lines VIII and IX are about parallel. Runs test: P = 0.30%. Fig. 13. The profile for 15 mm Hg can be well represented by 9 phases, with the transitions between-110 and -100 , at -86.2 , between -50 and -40, between -40 and -30, between -20 and -10 (jump), between 10 and 20 (jump), between 30 and, and between 40 and 50 mV. The data are insufficiently detailed in the range of phases IV and VIII for the lines to be resolved. Line II has a very high r value. Lines V and VI are about parallel, as are lines VII and IX. Runs test: P = 75.8%. The probability that the uneven distribution around the three profiles in Fig. 1L is due to chance is less than 2.5%. Fig. 14. Authors' Fig. 2C. Paddle chimera in on-cell patch (see authors' Fig. 2D) or whole-cell configuration (see authors' Fig. 2E) recorded from the same Sf-9 cell. Solid lines, Boltzmann functions. (But see reanalysis on next page.) 7 Fig. 15 (left above). Reanalysis of data in Fig. 14. The profile for the on-cell data can be well represented as pentaphasic, with the transitions at -71.9 and -58.2, between -40 and -30 (jump), and at 13.8 mV. Runs test: P = 2.51%. Fig. 16 (right above). Reanalysis of data in Fig. 14. The profile for the whole-cell data can also be well represented as pentaphasic, with the transitions between -50 and -40 mV, between -40 and -30 mV, and at 0 and 24.8 mV. The r value is very high for line IV, and exceedingly high for line III. Runs test: P = 56.6%. The probability that the uneven distribution around the two profiles in Fig. 14 is due to chance is less than 10%. Fig. 17. Authors' Fig. 4C. The effect of swelling by hypoosmotic shock. Boltzmann functions for Paddle chimera expressed in sF-9 cells before swelling (squares), at intermediate swelling state 1 (triangles, at intermediate swelling state 2 (diamonds), and at peak volume (circles). See also the original legend and reanalyses on the next page. 8 Fig. 18 (left above). Reanalysis of data in Fig. 17. The profile before swelling can be well represented as pentaphasic, with the transitions between -60 and -50 (jump), at -20, between 0 and 10 (jump), and at 24.2 mV. Lines I and II are parallel, as are lines III and IV. The r values for lines I, III and V are very high. Runs test: P = 22.6%. Fig. 19 (right above). The profile for intermediate swelling state 1 can be well represented by 7 phases, with the transitions at -60, -34.5, -13.8, 6.3 and 24.6, and between 40 and 50 mV. The r values are high for lines I and II. Runs test: P = 64.6%. Fig. 20 (left above). The profile for intermediate swelling state 2 can be well represented as hexaphasic, with the transitions between -80 and -70, between -40 and -30, between -30 and -20, at 3.6, and between 20 and 30 mV (jump). The data are insufficiently detailed in the range of phase III for the line to be resolved. The r values for lines II and VI are very high, that for line IV is exceedingly high. Lines V and VI are parallel. Runs test: P = 38.3%. Fig. 21 (right above). The profile during peak volume can also be well represented as hexaphasic, with the transitions between -70 and -60 (jump), between -40 and -30, between -30 and -20, at 2.1, and between 20 and 30 mV (jump). The data are insufficiently detailed in the range of phase III for the line to be resolved. Lines V and VI are about parallel. Runs test: P = 28.7%. The probability that the uneven distribution around the four profiles in Fig. 17 is due to chance is somewhat more than 10%. The profiles for the two independent experiments in Figs 20 and 21 have remarkably similar multiphasic patterns, with only a slight difference at high negative voltages. The essentially identical patterns, including the two parallel lines V and VI, constitute conclusive evidence that the patterns are indeed multiphasic. (Clearly, two so strikingly similar patterns cannot have arisen by chance.) 9 Fig. 22. Authors' Fig. 4F. The effect of swelling by hypoosmotic shock. Boltzmann functions for Paddle chimera expressed in Sf-9 cells during consecutive perfusion with iso-osmotic solution (squares), hypoosmotic solution (circles), and immediately after peaking volume (diamonds). See also the original legend and reanalyses below. Fig. 23. The profile can be well represented as tetraphasic, with the transitions at -33.9, -10 and 14.9 mV. Lines II-IV have very high r values. Runs test: P = 7.75%. Fig. 24. The profile can be well represented as hexaphasic, with the transitions at -46.0, -25.3, -4.5 and 20, and between 40 and 50 mV. The r values for lines IV and V are very high. Runs test: P = 3.90%. Fig. 25. The profile can be well represented as pentaphasic, with the transitions at -44.0, -17.3, 5.3 and 35.2 mV. The r values for lines II and IV are high. Runs test: P = 3.90%. The probability that the uneven distribution around the three profiles in Fig. 22 is due to chance is less than 1%. Notice also how the multiphasic profiles deviate from the curvilinear profiles in about the same way in the three plots. 10 Fig. 26. Authors' Fig. S2 and equation 1 (right). Kv2.1 channels in the same on-cell patch with 0 mm Hg (right plot in each panel), 5 mm Hg (center plot), and 15 mm Hg (left plot). The curvilinear lines are fits to equation 1 with the following constraints: (A) Vm and z independently fit for each pressure value, L constrained to be the same for all pressure values. (B) Vm and z constrained to be the same for all pressure values, L independently fit for each pressure value. (C) Vm, z and L independently fit for each pressure value. Fitting residuals are plotted above each panel. See also the original legend. Fig. 27. Reanalysis of data in Fig. 26. The profile can be well represented by 7 phases, with the transitions at -43.4 and -24.7, between 0 and 10, between 10 and 20, between 30 and 40 (jump), and between 50 and 60 mV (jump). The data are insufficiently detailed in the range of phase IV for resolution of the line. Lines VI and VII could possibly be a single line, but its r value (0.987) is low compared to that for line III. Runs test: P = 0.30%. Fig. 28. The profile can be well represented by 9 phases, with the transitions between -100 and -90, at -70, -50 and -27.4, between -10 and 0 (jump), between 10 and 20, between 30 and 40 (jump), and between 60 and 70 mV. Line IV has a very high r value. Runs test: P = 6.00%. Fig. 29. The profile can be well represented by 10 phases, with the transitions at -100 and -74.7, between -60 and -50, between -50 and -40, between -20 and -10 (jump), between 0 and 10 mV (jump), between 20 and 30 (jump), between 40 and 50 (jump), and between 60 and 70 mV. The data are insufficiently detailed in the range of phase IV for resolution of the line. Lines VI-IX are about parallel. Runs test: P = 15.1%. The combined probability that the uneven distribution around the three curvilinear profiles in Fig. S2A is due to chance is less than 0.5% (not shown). For Fig. S2B, the probability is less than 5% (not shown). With no constraints, i.e. for the three profiles in Fig. S2C, the combined probability is only somewhat above than 10%. (The multiphasic profiles for Figs S2A and S2B are the same as for Fig. S2C, and are not shown.) Thus, whereas the fits to the multiphasic profiles are very good, the fits to the authors' curvilinear profiles are poor even in the absence of constraints. 11 Fig. 30. Authors' Fig. S4. Global fit of all paddle chimera data sets. A global fit of 57 families of current extracted from 11 outside-out patches. Fitting residues are graphed above. See also original legend. The 57 data sets in Fig. 30 have all been reanalyzed and shown to be well represented by multiphasic profiles. The profiles are not shown here, except that the probabilities for the profiles being curvilinear are included in the count below. If the curvilinear profiles had completely represented the true situation, the experimental points should have been about equally distributed around the profiles. However, the points are, for the most part, unevenly distributed around the curves, quite often markedly so. As determined by the Runs test, an exact test, for 52 of the 75 profiles in the present analysis the probability that the uneven distribution is due to chance is less than 10%. For 43 of the profiles the probability is less than 5%, for 26 of the profiles the probability is less than 1%, and for 12 of the profiles the probability is less than 0.1%. (The probabilities for Fig. S1, reanalyzed in Nissen 2015a, are included in this count, but only the probabilities for Fig. S2C, not for Figs S2A and S2B.) The overall probability for the 75 profiles is of course vanishingly low. Conclusion 12 It is clear that the present data cannot be acceptably represented by curvilinear profiles. Thus, the points are often very unevenly distributed around the profiles, giving very low propabilities for the uneven distribution being due to chance.. The probability becomes vanishingly low if all profiles are considered together. In contrast, the data are well represented by multiphasic profiles. This conclusion also holds for other data for K+ channels (Nissen 2016) as well as for a large body of other data (Nissen 2015a,b and in preparation). Acknowledgment – I am very grateful to Bob Eisenberg for his continued interest and encouragement. References Fisher RA (1954) Statistical Methods for Research Workers, 12thed., section 21.1, Oliver Boyd, Edinburgh. Nissen (1971) Uptake of sulfate by root and leaf slices of barley. Physiol Plant 24: 315-324. Nissen P (174) Uptake mechanisms: Inorganic and organic. Annu Rev Plant Physiol. 25: 53-79. Nissen P (1991) Multiphasic uptake mechanisms in plants . Int Rev Cytol 126: 89-134. Nissen P (1996) Uptake mechanisms: Pp. 511-527 in: Waisel Y, Eshel S, Kafkafi U (eds.) Plant Roots. The Hidden Half (2.ed.).Marcel Dekker, Inc, New York. Nissen P (2015a). Discontinuous transitions: Multiphasic profiles for channels, binding, pH, folding and chain length. Posted on arXiv.org with Paper ID arXiv: 1511.06601. Nissen P (2015b) Multiphasic pH profiles for the reaction of tris-(hydroxymethyl).aminomethane with phenyl esters. Posted on arXiv.org with Paper ID arXiv: 1512.025601. Nissen P (2016) Profiles for voltage-activated currents are multiphasic, not curvilinear. Posted on arXiv.org with Paper ID arXiv: 1603.05144. Schmidt D, del Mármol J, MacKinnon R (2012) Mechanistic basis for low threshold mechanosensitivity in voltage-dependent K+ channels. Proc Natl Acad Sci USA 109: 10352-10357. Wald A, Wolfowitz J (1940) On a test whether two samples are from the same population. Ann Math Statistics 11: 147-162.
1203.1485	3	1203	2012-07-26T09:39:31	Long-Lived Electronic Coherence in Dissipative Exciton-Dynamics of Light-Harvesting Complexes	[ "physics.bio-ph", "physics.chem-ph", "quant-ph" ]	The observed prevalence of oscillatory signals in the spectroscopy of biological light-harvesting complexes at ambient temperatures has led to a search for mechanisms supporting coherent transport through larger molecules in noisy environments. We demonstrate a generic mechanism supporting long-lasting electronic coherence up to 0.3 ps at a temperature of 277 K. The mechanism relies on two properties of the spectral density: (i) a large dissipative coupling to a continuum of higher-frequency vibrations required for efficient transport and (ii) a small slope of the spectral density at zero frequency.	physics.bio-ph	physics	Long-Lived Electronic Coherence in Dissipative Exciton-Dynamics of Light-Harvesting Complexes Christoph Kreisbeck1 and Tobias Kramer1, 2 1Institute for Theoretical Physics, University of Regensburg, 93040 Regensburg, Germany. 2Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA The observed prevalence of oscillatory signals in the spectroscopy of biological light-harvesting complexes at ambient temperatures has led to a search for mechanisms supporting coherent transport through larger molecules in noisy environments. We demonstrate a generic mechanism supporting long-lasting electronic coherence up to 0.3 ps at a temperature of 277 K. The mechanism relies on two properties of the spectral density: (i) a large dissipative coupling to a continuum of higher- frequency vibrations required for efficient transport and (ii) a small slope of the spectral density at zero frequency. PACS numbers: 03.65.Yz, 78.47.nj, 05.60.Gg Long-lasting oscillatory signals in optically excited light-harvesting complexes (LHCs) point towards the prevalence of a coherent electronic dynamics in molec- ular networks at physiological temperatures [2 -- 5]. The experiments probe the electronic dynamics using two- dimensional (2d) echo-spectroscopy for a series of delay- times between the excitation pulse and the probe pulse. The 2d spectroscopy makes studies of the dynamics of dissipative systems possible and has found further ap- plications in mesoscopic systems such as molecular nan- otubes [6] and semiconductor devices [7]. Due to its known crystallographic structure and relative simplic- ity, the Fenna-Matthews-Olson (FMO) complex serves as the prototype system for studying the choreography of the energy transfer from the antenna to the reaction center of a light harvesting complex [8]. In 2d spectra long-lasting beatings are observed, ranging from 1.2 ps at T = 150 K to 0.3 ps at T = 277 K [9]. The interplay of coherent dynamics, which leads to a delocalization of an initial excitation arriving at the FMO network from the antenna, and the coupling to a vibronic environment with slow and fast fluctuations, has lead to studies of environmentally assisted transport in LHCs [10, 11]. An important open question is whether coherence plays a key-role in the functioning of light-harvesting complexes [8]. The theoretical understanding of the ex- periments is in its early stages and atomistic simulations based on molecular dynamics have not reached agree- ment [12, 13]. A calculation of 2d echo-spectra based on the molecular dynamics simulation [14] does not show clear coherent oscillations at T = 277 K. One key in- gredient for efficient transfer dynamics is the strong cou- pling to vibronic modes, which induces energy dissipation [10, 11, 15]. For the FMO complex the thermalization oc- curs within picoseconds and was observed by Brixner et al. by the decay of diagonal-peak amplitudes to lower en- ergies [16]. It has been proposed that the inclusion of the finite time scale of the reorganization process gives rise to long-lasting coherence in LHCs [17]. While a slug- gish bath relaxation leads to prolonged population beat- ings in the FMO network, calculations of 2d echo-spectra show oscillations of the exciton cross-peaks for only about 1/6th of the experimentally recorded time at T = 150 K [18]. For an even longer bath-relaxation cross-peak oscil- lations are absent already at T = 77 K [19]. Hence the search continues for theoretical models which support a long-lasting coherent dynamics leading to cross-peak os- cillations in 2d echo-spectra and simultaneously retain the fast dissipation. Recently, the coupling to a super- position of discrete vibronic modes rather than the exis- tence of a purely electronic coherence has been proposed as an explanation for the detected oscillations [20, 21]. However, no significant changes in the oscillations of the FMO complex have been seen in experiments which al- tered the vibronic modes by using native mutants, 13C substitution, and partial deuteration [22]. Here, we present an alternative mechanism which does yield long-lasting and purely electronic coherence in the presence of a strong dissipative coupling required for effi- cient transfer. The mechanism is independent of the spe- cific exciton-system and relies on properties of the contin- uum part of the vibronic spectral density as opposed to discrete vibronic modes. Realistic models of the exciton dynamics in the FMO complex have to include higher or- der phonon processes as well as the finite time scale of the reorganization process [17]. This requires to go beyond approximative rate equations [23] and non-perturbative techniques [24 -- 30] are necessary to study the dissipative transfer dynamics. One key parameter determining the duration of coher- ent oscillations is the spectral density, which encodes the the mode-dependent exciton-bath coupling. The spectral density has important implications for the two contribu- tions to the decoherence rate γ which is the sum of the relaxation rate γr and the pure-dephasing rate γd ([33], ch. 21.4.2, [34] ch. 3.8.2), γ = γr/2 + γd. (1) Before we discuss decoherence-rates of the FMO com- plex, we analyze the impact of the spectral density on a two-site system with site-energies ±ǫ/2, inter-site cou- plings d/2, and b2 = ǫ2 + d2. The two rates in Eq. (1) are given by γr ≈ d2S(b)/(2b2), γd = ǫ2S(0)/(2b2), and are linked to the spectral density J(ω) via S(ω) = J(ω) coth(¯hω/(2kBT )). At zero frequency, S(0) is pro- portional to the slope of the spectral density. We eval- uate the decoherence rate for two different models of the spectral density, (i) a super-Ohmic spectral density JSO ∼ ω, and a single-peak Drude-Lorentz spectral den- sity JDL(ω) = 2νλω/(ν2 + ω2), depicted in the Supple- mentary Material (SM) [1] Fig. S1. In the super-Ohmic case SSO(0) = 0 and the pure- dephasing term vanishes γd = 0 and decoherence arises only through relaxation. The opposite regime is present for JDL with SDL(0) = 4kBT λ/ν. For the parame- ters used in models of a sluggish bath-relaxation in the FMO complex [17] (bath-correlation time ν−1 = 50 fs, reorganization energy λ = 40 cm−1, ǫ = 150 cm−1, d = 200 cm−1) the pure-dephasing rate dominates γd,DL = 0.19 cm−1K−1, over the relaxation contribution γr,DL/2 = 0.08 cm−1K−1 to the decoherence rate. The spectral density for the FMO complex (circles in Fig. 1) has been extracted from fluorescence-line- narrowing experiments at T = 4 K [31, 32] and shows a super-Ohmic behaviour. However, previous calcula- tions of 2d-echo spectra of seven-site FMO complex have been often performed with a single-peak Drude-Lorentz spectral density [18, 19], and did not yield long-lasting cross-peak oscillations. The situation is different in the population dynamics, which has been analyzed for differ- ent spectral densities [26] and yields no significant differ- ences in the duration of population beatings for different spectral density. As we show next, 2d echo-spectra are not directly derivable from population dynamics and carry along ad- ditional information about the exciton dynamics [35]. To demonstrate this, we calculate the exciton-dynamics for the spectral densities J11peaks (J3peaks) shown in Fig. 1, which represent the experimentally determined spectral density by a superposition of n = 11 (n = 3) shifted Drude-Lorentz peaks [36] J(ω) = n X k=1 νkλkω k + (ω ± Ωk)2 . ν2 (2) The shifts allow us to vary the slope of the spectral den- sity J(ω) at ω = 0 and to mimic the super-Ohmic be- haviour seen in the experiment (JWendling). Compared to the single-peak Drude-Lorentz spectral-density dis- cussed before (λ = 40 cm−1, ν−1 = 50 fs, Ω = 0), S(0) is reduced by a factor of 13 (6). Both superpositions J{3,11}peaks provide a good approximation of the spectral density at the transition energies corresponding to the exciton energy differences. The Hamiltonian of the FMO complex is based on the Frenkel exciton model for LHCs [34] and consists of a 2 JWendling J11peaks J3peaks ) 1 − m c ( ) ω ( J 2 ω ¯h π = ) ω ( J 250 200 150 100 50 0 100 200 300 400 500 ¯hω (cm−1) Spectral density of the FMO complex. Cir- FIG. 1: cles: measured spectral density [31], parametrization [32]. Fits J{3,11}peaks are used for the GPU-HEOM calculation eq. (2). The marks on the frequency axis indicate the transi- tion energies corresponding to the difference of exciton fre- quencies. Parameters for J3peaks: λ = {10, 15, 13} cm−1, ν −1 = {250, 120, 65} fs, Ω = {85, 170, 300} cm−1, J11peaks: λ = {1.2, 6.4, 7.4, 15.6, 3.4, 1.8, 4, 2, 1.8, 1.9, 2} cm−1, ν −1 = {1600, 550, 400, 370, 750, 800, 750, 600, 750, 750, 500} fs, Ω = {53, 73, 117, 185, 235, 260, 285, 327, 363, 380, 480} cm−1. part describing the coherent exciton dynamics N N N Hex = X ǫma† mam + X m=1 m X n6=m Jmna† man, (3) where a† m is the creation operator of an electronic exci- tation of bacteriochlorophyll (BChl) m with site energy ǫm. The inter-site couplings between the N BChls are denoted by Jmn. The BChls are coupled to vibronic modes modeled by a set of independent harmonic os- cillators Hphon = Pm,i ¯hωib† i,mbi,m. The exciton-bath coupling is given by Hex−phon = Pm a† mam um with um = Pi ¯hωidi(bi,m + b† i,m). The reorganization energy Hreorg = Pm λa† i /2 is added to the site energy ǫm. For the seven-site FMO complex we take the site energies and inter-site couplings from Ref. [32]. For simplicity we assume identical vibronic- excitonic couplings at each site and neglect correlations of vibronic modes at different sites. The reorganization energy λ ≈ 40 cm−1 is of the same order as the exci- tonic energy-spacings, the bath-relaxation time, and the thermal energy kBT at T = 277 K [17, 32]. mam with λ = Pi ¯hωid2 The strong couplings of electronic and vibronic de- grees require to solve the density-matrix propagation with a non-perturbative and non-Markovian approach. The computation of the 2d echo-signal for the FMO com- plex necessitates about 103 more time-steps than the calculation of the population dynamics and has only re- cently been achieved for the non-Markovian case [18, 19]. We solve the exciton dynamics within the hierarchical (a) 12700 12500 12300 ) 1 − m c ( 3 ω 12100 (b) 12700 ) 1 − m c ( 3 ω 12500 12300 RP signal, FWHM= 0 cm−1 1 2 3 4 5 6 7 150 K T delay = 400 fs 7 6 5 4 3 2 1 CP(1,5) 12100 12300 12500 ω1 (cm−1) 12700 SE-RP signal, FWHM= 94 cm−1 1 2 3 4 5 6 7 150 K Tdelay = 400 fs A B 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 7 6 5 4 3 2 1 12100 C ArcSinh 12100 12300 12500 12700 ω1 (cm−1) FIG. 2: (color online) Calculated 2d echo-spectra of the FMO complex at T = 150 K for delay time Tdelay = 400 fs, spectral- density J3peaks. (a) Real part of the rephasing signal (sum of ground-state bleach, stimulated emission, and excited state- absorption) without static disorder (linear scale). The grid of exciton eigenenergies is labeled 1 − 7 and cross-peak CP(1,5) marked by the arrow. (b) Real part of the rephasing signal (stimulated emission) for static disorder FWHM= 94 cm−1. The ArcSinh scale facilitates comparison with [9], Fig. 1. equation of motion (HEOM) formalism [24, 25]. We gen- eralize this formalism from its original single-peak Drude- Lorentz term to a superposition of shifted peaks, eq. (2). The implementation on high-performance graphics pro- cessing units (GPU-HEOM) [15] allows us to advance 500 coupled hierarchical equations in lock-step and to obtain the computationally demanding 2d echo-spectra. We verified the convergence of GPU-HEOM by increas- ing the hierarchy-depth and inclusion of low-temperature correction terms [17, 18]. A typical 2d FMO-spectrum at T = 150 K and delay time Tdelay = 400 fs for the spectral density J3peak is shown in Fig. 2(a). The 2d echo-spectra are calculated as detailed in Ref. [18] by evaluating the third-order re- sponse function, including excited-state absorption [37] for the non-disorder averaged results. We mimic rota- tional averaging by probing along eight molecular orien- tations with respect to the laser-field polarization. 3 CP (1,5) CP (1,5)-SE 1 1.5 (a) T = 150 K (b) T = 150 K 1 0.5 4 3 2 1 (c) T = 277 K 0.5 0 −0.5 1 0.75 0.5 0.25 0 FWHM: 0 cm−1 FWHM: 47 cm−1 FWHM: 94 cm−1 (d) T = 277 K 0 0.2 0.4 0.6 0 0.2 0.4 0.6 Tdelay (ps) Tdelay (ps) FIG. 3: Cross-peak CP(1,5) oscillations of the FMO complex, real part of the rephasing signal. The sum of ground-state bleach, stimulated emission, and excited state-absorption is shown in (a) T = 150 K and (c) T = 277 K. (b) Stimulated emission pathway at T = 150 K for different disorder distri- butions FWHM= {0, 47, 94} cm−1. (d) Stimulated emission pathway at T = 277 K (no disorder). The inclusion of static disorder leads to a broadening of the peaks as shown in Fig. 2(b) for the stimulated- emission rephasing pathway and a Gaussian disorder- ensemble of 200 realizations with FWHM= 94 cm−1 added to the diagonal site-energies. The disordered peaks in Fig. 2(b) (which does not include the excited-state absorption) are somewhat narrower than in the exper- imental data [9], Fig. 1C. In Fig. 2, we reference the peaks on the diagonal (DP) and the cross-peaks (CP) by their location on the exciton eigenenergy-grid. Disorder leads to a broad central region where the peaks coalesce (compare Fig. 2(a) and (b)), and amplitude is present in regions A, B, and C, in reasonable agreement the exper- imental data in [9], Fig. 1C. The calculated 2d-spectra differ significantly from the single-peak Drude-Lorentz results shown in the SM [1], Fig. S5 and in [18]. Besides the broader line-shapes in the single-peak Drude-Lorentz case, the dynamics of the diagonal-peak amplitudes is al- tered and an upward amplitude transfer from DP(2,2) to DP(3,3) is visible for J3peaks, which is in line with the experimental data at T = 77 K [16], but does not exist in the single-peak Drude-Lorentz case. To determine the coherence-time of cross-peak oscil- lations, we analyze the changes in the amplitude of the cross-peak CP(1,5) marked with the arrow in Fig. 2(a). The oscillatory signal of CP(1,5) for J3peaks shows mod- ulations for the rephasing pathways (Fig. 3(a)). The coherent nature of the oscillations is best analyzed in the stimulated-emission (SE) pathway in Fig. 3(b) which shows oscillations at T = 150 K (277 K) for delay-times Tdelay up to 0.6 ps (0.3 ps). The excited-state absorption ρE1,E5 1 0.5 0 J3peaks J11peaks −0.5 T = 150 K ρE1,E5 1 0.5 0 −0.5 J3peaks J11peaks T = 277 K 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 t (ps) t (ps) FIG. 4: Exciton dynamics for the FMO with the spectral densities J3peaks and J11peaks. Time evolution of the coher- ence ρE1,E5 (t) = hE1ρ(t)E5i. The duration of coherent os- cillations of cross-peak CP(1,5) shown in Fig. 3 is in good agreement with the coherence life times ρE1,E5 (t). included in Fig. 3(a) neglects correlation between exci- tons and almost compensates the stimulated-emission, resulting in a larger non-oscillatory background. The addition of static disorder diminishes the oscillations to some extent, but the oscillatory signal prevails as shown for the SE-pathways in Fig. 3(b) at T = 150 K. In the single-peak Drude-Lorentz case discussed in [18] and the SM [1], Fig. S4, the oscillations decay two times faster and no long-lasting oscillations remain in the stimulated- emission signal at T = 277 K. The information of the exciton lifetime is already in- corporated in the simpler to calculate single exciton dy- namics. To this end, we prepare the system in an ini- tial density matrix ρ(t = 0) = EiihEj given in terms of two eigenstates {Eii, Eji} of Hex. In the 2d echo- spectra setup such a delocalized exciton state is reached after the first two laser pules illuminate the sample. The life time of the oscillatory components of cross-peak CP(1,5) of the 2d echo-spectra in Fig. 3 is in good agree- ment with the life time of the corresponding coherences ρE1,E5(t) = hE1ρ(t)E5i shown in Fig. 4. To isolate the impact of the vibronic mode-distribution we compare the coherences for the spectral densities J3peaks and J11peaks in Fig. 4. The coherences are al- most unchanged by switching from J3peaks to J11peaks. Specific modes in the peaked spectral density as the one near 180 cm−1 induce beatings in the coherences with small amplitude ≤ 0.01. In contrast to δ-shaped vi- bronic modes, the peaks in J11peaks which mimic the experimental data, have only a minor influence to the coherences. However, this does not imply that higher- frequency modes are unimportant. The relaxation to thermal equilibrium, which is prominently visible in 2d- echo spectra, relies on a large enough coupling of the exciton-system to the vibronic continuum. The spec- tral densities studied have a reorganization energy λ ≥ 35 cm−1 which lead to a thermalization after 5 ps at T = 77 K as observed for the FMO complex [16]. 4 For efficient energy-transfer it is important to verify that the decoherence rate γ still supports a thermaliza- tion of the exciton system within a few picoseconds. This is indeed the case for J3peaks and J11peaks and shows that despite the reduced slope of the spectral density, the sys- tem is still strongly coupled to the vibronic environment. Compared to the single-peak Drude-Lorentz form, the spectral densities J{3,11}peaks support even a faster trans- fer from the initial BChl 6 to the target BChl 3. In agree- ment with the experimental observations, the 2d spectra calculated here for the spectral density J3peaks do support long-lasting oscillations of the cross-peaks. In conclusion, we have shown that two seemingly contradictory requirements, namely (i) strong coupling needed for a fast thermalization and (ii) the observation of long-lasting coherent oscillations are two sides of the same coin. The careful modeling of the continuous part of the spectral density towards zero frequency is essen- tial for the prevalence of coherence in molecular networks since it determines the pure-dephasing rate γd. By com- bining the exciton Hamiltonian with a spectral density with the small slope at ω = 0 seen in experiments, we find a much better agreement of the theoretically calcu- lated 2d spectra and the experimental data once static disorder is included. The vanishing pure-dephasing rate of a super-Ohmic spectral density implies that the deco- herence is driven by the relaxation rate γr. This is ex- actly the opposite behaviour compared to models based on the single-peak Drude-Lorentz spectral density [17]. Furthermore, we find only a very small influence of the additional peaks in the spectral density J11peaks on the coherences. The cross-peak oscillations require a calcu- lation of 2d-echo spectra beyond the weak-coupling ex- pansion for structured spectral densities, which becomes feasible by the efficient GPU-HEOM algorithm. We thank B. Hein for help with calculating 2d-spectra and M. Rodr´ıguez for helpful discussions. This work is supported by the DFG within the Emmy-Noether pro- gram (KR 2889). Time on the Harvard SEAS "Reso- nance" GPU cluster and support by the SEAS Academic Computing team are gratefully acknowledged. [1] See Supplemental Material in the Appendix. [2] G. S. Engel et al., Nature 446, 782 (2007). [3] H. Lee, Y.-C. Cheng, and G. R. Fleming, Science 316, 1462 (2007). [4] E. Collini et al., Nature 463, 644 (2010). [5] T. R. Calhoun et al., J. Phys. Chem. B 113, 16291 (2009). [6] D. Abramavicius et al., Phys. Rev. Lett. 108, 067401 (2012). [7] W. Kuehn et al., Phys. Rev. Lett. 107, 067401 (2011). [8] G. D. Scholes, G. R. Fleming, A. Olaya-Castro, and R. van Grondelle, Nature Chemistry 3, 763 (2011). [9] G. Panitchayangkoon et al., Proc. Natl. Acad. Sci. 107, 5 12766 (2010). [26] P. Nalbach, D. Braun, and M. Thorwart, Phys. Rev. E [10] P. Rebentrost et al., New J. Phys. 11, 033003 (2009). [11] M. B. Plenio and S. F. Huelga, New. J. Phys. 10, 113019 84, 041926 (2011). [27] L. Muhlbacher and J. Ankerhold, New J. Phys. 10, (2008). 065023 (2008). [12] C. Olbrich, J. Strumpfer, K. Schulten, and U. [28] W. Koch, F. Grossmann, J. T. Stockburger, and J. Anker- Kleinekathofer, J. Phys. Chem. Lett. 2, 1771 (2011). hold, Phys. Rev. Lett. 100, 230402 (2008). [13] S. Shim, P. Rebentrost, S. Valleau, and A. Aspuru-Guzik, [29] J. Roden, A. Eisfeld, W. Wolff, and W. T. Strunz, Phys. Biophys. J. 102, 649 (2012). Rev. Lett. 103, 058301 (2009). [14] C. Olbrich et al., J. Phys. Chem. B 115, 8609 (2011). [15] C. Kreisbeck, T. Kramer, M. Rodr´ıguez, and B. Hein, J. Chem. Theory Comput. 7, 2166 (2011). [16] T. Brixner et al., Nature 434, 625 (2005). [17] A. Ishizaki and G. R. Fleming, Proc. Natl. Acad. Sci. 106, 17255 (2009). [18] B. Hein, C. Kreisbeck, T. Kramer, and M. Rodr´ıguez, New J. Phys. 14, 023018 (2012). [19] L. Chen, R. Zheng, Y. Jing, and Q. Shi, J. Chem. Phys. 134, 194508 (2011). [30] A. Kolli, A. Nazir, and A. Olaya-Castro, J. Chem. Phys. 135, 154112 (2011). [31] M. Wendling et al., J. Phys. Chem. B 104, 5825 (2000). [32] J. Adolphs and T. Renger, Biophys. J. 91, 2778 (2006). [33] U. Weiss, Quantum dissipative systems, 3rd ed. (World Scientific Publishing Co. Pte. Ltd., Singapore, 2008). [34] V. May and O. Kuhn, Charge and Energy Transfer Dy- namics in Molecular Systems (Wiley-VCH, Weinheim, 2004). [35] J. Yuen-Zhou and A. Aspuru-Guzik, J. Chem. Phys. 134, [20] N. Christensson, H. F. Kauffmann, T. Pullerits, and T. 134505 (2011). Mancal, arXiv:1201.6325v1 (2012). [36] C. Meier and D. J. Tannor, J. Chem. Phys. 111, 3365 [21] A. Chin et al., arXiv:1203.0776v1 (2012). [22] D. Hayes et al., Faraday Discuss. 150, 459 (2011). [23] C. W. Gardiner and P. Zoller, Quantum Noise (Springer- (1999). [37] S. Mukamel, Principles of Nonlinear Optical Spectroscopy (Oxford University Press, Oxford, 1999). Verlag, Berlin Heidelberg, 2004). [38] B. Bruggemann, P. Kjellberg, and T. Pullerits, Chem. [24] Y. Tanimura and R. Kubo, J. Phys. Soc. Jpn. 58, 101 Phys. Lett. 444, 192 (2007). (1989). [25] Y. Tanimura, J. Phys. Soc. Jpn. 75, 082001 (2006). Appendix Comparison of exciton dynamics for different spectral densities The exciton dynamics is strongly affected by the spectral density and the dephasing rate in Eq. (1) depends on the slope of the spectral density towards zero frequency. Fig. 5 shows a close up of spectral densities discussed in the main text, and in additions includes for convenient reference the single-peak Drude-Lorentz density (dotted line, JDL), used in some previous studies [17 -- 19]. JWendling J11peaks J3peaks JDL ) 1 − m c ( ) ω ( J 2 ω ¯h π = ) ω ( J 6 4 2 0 JWendling J11peaks J3peaks JDL ) 1 − m c ( ) ω ( J 2 ω ¯h π = ) ω ( J 250 200 150 100 50 0 0 2 4 6 8 10 0 100 200 300 400 500 ¯hω (cm−1) ¯hω (cm−1) FIG. 5: Left panel: low frequency onset of the spectral density of the FMO complex. Right panel: global view. Circles: measured spectral density [31], parametrization [32]. Solid line J3peaks, dashed line J11peaks, dotted line JDL, parameters given in the main text. 6 (a) ρE1,E5 0.02 0 -0.02 J3peaks J11peaks JDL T = 150 K (b) ρE1,E5 0.02 0 -0.02 J3peaks J11peaks JDL T = 277 K 0.5 0.75 1 1.25 1.5 0.25 0.5 0.75 1 1.25 t [ps] t [ps] FIG. 6: Detail of the long-time evolution of the coherence ρE1,E5 (t) = hE1ρ(t)E5i, which shows small-amplitude oscillations for J11peaks related to the peaks of the spectral density J11peaks located at higher frequencies. For JDL (dotted line), almost no coherences prevail. Population dynamics and coherences The additional peaks present in J11peaks at higher frequencies, which are not contained in J3peaks, affect the coher- ences of the off-diagonal elements ρE1,E5 only weakly, Fig. 6. The influence of the additional peaks in the spectral density J11peaks becomes visible as small-amplitude oscillations for longer delay times (Fig. 6). This shows how the initially electronic coherence gets transformed into oscillations related to the details of the sharp peaks in the spectral density J11peaks. This finding is in-line with the theoretical calculations for spectral-densities with added δ-peaked vibronic modes [20]. The same observation applies to the relaxation process, which is shown in Fig. 7 for a population dynamics starting either at site 1 or at site 6. For comparison, the results for the Drude-Lorentz spectral density (dotted curve) is also included, which deviates most significantly for ρ66. (a) ρ1 1 , 1 (b) ρ1 1 , J3peaks J11peaks JDL J3peaks J11peaks JDL T = 150 K T = 277 K 0.5 0 (c) ρ6 6 , 1 0.5 0 (d) ρ6 6 , J3peaks J11peaks JDL J3peaks J11peaks JDL T = 150 K T = 277 K 0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2 t [ps] t [ps] t [ps] t [ps] FIG. 7: Population dynamics for an initial population at sites 1 and 6 for the 3 spectral densities shown in Fig. 5 and for temperatures T = 150 K and T = 277 K. The relaxation proceeds on very-similar time-scales for the two spectral densities J3peaks and J11peaks. 7 (a) CP(1,5) 2 T = 150 K 1.5 1 (b) CP(1,5) (c) CP(1,5)-SE 1 (d) CP(1,5)-SE 1 2 T = 277 K T = 150 K T = 277 K 1.5 1 J3peaks JDL J3peaks JDL 0.5 0 0.5 0 J3peaks JDL J3peaks JDL 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0 0.2 0.4 0.6 t (ps) t (ps) t (ps) t (ps) FIG. 8: Exciton dynamics for the FMO complex for two different spectral densities J3peaks and JDL. (a) Real part of cross- peak CP(1,5) of the signal (rephasing and non-rephasing) at T = 150 K and T = 277 K. (b) Coherent oscillations of the stimulated emission (SE) part of cross-peak CP(1,5). Differences in 2d spectra The cross-peak oscillations depend strongly on the initial slope of the spectral densities towards zero frequency. This is reflected in Fig. 8, which compares the cross-peak oscillations for the two different spectral densities J3peaks and JDL, for which the dephasing- and relaxation contributions to the decoherence rate are reversed in magnitude. Fig. 8(b) shows that for JDL (dotted lines) the cross-peak oscillations are largely reduced at T = 150 K and almost completely absent at T = 277 K, while the super-Ohmic onset of J3peaks still leads to visible oscillations at T = 277 K (solid lines). Moreover, also the amplitudes of the diagonal-peaks show pronounced differences for the two spectral densities. The spectral density J3peaks transfers the highest diagonal-peak amplitude upwards with increasing delay time from DP(2,2) to DP(3,3) as illustrated in the left panels of Fig. 8. This shift is also seen in the experimental data at T = 77 K [16], Figure 1, panels a,b. In the Drude-Lorentz case (Fig. 8, right panels) the highest amplitude is located at diagonal-peak DP(2,2) for all delay-times, which corresponds to previous models (see Fig. 2 in [38]), but differs from the experimental data and the result for J3peaks. (a) 12700 12500 12300 ) 1 − m c ( 3 ω 12100 12700 ) 1 − m c ( 3 ω 12500 12300 12100 Tdelay = 40 fs Tdelay = 40 fs 1 2 3 4 5 6 7 1 2 3 4 5 6 7 (b) 7 6 5 4 3 2 1 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 12700 12500 12300 ) 1 − m c ( 3 ω 12100 12100 12300 12500 12700 12100 12300 12500 12700 ω1 (cm−1) ω1 (cm−1) Tdelay = 100 fs Tdelay = 100 fs 1 2 3 4 5 6 7 1 2 3 4 5 6 7 7 6 5 4 3 2 1 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 12700 ) 1 − m c ( 3 ω 12500 12300 12100 7 6 5 4 3 2 1 7 6 5 4 3 2 1 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 12100 12300 12500 12700 12100 12300 12500 12700 ω1 (cm−1) ω1 (cm−1) FIG. 9: (color online) Calculated 2d echo-spectrum of the FMO complex at T = 277 K for the delay times Tdelay = {40, 100} fs. Shown is the real part of the rephasing and non-rephasing signal. The square marks the diagonal peak (DP) with the highest amplitude. Left panels: spectral density J11peaks, showing amplitude shifts from DP(2,2) at Tdelay = 40 fs to DP(3,3) for Tdelay = 100 fs, right panels: spectral density JDL, highest amplitude remains at DP(2,2).
1312.4056	2	1312	2015-03-12T18:46:01	Quantum Model for a Periodically Driven Selectivity Filter in K$^{+}$ Ion Channel	[ "physics.bio-ph", "quant-ph" ]	In this work, we present a quantum transport model for the selectivity filter in the KcsA potassium ion channel. This model is fully consistent with the fact that two conduction pathways are involved in the translocation of ions thorough the filter, and we show that the presence of a second path may actually bring advantages for the filter as a result of quantum interference. To highlight interferences and resonances in the model, we consider the selectivity filter to be driven by a controlled time-dependent external field which changes the free energy scenario and consequently the conduction of the ions. In particular, we demonstrate that the two-pathway conduction mechanism is more advantageous for the filter when dephasing in the transient configurations is lower than in the main configurations. As a matter of fact, K$^+$ ions in the main configurations are highly coordinated by oxygen atoms of the filter backbone and this increases noise. Moreover, we also show that, for a wide range of dephasing rates and driving frequencies, the two-pathway conduction used by the filter leads indeed to higher ionic currents when compared with the single path model.	physics.bio-ph	physics	Quantum Model for a Periodically Driven Selectivity Filter in K+ Ion Channel Centro de Ciencias Naturais e Humanas, Universidade Federal do ABC, 09210-170, Santo Andr´e, Sao Paulo, Brazil A. A. Cifuentes and F. L. Semiao In this work, we present a quantum transport model for the selectivity filter in the KcsA potassium ion channel. This model is fully consistent with the fact that two conduction pathways are involved in the translocation of ions thorough the filter, and we show that the presence of a second path may actually bring advantages for the filter as a result of quantum interference. To highlight interferences and resonances in the model, we consider the selectivity filter to be driven by a controlled time-dependent external field which changes the free energy scenario and consequently the conduction of the ions. In particular, we demonstrate that the two-pathway conduction mechanism is more advantageous for the filter when dephasing in the transient configurations is lower than in the main configurations. As a matter of fact, K+ ions in the main configurations are highly coordinated by oxygen atoms of the filter backbone and this increases noise. Moreover, we also show that, for a wide range of dephasing rates and driving frequencies, the two-pathway conduction used by the filter leads indeed to higher ionic currents when compared with the single path model. PACS numbers: 03.65.Yz, 87.15.A-, 05.60.Gg, I. INTRODUCTION Quantum biology is an emerging field of research which aims at investigating the possibility of a functional role for quantum mechanics or coherent quantum effects in biological systems. Notably, much of the work made in the last years have been related to transport, being the Fenna-Matthews- Olson (FMO) complex an important example [1]. This pigment-protein complex is present in green sulfur bacteria and its function is to channel excitons from the chlorosome antenna complex, where light is harvested, to the reaction cen- ter which execute the primary energy conversion reactions of photosynthesis. Much of the attention given to this complex, and to quantum biology, in general, arose from the experimen- tal observation of long-lived oscillatory features using ultra- fast 2-D spectroscopy [2]. Such oscillations were interpreted as evidence for the presence of long-lived electronic coher- ence, something not trivial given the complexity of these open systems. After these experimental observations, many efforts have been made to explain the origin of those coherences [3], and also to investigate their possible relation with entangle- ment and other quantum related effects [4]. Due to its particular features and efficiency, ion channels constitute another biological system where non trivial quan- tum effects may appear and be functional [5]. These channels are transmembrane proteins and they have an important role in the production of electric signals in biological systems [6]. Their structure gives rise to a selectivity filter which is a very narrow channel which catalyses the dehydration, transfer, and rehydration of the ions in a very efficient way, achieving a flux of about 108 ions per second [7]. Throughly crystallographic studies and free energy computer simulations showed, more than ten years ago, that ion translocation in the filter involves transitions between two main states, and that these transitions occur through two physically distinct pathways of conduction [8, 9]. These pathways involve either two or three K+ ions occupying the selectivity filter. Details about the experiment and the simulations demonstrating that ion translocation in the channel unmistakably follows two distinct paths can be found in Methods section of [8]. In this work, we consider a quantum model with two con- duction pathways in accordance with experimental results and simulations in potassium channels [8, 9]. Consequently, the transport in this system constitutes a two-path problem where quantum superposition effects coming from the competing paths of conduction play a decisive role. Since we treat here the ionic current in the filter, the results predicted in this work can in principle be experimentally accessed with physiolog- ical techniques [5]. This requires the filter to be driven by a periodic time dependent electric field which is also included in our analyses. In the following, we present the basic elements of the model and the inclusion of the driving field. We then study the ionic conduction or current, highlighting the pos- sible advantages of having two and not just one conduction path. In particular, we study the role of having non uniform dephasing in the topology, given that this is the most likely physical picture in the selectivity filter. II. MODELING CONFIGURATIONS AND PATHS IN THE CHANNEL In the systematically studied KcsA potassium channel from soil bacteria Streptomyces lividians, whose structure is very well known [10], K+ ions loose their hydrating water molecules to enter the selectivity filter and carbonyl oxygen atoms in its backbone replace the water molecules. This al- lows the formation of a series of coordination shells through which the K+ ions can move. Qualitatively distinguishable configurations of ions and water molecules in the selectivity filter correspond to different configurations which will be rep- resented here as two-level systems. To be more specific, if a site k is populated i.e., in state 1(cid:105)k, this configuration is ac- tive. Otherwise, if it is in state 0(cid:105)k, this configuration is not active or not participating in the ionic conduction. In a classi- cal hoping mechanism, we would never find superpositions in one configuration (being and not being used) or entanglement between different configurations (different sites). Such quan- tum coherent events lead to resonances in the ionic current which might be measured directly using physiological tech- 5 1 0 2 r a M 2 1 ] h p - o i b . s c i s y h p [ 2 v 6 5 0 4 . 2 1 3 1 : v i X r a niques [5]. Ion translocation in the KcsA consists of K+ ions that pro- ceed along the pore axis of the selectivity filter in a single fash- ion with water molecules intercalating them. This gives rise to two main configurations, usually denoted as 1-3 and 2-4, representing the position of a pair of K+ ions in the selectivity filter as depicted in Figure 1. The smaller these numbers are, the closer to the extracellular side the ions are. For the sake of simplicity in notation, we will denote here 2-4 by s (source) and 1-3 by d (drain), indicating that we will assume that 2-4 is in part driven incoherently by interactions of intracellular K+ ions with the carbonyl oxygens in the entrance of the channel, and 1-3 can decay incoherently to another configuration cul- minating with an ion leaving the cell. These processes can be described by Lindblad superoperators in the form [11] − s ρσ s − d ρσ+ d − s σ+ − σ+ d , ρ d σ Ls(ρ) = Γs Ld(ρ) = Γd + 2σ (1) (2) where Ls(ρ) (Ld(ρ)) causes incoherent pump (decay) in the source (drain), Γs (Γd) is the incoherent pump (decay) rate for the source (drain), and σ+ k ) is the two-level raising and lowering operator which create (destroy) excitations in site k = s, d. Hereafter, {(cid:63), ρ} denotes the anticommutator {(cid:63), ρ} ≡ (cid:63) ρ + ρ (cid:63) and ρ is the density operator for the four sites. k (σ− , (cid:0)−(cid:8)σ (cid:16)−(cid:110) s , ρ(cid:9) + 2σ+ (cid:111) (cid:1) , (cid:17) FIG. 1: In the selectivity filter, ions move outwards the cell in the presence of negative oxygen atoms (red segments) of the carbonyl groups in the lateral backbones. Just two of the four backbones are shown. The concerted motion involves alternating potassium ions K+ and water molecules H2O (not shown). S 1 − S 4 are binding sites which are numbered according to convention that numbers decrease when the ion proceeds from intra- to extracellular space. There are two optimal pathways connecting s and d [8, 9], and they will be denoted here by numbers 1 and 2, with no risk of confusing them with the main configurations since these are now denoted as s and d. Figure 2 depicts this two-path network on the left panel. On the right panel, we present a linear single-path chain which will serve as a benchmark to test for a possible quantum advantage of the two pathway conduction. Along pathway 1, an ion first approaches the in- tracellular entrance to the selectivity filter, and then pushes the two ions in the filter, causing the outermost ion to leave the channel into the extracellular side. This is known as concentration-dependent path [8]. In the second optimal path- way, the concentration-independent path, the two ions in the filter move first, leaving a gap in the selectivity filter which 2 Hhop = c(σ+ will attract an incoming ion from intracellular space. All mi- croscopic elementary steps involved in these transitions can occur reversibly [9]. For this reason, we represent the situa- tion by a hopping term in the Hamiltonian ( = 1), − − − − 1 σ+ 1 ) + c(σ+ s σ+ d ) + σ + σ s σ 1 σ 1 d − − − − 2 σ+ 2 ) + βc(σ+ +βc(σ+ s σ+ d ), (3) + σ + σ 2 σ s σ 2 d k (σ− where c is a hopping rate and σ+ k ) with k = 1, 2 is the two- level raising (lowering) operator for site k. For the two (single) pathway topology we take β = 1 (β = 0). Of course, these configuration changes do not take place at the same rate, but we kept them equal in this treatment for the sake of simplicity. Our main goal is to discuss the general aspects of the problem. FIG. 2: (left) Two-path topology where the source s is incoherently driven and excitations hop thorough the network until it is incoher- ently dissipated through the drain d. (right) Single-path topology. All sites are subjected to dephasing (wavy arrows). The selectivity filter in potassium ion channels uses a two-path configuration for changing between s and d. III. DRIVING THE CHANNEL The last section, and especially Hamiltonian (3), refers to the channel under natural conditions in the cell membrane. In this case, there are only small electric fields due to concen- trations of different ions inside and outside the cell and also charged residues of aminoacids forming the filter. Now, we will consider a particular technique which allows one to probe individual channels in the membrane and to subject them to different electric fields and chemical environments. This tech- nique, called patch clamping [12], enables one to subject the ions in the channel to constant and time-dependent potentials due to applied electric fields [5]. Consequently, this changes the free energy scenario which rules the translocation of ions in the filter [8]. Following [5], we will consider the field to be engineered so that the configurations follow − s + 2(σ+ Hext = (Ω0 + Ω1 cos ωt)[σ+ + βσ+ s σ − 2 ) 2 σ 1 σ − 1 +3σ+ d σ − d ], (4) where Ω0 and Ω1 are essentially the amplitudes of the dc and ac parts of the field, respectively, and ω the angular frequency of the ac part. Similar changes in the free energy scenario oc- cur due to long-range coupling mechanisms in response to a perturbation at a large distance in the protein. This is what happens, for instance, in ion pumps due to the binding of Adenosine triphosphate (ATP). The full Hamiltonian for the periodically driven selectiv- ity filter is then H = Hhop + Hext. And this does not refer to ds121sdthe channel under natural condition but rather to the channel being probed in the patch clamping setup. Interaction with the environmental degrees of freedom, especially vibrations of the carbonyl groups of the selectivity filter backbone [3], surely induces dephasing noise. In this first treatment of the problem, we will assume the simplest model where this noise is local and memoryless, i.e., we will use the following Lind- blad superoperator (cid:1) , − i ρσ+ i σ − i i σ (5) (cid:88) γi i=s,d,1,2 (cid:0)−(cid:8)σ+ i , ρ(cid:9) + 2σ+ − i σ Ldeph(ρ) = where γi is a time independent positive dephasing rate. Usu- ally, the drain excitation probability or population is used to quantify transport efficiency in coupled quantum systems [13 -- 16]. Here, we are interested in the time average of this quan- tity or current I which reads [5] (cid:90) T 0 I = lim T→∞ 1 T 2 Γdρd(t) dt, (6) where ρd(t) is the reduced state of the drain which is obtained by tracing out all other sites. Both, the concentration dependent and independent paths, appear in the experiments and simulations as pathways con- necting site s to site d, and then contribute to ionic conduction [8, 9]. In the quantum regime, these different paths may com- pete leading to interference effects in the observable current I. We now present our findings about the main trends followed by this current. In particular, we investigate a possible quan- tum advantage of having two conduction paths linking s and d by comparing the current I produced with the topologies shown in Figure 2. Although we are treating transport in the context of a biological system, it is worthwhile noticing that coupled two-level systems also appears in a great variety of physical scenarios including, for example, quantum dots [15] and superconducting qubits [17]. Consequently, the results presented here might be of value for quantum technologies using qubits, where our results might even be promptly simu- lated and experimentally observed [18]. IV. ESTIMATION OF PARAMETERS AND DECOHERENCE IN THE SYSTEM In this section, we provide a more detailed physical discus- sion about the choice of parameters used in the simulations presented later on in this work. In particular, we try to pre- dict the order of magnitude of the decoherence rates in order to compare it with estimated frequencies of the system. It is important to remark that our work is based on an effective model. The same is done in recent descriptions of the FMO [13, 14]. In this case, nonlinear spectroscopy techniques gives direct information on coupling constants and frequencies ap- pearing in an effective description based on the occupation of two-level sites. For the ion channels, however, there is not yet an experimental technique which provides information on the parameters used in effective models. Consequently, we have to estimate the order of magnitude of the parameters involved 3 in our model using indirect available experimental data. This is of course not the most suitable scenario to do predictions, but we think that this is not a reason to prevent serious inves- tigations which pave the way for advances and motivate the proposition of new experimental techniques to verify or fal- sify the findings of the models. The physical constants should be chosen as to fullfill the expected (and measured) current which is transported by the channel under regular conditions, i.e., 108 ions/sec [7]. To achieve this, we basically follow the reasoning presented in [5]. From (3) and (4), it follows that Ω0 is the energy dif- ference between the configurations and c is the hopping rate between them. Let us consider the case c << Ω0. In or- der to active a transfer rate of 108 ions/sec, perturbation the- ory tell us that we must have c2/Ω0 ≈ 108ions/sec. On the other hand, this time dependent transport model presents res- onances when Ω0 = nω with n integer and ω the frequency of the drive which appears in (4). We will set the system to work near theses resonances. By defining the constant ω0 = 108s−1, these conditions are fulfilled by choosing c ≈ ω0 and n >> 1. For this reason, the simulations are run using c = 8ω0 and Ω0 = 256ω0. The incoherent pump and disposal of configura- tions given by Γs and Γd, respectively, depend on the specific conditions under which the experiment will be set. It is natural to think them as a monotonic function of the concentration of ions inside and outside the cell, as supposed in [5]. Numerical simulations show that variations of these constants only limit the total current but not its dependence on other parameters. So, it does no harm to fix these constants to be the same order as the rate 108 ions/sec, i.e, Γs = 2ω0 and Γd = ω0. Finally, one have some freedom to set Ω1 (the drive amplitude) at any desired value because it is an externally controlled parameter. Since the behavior of the current as a function of Ω1 consists of a sequence of maxima and minima [5], we will set this am- plitude such that one has the first minimum of current. This is achieved with Omega1 = 284.92ω0. Giving the complexity of the system, it is not easy to an- ticipate the decoherence rate. For this reason, in the first simulation shown in Figure 3, we vary the dephasing rate γ over a wide range. Consequently, we can draw conclusions in regimes such as pure quantum transport (small decoher- ence) and highly classical transport (massive decoherence). In what regime precisely the channel works is a question to be answered experimentally such as happened to some photo- syntetic complexes which were shown to keep track of some quantumness due to partial preservation of quantum coher- ences [2]. However, it is interesting to see that it is possi- ble, through simplified assumptions, to provide a rough or- der of magnitude of the dephasing rate. Again, we follow the reasoning originally shown in [5]. One can assume that the dominant form of noise comes from the stretch mode of the carbonyl groups in the ion channel. This naturally changes the width of the wells forming the trap sites, causing the fre- quency or energy of the stable configurations to fluctuate. The simplified model considering just one trap site and one mode of stretch is then described by the Hamiltonian Ht,CO = ωtb†b + ωCOa†a + λb†b(a† + a), (7) where ωt is the frequency of ion in the trapping well and ωCO is the frequency of the stretch mode. If one consider the trans- port of ions as a sequence of tunelling events through barriers separating the wells, it is not hard to show that the frequency of motion in each well must be around ωt ≈ 1012s−1 in or- der to obtain a tunneling rate of 108 ions/sec, as shown in [5]. Concerning the mode, given its typical high frequency ωCO ≈ 1014s−1, only the ground state is appreciably popu- lated in room temperature and the corresponding mean square deviation of the position of the oxygen atoms in the carbonyl groups is of the order of 0.02Å. One can then attempt to nu- merically find the order of magnitude of the fluctuations in ωt induced by oscillations of amplitude 0.02Å. This fluctuation turns out to be about one order of magnitude smaller than ωt [5]. Consequently, we can roughly consider λ = ωt/10 in (7). In the scope of this simplified model, an initial super- position state of the ion such as (0(cid:105) + 1(cid:105))/ 2 times the ground state of the mode, would evolve such that after a time td ≈ 103ω−1 t = 1ns, the fidelity with the initial state will drop to 1/e. This suggests a decoherence rate about γ ≈ 1/td = 109s−1. Therefore, in this crude estimation, one obtains that the dephasing rate γ is approximatelly one or- der of magnitude stronger than ω0. But this is not enough to discard the analyses of regimes where γ is slightly bigger or smaller than ω0. In fact, we will show in next section that the two-path topology is more efficient than the single path one even for high dephasings. As said before, it is likely that only an experiment will be able to precisely determine γ. In [5], it is proposed the use of physiological techniques to experimen- tally estimate γ from measurements of current. √ V. SIMULATIONS We first analyze the role of dephasing in the transport, es- pecially in the conduction pathways embodied by sites 1 and 2. It is well known that in the selectivity filter, the config- urations s and d have K+ ions residing near the center of a box formed by eight carbonyl oxygens, while in the interme- diate sites 1 and 2 the coordination is reduced to six oxygen atoms, with just four of them provided by the carbonyl groups of the backbone [8]. Since coupling to the stretch mode of the carbonyl groups is expected to be the main cause of de- coherence, we expect that the intermediate sites will possess lower decoherence rates. It is then interesting to see whether having less dephasing in the intermediate configurations helps conduction. In order to investigate this point, we compare both topolo- gies considering fixed dephasing in sites s and d (γs = γd = γ = 0.4ω0), and varying the dephasing γ in 1 and 2 (γ1 = γ2 = γ). The current I(γ) for this configuration is presented in Fig- ure 3, where we subtracted the current I(γ) which corresponds to the case of invariant dephasing. Interestingly enough, it is clear that the two pathway topology benefits from the passing through configurations of reduced dephasing (γ < γ = 0.4ω0) which is consistent with the fact that the intermediate sites are less coordinated. It is import to remark that Figure 3 does not allow us to de- 4 FIG. 3: Current I(γ) obtained by fixing dephasing γ = 0.4ω0 in sites s and d and varying γ which is the dephasing in sites 1 and 2. We subtracted I(γ) which is the current for equal dephasing in all sites. Solid line corresponds to the single-pathway topology and dashed line to the two-pathway. We used ω = 4ω0 and Ω1 = 284.92ω0. cide which topology is best suited for transport for a given value of γ. It only allows us to evaluate the advantage of having intermediate sites of lower decoherence rates than the binding sites. In figure 4, we address this point by fixing the ration γ/γ and varying γ. We show the behavior of the cur- rent in the cases where decoherence in the intermediate sites is lower or stronger than in the binding sites. It is now quite clear that, in respect to different decoherence regimes, the two-pathway topology is more likely to bring transport ad- vantages to the channel compared to the linear topology. In general, for a giving value of the driving field frequency ω, models such as the one considered here present resonances when varying the driving field amplitude Ω1 [5]. In order to gain more information about the advantages of having one or two conduction pathways in the filter, we now study the global maximum of current Imax(Ω1) as a function of ω. The result is shown in Figure 5. It is now clear that for most cases the two-path topology can offer advantages with and without de- phasing i.e., the range of ω for which the two-path supersedes the single path is quite wide. The results are actually quite convincing in favor of the two-path topology. Let us take the case ω = 10ω0, for instance. The current with no dephasing using the two-path topology is more than twice the current observed in the single-path topology. This is a clear evidence of constructive quantum interference arising from competing conduction paths. Even in the presence of dephasing, the ad- vantage of the two-path topology at ω = 10ω0 is much more pronounced than advantage found with the single-path for small ω. Therefore, having two competing paths of conduc- tion gives in general advantages for the filter, and this might have actually been used by channel to help it operate under a real noisy environment. As a final comment, it is interesting to see that about ω = 12ω0, dephasing helps conduction in the two-path topology. This is a phenomenon called dephasing- assisted transport in the literature [13, 14]. The same happens for the single-path topology for ω bigger than about 9ω0. γ /ω0010203040∼0.000.020.040.060.080.100.12-0.02I(γ)/ω0 - I(γ)/ω0∼∼I(γ)/ω0 - I(γ)/ω0γ /ω0∼0.00.20.40.60.81.0-0.02-0.010.000.010.02∼VI. FINAL REMARKS 5 In this work, we presented a simple quantum model which takes into account the most significant features of the KcsA potassium channel. In particular, we included the fact that this system employs two pathways of conduction. From a quan- tum point of view, this could be a big advantage given that possible constructive interference effects can play a role. We then studied the role played by a second pathway of conduc- tion, and we found that there might be indeed some advan- tage for the filter to have it. This advantage appears in both the closed system dynamics, which is certainly not the case in real ion channels, and in the open system scenario where system functions. It is important to remark that, from the ex- perimental side, it is still necessary to wait for advances to discover whether or not this system operates in this moderate noisy regime where the two-path topology confers advantages over the single-path. In other words, it is still an open question whether or not quantum coherence is present in this interesting biological system. However, the measure of the current using the scheme proposed in [5] would be enough to decide on the validity of the model. On the other hand, given that quantum transport is a very important topic for modern technologies, our results may still find applications in a great variety of cou- pled quantum systems such as arrays of quantum dots, trapped ions, and other systems alike. ] FIG. 4: (Above) Current I(γ) obtained by fixing γ/γ = 0.3 and vary- ing γ which is the dephasing in sites s and d. The quantity γ is the dephasing in the intermediate configurations 1 and 2. Solid line corresponds to the single-pathway topology and dashed line to the two-pathway. (Below) The same for γ/γ = 3. Solid line corresponds to the single-pathway topology and dashed line to the two-pathway. We used ω = 4ω0 and Ω1 = 284.92ω0. FIG. 5: Behavior of the global maximum of current Imax(Ω1) as a function of ω. Squares refer to the case with no dephasing and tri- angles to dephasing with γs = γd = 0.4ω0 and γ1 = γ2 = 0.1ω0. Empty shapes in dashed lines refer to the two-path topology and filled shapes in solid lines refer to the linear single-path topology. Acknowledgments. A.A.C. acknowledges Fundac¸ao de Amparo a Pesquisa do Estado de Sao Paulo (FAPESP) Grant No. 2012/12624-6, Brazil. F.L.S. acknowledge participa- tion as member of the Brazilian National Institute of Science and Technology of Quantum Information (INCT/IQ). F.L.S. also acknowledges partial support from CNPq under grant 308948/2011 − 4, Brazil. [1] R. E. Fenna and B. W. Matthews, Nature 258, 573 (1975); J. Adolphs and T. Renger, Biophys. J. 91, 2778 (2006). [2] G. S. Engel, T. R. Calhoun, E. L. Read, T. -K. Ahn, T. Man- cal, Y. -C. Cheng, R. E. Blankenship, and G. R. Fleming, Na- ture 446, 782 (2007); G. Panitchayangkoon, D. Hayes, K. A. Fransted, J. R. Caram, E. Harel, J. Wen, R. E. Blankenship, G. S. Engel, Proc. Natl. Acad. Sci. USA 107, 12766 (2010). [3] A. W. Chin, J. Prior, R. Rosenbach, F. Caycedo-Soler, S. F. Huelga, and M. B. Plenio, Nature Phys. 9, 113 (2013); T. Renger and F. Muh, Phys. Chem. Chem. Phys. 15, 3348 (2013). [4] M. Sarovar, A. Ishizaki, G. R. Fleming, and K. B. Whaley, Nat. Phys. 6, 462 (2010); K. Bradler, M. M. Wilde, S. Vinjanampa- thy, and D. B. Uskov, Phys. Rev. A 82, 062310 (2010). [5] A. Vaziri and M. B Plenio, New J. Phys. 12, 085001 (2010). [6] D. L. Nelson and M. M. Cox, Lehninger Principles of Biochem- istry (W. H. Freeman and Company, 2008). [7] E. Gouaux and R. MacKinnon, Science 310, 1461 (2005). [8] J. Morais-Cabral, Y. Zhou, and R. MacKinnon, Nature, 414 37 010203040γ /ω00.150.200.250.350.30I(γ)/ω0 010203040γ /ω00.150.200.250.350.30I(γ)/ω0 2 8 6 4 10 12ω/ω00.30.40.50.60.70.80.91.00.10.2Imax(Ω1)/ω0(2001). [9] S. Bern`eche and B. Roux, Nature, 414 73 (2001). [10] D. A. Doyle, J. M. Cabral, R. A. Pfuetzner, A. Kuo, J. M. Gul- bis, S. L. Cohen, B. T. Chait, R. MacKinnon, Science, 280, 69 (1998). [11] G. Lindblad, Commun. Math. Phys., 48 119 (1976); D. Walls and G. Milburn Quantum Optics (Springer 1994). [14] P. Rebentrost, M. Mohseni, I. Kassal, S. Lloyd, and A. Aspuru- Guzik, New J. Phys., 11 033003 (2009)). [15] F. L. Semiao, K. Furuya, and G. J. Milburn, New J. Phys., 12 083033 (2010)). [16] A. W. Chin, A. Datta, F. Caruso, S. Huelga, and M. B. Plenio, New J. Phys., 12 65002 (2010)). [17] Y. Makhlin, G. Schn, and A. Shnirman, Rev. Mod. Phys. 73, [12] B. Sakmann and E. Neher, Single-Channel Recording 357 (2001). (Springer, 2009). [13] K. B. Plenio and S.F. Huelga , New J. Phys., 10 113019 (2008). [18] A. A. Cifuentes and F. L. Semiao, in preparation. 6
1103.6109	1	1103	2011-03-31T07:25:34	Electron-conformational transformations in nanoscopic RyR channels govern both the heart's contraction and beating	[ "physics.bio-ph", "cond-mat.mes-hall", "cond-mat.soft" ]	We show that a simple biophysically based electron-conformational model of RyR channel is able to explain and describe on equal footing the oscillatory regime of the heart's cell release unit both in sinoatrial node (pacemaker) cells under normal physiological conditions and in ventricular myocytes under Ca$^{2+}$ SR overload.	physics.bio-ph	physics	Electron-conformational transformations in nanoscopic RyR channels govern both the heart's contraction and beating A.S. Moskvin,1 A.M. Ryvkin,2 O.E. Solovyova,2, 3 and V.S. Markhasin2 1Department of Theoretical Physics, Ural State University, 620083 Ekaterinburg, Russia 2Institute of Immunology and Physiology, Ural Branch of RAS, 620049, Ekaterinburg, Russia 3Department of Computational Mathematics, Ural State University, 620083 Ekaterinburg, Russia (Dated: October 30, 2018) We show that a simple biophysically based electron-conformational model of RyR channel is able to explain and describe on equal footing the oscillatory regime of the heart's cell release unit both in sinoatrial node (pacemaker) cells under normal physiological conditions and in ventricular myocytes under Ca2+ SR overload. i Calcium (Ca2+) dynamics is of a principal impor- tance for functioning of different heart's cells from atrial and ventricular cardiomyocites to sinoatrial node cells (SANC) though the former are responsible for the heart's contraction while the latter for primary heart's pacemak- ing, respectively 1. Cardiac contraction in cardiomyocites is activated by an increase in intracellular calcium con- centration (Ca2+ ), most of which comes from a specific calcium cistern of sarcoplasmic reticulum (SR). Ca2+ is released via the ryanodine receptors (RyR) in response to Ca2+ entering the cell via the L-type channels (see Fig. 1). The cardiac type RyR is the common major Ca2+ release channel type in SANC and ventricular myocytes. It has been experimentally documented in chemically skinned and voltage-clamped SANC, in which effects of voltage-activated sarcolemmal ion currents are excluded, that the isolated SR is capable to spontaneously and rhythmically release Ca2+ via RyRs 2,3. These spon- taneous, rhythmic, local subsarcolemmal Ca2+ releases (Ca2+ clock), which occur in SANCs, interact somehow with the classic sarcolemmal voltage oscillator (mem- brane clock 4). At present there is a general consensus about the importance of Ca2+ oscillator for SANC rate 5, however, an important discussion still remains whether it is a dominant or critical factor for cardiac pacemaker cell functioning. Furthermore, the very existence of the intracellular Ca2+ clock is not captured by the most part of existing essentially membrane-delimited cardiac pace- maker cell numerical models. Recently, Maltsev and Lakatta 6 have developed a new numerical SANC model (ML-model) featuring interac- tions of SR-based Ca2+ and membrane clocks to explore novel mechanistic insights into cardiac impulse initia- tion. They started with a well-known simplified model of the cell structure consisting of four compartments: sub-sarcolemmal space (subspace), cytosol, network SR (nSR), and junctional, or luminal SR (jSR) (Fig. 1). As in most existing models the authors used an effective medium theory, where Ca2+ concentrations in the sub- space and in jSR (CaSS and CajSR) are main governing parameters that obey standard reaction-diffusion equa- tions, while RyR gating is usually considered in a simpli- fied manner through a dependence of the release on the Ca2+ concentrations. The ML-model adopted the for- FIG. 1: Fig. 1. Schematic illustration of the cell compart- ments and Ca2+ fluxes in the SR Ca2+ clock toy model (see below Eqs.(7)-(10). SERCA is a sarco/endoplasmic reticulum Ca2+-ATPase that transfers Ca2+ from the cytosol to the SR. mulation of cardiac RyR function developed by Shannon et al. 7) and the Kurata et al. model 8 of primary rab- bit SANC. Finally the model was formulated in terms of a system of 29 first-order differential equations. The isolated SR can indeed operate as a self-sustained Ca2+ oscillator, described by a simple "release-pumping-delay" mechanism: a small spontaneous Ca2+ release from jSR to the subspace occurs as the primary or initiating event. When CaSS increases to a sufficient level, it amplifies the Ca2+ release via the mechanism of the Ca2+-induced Ca2+ release (CICR) 1; this relatively strong, secondary Ca2+ release simultaneously depletes (i.e., resets) jSR. The released Ca2+ is pumped into the nSR. The delay between releases is determined by the Ca2+ pumping rate and Ca2+ diffusion from the subspace to cytosol and also from nSR to jSR. As CajSR slowly increases, RyRs are restituted, and the next release is ultimately initiated, etc. The ML-model 6 of coupled oscillators seems to repro- duce basically all recently discovered new behavioral de- tails of cardiac pacemaker cell function, however, this phenomenological integrative model ignores many impor- tant physiological features of the cardiac cells, in par- ticular, the fine spatiotemporal structure of the Ca2+ release. The model of integrated Ca2+ dynamics does locally propagating Ca2+ re- not describe stochastic, leases within the subsarcolemmal space. Indeed, RyRs in SANC, as in other cardiomyocites, seem to be arranged in clusters under sarcolemma (Fig. 1) and thus proba- bly form subsarcolemmal calcium release units (CRUs). In this instance, RyRs release Ca2+ into a relatively small volume of subspace where individual jSRs of CRUs approach sarcolemma 1. Thus a realistic modeling of the Ca2+ oscillator and SANC function should include a stochastic mechanism of a local Ca2+ release gener- ation by CRUs. Furthermore, main assumption of the ML-model 6, that is the Ca2+ released from RyR chan- nels activates the RyR channels as like as the trans- sarcolemmal Ca2+ from the L-type channel in a close apposition, seems to be questionable. Recently we have applied well-known electron- conformational (EC) model (see, e.g.,Ref.9) for a single RyR channel 10 -- 12 that was shown to capture important features of the individual and cooperative behaviour of RyRs in ventricular myocytes. The EC model of RyR functioning under Ca2+ stimuli is based on a biophysi- cal adaptation of the well-known theory of photo-induced structural phase transitions, which has been successfully applied to different solids 13. Hereafter, in the Letter we will show that EC model of RyR channel is able to ex- plain and describe on equal footing a puzzling sponta- neous oscillatory regime of the release unit both in SANC under normal physiological conditions and in ventricular myocytes under Ca2+ SR overload. The ion-activated RyR channel is a giant (30 × 30 nm) macromolecular protein complex comprising 4 subunits of 565 000 Daltons each 1. As other ion channels it has a great many of internal electron and conformational degrees of freedom and exhibits remarkable complexi- ties that need to be considered when developing realistic models of ion permeation. Nevertheless, until recently most modelling efforts for RyR channels were focused on a simple "hole in the wall" type model with a set of dif- ferent (open, closed) states. Our knowledge of molecular mechanisms of RyR channel functioning is limited; hence we are forced to start with the most general "physicists'" approach, which is typical for protein biophysics. Such an approach to the modelling of biomolecular system im- plies its simplifying to bare essentials with guidance from experimental data. Modelling the RyR we start with a simple and a lit- tle bit naive picture of the massive nanoscopic chan- nel like an elastic rubber tube with a varying cross- section governed by a conformational coordinate Q and a light "electronic" plug switched due to Ca2+-RyR bind- ing/unbinding in the subspace 14. This electronic plug in- teracts with the conformational coordinate and acts as a trigger to stimulate its change and related channel cross- section/conductivity. In other words, we reduce a large variety of RyR degrees of freedom to only two: a fast and a slow one, conventionally termed as electronic and 2 FIG. 2: Fig. 2. Left panel: Naive "tube-and-plug" model of RyR channel in a closed ground state. Small letters c, o are used for electronically closed and open states, respectively, and capital letters C, O for conformationally closed and fully open states, respectively. Right panel: Corresponding adia- batic potentials in EC-model of RyR. Vertical arrows point to Ca2+-induced Franck-Condon (FC) electronic transitions, horizontal arrow points to a non-FC tunneling transition, downhill arrows point to a conformational dynamics. conformational one, respectively. Both degrees of free- dom are implied to be coupled to realize an EC trans- formation that is the electronic control of the slow con- formational motion. Bearing in mind the main function of RyR channels, we assume only two actual electronic RyR states: "open" and "closed", and a single confor- mational degree of freedom, Q, described by a classical continuous variable. Figure 2 illustrates the model with a set of representative states of the system. Change in the electronic and conformational states reg- ulate the main RyR channel function, i.e. determines whether the channel is "open" and permeable for Ca2+ ions or "closed" and impermeable to ions. Hereafter we assume that the conformational variable Q specifies the RyR channel "cross-section" or, more precisely, a perme- ability for Ca2+, while the dichotomic electronic variable determines its opening and closure. This allows us to describe the Ca2+ flux through the RyR as follows: JRyR = D(Q)(CajSR − CaSS), (1) if the channel is electronically open, and JRyR = 0, if it is closed. Here, the permeability coefficient D(Q) reflects the ease with which Ca2+ passes through an open RyR. Its functional dependence on the conformational coordi- nate should be one of the essential model assumptions. D(Q) is assumed to be an increasing function of the con- formational coordinate, varying from zero or small leak- age value to some saturated value D0 (0 < D(Q) < D0), when Q varies from large negative to large positive mag- nitudes, passing through some subconductive state at Q = 0. As a simplest limiting case, we may consider the step-like dependence D(Q < 0) = 0, D(Q ≥ 0) = D0. Hereafter, we shall assume a simple harmonic approx- imation for the conformational energy and use a Hookes harmonic law for elastic potential energy: E = 1 2 KQ2, where K is the effective "elastic" constant and Q = 0 relates to a base state with "unstrained tube" and a bare cross-section. It is worth noting that namely the EC model introduces the energy to be an important factor of the RyR functioning. As a starting point of the EC model algebra we in- troduce a simple effective Hamiltonian for a single RyR channel as follows 10 -- 12 Hs = −∆sz − hsx − pQ + K 2 Q2 + aQsz , (2) where sz and sx are well-known Pauli matrices, and the first term describes the bare energy splitting of "up" and "down" (electronically "open" and "closed") states with an energy gap ∆, while the second term describes their mixing. It is worth noting that given h = 0 we arrive at a classical approach with a dichotomic electronic vari- able. The third and fourth terms in (2) describe the linear and quadratic contributions to the conformational energy. Here, the linear term formally corresponds to the energy of an external conformational stress, described by an effective stress parameter p. The last term describes the EC interaction with the coupling parameter a. Here- after we make use of the dimensionless conformational variable Q; therefore all of the model parameters (∆, h, p, K, a) are assigned energy units. Two eigenvalues of our Hamiltonian E±(Q) = K 2 Q2 − pQ ± 1 2 (cid:2)(∆ − aQ)2 + h2(cid:3)1/2 (3) define two branches of the adiabatic, or conformational potential (CP), attributed to electronically closed (E−) and electronically open (E+) states of the RyR, respec- tively (see Fig. 2). The classical dynamics of the conformational coordi- nate we assume to obey a conventional Langevin equa- tion of motion M Q = − ∂ ∂Q E(Q) − M Γ Q + R(t), (4) where first term describes a total systematic conforma- tional force with M being an effective RyR mass (below M let to be unity), Γ is an effective dimensionless friction damping constant (relaxation rate), and R is the thermal fluctuation force (Gaussian-Markovian noise). The last two terms reflect the coupling to an external environment of channels. The thermo-activated transitions are caused by the thermal fluctuating force and free from the Franck- Condon principle. As a result of thermal transitions the whole system will finally relax to the thermal equilibrium. The temperature plays an important role in overcoming the energy barriers. In fact we deal with a hybrid "over" (thermal activation) and "through" (quantum tunneling) barrier transfer transitions (reactions). Quantum tun- neling is often addressed to specify the low-temperature limit of the barrier transfer transition probability that points to the possible way to uncover non-classical be- haviour for the RyR channel. We assumed the resonant quantum tunneling takes place between two branches of 3 conformational potential starting within a tunneling zone δQ = ǫ centered at the CP minimum with the probability obeyed the effective Gamov law Ptun = P0e−Atun∆Q√∆E , (5) where ∆Q is the width, ∆E is the height of the energy barrier, or the energy separation between the tunneling points and the point of the two branch intersection (see Fig. 2), and P0 effective tunneling attempt frequency. At present there is no sufficient understanding of the mechanisms that regulate local Ca2+ signaling in heart's cell despite persistent efforts to discriminate between the cytosolic and luminal Ca2+ activation hypotheses. Cal- cium enters the subspace by two pathways: across the sarcolemma via L-type channels and from the SR via RyR channels. Activation of RyRs by CajSR has been attributed to either Ca2+ feedthrough to high-affinity cy- toplasmic Ca2+ activation sites or to Ca2+ regulatory sites on the luminal side of the RyR. However, most of experimental observations on cardiac RyRs more difficult to reconcile with the Ca2+ feedthrough effect 15. A lu- minal Ca2+ sensor appears to continuously regulate the functional activity of the SR Ca2+ stores by linking SR Ca2+ content to the activity of the RyRs 15. However, in contrast with purely electronic effect of CaSS, the effect of relatively slowly varying CajSR on the RyR channels is likely to be purely "mechanical" one, through the re- spective conformational strain applied to RyR channels. The effect can be naturally incorporated into EC model, if we assume the strain parameter p in the RyR Hamil- tonian to be a function of CajSR or the jSR-to-subspace Ca2+ gradient. We assume p to rise with the luminal Ca2+ concentration in accordance with the Hill curve 11: p = 2 [CajSR]n [CajSR]n + K n Ca − 1; −1 ≤ p ≤ +1 , (6) where KCa is the half maximal value, n is a Hill coeffi- cient. Rise of p in the interval (-1,+1) leads to a crucial modification of CP from that of stabilizing closed RyR state to that of stabilizing open RyR state. It should be noted that in terms of the EC model one might in- troduce a critical SR load that specifies a critical ef- fective strain pcr = (cid:0) a a ∆(cid:1) when the minimum of the CPc branch for the electronically closed RyR state crosses CPo branch thus destroying the RyR bistability conditions and making RyRs stay close to their activation threshold where their cC→oO activation can presumably be easily provoked. − K 2 Cooperative dynamics of the RyR clusters in CRU has been studied in a series of model simulations for 11×11 square RyR lattice in our prior paper 12. We simulated both "in vitro" dynamics of RyR lattice when the calcium surrounding was assumed to be simply as a source of effective external fields, such as effective strain, and "in vivo" situation when the RyR lattice dynam- ics is incorporated into the whole cell calcium dynamics (EC-CRU model). In such a case the EC dynamics of RyR lattice is assumed to specify the number of open channels Nopen which in its turn specifies the release flux from the SR to subspace through RyR channels: Jrel = krelNopen(CajSR − CaSS), where krel is a sin- gle RyR channel release velocity coefficient. The system demonstrated four different modes of behaviour, depend- ing on the SR load 12: inactivation mode I at a rather small SR load (heavily underloaded SR); single channel activated mode II (Ca synapse, or quark mode) at under- loaded SR; domino-like firing-termination mode III with high degree of cooperativity (cluster bomb mode) at opti- mally loaded SR. The cluster bomb mode is characterized by a step-by-step opening of the neighboring RyRs with formation of a cluster composed of up to 40% open RyRs. The process results in an effective high-gain Ca2+ release from the SR. However, the decrease in SR load leads to a lowering of the effective strain, accompanied by the shift of the system to preferable channel's closure. This negative feedback effect results, first, in a slowing down of the nucleation process and, second, in an evolution of the inverse domino-like effect with full collapse of the cluster of open RyRs and termination of Ca2+ release. Puzzlingly, the CRU simulation 12 showed that SR over- load can result in the excitation of the RyR lattice auto- oscillations when the domino-like opening of a cluster of RyR channels resulted in effective Ca2+ release that, in turn, caused a lowering of the effective strain and simul- taneous closure of RyRs before the SR load started ris- ing. However, until CajSR approached the initial value, close to a critical concentration all channels re-opened si- multaneously, and Ca2+ release repeated spontaneously. This behaviour is repetitive, i.e. the system turned out to behave in an auto-oscillatory mode IV with a sponta- neous SR Ca2+ release. It should be emphasized that in contrast with the ML model, the auto-oscillation mode IV occurs as a result of the CajSR-dependent shape of the CP and a purely conformational cC ↔ oO transfor- mation without any L-type channel activity and Ca2+- induced electronic c ↔ o transitions. To the best of our knowledge it was a first quantitative model for the CRU oscillatory mode. Obviously the abnormal Ca2+ automaticity in SR over- loaded ventricular myocytes suggests the normal pace- maker SANC activity may have some similar features and actually proceed with elevated Ca2+ level. Indeed, Vino- gradova et al. 3 found a minimum diastolic Ca2+ level in rabbit SANC of ∼ 200 nM, that is twice the diastolic Ca2+ level in resting ventricular myocytes. Hereafter in the paper we address a more realistic EC- CRU model relevant for the SANCs which incorporates two main features of the Ca2+ machinery, that is a lo- cal and stochastic character of the Ca2+ release. As in our prior paper 12 we address the CRU model with 11×11 square lattice of RyRs, however, for simplicity we neglect any RyR-RyR coupling. At variance with12 each RyR channel was described by a diabatic CP with non-FC tun- neling transition between two CP branches. The calcium fluxes in a simplified cell model (Fig. 1) were assumed to obey a standard system of four differential equations 16: 4 (7) (8) (9) dCaSS dt dCai dt = (kjSR SS Jrel − Jdif f ); = (kSS i Jdif f − Jup); = (ki nSRJup − Jref ); dCanSR dt dCajSR dt = (knSR jSR Jref − Jrel) , (10) SS = 0.12, kSS i = 0.022, ki Jrel jSR = 9.7). = kref (CanSR − CajSR), where Jref = Nopenkrel (CajSR − CaSS), Jdif f = kdif f (CaSS − Cai) are diffusion fluxes between the nSR and the jSR, between the jSR and subspace, between subspace and cytosol, respectively, Jup = Pup Cai is uptake flux; Kup+Cai kα β are volume ratio constants between α and β cell compartments (kjSR nSR = 40, knSR It should be noted that all the parameters were chosen rather as typical for integrative cell models than for a single CRU model. In other words, all the cell RyRs we consider to form a system of identical CRU's functioning in concert. We set initial conditions as follows: Cai = 0.1 µM , CaSS = 0.5 µM , CajSR = 0.05 mM , CanSR = 1.5 mM , for other Ca2+ dy- namics parameters:Kup = 0.0006 mM ; kdif f = 5 ms−1. As in the ML model the CaSS and CajSR time courses strongly depend on the refill and release rate constants, that are expected both to change from their original values of ventricular myocytes and likely mediate the regulation of SR Ca2+ clock ticking speed 6. Main results of numerical simulations are presented in Fig. 3. As in ML model, three types of the steady-state were found: 1) steady rhythmic oscillations; 2) no oscilla- tions or damped oscillations; and 3) chaotic oscillations. The highest rates are reached when the oscillator ap- proaches to dynamic equilibrium (steady release, Fig. 3d, right part). The lowest rates are reached when the os- cillator approaches to static equilibrium. Static equilib- rium occurs when either the release rate is too small or Ca2+ pumping rate is too fast, i.e., jSR becomes highly loaded. Fig. 3 distinctly shows that CRU can operate as a self-sustained Ca2+ oscillator, however, the triggering of the Ca2+ release is determined by the bringing CajSR near the critical concentration, or activation threshold rather than by a small initial spontaneous Ca2+ release from jSR as in the ML-model 6. Maximal CajSR appears to slightly rise with kref , while its oscillation amplitude strongly depends on kref , however, in an absolutely dif- ferent way than in ML-model. Indeed, for a rather wide range of kref the CajSR oscillation amplitude rises with the refill rate, however, without full depletion of the jSR as it occurs in ML-model. Furthermore, in contrast with 5 FIG. 3: Fig. 3. A: Illustration of a repetitive shaping of the RyR's CP branches during refill-release mode and a flash pattern of open and closed RyRs in our 11×11 cluster. B: Model simulations illustrating that CRU can operate as a self-sustained Ca2+ oscillator. C: SR Ca2+ oscillator operates following a Bowditch-like pattern: "the faster rate, the stronger release". ÷ 5 · 10−2 ms−1 at krel = 0.01 ms−1, Bottom panel D: CaSS and CajSR courses given different refill rate constant kref = 10−4 kdif = 5 ms−1, Pup = 1 µM/ms. We make use of the EC model parameters as follows: ∆ = 0, K = 12, a = 5 , Γ = 7, P 0 tun = 0.1, Atun = 20, ǫ = 0.001. the ML-model the EC-CRU model reveals the Bowditch- like behavior: "the faster rate, the stronger release". In- deed, within oscillatory regime both the frequency and amplitude of CaSS do increase with kref . A detailed analysis shows this can be explained as a result of some retardation effect due to a slow conformational dynam- ics near the CP minima; optimal condition for the inter- branch tunneling and Ca2+ release are realized at higher kref and consequently, at higher CajSR. Indeed, a high- speed increase of the CajSR results in a rapid rise of effective stress p with a shift of the CP branches with its minima up and right (see Fig. 3). Given rather slow conformational dynamics the CP shift leads to a tunnel- ing retardation effect as the conformational coordinate Q needs a time to reach the new tunneling zone ǫ in the vicinity of new CP minimum. In other words, the EC system is not in time for adjusting to the rapid shift of the CP branches. However, the higher CajSR the slower the CP shift, that allows conformational coordinate Q to run tunneling zone down, and provide the optimal con- dition for the inter-branch tunneling and Ca2+ release. Despite the tunneling retardation, the bigger kref the shorter time to reach the critical CajSR level. Just the opposite behavior, "the higher the rate, the lower the amplitude," represents the major limitation of the isolated Ca2+ oscillator in frames of ML-model: it is unable to generate high-amplitude oscillations at higher rates. However, the authors 6 have shown that the inter- actions of the SR and sarcolemma clocks overcomes the limitation of the isolated SR Ca2+ clock, i.e., the oscilla- tion amplitude of the full system Ca2+ clock raises as the oscillation rate increases, i.e., the Bowditch phenomenon restores. At the same time the "Bowditch-like" behavior (the faster the stronger) for the both CaSS and CajSR appears to be a distinct feature of the EC-CRU model. In other words, it seems the SR Ca2+ oscillator can serve as a dominant source of persistent heart's beating. Thus the interrelation between two types of cell oscillators needs 6 in a futher examination. In summary, at variance with the integrative model by Maltsev and Lakatta simple biophysically based EC model of RyR channel describes stochastic local Ca2+ releases within the subsarcolemmal space as a result of conformational transformations followed by a tunneling between two CP branches. The model is able to explain and describe the spontaneous oscillatory regime of the CRU both in pacemaker cells under normal physiological conditions and in ventricular myocytes under Ca2+ SR overload including its subtle features such as the ampli- tude and frequency fluctuations. At variance with the integrative ML-model, the oscillation amplitude of the intracellular Ca2+ clock raises as the oscillation rate in- creases, thus providing the Bowditch law functioning of the pacemaker cell without any membrane clock assis- tance. Despite the EC model is intentionally simplistic, it offers novel insight into potential mechanisms govern- ing by the Ca2+ fluxes and may thus provide a starting point for further exploration of physical principles guid- ing cardiac cell functioning in vitro and in vivo. We acknowledge the support by Ural Branch of RAS under grant No.09-M-14-2001. 1 D.M. Bers, Excitation-Contraction Coupling and Cardiac Contractile Force. Second Edition, Kluwer Academic Pub- lishers, New York, 2002, 420 pp. 2 A. Fabiato, Basic Res. Cardiol., 80, suppl.2, 83 (1985). 3 T.M. Vinogradova, Y.Y. Zhou, V. Maltsev, A. Lyashkov, M. Stern, E.G. Lakatta, Circ Res 94, 802, (2004). 4 D. Noble., J. Physiol. 353, 1 (1984). 5 R.W. Tsien, R.S. Kass, R. Weingart, J. Exp. Biol. 81, 205, (1979); S.L. Lipsius, D.M. Bers, J. Mol. Cell Cardiol. 35, 891 (2003). 6 V.A. Maltsev and E.G. Lakatta, Am. J. Physiol. Heart Circ. Physiol. 296, 594, (2009) 7 T.R. Shannon, F. Wang, J. Puglisi, C. Weber, D.M. Bers, Biophys. J. 87, 3351 (2004). 8 Y. Kurata, I. Hisatome, S. Imanishi, T. Shibamotop, Am. J. Physiol. Heart. Circ. Physiol. 283 H2074 (2002). 9 A.B. Rubin, Biofizika. Teoreticheskaya biofizika, V.I, URSS, 2004, 464 p.(in Russian). 10 A.S. Moskvin, M.P. Philipiev, O.E. Solovyova, P. Kohl, V.S. Markhasin, Doklady Biochemistry and Biophysics, 400, 32 (2005). 11 A.S. Moskvin, M.P. Philipiev, O.E. Solovyova, V.S. Markhasin, Journal of Physics: Conference Series, 21, 195 (2005). 12 A.S. Moskvin, M.P. Philipiev, O.E. Solovyova, P. Kohl, V.S. Markhasin, Progress in Biophysics and Molecular Bi- ology, 90, 88 (2006). 13 K. Koshino., T. Ogawa, Journal of Luminescence, 87-89, 642 (2000). 14 At variance with the ion-activated RyR channel the potential-activated channels can be described by a single conformational coordinate governed by the electric field. 15 I. Gyorke and S. Gyorke, Biophys. J. 75 (6), 2801 (1998). 16 E.A. Sobie, K.W. Dilly, J. dos Santos Cruz, W.J. Lederer, M.S. Jafri, Biophys. J. 83, 59 (2002).
1707.02566	1	1707	2017-07-09T11:50:26	Continuum gating current models computed with consistent interactions	[ "physics.bio-ph" ]	The action potential signal of nerve and muscle is produced by voltage sensitive channels that include a specialized device to sense voltage. Gating currents of the voltage sensor are now known to depend on the back-and-forth movements of positively charged arginines through the hydrophobic plug of a voltage sensor domain. Transient movements of these permanently charged arginines, caused by the change of transmembrane potential, further drag the S4 segment and induce opening/closing of ion conduction pore by moving the S4-S5 linker. The ion conduction pore is a separate device from the voltage sensor, linked (in an unknown way) by the mechanical motion and electric field changes of the S4-S5 linker. This moving permanent charge induces capacitive current flow everywhere. Everything interacts with everything else in the voltage sensor so everything must interact with everything else in its mathematical model, as everything does in the whole protein. A PNP-steric model of arginines and a mechanical model for the S4 segment are combined using energy variational methods in which all movements of charge and mass satisfy conservation laws of current and mass. The resulting 1D continuum model is used to compute gating currents under a wide range of conditions, corresponding to experimental situations. Chemical-reaction-type models based on ordinary differential equations cannot capture such interactions with one set of parameters. Indeed, they may inadvertently violate conservation of current. Conservation of current is particularly important since small violations (<0.01%) quickly (<< 10-6 seconds) produce forces that destroy molecules. Our model reproduces signature properties of gating current: (1) equality of on and off charge in gating current (2) saturating voltage dependence in QV curve and (3) many (but not all) details of the shape of gating current as a function of voltage.	physics.bio-ph	physics	CONTINUUM GATING CURRENT MODELS COMPUTED WITH CONSISTENT INTERACTIONS Tzyy-Leng Horng1 , Robert S. Eisenberg2, Chun Liu3, Francisco Bezanilla4 1Department of Applied Mathematics, Feng Chia University, Taichung, Taiwan 40724 2Department of Molecular Biophysics and Physiology, Rush University, Chicago, Illinois 60612, United 3Department of Mathematics, Pennsylvania State University, University Park, Pennsylvania 16802, United States 4Department of Biochemistry and Molecular Biology, University of Chicago, Chicago, Illinois 60637, States United States Abstract The action potential signal of nerve and muscle is produced by voltage sensitive channels that include a specialized device to sense voltage. Not surprisingly the voltage sensor depends on the movement of permanent charges of basic side chains in the changing electric field, as suggested long ago by Hodgkin and Huxley and Bezanilla and Armstrong, for example. Gating currents of the voltage sensor are now known to depend on the back-and-forth movements of positively charged arginines through the hydrophobic plug of a voltage sensor domain. Transient movements of these permanently charged arginines, caused by the change of transmembrane potential, further drag the S4 segment and induce opening/closing of ion conduction pore by moving the S4-S5 linker. The ion conduction pore is a separate device from the voltage sensor, linked (in an unknown way) by the mechanical motion and electric field changes of the S4-S5 linker. This moving permanent charge induces capacitive current flow everywhere. Everything interacts with everything else in the voltage sensor so everything must interact with everything else in its mathematical model, as everything does in the whole protein. A PNP-steric model of arginines and a mechanical model for the S4 segment are combined using energy variational methods in which all movements of charge and mass satisfy conservation laws of current and mass. The resulting 1D continuum model is used to compute gating currents under a wide range of conditions, corresponding to experimental situations. Conservation laws are partial differential equations in space and time. Chemical-reaction-type models based on ordinary differential equations cannot capture such interactions with one set of parameters. Indeed, they may inadvertently violate conservation of current. Conservation of current is particularly important since small violations (<0.01%) quickly (<< 10-6 seconds) produce forces that destroy molecules. Our model reproduces signature properties of gating current: (1) equality of on and off charge in gating current (2) saturating voltage dependence in QV curve and (3) many (but not all) details of the shape of gating current as a function of voltage. The model computes gating current flowing in the baths produced by arginines moving in the voltage sensor. The model also captures the capacitive pile up of ions from the bulk solution in the form of electric double layers in vestibules adjacent to both ends of the hydrophobic plug. This pile up is coupled to the movement of arginines, and appears as a spike in the 1 beginning of gating current that have been already observed as the early transient gating current in experiments. Our results agree qualitatively well with experiments, and can obviously be improved by adding more details of the structure and its correlated movements. The proposed continuum model is a promising tool to explore the dynamics and mechanism of the voltage sensor that avoids the sampling constraints that so limit the representation of biological concentrations (10-7- 5×10-1 M) and currents (time scale > 10-4 sec) in molecular dynamics simulations that compute atomic motion on the 10-15 sec time scale. Keywords: ion channel; voltage sensor; gating current; arginine; S4; PNP-steric model 1. Introduction Much of biology depends on the voltage across cell membranes. The voltage across the membrane must be sensed before it can be used by proteins. Permanent charges1 move in the strong electric fields within membranes, so carriers of gating and sensing charge were proposed as voltage sensors even before membrane proteins were known to span lipid membranes [18-19, 21-23]. Movement of permanent charges of the voltage sensor is gating current and movement is the voltage sensing mechanism. Measurements of gating current were made much easier by biological preparations with artificially increased densities of voltage sensors and blocked conduction pores. Knowledge of membrane protein structure has allowed us to identify and look at the atoms that make up the voltage sensor. Protein structures do not include the membrane potentials and macroscopic concentrations that power gating currents, and therefore simulations are needed. Atomic level simulations like molecular dynamics do not provide an easy extension from the atomic time scale 2×10-16 sec to the biological time scale of gating currents that starts at 50×10-6 sec and reaches 50×10-3 sec. Calculations of gating currents from simulations must average the trajectories (lasting 50×10-3 sec sampled every 2×10-16 sec) of ~106 atoms all of which interact through the electric field to conserve charge and current, while conserving mass. It is difficult to enforce continuity of current flow in simulations of atomic dynamics because simulations compute only local behavior while continuity of current is global, involving current flow far from the atoms that control the local behavior. It is impossible to enforce continuity of current flow in calculations that assume equilibrium (zero flow) under all conditions. Periodic boundary conditions are widely used in simulations. Such conditions take a box of material and replicate it identically, so the potential at the corresponding edges of the box are identical. If the potentials are identical, current will not flow. Periodic boundary conditions of this sort are incompatible with current flow from one boundary to the other. Voltage clamp experiments, and natural biological function involve current flow from one boundary to another. A hybrid approach is needed, starting with the essential knowledge of structure, but computing only those parts of the structure used by biology to sense voltage. In close packed ('condensed') systems like the voltage sensor, or ionic solutions, 'everything interacts with everything else' because electric fields are long ranged as well as exceedingly strong [14]. In ionic solutions, ion channels, even in 1 Permanent charge is our name for a charge or charge density independent of the local electric field, for example, the charge and charge distribution of Na+ but not the charge in a highly polarizable anion like Br or the nonuniform charge distribution of H2O in the liquid state with its complex time dependent (and perhaps nonlinear) polarization response to the local electric field. 2 enzyme active sites, steric interactions that prevent the overfilling of space in well-defined protein structures are also of great importance as they produce long range correlations [17]. Closely packed charged systems are best handled mathematically by energy variational methods. Energy variational methods guarantee that all variables satisfy all equations (and boundary conditions) at all times and under all conditions and are thus always consistent. We have used the energy variational approach developed in [12, 20] to derive a consistent model of gating charge movement, based on the basic features of the structure of crystallized channels and voltage sensors. The schematic of the model is shown below. The continuum model we use simulates the mechanical dynamics in a single voltage sensor, while the experimental data from [7] is from many independent voltage sensors. Ensemble averages of independent voltage sensors recording is roughly equivalent to macroscopic continuum modeling in a single voltage sensor. It is unlikely one can do better until experiments determine the structural source of the variance in the voltage sensor and its time, voltage, and chemical dependence on ion type and concentration. 3. Mathematical model The reduced mechanical model for a voltage sensor is shown in Fig. 1 with four arginines Ri, i=1, 2, 3, 4, each attached to S4 helix by identical spring with spring constant K. The electric field will drag these four arginines since each arginine carries +1 charge. In addition, the charged arginines move the group itself. S4 connect to S3 and S5 at its two ends by identical springs with spring constant KS4/2. Once the membrane is depolarized from say 90mV inside negative, to +10mV inside positive, arginines together with S4 will be driven towards extracellular side. When the membrane is repolarized from say +10mV volts to 90mV, the arginines move to the intracellular side. This movement is the basic voltage sensing mechanism of ion channel. The movement of S4 triggers the opening or closing of the lower gate mainly consisting of S6 of the permeant ion channel by a mechanism widely assumed to be mechanical, although we hasten to say that electrical aspects of the linker motion are likely to be involved, as well, perhaps acting as a deterministic trigger and control for the otherwise stochastic sudden discontinuous change of current seen in single channel measurements of ion flow through the permeation channel. When arginines are driven by electric field, they are forced to move through a hydrophobic plug, composed by several nonpolar amino acids from S1, S2 and S3 [25]. Arginines moving from vestibule, a hydrated lumen, on one side, though the hydrophobic plug, and entering into the vestibule on another side would involve dehydration when passing through the hydrophobic plug. Therefore, arginines will encounter a barrier in the potential of mean force, an energy barrier, mainly dominated by the solvation energy [42]. Note that Na+ and Cl (used here as the only solution ions, for simplicity) can only exist in vestibules and are not allowed into the hydrophobic plug. The bottoms of the two vestibules on each side of the system act as impermeable walls for Na+ and Cl. When the voltage is turned on and off, these two walls will form a pair of capacitors storing/releasing net ion charges in their electric double layers (EDL). Molecular dynamics considers every particle individually and must deal with its enormous thermal motion (more or less at the speed of sound 1,000 m/s or 1 nm per nanosecond), as it seeks to compute gating currents on the biological time scale. We use a continuum treatment that deals with the thermal motion on the average, as temperature or diffusion. Such methods often include the correlations of the average electric field, e.g., Poisson-Nernst-Planck (PNP), but not those that differ from atom to atom, as 3 a glance at the structure of any protein shows is important. Here we include a steric effect in the PNP model [20], and so deal with many of the atomic scale correlations ignored in classical PNP. We hasten to add that it is not at all clear how well molecular dynamics deals with correlations produced by the average electric field, particularly when they include nonequilibrium components that drive current flow. Classical molecular dynamics assumes equilibrium and uses periodic boundary conditions that have certain difficulties dealing with the current flow and nonperiodic electric fields found in real devices and channels, as we have previously mentioned. Continuum models can adequately describe the time behavior of a voltage sensor in action, in which the time span is on the order of several hundreds of milliseconds. The four arginines Ri, i=1, 2, 3, 4, are described by their individual density distributions (concentrations) ci, i=1, 2, 3, 4, allowing the arginines to interact with Na+ and Cl in vestibules. The density (concentration) distributions represent probability density functions as shown explicitly in the theory of stochastic processes, used to derive such equations in [29] using the general methods of [28]. This practice is used widely in quantum mechanics [3, 34]. The important issue here is how well the correlations are captured in the continuum model. Some are more likely to be faithfully captured in molecular dynamics simulations (e.g., more or less local hard sphere interactions), others in continuum models (e.g., correlations induced by far field boundary conditions). Here we treat the S4 itself as a rigid body so we can capture the basic mechanism of a voltage sensor without considering the full details of structure, which might lead to a three dimensional model hard to compute in the time available. We construct an axisymmetric 1D model with a three-zone geometric configuration illustrated in Fig. 2 following Fig. 1. Zone 1 with 𝑧 ∈ [0, 𝐿(cid:3019)] is the intracellular vestibule; zone 2 with 𝑧 ∈ [𝐿(cid:3019), 𝐿(cid:3019) + 𝐿] is the hydrophobic plug; zone 3 with 𝑧 ∈ [𝐿(cid:3019) + 𝐿, 2𝐿(cid:3019) + 𝐿] is the extracellular vestibule. Arginines, Na+, and Cl can all reside in zone 1 and 3. Zone 2 only allows the residence of arginines, albeit with a severe hydrophobic penalty because of their permanent charge, in a region of low dielectric coefficient hence called hydrophobic. Spatial variation of dielectric coefficient allows construction of many useful devices [15, 41]. Types of boundary conditions and relevant sizes of this geometric configuration are also specified in Fig. 2. Note the no-flux boundary conditions specified in Fig. 2. One prevents Na+ and Cl from entering the hydrophobic plug (zone 2) with low dielectric coefficient. The other boundary condition constrains S4 motion and so prevents the arginines from leaving the vestibules into intracellular/extracellular domains. We first non-dimensionalize all physical quantities as follows, 𝑐(cid:3036) = (cid:3030)(cid:3284) (cid:3030)(cid:3116) , 𝜙(cid:3560) = (cid:3109) (cid:3038)(cid:3251)(cid:3021)/(cid:3032) , 𝑈(cid:3561) = (cid:3022) (cid:3038)(cid:3251)(cid:3021) , 𝑠 = (cid:3046) (cid:3019) , 𝑡 = (cid:3047) (cid:3019)(cid:3118)/(cid:3005)(cid:3299) , 𝐷(cid:3561)(cid:3036) = (cid:3005)(cid:3284) (cid:3005)(cid:3299) , 𝑔(cid:3556)(cid:3036)(cid:3037) = (cid:3034)(cid:3284)(cid:3285) (cid:3038)(cid:3251)(cid:3021)/(cid:3030)(cid:3116) , 𝐽(cid:4634)(cid:3036) = (cid:3011)(cid:3284) (cid:3030)(cid:3116)(cid:3005)(cid:3299)/(cid:3019) , 𝐼(cid:4634) = (cid:3010) (cid:3032)(cid:3030)(cid:3116)(cid:3005)(cid:3299)(cid:3019) , where 𝑐(cid:3036) is concentration of species i, with i=Na+, Cl, 1, 2, 3, and 4. Each is scaled by 𝑐(cid:2868) which is the bulk concentration of NaCl at the intracellular/extracellular domains. Here 𝑐(cid:2868) is set to be 184mM, equal on both sides, so that the Debye length 𝜆(cid:3005) = (cid:3495) (cid:3084)(cid:3293)(cid:3084)(cid:3116)(cid:3038)(cid:3251)(cid:3021) (cid:3030)(cid:3116)(cid:3032)(cid:3118) is 1nm when the relative permittivity 𝜀(cid:3045) = (cid:3013)(cid:2878)(cid:2870)(cid:3013)(cid:3019) 80. Note that 𝑐(cid:3036), i=1, 2, 3, and 4 needs to satisfy the following additional constraint ∫ (cid:2868) 𝐴(𝑧)𝑐(cid:3036)𝑑𝑧 = 1 due to uniqueness of each arginine. 𝜙 is the electric potential scaled by 𝑘(cid:3003)𝑇/𝑒 with 𝑘(cid:3003) being the Boltzmann constant; 𝑇 the temperature; e the elementary charge. All relevant external potentials U are scaled by 𝑘(cid:3003)𝑇. All sizes s are scaled by R, which is the radius of vestibule as shown in Fig. 2. R=1nm here. The time t is scaled by 𝑅(cid:2870)/𝐷(cid:3051), with 𝐷(cid:3051) being a diffusion coefficient that can be adjusted later to be 4 consistent with the time spans of on/off currents measured in experiments (caused by the movement of arginines). The diffusion coefficient of species i is scaled by 𝐷(cid:3051) . The coupling constant 𝑔(cid:3036)(cid:3037) of PNP-steric model based on combining rules of Lennard Jones, representing the strength of steric interaction between species i and j, is scaled by 𝑘(cid:3003)𝑇/𝑐(cid:2868) [20]. Note that here we only consider steric interaction among arginines, which is the most significant since arginines are generally crowded in hydrophobic plug and vestibules. For simplicity, we assume 𝑔(cid:3036)(cid:3037) = (cid:3420) 𝑔, ∀𝑖 ≠ 𝑗 0, ∀𝑖 = 𝑗 , 𝑖, 𝑗 = 1,2,3,4. The flux density of species i, 𝐽(cid:3036), is scaled by 𝑐(cid:2868)𝐷(cid:3051)/𝑅, and therefore the electric current I is scaled by 𝑒𝑐(cid:2868)𝐷(cid:3051)𝑅. For simplicity of notation, we will drop ~ for all dimensionless quantities from here on. Based on Fig. 2, the governing 1D dimensionless PNP-steric equations are expressed below. The first one is Poisson equation: − (cid:2869) (cid:3031) (cid:3002) (cid:3031)(cid:3053) (cid:4672)Γ𝐴 (cid:3031)(cid:3109) (cid:3031)(cid:3053) (cid:4673) = ∑ 𝑧(cid:3036)𝑐(cid:3036) (cid:3015) (cid:3036)(cid:2880)(cid:2869) , 𝑖 = Na, Cl, 1, 2, 3, 4, (1) with 𝑧(cid:3015)(cid:3028) = 1, 𝑧(cid:3004)(cid:3039) = −1, 𝑧(cid:3036) = 𝑧(cid:3028)(cid:3045)(cid:3034) = 1, 𝑖 = 1, 2, 3, 4, Γ = (cid:3118) (cid:3090)(cid:3253) (cid:3019)(cid:3118) and A(z) being the cross-sectional area. For zones 1 and 3, Γ = 1 since here the arginines are fully hydrated with 𝜀(cid:3045) = 80. For zone 2, we assume a hydrophobic environment 𝑤𝑖𝑡ℎ 𝜀(cid:3045) = 8, and therefore Γ = 0.1. The second equation is the species transport equation based on conservation law: (cid:3105)(cid:3030)(cid:3284) (cid:3105)(cid:3047) + (cid:2869) (cid:3105) (cid:3002) (cid:3105)(cid:3053) (𝐴𝐽(cid:3036)) = 0, 𝑖 = Na, Cl, 1 ,2 , 3, 4. (2) with the content of Ji expressed below based on Nernst-Planck equation for Na+ and Cl-: 𝐽(cid:3036) = −𝐷(cid:3036) (cid:4672) (cid:3105)(cid:3030)(cid:3284) (cid:3105)(cid:3053) + 𝑐(cid:3036)𝑧(cid:3036) (cid:3105)(cid:3109) (cid:3105)(cid:3053) (cid:4673) , 𝑖 = Na, Cl, 𝑧 ∈ zones 1 and 3, (3) and based on Nernst-Planck equation with steric effect and some imposed potentials for 4 arginines ci, i=1, 2, 3 and 4, based on Fig. 2, 𝐽(cid:2869) = −𝐷(cid:2869) (cid:3436) (cid:2986)(cid:3004)(cid:3117) (cid:3105)(cid:3053) + 𝑧(cid:3028)(cid:3045)(cid:3034)𝐶(cid:2869) (cid:3105)(cid:3109) (cid:3105)(cid:3053) + 𝐶(cid:2869) (cid:4672) (cid:3105)(cid:3023)(cid:3117) (cid:3105)(cid:3053) + (cid:3105)(cid:3023)(cid:3277) (cid:3105)(cid:3053) (cid:4673) + 𝑔𝐶(cid:2869) (cid:4672) (cid:2986)(cid:3004)(cid:3118) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3119) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3120) (cid:3105)(cid:3053) (cid:4673)(cid:3440) , 𝑧 ∈ all zones, (4) 𝐽(cid:2870) = −𝐷(cid:2870) (cid:3436) (cid:2986)(cid:3004)(cid:3118) (cid:3105)(cid:3053) + 𝑧(cid:3028)(cid:3045)(cid:3034)𝐶(cid:2870) 𝐽(cid:2871) = −𝐷(cid:2871) (cid:3436) (cid:2986)(cid:3004)(cid:3119) (cid:3105)(cid:3053) + 𝑧(cid:3028)(cid:3045)(cid:3034)𝐶(cid:2871) 𝐽(cid:2872) = −𝐷(cid:2872) (cid:3436) (cid:2986)(cid:3004)(cid:3120) (cid:3105)(cid:3053) + 𝑧(cid:3028)(cid:3045)(cid:3034)𝐶(cid:2872) (cid:3105)(cid:3109) (cid:3105)(cid:3053) (cid:3105)(cid:3109) (cid:3105)(cid:3053) (cid:3105)(cid:3109) (cid:3105)(cid:3053) + 𝐶(cid:2870) (cid:4672) (cid:3105)(cid:3023)(cid:3118) (cid:3105)(cid:3053) + (cid:3105)(cid:3023)(cid:3277) (cid:3105)(cid:3053) (cid:4673) + 𝑔𝐶(cid:2870) (cid:4672) (cid:2986)(cid:3004)(cid:3117) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3119) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3120) (cid:3105)(cid:3053) (cid:4673)(cid:3440) , 𝑧 ∈ all zones, (5) + 𝐶(cid:2871) (cid:4672) (cid:3105)(cid:3023)(cid:3119) (cid:3105)(cid:3053) + (cid:3105)(cid:3023)(cid:3277) (cid:3105)(cid:3053) (cid:4673) + 𝑔𝐶(cid:2871) (cid:4672) (cid:2986)(cid:3004)(cid:3117) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3118) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3120) (cid:3105)(cid:3053) (cid:4673)(cid:3440) , 𝑧 ∈ all zones, (6) + 𝐶(cid:2872) (cid:4672) (cid:3105)(cid:3023)(cid:3120) (cid:3105)(cid:3053) + (cid:3105)(cid:3023)(cid:3277) (cid:3105)(cid:3053) (cid:4673) + 𝑔𝐶(cid:2872) (cid:4672) (cid:2986)(cid:3004)(cid:3117) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3118) (cid:3105)(cid:3053) + (cid:2986)(cid:3004)(cid:3119) (cid:3105)(cid:3053) (cid:4673)(cid:3440) , 𝑧 ∈ all zones. (7) The first and second terms in Eqs. (3-7) describe diffusion and electro-migration term respectively. The third term in Eqs. (4-7) are external potential terms with Vi, i=1, 2, 3 and 4 being the constraint potential for the 4 arginines ci to S4 represented here by a spring connecting each arginine ci to S4 as shown in Fig. 1. It is expressed as 𝑉(cid:3036)(𝑧, 𝑡) = 𝐾(𝑧 − (𝑧(cid:3036) + 𝑍(cid:3020)(cid:2872)(𝑡)))(cid:2870), (8) where K is the spring constant, zi is the fixed anchoring position of the spring for each arginine ci on S4, 𝑍(cid:3020)(cid:2872)(𝑡) is the center-of-mass z position of S4 by treating S4 as a rigid body. Here we set z1=0.6, z2=0.2, z3=-0.2, z4=-0.6 given from the structure that gives the arginine anchoring interval on S4 as 0.4nm. 𝑍(cid:3020)(cid:2872)(𝑡) follows the motion of equation based on spring-mass system: 5 𝑚(cid:3020)(cid:2872) (cid:3031)(cid:3118)(cid:3027)(cid:3268)(cid:3120) (cid:3031)(cid:3047)(cid:3118) + 𝑏(cid:3020)(cid:2872) (cid:3031)(cid:3027)(cid:3268)(cid:3120) (cid:3031)(cid:3047) + 𝐾(cid:3020)(cid:2872)(𝑍(cid:3020)(cid:2872) − 𝑍(cid:3020)(cid:2872),(cid:2868)) = ∑ 𝐾 (cid:4672)𝑧(cid:3036),(cid:3004)(cid:3014) − (𝑧(cid:3036) + 𝑍(cid:3020)(cid:2872))(cid:4673) , (cid:2872) (cid:3036)(cid:2880)(cid:2869) (9) where 𝑚(cid:3020)(cid:2872), 𝑏(cid:3020)(cid:2872) and 𝐾(cid:3020)(cid:2872) are mass, damping coefficient and restraining spring constant for S4. 𝑍(cid:3020)(cid:2872),(cid:2868) is the natural position of 𝑍(cid:3020)(cid:2872)(𝑡) when the net force on the right hand side of Eq. (9) vanishes. Here, 𝑧(cid:3036),(cid:3004)(cid:3014) is the center of mass for the set of arginines ci , which can be calculated by 𝑧(cid:3036),(cid:3004)(cid:3014) = (cid:3261)(cid:3126)(cid:3118)(cid:3261)(cid:3267) ∫ (cid:3116) (cid:3002)((cid:3053))(cid:3053)(cid:3030)(cid:3284)(cid:3031)(cid:3053) (cid:3261)(cid:3126)(cid:3118)(cid:3261)(cid:3267) ∫ (cid:3116) (cid:3002)((cid:3053))(cid:3030)(cid:3284)(cid:3031)(cid:3053) , i=1, 2, 3, 4. (10) We assume that the spring mass system for S4 is over-damped, which means the inertia term in Eq. (9) can be neglected. Another external potential in the third term of Eqs. (4-7) is the energy barrier 𝑉(cid:3029), which is nonzero only in zone 2 representing mainly the solvation energy barrier from hydrophobicity. Generally, the accurate shape of 𝑉(cid:3029) in zone 2 requires a calculation of potential mean force (PMF) using molecular dynamics simulations by the structure of protein and the possible change in the PMF with membrane potential, ionic conditions, consistent with the distribution of charge and boundary conditions, and so on. However, here we simply assume a hump shape for the PMF, while we await proper calibrated calculations of the PMF and its dependence on membrane potential, etc. 𝑉(cid:3029),(cid:3040)(cid:3028)(cid:3051)(cid:3435)tanh(cid:3435)5(𝑧 − 𝐿(cid:3019))(cid:3439) − 𝑡𝑎𝑛ℎ(cid:3435)5(𝑧 − 𝐿 − 𝐿(cid:3019))(cid:3439) − 1(cid:3439), 𝑧 ∈ zone 2, (cid:3420) (11) 0, 𝑧 ∈ zones 1 and 3. with 𝑉(cid:3029),(cid:3040)(cid:3028)(cid:3051) set to be 5 in the current situation [42]. If we set 𝑉(cid:3029),(cid:3040)(cid:3028)(cid:3051) too large, the gating current would be very small since it would be very difficult for arginines to move across this barrier. Note that the tanh function is designated to smooth the otherwise top-hat-shape barrier profile, with its awkward infinite slopes not good for (numerical) differentiation. It is also generally believed that energy barrier in a protein structure does not have a jump. The last term in Eqs. (4-7) is the steric term that accounts for steric interaction among arginines [20]. Here we set g=0.5. Larger g implies larger steric effect, but g cannot be arbitrarily large due to the limitation of stability. Governing equations Eqs. (1-7) were derived by energy variational method based on the following energy (in dimensional form): 𝐸 = (cid:3505) (cid:4680)𝑘(cid:3003)𝑇 (cid:3533) 𝑐(cid:3036)𝑙𝑜𝑔𝑐(cid:3036) − (cid:3023) (cid:3028)(cid:3039)(cid:3039) (cid:3036) + (cid:3533) (cid:3028)(cid:3045)(cid:3034)(cid:3036)(cid:3041)(cid:3036)(cid:3041)(cid:3032)(cid:3046) (cid:3036),(cid:3037) 𝜀(cid:2868)𝜀(cid:3045) 2 𝑔(cid:3036)(cid:3037) 2 𝑐(cid:3036)𝑐(cid:3037) (cid:4681) 𝑑𝑉, ∇𝜙(cid:2870) + (cid:3533) 𝑧(cid:3036)𝑒 (cid:3028)(cid:3039)(cid:3039) (cid:3036) 𝑐(cid:3036)𝜙 + (cid:3533) (cid:3028)(cid:3045)(cid:3034)(cid:3036)(cid:3041)(cid:3036)(cid:3041)(cid:3032)(cid:3046) (𝑉(cid:3036) + 𝑉(cid:3029)) 𝑐(cid:3036) where the first term is entropy; second and third terms are electrostatic energy; fourth term is constraint and barrier potential for arginines; last term is the steric energy term based on Lennard-Jones potential [12, 20]. The Poisson equation Eq. (1) is derived by 𝛿𝐸 𝛿𝜙 = 0, and species flux densities in Eqs. (3-7) are derived by where 𝜇(cid:3036) is the chemical potential of species i. 𝜇(cid:3036) = 𝛿𝐸 𝛿𝑐(cid:3036) , 𝐽(cid:3036) = − 𝐷(cid:3036) 𝑘(cid:3003)𝑇 𝑐(cid:3036)∇𝜇(cid:3036), Also, here we assume quasi-steady state for Na+ and Cl-, which means (cid:3105)(cid:3030)(cid:3284) (cid:3105)(cid:3047) = 0, 𝑖 = Na, Cl. This is 6 justified by the fact that the diffusion coefficients of Na+ and Cl in vestibules are much larger than the diffusion coefficient of arginine based on the very narrow time span of the leading spike of gating current measured in experiments. The spike comes from the linear capacitive current of vestibule when the command potential suddenly rises or drops. This quasi-steady state assumption is essential. Otherwise using realistic diffusion coefficients for Na+ and Cl- would render Eqs. (1-7) too stiff to integrate in time. The spike contaminating the gating current can be removed by a simple technique called P/n leak subtraction in experiments (see Section 5.4). How to do this computationally will be discussed in section 5.3. Boundary and interface conditions for electric potential 𝜙 are 𝜙(0) = 𝑉, 𝜙(𝐿(cid:3019) (cid:2879)) = 𝜙(𝐿(cid:3019) (cid:2878)), Γ(𝐿(cid:3019) (cid:2879))𝐴(𝐿(cid:3019) (cid:2879)) 𝑑𝜙 𝑑𝑧 (𝐿(cid:3019) (cid:2879)) = Γ(𝐿(cid:3019) (cid:2878))𝐴(𝐿(cid:3019) (cid:2878)) 𝑑𝜙 𝑑𝑧 (𝐿(cid:3019) (cid:2878)), 𝜙(𝐿(cid:3019) + 𝐿(cid:2879)) = 𝜙(𝐿(cid:3019) + 𝐿(cid:2878)), Γ(𝐿(cid:3019) + 𝐿(cid:2879))𝐴(𝐿(cid:3019) + 𝐿(cid:2879)) (cid:3031)(cid:3109) (cid:3031)(cid:3053) (𝐿(cid:3019) + 𝐿(cid:2879)) = Γ(𝐿(cid:3019) + 𝐿(cid:2878))𝐴(𝐿(cid:3019) + 𝐿(cid:2878)) (cid:3031)(cid:3109) (cid:3031)(cid:3053) (𝐿(cid:3019) + 𝐿(cid:2878)), 𝜙(2𝐿(cid:3019) + 𝐿) = 0. (12) (cid:2878), 𝑡) = 𝑐(cid:3036)(𝐿(cid:3019) These are Dirichlet boundary conditions at both ends and continuity of electric potential and displacement at the interfaces between zones. Boundary and interface conditions for arginine are 𝐽(cid:3036)(0, 𝑡) = 𝐽(cid:3036)(2𝐿(cid:3019) + 𝐿, 𝑡) = 0, 𝑐(cid:3036)(𝐿(cid:3019) 𝑐(cid:3036)(𝐿(cid:3019) + 𝐿(cid:2879), 𝑡) = 𝑐(cid:3036)(𝐿(cid:3019) + 𝐿(cid:2878), 𝑡), 𝐴(𝐿(cid:3019) + 𝐿(cid:2879))𝐽(cid:3036)(𝐿(cid:3019) + 𝐿(cid:2879), 𝑡) = 𝐴(𝐿(cid:3019) + 𝐿(cid:2878))𝐽(cid:3036)(𝐿(cid:3019) + 𝐿(cid:2878), 𝑡), 𝑖 = 1,2,3,4. (13) No-flux boundary conditions are placed at both ends of the gating pore to prevent arginines and S4 from entering intracellular/extracellular domains. The others are continuity of concentration and flux at interfaces between zones. Boundary conditions for Na+ and Cl are 𝑐(cid:3015)(cid:3028)(0, 𝑡) = 𝑐(cid:3004)(cid:3039)(0, 𝑡) = 𝑐(cid:3015)(cid:3028)(2𝐿(cid:3019) + 𝐿, 𝑡) = 𝑐(cid:3004)(cid:3039)(2𝐿(cid:3019) + 𝐿, 𝑡) = 1, 𝐽(cid:3015)(cid:3028)(𝐿(cid:3019), 𝑡) = 𝐽(cid:3004)(cid:3039)(𝐿(cid:3019), 𝑡) = 𝐽(cid:3015)(cid:3028)(𝐿(cid:3019) + 𝐿, 𝑡) = 𝐽(cid:3004)(cid:3039)(𝐿(cid:3019) + 𝐿, 𝑡) = 0. (14) (cid:2879), 𝑡), 𝐴(𝐿(cid:3019) (cid:2879))𝐽(cid:3036)(𝐿(cid:3019) (cid:2879), 𝑡) = 𝐴(𝐿(cid:3019) (cid:2878), 𝑡), (cid:2878))𝐽(cid:3036)(𝐿(cid:3019) Dirichlet boundary conditions are placed at both ends of the gating pore to describe the concentrations for Na+ and Cl are bulk concentration over there and no-flux boundary conditions at both ends of hydrophobic plug to describe the impermeability of Na+ and Cl into hydrophobic plug. Besides the main input parameter 𝑉, which is the voltage bias (command potential) applied, other parameters like Di, i=1, 2, 3, 4, K, KS4, bS4 are also required. Results are especially sensitive to the values of K, KS4, bS4. We have tried and found Di=50, i=1,2,3,4, K=3, KS4=3, bS4=1.5 fit best with experiment [7]. We compare computational outputs with important experiments: (1) gating current versus voltage curve (IV) and (2) gating charge versus voltage curve (QV) as well as (3) gating current vs. time curves. Usually the electric current in the ion channel is treated simply as the flux of charge and is uniform in z when steady. This is not so in the present non-steady dynamic situation, since storing and releasing of charge in vestibules and the channel are involved. Here the flux of charge at the middle of hydrophobic plug, z= LR+L/2, was computed to estimate the experimentally observed gating current. However, it is actually impossible to experimentally measure the current within the channel since it is impossible to put a probe in the middle of the hydrophobic plug and optical methods to record these variables are so far unavailable. In experiments, the voltage clamp technique is used, and on/off gating current through the membrane is measured in experiments, which should be equal to the flux of charge at z=0 in the present frame work as shown in Fig. 2. The flux of charges at any z position 𝐼(𝑧, 𝑡) can be related to the flux of charges at z=0, 𝐼(0, 𝑡), simply by charge conservation: (cid:3105) (cid:3105)(cid:3047) 𝑄(cid:3041)(cid:3032)(cid:3047)(𝑧, 𝑡) = 𝐼(0, 𝑡) − 𝐼(𝑧, 𝑡), (15) where 7 𝑄(cid:3041)(cid:3032)(cid:3047)(𝑧, 𝑡) = ∫ 𝐴(𝜉) ∑ 𝑧(cid:3036)𝑐(cid:3036)𝑑𝜉, all (cid:3036) (16) (cid:3053) (cid:2868) and flux of charges at any z position 𝐼(𝑧, 𝑡) is defined by 𝐼(𝑧, 𝑡) = 𝐴(𝑧) ∑ 𝑧(cid:3036)𝐽(cid:3036)(𝑧, 𝑡) all (cid:3036) . (17) We may as well identify (cid:3105) (cid:3105)(cid:3047) 𝑄(cid:3041)(cid:3032)(cid:3047)(𝑧, 𝑡) as the displacement current, and denote it as 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043)(𝑧, 𝑡), since it will be shown later that Eq. (15) is equivalent to Ampere's law in Maxwell's equations, and (cid:3105) (cid:3105)(cid:3047) 𝑄(cid:3041)(cid:3032)(cid:3047)(𝑧, 𝑡) is exactly the displacement current in Ampere's law. A general discussion can be found in [10]. Hence, Eq. (15) can be simply re-written as 𝐼(cid:3047)(cid:3042)(cid:3047)(𝑧, 𝑡) = 𝐼(𝑧, 𝑡) + 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043)(𝑧, 𝑡) = 𝐼(0, 𝑡), (18) where we define the sum of displacement current and flux of charges as the total current 𝐼(cid:3047)(cid:3042)(cid:3047)(𝑧, 𝑡). The z-distribution of the total current should be uniform by Kirchhoff's law, which will be computationally verified in section 5.3. We are also interested in observing the net charge at vestibules. Taking net charge at intracellular vestibule, 𝑄(cid:3041)(cid:3032)(cid:3047)(𝐿(cid:3019), 𝑡), as an example, the net charge consists of arginine charges and their counter charges formed by the EDL of ionic solution over there. Generally, electro-neutrality is approximate but will not be exact there. Note that the existence of EDL's at vestibules, acting as a pair of capacitors, is an unavoidable consequence of discontinuities in constitutive properties like permanent charge density and dielectric coefficients. Realistic atomic scale simulations will have EDL and capacitive currents. More about flux of charge, displacement current and net charge at vestibules will be discussed in section 5.3. Eq. (15) is consistent with Ampere's law in Maxwell's equations: or equivalently, ∇× (cid:4672) (cid:3003)(cid:4652)⃗ (cid:3091)(cid:3116) (cid:4673) = 𝜀(cid:2868)𝜀(cid:3045) (cid:3105)(cid:3006)(cid:4652)⃗ (cid:3105)(cid:3047) + 𝐽⃗, (19) ∇ ∙ (cid:4672)𝜀(cid:2868)𝜀(cid:3045) (cid:3105)(cid:3006)(cid:4652)⃗ (cid:3105)(cid:3047) + 𝐽⃗(cid:4673) = 0, (20) where 𝐸(cid:4652)⃗ is the electric field and 𝐽⃗ is flux density of charge (current density). Eq. (20) tells us that the current is conserved everywhere and it consists of flux of charges 𝐽⃗ and displacement current 𝜀(cid:2868)𝜀(cid:3045) (cid:3105)(cid:3006)(cid:4652)⃗ (cid:3105)(cid:3047) Eq. (20) can be derived by Poisson equation and species transport equation like Eq. (1) and Eq. (2). Starting from Poisson equation in dimensional form: . or equivalently −∇ ∙ (𝜀(cid:2868)𝜀(cid:3045)∇𝜙) = 𝜌 + ∑ 𝑧(cid:3036)𝑒𝑐(cid:3036) (cid:3036) , (21) ∇ ∙ (cid:3435)𝜀(cid:2868)𝜀(cid:3045)𝐸(cid:4652)⃗(cid:3439) = 𝜌 + ∑ 𝑧(cid:3036)𝑒𝑐(cid:3036) (cid:3036) . (22) Taking time derivative of Eq. (22), and using species transport equation based on mass conservation, ∇ ∙ (cid:4672)𝜀(cid:2868)𝜀(cid:3045) (cid:3105)(cid:3006)(cid:4652)⃗ (cid:3105)(cid:3047) (cid:4673) = ∑ 𝑧(cid:3036)𝑒 (cid:3036) (cid:3105)(cid:3030)(cid:3284) (cid:3105)(cid:3047) , (23) (cid:3105)(cid:3030)(cid:3284) (cid:3105)(cid:3047) + ∇ ∙ 𝐽⃗(cid:3036) = 0, (24) then 8 ∇ ∙ (cid:4672)𝜀(cid:2868)𝜀(cid:3045) (cid:3105)(cid:3006)(cid:4652)⃗ (cid:3105)(cid:3047) (cid:4673) = ∑ 𝑧(cid:3036)𝑒 (cid:3036) (cid:3105)(cid:3030)(cid:3284) (cid:3105)(cid:3047) = − ∇ ∙ ∑ 𝑧(cid:3036)𝑒𝐽⃗(cid:3036) = −∇ ∙ 𝐽⃗ (cid:3036) , (25) which is exactly Eq. (20) by defining Casting Eq. (20) into the present 1D framework and integrate in space from 0 to z, we have 𝐽⃗=∑ 𝑧(cid:3036)𝑒𝐽⃗(cid:3036) (cid:3036) . (26) 𝜀(cid:2868)𝜀(cid:3045)𝐴(𝑧) (cid:3105)(cid:3006)((cid:3053),(cid:3047)) (cid:3105)(cid:3047) + 𝐼(𝑧, 𝑡) = 𝜀(cid:2868)𝜀(cid:3045)𝐴(0) (cid:3105)(cid:3006)((cid:2868),(cid:3047)) (cid:3105)(cid:3047) + 𝐼(0, 𝑡). (27) Comparing with Eq. (18), 𝜀(cid:2868)𝜀(cid:3045)𝐴(𝑧) (cid:3105)(cid:3006)((cid:3053),(cid:3047)) (cid:3105)(cid:3047) − 𝜀(cid:2868)𝜀(cid:3045)𝐴(0) (cid:3105)(cid:3006)((cid:2868),(cid:3047)) (cid:3105)(cid:3047) = 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043)(𝑧, 𝑡), (28) which justifies the naming of displacement current in Eq. (18). To construct the QV curve, we calculate 𝑄(cid:2869) = ∫ 𝐴(𝑧) ∑ 𝑐(cid:3036)𝑑𝑧 (cid:2872) (cid:3036)(cid:2880)(cid:2869) (cid:3013)(cid:3267) (cid:2868) (cid:3013)(cid:3267)(cid:2878)(cid:3013) , 𝑄(cid:2870) = ∫ (cid:3013)(cid:3267) 𝐴(𝑧) ∑ 𝑐(cid:3036)𝑑𝑧 (cid:2872) (cid:3036)(cid:2880)(cid:2869) , 𝑄(cid:2871) = ∫ (cid:2870)(cid:3013)(cid:3267)(cid:2878)(cid:3013) (cid:3013)(cid:3267)(cid:2878)(cid:3013) 𝐴(𝑧) ∑ 𝑐(cid:3036)𝑑𝑧 (cid:2872) (cid:3036)(cid:2880)(cid:2869) , which are the amounts of arginine found in zone 1, 2 and 3, respectively. Usually 𝑄(cid:2870) ≈ 0 due to the energy barrier 𝑉(cid:3029) in zone 2. Arginines tend to jump across zone 2 when driven from zone 1 to zone 3 when the voltage V is turned on. The number of arginines that move and settles at zone 3 depends on the magnitude of 𝑉. Besides IV and QV curves, the time course of the movement of the arginines and of gating charge (and 𝑧(cid:3036),(cid:3004)(cid:3014)(𝑡) and 𝑍(cid:3020)(cid:2872)(𝑡)), are important to report here. The movement of arginines can (almost) be recorded in experiments nowadays by optical methods. Many qualitative models accounting for the movement of S4 and conformation change of the voltage sensor have been proposed. Readers are referred to the review paper [37] for more details. 4. Numerical method High-order multi-block Chebyshev pseudospectral methods are used here to discretize Eqs. (1-2) in space [38]. The resultant semi-discrete system is then a set of coupled ordinary differential equations in time and algebraic equations (an ODAE system) [2]. The ordinary differential equations are chiefly from Eq. (2), and algebraic equations are chiefly from Eq. (1) and boundary/interface conditions Eq. (12-14). This system is further integrated in time by an ODAE solver (ODE15S in MATLAB [31-32]) together with appropriate initial condition. ODE15S is a variable-order-variable-step (VSVO) solver, which is highly efficient in time integration because it adjusts the time step and order of integration. High-order pseudospectral methods generally provide excellent spatial accuracy with economically practicable resolutions. A combination of these two techniques makes the whole computation very efficient. This is particularly important, since numerous computations have to be tried during the tuning of parameters. 5. Results and discussions Here we explored several parameter values to obtain charge movement with kinetics and steady state properties similar to the experimentally recorded gating currents. Numerical results based on the mathematical model described above were calculated and compared with experimental measurements [7]. Our one dimensional continuum model has advantages and disadvantages. The lack of 9 three-dimensional structural detail means of course that some details of the gating current and charge cannot be reproduced. It should be noted however that to reproduce those, one needs more than just static structural detail. One must also know how the structures (particularly their permanent and polarization charge) change after a command potential is applied, in the ionic conditions of importance. Structural detail that does not depend on time can be expected to give an incomplete, probably misleading image of mechanism (imagining studying an internal combustion engine without seeing the pistons or valves moving). The 1D model has the important advantages that it computes the actual experimental results on the actual experimental time scale, in realistic ionic solutions and with far field boundary conditions actually used in voltage clamp experiments. It also conserves current as we will discuss later 5.1 QV curve When the membrane and voltage sensor is held at a large inside negative potential (e.g., hyperpolarized to inside negative say -90mV), S4 is in a down position and all arginines stay in the intracellular vestibule. When the potential is made more positive (e.g., depolarized to +10mV, inside positive), S4 is in the up position and all arginines are at the extracellular vestibule. The voltage dependence of the charge (arginines) transferred from intracellular vestibule to extracellular vestibule is characterized as a QV curve in experimental papers and it is sigmoidal in shape [7]. Fig. 3(a) shows that our computed QV curve- the dependence of 𝑄(cid:2871) on V-is in very good agreement with the experiment [7]. Fig. 3(b) shows the steady-state distributions of Na+, Cl and arginines in the inside negative, hyperpolarized situation (V=-90mV). As we can see, all arginines stay at intracellular vestibule, and none of arginines transferred to extracellular vestibule (𝑄(cid:2871) ≈ 0). Fig. 3(c) shows the situation at V=-48mV, which is the midpoint of QV curve. As we can see, each vestibule has half of the arginines staying in it (𝑄(cid:2871) = 2): R1 and R2 are expected in extracellular vestibule, and R3 and R4 in intracellular vestibule. The concentration distributions of arginines shown in Fig. 3(c), interpreted as individual probability density functions, show R1 and R2 residing more in extracellular vestibule, and R3 and R4 on the contrary more in intracellular vestibule. Note that there are almost no arginines in zone 2 (hydrophobic plug) due to the energy barrier in it. If the command potential at the midpoint (𝑄(cid:2871) = 2) is V=0mV, the unforced natural position of S4 is at the middle of gating pore, i.e., 𝑍(cid:3020)(cid:2872),(cid:2868) = 𝐿(cid:3019) + 0.5𝐿. In experimental measurement [7], V is actually -48mV instead of 0mV for the midpoint. This requires 𝑍(cid:3020)(cid:2872),(cid:2868) to be biased from 𝐿(cid:3019) + 0.5𝐿 to 𝑍(cid:3020)(cid:2872),(cid:2868) = 𝐿(cid:3019) + 0.5𝐿 + 1.591nm. Fig. 3(d) shows the situation at full depolarization (V=-8mV). As we can see, all arginines move to extracellular vestibule (𝑄(cid:2871) ≈ 4). 5.2 Gating current Fig. 4 shows the time course of gating currents, observed as flux of charge at the middle of gating pore 𝐼(𝐿(cid:3019) + 𝐿/2, 𝑡), due to the movement of arginines when the membrane is largely depolarized, and partially depolarized. In the case of large depolarization, V rises from -90mV to -8mV at t=10, holds on till t=150, and drops back to -90mV as shown in Fig. 4(a). Time course of gating current and contributions of individual arginines are shown in Fig. 4(b). We observe that the rising order of each current component follows the moving order of R1, R2, R3 and R4 when depolarized, and that order is reversed when repolarized later. The area under the gating current is the amount of charge moved. Since arginines move forward and backward in this depolarization/repolarization scenario, the areas 10 under the ON current (arginines moving forward) and the OFF current (arginines moving backward) are same. The areas are equal for each component of current as well. The equality of area is an important signature of gating current that contrasts markedly with the properties of ionic current. Equality of area has in fact been used as a signature that identifies gating current and separates it from ionic current [26, 5]. Since almost all arginines move from intracellular to extracellular vestibule when the command potential is suddenly made more positive in a large depolarization, the area under each current component is then very close to 1. In the case of partial depolarization, V rises from -90mV to -50mV at t=10, holds on till t=150, and drops back to -90mV as shown in Fig. 4(c). The time course of gating current and its four components contributed by each arginine for this situation is shown in Fig. 4(d). Under this partial polarization, not all arginines move past the middle of hydrophobic plug due to weaker driving force in partial polarizations compared with large depolarization case. This can be seen in Fig. 4(d), where areas under each component current are different, and based on these various areas R1 moves more towards extracellular vestibule than R2; R2 more so than R3; and etc. The gating currents shown above can be better understood by looking at a sequence of snapshots showing the spatial distribution of electric potential, species concentration and electric current. The distributions at several times are shown in Fig. 5 for the case of sudden change in command voltage to a more positive value, a large depolarization, and Fig. 6 for the case of small positive going change in potential, a partial depolarization. Fig. 5 shows that almost all arginines are driven from intracellular to extracellular vestibules in case of large depolarization, but only part of the arginines move in case of partial depolarization (Fig. 6). The electric potential profiles at t=13 and t=148 show that the profile of electric potential as arginines move from left to right even though the voltage is maintained constant across the sensor. This is not a constant field system at all! Note that slight bulges in electric potential profile exist wherever arginines are dense. This can be easily understood by understanding the effect of Eq. (1) on a concave spatial distribution of electric potential. In Figs. 5 and 6, total current defined in Eq. (18), though changing with time, is always constant in z at all times, which is no surprise because it must satsify Kirchhoff's law. At t=13, a time that gating current is not quiescent as shown in Fig. 4(b) and 4(d), we can particularly visualize the z-distributions of flux of charges 𝐼(𝑧, 𝑡), displacement of current 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043)(𝑧, 𝑡) and total current 𝐼(cid:3047)(cid:3042)(cid:3047)(𝑧, 𝑡) individually in Figs. 5 and 6. 5.3 Flux of charges at different locations Flux of charges 𝐼(𝑧, 𝑡), together with displacement of current 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043)(𝑧, 𝑡) and total current 𝐼(cid:3047)(cid:3042)(cid:3047)(𝑧, 𝑡), depicted in Figs. 5 and 6 deserves more discussions here. Flux of charges at the middle of gating pore, 𝐼(𝐿(cid:3019) + 𝐿/2, 𝑡) , and both ends of gating pore, 𝐼(0, 𝑡) and 𝐼(2𝐿(cid:3019) + 𝐿, 𝑡) , should be computed by 𝐼 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:2870) , 𝑡(cid:4673) = 𝐴 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:4673) ∑ (cid:2870) (cid:3028)(cid:3045)(cid:3034)(cid:3036)(cid:3041)(cid:3036)(cid:3041)(cid:3032)(cid:3046) 𝑧(cid:3036)𝐽(cid:3036) (cid:4672)𝐿(cid:3019) + , 𝑡(cid:4673) , (29) (cid:3013) (cid:2870) 𝐼(0, 𝑡) = 𝐴(0) ∑ 𝐼(2𝐿(cid:3019) + 𝐿, 𝑡) = 𝐴(2𝐿(cid:3019) + 𝐿) ∑ 𝑧(cid:3036)𝐽(cid:3036)(0, 𝑡) (cid:3036)(cid:2880)(cid:3015) ,(cid:3004)(cid:3039) , (30) . (31) 𝑧(cid:3036)𝐽(cid:3036)(2𝐿(cid:3019) + 𝐿, 𝑡) (cid:3036)(cid:2880)(cid:3015)(cid:3028),(cid:3004)(cid:3039) Except 𝐼 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:2870) , 𝑡(cid:4673) , 𝐼(0, 𝑡) and 𝐼(2𝐿(cid:3019) + 𝐿, 𝑡) are trivially zero due to the implement of quasi-steadiness (cid:3105)(cid:3030)(cid:3284) (cid:3105)(cid:3047) = 0, 𝑖 = Na, Cl, in vestibules, which causes 𝐽(cid:3015)(cid:3028) and 𝐽(cid:3004)(cid:3039) to be uniform in 11 vestibules by Eq. (2), and further become zero by the no-flux boundary conditions for 𝑁𝑎(cid:2878) and 𝐶𝑙(cid:2879) at the bottom of vestibules as described in Eq. (14). We have to alternatively reconstruct 𝐼(0, 𝑡) and 𝐼(2𝐿(cid:3019) + 𝐿, 𝑡) by charge conservation of 𝑁𝑎(cid:2878) and 𝐶𝑙(cid:2879), 𝐼(0, 𝑡) = (cid:3031) (cid:3031)(cid:3047) (cid:3013) ∫ 𝐴(𝑧) ∑ (cid:2868) Na,Cl 𝑧(cid:3036)𝑐(cid:3036)𝑑𝑧 , (32) 𝐼(2𝐿(cid:3019) + 𝐿, 𝑡) = − (cid:3031) (cid:3031)(cid:3047) (cid:3013)(cid:2878)(cid:2870)(cid:3013)(cid:3267) ∫ (cid:3013)(cid:2878)(cid:3013)(cid:3267) 𝐴(𝑧) ∑ Na,Cl 𝑧(cid:3036)𝑐(cid:3036)𝑑𝑧 . (33) After obtaining 𝐼(0, 𝑡) and 𝐼(2𝐿(cid:3019) + 𝐿, 𝑡), we can further reconstruct flux of charges 𝐼(𝑧, 𝑡) at zone 1 (intracellular vestibule) and zone 3 (extracellular vestibule) by charge conservation of 𝑁𝑎(cid:2878) and 𝐶𝑙(cid:2879) again, 𝐼(𝑧, 𝑡) = 𝐼(0, 𝑡) − (cid:3031) (cid:3031)(cid:3047) (cid:3053) ∫ 𝐴(𝑧) ∑ (cid:2868) 𝑧(cid:3036)𝑐(cid:3036)𝑑𝑧 , 𝑧 ∈ [0, 𝐿(cid:3019)], (34) Na,Cl 𝐼(𝑧, 𝑡) = 𝐼(2𝐿(cid:3019) + 𝐿, 𝑡) + (cid:3031) (cid:3031)(cid:3047) (cid:2870)(cid:3013)(cid:3267)(cid:2878)(cid:3013) ∫ (cid:3053) 𝐴(𝑧) ∑ Na,Cl 𝑧(cid:3036)𝑐(cid:3036)𝑑𝑧 , 𝑧 ∈ [𝐿(cid:3019) + 𝐿, 2𝐿(cid:3019) + 𝐿]. (35) Once flux of charges is analyzed, we can then compute the displacement current based on finding the time derivative of Eq. (16). Finally we sum up flux of charges 𝐼(𝑧, 𝑡) and displacement of current 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043)(𝑧, 𝑡) to obtain total current 𝐼(cid:3047)(cid:3042)(cid:3047)(𝑧, 𝑡) and verify that it satisfies Eq. (18) and is uniform in z. This verification is shown in Figs. 5 and 6 at several various times, and is fact true at any time as well. In the bottom rows of Figs. 5 and 6 at t=13, we observe that 𝐼(𝑧, 𝑡) is generally non-uniform in z and is accompanied by congestion/decongestion of arginines in between. However 𝐼(𝑧, 𝑡) is almost uniform at zone 2 (hydrophobic plug), which means almost no congestion/decongestion of arginines occurs there, and therefore no contribution to the displacement current (cid:3031) (cid:3031)(cid:3047) 𝑄(cid:3041)(cid:3032)(cid:3047)(𝑧, 𝑡) from zone 2. This is no surprise since arginines can hardly reside at zone 2 due to energy barrier in it. Because of the energy barriers, once arginines leave zone 1 (intracellular vestibule), they immediately jump across zone 2 and enter into zone 3 (extracellular vestibule). Several things are worth noting in the time courses of 𝐼 (cid:4672)𝐿(cid:3019) + , 𝑡(cid:4673), and 𝐼(0, 𝑡) illustrated in Fig. (cid:3013) (cid:2870) 7(a) under the case of large depolarization. First, 𝐼 (cid:4672)𝐿(cid:3019) + , 𝑡(cid:4673) is noticeably larger than 𝐼(0, 𝑡) in ON (cid:3013) (cid:2870) period. This is because their difference, exactly the displacement current 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043), is always negative at zone 2 when depolarized, since arginines are leaving zone 1 and make (cid:3031) (cid:3031)(cid:3047) 𝑄(cid:3041)(cid:3032)(cid:3047) < 0 for zone 2. It is expected the area under the time course of 𝐼 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:2870) , 𝑡(cid:4673) would be very close to 4e, as verified by the time courses of Q3 in Fig. 7(b), while the counterpart area of experimentally measurable 𝐼(0, 𝑡) would be less than 4e due to its smaller magnitude compared with 𝐼 (cid:4672)𝐿(cid:3019) + , 𝑡(cid:4673). This may account for some (cid:3013) (cid:2870) experimental observations that at most 13e [4, 27, 30], instead of 16e, are moved during full depolarization in 4 voltage sensors (for a single ion channel) based on calculating the area under voltage-clamp gating current. Therefore, flux of charge at any location of zone 2, though impossible to measure in experiments so far, will give us amount of arginines moved during depolarization more reliably than the measurable 𝐼(0, 𝑡). 12 Second, we see in Fig. 7(a) with magnification in its inset plot that, as in experiments, 𝐼(0, 𝑡), but not 𝐼 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:2870) , 𝑡(cid:4673), has contaminating leading spikes in ON and OFF parts of the current. These spikes are linear capacitive currents from solution EDL of vestibules caused by sudden rising and dropping of command potential, and are called the early transient gating current in experiments [33, 35-36]. In voltage-clamp experiments, subtracting this linear capacitive component and removing the spike from gating current is done by 'leak subtraction', in various forms, e.g., P/4 (see Section 5.4) In reality, this linear capacitive current that is subtracted in this procedure comes from both the lipid bilayer membrane in parallel with the channel and vestibule solution in series with the channel. Here, we only considered capacitive current from solution EDL of vestibule and ignored the capacitive current of the membrane in parallel with the channel (which is actually much larger than vestibule capacitive current) because we use Dirichlet boundary conditions for 𝜙 at both ends of gating pore in Eq. (12). Following the idea of experiment [7], we calculated 𝐼(0, 𝑡) with V rising from -150mV to -140mV at t=10, holding on till t=150, and dropping back to -150mV. The reason to choose the rising range of V from -150mV to -140mV is that essentially none of the arginines move in this hyperpolarized situation. The voltage step only quickly charges and discharges solution EDL in vestibules, and the computational time course of 𝐼(0, 𝑡) is just two spikes when the command potential rises suddenly and drops suddenly later. Subtracting this hyperpolarized 𝐼(0, 𝑡), multiplied by a proportion factor, from its original counterpart will then remove the spikes, and the unspiked 𝐼(0, 𝑡) is shown in Fig. 7(a). Third, in Fig. 7(b), the time courses of Q1 and Q3 show that arginines move from the intracellular vestibule to the extracellular one when depolarized and move back to intracellular vestibule again when repolarized later. As arginines move from one vestibule to another, the concentrations of Na+ and Cl also correspondingly change with time at the vestibules, form counter charges through EDL, and balance arginine charges at vestibules. However, these movement only maintain approximate, not exact, electroneutrality as shown in Fig. 7(b). The violation of electroneutrality is produced by the displacement current, that is not negligible we note. As mentioned above, here we used flux of charges at the middle of hydrophobic plug, 𝐼 (cid:4672)𝐿(cid:3019) + , 𝑡(cid:4673), instead of experimentally measurable 𝐼(0, 𝑡) to represent 'the gating current' in discussions. This (cid:3013) (cid:2870) 'gating current' just leaves out the displacement current 𝐼(cid:3031)(cid:3036)(cid:3046)(cid:3043) (cid:4672)𝐿(cid:3019) + , 𝑡(cid:4673). We use this definition of (cid:3013) (cid:2870) 'gating current' for several reasons: (1) the area under time course of 𝐼 (cid:4672)𝐿(cid:3019) + , 𝑡(cid:4673) gives us the amount (cid:3013) (cid:2870) of arginines moved during depolarization more faithfully than 𝐼(0, 𝑡) as explained above. The fluxes of charge for each arginine shown in Fig. 4 (b) and (d) carry important information about how each arginine is moved by the electric field, that will be illustrated in Fig. 8. All these will not be easy to display and comprehend if we use 𝐼(0, 𝑡) instead. (2) Using 𝐼(0, 𝑡) as a definition of gating current would require a decontamination by removing the leading spikes in it, which is computationally costly by the procedure described above. Especially, it would pose a heavy numerical burden when doing parameter fitting, where numerous repeated computations need to be conducted; (3) The shape of 𝐼 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:2870) , 𝑡(cid:4673) is actually close to the unspiked 𝐼(0, 𝑡). Hence here we used the computationally 13 handy 𝐼 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:2870) , 𝑡(cid:4673) to replace 𝐼(0, 𝑡) as the gating current. 5.4 Comparison with experimental records. Our computations have limited fidelity at short times because of time step limitation in integrating stiff systems. The spike artifacts are one example, described previously. Experimental measurements [16, 33] of the fast transient gating current are fascinating but our calculations must be extended to deal with them. A more general consideration is the subtraction procedure used in experiments to isolate gating current from currents arising from other sources. Channels and their voltage sensors are always accompanied by large amounts of lipid membranes. Currents through channels and in voltage sensors are 'in parallel' with large capacitive currents through lipid membranes. The lipid membrane of cells introduces a large capacitance ( 𝐶(cid:3039)(cid:3036)(cid:3043)(cid:3036)(cid:3031) ≅ 6×10(cid:2879)(cid:2875) farads/cm(cid:2870) ) that has nothing to do with the capacitive (i.e., displacement) currents produced by charge movement in the voltage sensor. Fortunately, this capacitance 𝐶(cid:3039)(cid:3036)(cid:3043)(cid:3036)(cid:3031) is a nearly ideal circuit element. Current through the membrane capacitance is entirely a displacement current accurately described by 𝑖(cid:3030)(cid:3028)(cid:3043) = 𝐶(cid:3039)(cid:3036)(cid:3043)(cid:3036)(cid:3031) (cid:2986)(cid:2906) (cid:2986)(cid:2930) with a single constant 𝐶(cid:3039)(cid:3036)(cid:3043)(cid:3036)(cid:3031). V is the voltage across the lipid capacitor. Note that 𝑖(cid:3030)(cid:3028)(cid:3043) does not include any current carried by the translocation of ions across the lipid. In experimental measurements, this displacement current 𝑖(cid:3030)(cid:3028)(cid:3043) is always present, in large amounts, because voltage sensors (and channels) are always embedded in lipid membranes. Experimental measurements mix the displacement current of the lipid membrane and the displacement current of the voltage sensor. Indeed, the lipid membrane current dominates the measurement of displacement current in native preparations and remains large in systems mutated to have unnaturally large numbers of voltage sensors. A procedure to remove the lipid membrane current is needed if the gating current of the voltage sensor is to be measured. The procedure introduced by Schneider and Chandler [26] has been used ever since in the improved P/4 version developed by Armstrong and Bezanilla [1] reviewed and discussed in [8]. Also, see another approach in [6, 13]. Schneider and Chandler's procedure [26] estimates the so-called linear current 𝑖(cid:3051) = 𝐶(cid:3051) (cid:2986)(cid:2906) (cid:2986)(cid:2930) in conditions in which the voltage sensor and 𝐶(cid:3051) behaves as an ideal circuit element. An ideal capacitor has a capacitance 𝐶(cid:3051) independent of voltage, time, current, or ionic composition. The Schneider procedure then subtracts that linear current 𝑖(cid:3051) from the total current measured in conditions in which the voltage sensor does not behave as an ideal capacitor. The leftover estimates the nonlinear properties of the charge movement in the voltage sensor. That is to say it estimates the charge movement of the voltage sensor that is not proportional to the size of the voltage step used in the measurement. The leftover is called gating current here and in experimental papers. The gating current reported in experiments however is missing a component of the displacement current of the voltage sensor, because of the Schneider and Chandler procedure that estimates 𝑖(cid:3051). The Schneider and Chandler procedure does not measure 𝑖(cid:3030)(cid:3028)(cid:3043) by itself. The linear current 𝑖(cid:3051) estimated by the Schneider and Chandler procedure has more in it than the current through the lipid membrane capacitor 𝑖(cid:3030)(cid:3028)(cid:3043). Rather, it contains the lipid membrane current 𝑖(cid:3030)(cid:3028)(cid:3043) plus current through any structures in the membrane ('in parallel') in which current follows the law 𝑖(cid:3051) = 𝐶(cid:3051) 14 (cid:2986)(cid:2906) (cid:2986)(cid:2930) . Clearly, some of the current produced by movements of the arginines in the voltage sensor will be a linear displacement current, a linear component of gating current. But that component is likely to be included in 𝑖(cid:3051) and so would not be present in the reported gating current. The linear component of gating current is removed by the experimental procedure that is needed to remove the lipid membrane displacement current 𝑖(cid:3030)(cid:3028)(cid:3043). Other systems may contribute to the linear displacement current as well, for example, (1) all sorts of experimental and instrumentation artifacts and (2) displacement current in the conduction channel itself. The conduction channel of field effect transistors produces a large capacitance described by drift diffusion equations quite similar to the PNP equations of the open conduction channel. Our procedure in the computations reported here subtracts a hyperpolarized current with arginines not moving in this situation, thereby removing all the currents carried by arginine movement, linear and nonlinear in voltage. Most systems have substantial motions that are linear in voltage (even if the system is labelled 'nonlinear'). The response of most systems to a voltage step can be written as a sum of terms, each with a different dependence on the size of the applied step of voltage V. The first term in that sum changes linearly if the voltage step is changed. Higher order terms exhibit a nonlinear dependence on V. The linear term is present in most systems, just as it is present in most Taylor expansions of nonlinear functions. The linear component missed in experiments, and removed in these calculations, may have functional and structural significance. The voltage sensor works by sensing voltage, for example, by producing a motion of arginines. That motion-the response of the voltage sensor in this model-includes a linear component. The signal passed to the conduction channel, to control gating, is likely to include the linear component of sensor function, including the linear component of the motion of the arginines. Confusion will result if the linear component is ignored when a model is created that links the voltage sensor to the gating process of the conduction channel. Direct measurements of the movement of arginines (e.g., with optical methods) are likely to include the linear component and so should not agree with experimental measurements of gating current or with the currents reported here. 5.5 Time course of arginine and S4 translocation Time course of Q (amount of arginines moved to extracellular vestibule, equal to Q3 here) and center-of-mass trajectories of individual arginines (𝑧(cid:3036),(cid:3004)(cid:3014), i=1, 2, 3, 4) and S4 segment (𝑍(cid:3020)(cid:2872)) are shown in Fig. 8 with (a) and (b) for the case of large depolarization and (c) and (d) for the case of partial depolarization. In the case of large depolarization, from Fig. 8(b), we can see arginines and S4 are driven towards extracellular vestibule once the electric field is turned on. Their z-positions quickly reach individual steady states with almost all arginines transferred to extracellular vestibule as previously shown in Fig. 5 and therefore Q is close to 4 as shown in Fig. 8(a). Arginines and S4 move back to intracellular vestibule once the voltage drops back to -90mV. From Fig. 8(b), the forward moving order of arginines is R1, R2, R3 and R4, and the backward moving order is on the opposite R4, R3, R2 and R1. This obviously coincides with the structure. Note that S4 is initially farthest to the right but lags behind R1 and R2 during movement in depolarization as shown in Fig. 8(b). This is certainly because S4 is relaxed to an almost unforced situation close to its natural position 𝑍(cid:3020)(cid:2872),(cid:2868). We can further calculate the displacements of each arginine and S4 during this full depolarization, and find Δ𝑧(cid:2869),(cid:3004)(cid:3014) ≈ Δ𝑧(cid:2870),(cid:3004)(cid:3014) ≈ Δ𝑧(cid:2871),(cid:3004)(cid:3014) ≈ 1.93nm, Δ𝑧(cid:2872),(cid:3004)(cid:3014)=1.76nm, Δ𝑍(cid:3020)(cid:2872)=1.51nm. Besides almost the same displacements for R1, R2 and R3, their average moving velocities are also very close to each other. This seems to suggest a 15 synchronized movement among R1, R2 and R3. Also, we can see the movements of arginines contribute significantly to the movement of S4 segment. This can be seen by steady state z-position of S4 derived from Eq. (9), 𝑍(cid:3020)(cid:2872) = (cid:3012) (cid:3012)(cid:3268)(cid:3120)(cid:2878)(cid:2872)(cid:3012) ∑ (cid:3435)𝑧(cid:3036),(cid:3004)(cid:3014) − 𝑧(cid:3036)(cid:3439) + (cid:2872) (cid:3036)(cid:2880)(cid:2869) (cid:3012)(cid:3268)(cid:3120) (cid:3012)(cid:3268)(cid:3120)(cid:2878)(cid:2872)(cid:3012) 𝑍(cid:3020)(cid:2872),(cid:2868) = (cid:2869) (cid:2873) (cid:3427)𝑍(cid:3020)(cid:2872),(cid:2868) + ∑ 𝑧(cid:3036),(cid:3004)(cid:3014) (cid:2872) (cid:3036)(cid:2880)(cid:2869) (cid:3431). Experimental estimates of S4 displacement during full depolarization ranged from 2-20 A(cid:466) [24, 37]. This large deviation accommodates several models for voltage sensing: transporter model, helical screw model and paddle model [37]. Our Δ𝑍(cid:3020)(cid:2872)=1.51nm here is large enough that seems to agree better with experimental estimates requiring large displacements, such as the paddle model. However, our model is more based on helical screw model, which is known to have shorter displacements. A plausible explanation for our over-estimated Δ𝑍(cid:3020)(cid:2872) is that our 1D model has made the known helical path into a linear path. It would be shorter for displacement perpendicular to the plane of the membrane. The diameter of an alpha helix between alpha carbons is approximately 1 nm, therefore if the rotation were 90o, the displacement in 1D would be extended by 0.5×0.5×π=0.78 nm, making the linear translation perpendicular to the membrane only 0.73 nm. This number is very close to the displacement perpendicular to the membrane that is estimated when comparing the open-relaxed state crystal structure of Kv1.2 [9] and the consensus closed structure that has been derived from experimental measurements [40]. For partial depolarization, arginines and S4 are the same driven towards extracellular vestibule once the electric field is turned on and return when voltage drops. However, since the driving force is weaker than in a large depolarization, their z-positions do not reach steady states as they do during a full depolarization before they return to intracellular vestibule once the voltage drops. This behavior can be seen from Fig. 8(c), that Q reaches 1.57 at most which should be 2 instead if equilibrium, which can be reached if time is long enough, as shown in QV curve of Fig. 3(a). Fig. 8(d) shows that S4 initially farthest to the right lags behind R1 during movement and is almost caught up by R2. The maximum displacements of arginines and S4 calculated from Fig. 8(d) are Δ𝑧(cid:2869),(cid:3004)(cid:3014) =1.36nm, Δ𝑧(cid:2870),(cid:3004)(cid:3014) =0.966nm, Δ𝑧(cid:2871),(cid:3004)(cid:3014) =0.459nm, Δ𝑧(cid:2872),(cid:3004)(cid:3014) =0.316nm, and Δ𝑍(cid:2872),(cid:3004)(cid:3014) =0.616nm. 5.6 Family of gating currents for a range of voltages Fig. 9(a) shows the time courses of gating current for a range of voltages V, ranging from -62mV to -8mV. We can see the area under gating current, for both on and off parts, increases with V since more arginines are transferred to extracellular vestibule as V increases. However, this area will saturate with further increasing V since all arginines would then be transferred to the extracellular vestibule. The shapes of this family of gating currents agree well with experiment [7] in both magnitude and time span. We can characterize the time span by fitting the time course (only the decay part) of a gating current by 𝑎𝑒(cid:2879)(cid:3047) (cid:3099)(cid:3117)⁄ + 𝑏𝑒(cid:2879)(cid:3047) (cid:3099)(cid:3118)⁄ , 𝜏(cid:2869) < 𝜏(cid:2870), as generally conducted in experiment [7], where 𝜏(cid:2869) is the fast time constant and 𝜏(cid:2870) is the slow time constant. Usually the movement of arginines is dominated by 𝜏(cid:2870). Here 𝜏(cid:2870) was calculated from simulation and compared with experiment [7] as shown in Fig. 9(b). Since in our computation the time is in arbitrary unit, we have scaled the time to have the maximum 𝜏(cid:2870) to fit with its counterpart in experiment [7]. The trend of 𝜏(cid:2870) versus V in our result agrees well with experiment [7]. For the left branches to the maximum point in Fig. 9(b), simulation result fits very well with experiment. For the right branches to the maximum point, simulation result overestimates 𝜏(cid:2870) compared with experiment. This is consistent with the observation that the amount of transferred 16 charges Q saturates faster in experiment data than in present simulation as V increases as shown in the QV curve of Fig. 3(a). This phenomenon is related to cooperativity of movement among arginines, that will be further discussed below. 5.7 Effect of voltage pulse duration Fig. 10 shows the effect of voltage pulse duration with Fig. 10(a) for the case of partial depolarization and Fig. 10(b) for the case of full depolarization. Magnitude and time span of gating current are influenced by pulse duration in both cases, but the shape will asymptotically approach the same as pulse duration increases. This behavior occurs because driving arginines towards extracellular vestibule by the command potential takes time. If the pulse duration is long enough, the time course of Q will approach its steady state like in Fig. 8(a) for large depolarization. Partial depolarization takes longer time to reach its steady state as demonstrated in Fig. 8(c). Again these shapes of gating currents in Fig. 10 compare favorably with experiment [7]. 6. Conclusions The present 1D mechanical model of the voltage sensor tries to capture the essential structural details that are necessary to reproduce the basic features of experimentally recorded gating currents. After finding appropriate parameters, we find that the general kinetic and steady-state properties are well represented by the simulations. The continuum approach seems to be a good model of voltage sensors, provided that it takes into account all interactions, and satisfies conservation of current. The continuum approach used here describes the mechanical behavior of a single voltage sensor, but experimental measurements are ensemble averages of multiple ion channels (and hence multiple voltage sensors). There is no conflict, however. Experiments average of measurements of large numbers of voltage sensors as does our macroscopic model. Here we simplify the profile of energy barrier in hydrophobic plug since the potential of mean force (PMF) in that region, and its variation with potential and conditions, is unknown. Therefore the next step is to model the details of interactions or the moving arginines with the wall of the hydrophobic plug. There is plenty of detailed information on the amino acid side chains in the plug and how each one of them has important effects in the kinetics and steady-state properties of gating charge movement [25]. The studied side chains reveal steric as well as dielectric interactions with the arginines that the present model does not have. On the other hand, the power of the present mathematical modeling is precisely the implementation of interactions, therefore we believe that when we add the dielectric details of the channel a better prediction of the currents should be attained. Also, here the plug energy barrier Vb is assumed to be independent of time. However, once the first arginine enters the hydrophobic plug by carrying some water with it, this partial wetting of pore will lower Vb, chiefly consisting of solvation energy, and enable the next arginine to enter the plug with less difficulty. This might explain the cooperativity of movement among arginines when they jump through the plug. This remains to be included in the future modeling. Besides, it has been reported that a very strong electric field might affect the hydration equilibrium of the hydrophobic plug and would lower its hydration energy barrier as well [39]. This cooperativity of movement may help explain the quick saturation in the upper right branch of QV curve (and smaller τ2). This also remains to be considered in the future 17 for ordinary differential modeling. Also, further work must address the mechanism of coupling between the voltage sensor movements and the conduction pore. It seems likely that the classical mechanical models will need to be extended to include coupling through the electrical field. It is possible that the voltage sensor modifies the stability of conduction current by triggering sudden transitions from closed to open state, in a controlled process reminiscent of Coulomb blockade. ACKNOWLEDGMENTS This research was sponsored in part by the Ministry of Science and Technology of Taiwan under grant no. MOST-102-2115-M-002-015-MY4 (T.L.H.) and MOST-104-2115-M-035 -002 -MY2 (T.L.H.). Dr. Horng also thanks the support of National Center for Theoretical Sciences Mathematical Division of Taiwan (NCTS/MATH), and Dr. Ren-Shiang Chen for the long-term benefiting discussions. REFERENCES [1] Armstrong, C.M., and F. Bezanilla. 1974. Charge movement associated with the opening and closing of the activation gates of the sodium channels. J. Gen. Physiol. 63:533-552. [2] Ascher, U. M., and L. R. Petzold. 1998. Computer methods equations and differential-algebraic equations. SIAM, Philadelphia. [3] Benseny, A., G. Albareda, A(cid:437) . S. Sanz, J. Mompart, and X. Oriols. 2014. Applied Bohmian mechanics. The European Physical Journal D. 68(10):286. [4] Bezanilla, F. 2008. How membrane proteins sense voltage. Nat. Rev. Mol. Cell Biol. 9:323–332. [5] Bezanilla, F., and C. M. Armstrong. 1975. Properties of the sodium channel gating current. Cold Spring Harb. Symp. Quant. Biol. 40: 297-304. [6] Bezanilla, F., and C. M. Armstrong. 1977. Inactivation of the sodium channel. I. Sodium current experiments. J. Gen. Physiol. 70(5):549-566. [7] Bezanilla, F., E. Perozo, and E. Stefani. 1994. Gating of K+ channels : II. The components of gating currents and a model of channel. Biophy. J. 66:1011-1021. [8] Bezanilla, F., and J. Vergara. 1980. Properties of Excitable membranes. In: Membrane structure and Function. E.E. Bittar, Ed. Vol. II, Ch. 2. J. Wiley & Sons, New York, pp. 53-113. [9] Chen, X., Q. Wang, F. Ni, and J. Ma. 2010. Structure of the full-length Shaker potassium channel Kv1.2 by normal-mode-based X-ray Sci. USA. 107(25):11352-11357. [10] Eisenberg, B. 2016. Conservation of Current and Conservation of Charge. arXiv. https://arxiv.org/abs/1609.09175. [11] Eisenberg, B. 2016. Maxwell Matters. arXiv. https://arxiv.org/pdf/1607.06691. [12] Eisenberg, B., Y. K. Hyon, and C. Liu. 2010. Energy variational analysis EnVarA of ions in water and channels: field theory for primitive models of complex ionic fluids. Journal of Chemical Physics. 133:104104. [13] Fernandez, J. M., F. Bezanilla, and R. E. Taylor. 1982. Distribution and kinetics of membrane dielectric polarization. II. Frequency domain studies of gating currents. J. Gen. Physiol. 79(1):41-67. [14] Feynman, R. P., R. B. Leighton, and M. Sands. 1963. The Feynman lectures on physics, Vol. 2: mainly electromagnetism and matter. Addison-Wesley, New York. [15] Fiedziuszko, S. J., I. C. Hunter, T. Itoh, Y. Kobayashi, T. Nishikawa, S. N. Stitzer, and K. Wakino. 2002. refinement. Proc. Natl. Acad. crystallographic 18 Dielectric materials, devices, and circuits. IEEE Transactions on Microwave Theory and Techniques. 50(3):706-720. [16] Forster, I. C., and N. G. Greeff. 1992. The early phase of sodium channel gating current in the squid giant axon. Characteristics of a fast component of displacement charge movement. Eur. Biophys. J. 21(2): 99-116. [17] Jimenez-Morales, D., J. Liang, and B. Eisenberg. 2012. Ionizable side chains at catalytic active sites of enzymes. European Biophysics Journal. 41(5):449-460. [18] Hodgkin, A. L. 1992. Chance and Design. Cambridge University Press, New York. [19] Hodgkin, A. L., and A. F. Huxley. 1952. A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. 117:500-544. [20] Horng, T. L., T. C. Lin, C. Liu, and B. Eisenberg. 2012. PNP equations with steric effects: a model of ion flow through channels. The Journal of Physical Chemistry B. 116(37):11422-11441. [21] Huxley, A. 2000. Sir Alan Lloyd Hodgkin, O. M., K. B. E. 5 February 1914-20 December 1998. Biograph. Mem. Roy. Soc. 46: 221-241. [22] Huxley, A. F. 1992. Kenneth Stewart Cole. Biograph. Mem. Roy. Soc. 38:99–110. [23] Huxley, A. F. 2002. From overshoot to voltage clamp. Trends in neurosciences. 25(11):553-558. [24] Kim, D. M. and C. M. Nimigean. 2016. Voltage-gated potassium channels: a structural examination of selectivity and gating. Cold Spring Harb. Perspect. Biol. 8:a029231. [25] Lacroix, J. J., H. C. Hyde, F. V. Campos, and F. Bezanilla. 2014. Moving gating charges through the gating pore in a Kv channel voltage sensor. Proc. Natl. Acad. Sci. USA. 111:1950-1959. [26] Schneider, M. F., and W. K. Chandler. 1973. Voltage dependent charge movement in skeletal muscle: a possible step in excitation contraction coupling. Nature. 242:244-246.   [27] Schoppa, N. E., K. McCormack, M. A. Tanouye, and F. J. Sigworth. 1992. The size of gating charge in wild-type and mutant shaker potassium channels. Science. 255:1712–1715.   [28] Schuss, Z. 2009. Theory And Applications Of Stochastic Processes: An Analytical Approach. Springer, New York. [29] Schuss, Z., B. Nadler, and R. S. Eisenberg. 2001. Derivation of Poisson and Nernst-Planck equations in a bath and channel from a molecular model. Phys. Rev. E. 64:036116. [30] Seoh, S.-A., D. Sigg, D. M. Papazian, and F. Bezanilla. 1996. Voltage-sensing residues in the S2 and S4 segments of the Shaker K+ channel. Neuron. 16:1159–1167.   [31] Shampine, L. F., and M. W. Reichelt. 1997. The MATLAB ODE Suite. SIAM Journal on Scientific Computing. 18:1–22. [32] Shampine, L. F., M. W. Reichelt, and J. A. Kierzenka. 1999. Solving Index-1 DAEs in MATLAB and Simulink. SIAM Review. 41:538–552. [33] Sigg, D., F. Bezanilla, and E. Stefani. 2003. Fast gating in the Shaker K+ channel and the energy landscape  of activation. Proc. Natl. Acad. Sci. USA. 100:7611-7615.   [34] Sole, A., X. Oriols, D. Marian, and N. Zanghı. 2016. How does quantum uncertainty emerge from deterministic Bohmian mechanics? Fluct. and Noise Lett. 15(03):1640010. [35] Stefani, E., and F. Bezanilla. 1997. Voltage dependence of the early events in voltage gating. Biophys. J. 72:131.   [36] Stefani, E., D. Sigg, and F. Bezanilla. 2000. Correlation between the early component of gating current  and total gating current in Shaker K channels. Biophys. J. 78:7.   [37] Tombola, F., M. M. Pathak, and E. Y. Isacoff. 2006. How does voltage open an ion channel? Annu. Rev. 19 Cell Dev. Biol. 22:23-52. [38] Trefethen, L. N. 2000. Spectral Methods in MATLAB. SIAM, Philadelphia. [39] Vaitheeswaran, S., J. C. Rasaiah, and G. Hummer. 2004. Electric field and temperature effects on water in the narrow nonpolar pores of carbon nanotubes. J. Chem. Phys. 121(16):7955-7965. [40] Vargas, E., F. Bezanilla, and B. Roux. 2011. In search of a consensus model of the resting state of a voltage-sensing domain. Neuron. 72(5):713-20. [41] Varsos, K., J. Luntz, M. Welsh, and K. Sarabandi. 2011. Electric Field-Shaping Microdevices for Manipulation of Collections of Microscale Objects. Proceedings of the IEEE. 99(12): 2112-2124. [42] Zhu, F., and G. Hummer. 2012. Drying transition in the hydrophobic gate of the GLIC channel blocks ion conduction. Biophys. J. 103:219-227. Figure 1. Geometric configuration of the model including the attachments of arginines to the S4 segment. 20 Figure 2. Following Figure 1, an axisymmetric 3-zone domain shape is designated in r-z coordinate for the current 1D model. Here the diameter of hydrophobic plug is 0.3nm (arginine's diameter); L=0.7nm; LR=1.5nm; radius of vestibule is R=1nm. BC means Boundary Condition Figure 3. (a) QV curve and comparison with [7]. Steady-state distributions for Na, Cl and arginines at (b) V=-90mV, (c) V=-48mV, (d) V=-8mV. 21 Figure 4. (a) Time course of V rising from -90mV to -8mV at t=10, holds on till t=150, and drops back to -90mV, (b) time course of gating current and its components corresponding to (a), (c) time course of V rising from -90mV to -50mV at t=10, holds on till t=150, and drops back to -90mV, (d) time course of gating current and its components corresponding to (c). Figure 5. The four panels on the top row are species concentration distributions at t=0, 13, 148, for the case of large depolarization with V rising from -90mV to -8mV at t=10, holding on till t=150, and dropping back to -90mV. The four panels on the middle row are concurrent electric potential profiles. The four panels on the bottom row are concurrent electric current profiles with components of flux of charge, displacement current and total current. 22 Figure 6. The four panels on the top row are species concentration distributions at t=0, 13, 148, for the case of large depolarization with V rising from -90mV to -50mV at t=10, holding on till t=150, and dropping back to -90mV. The four panels on the middle row are concurrent electric potential profiles. The four panels on the bottom row are concurrent electric current profiles with components of flux of charge, displacement current and total current. Figure 7. (a) Time courses of 𝐼 (cid:4672)𝐿(cid:3019) + (cid:3013) (cid:2870) , 𝑡(cid:4673) , 𝐼(0, 𝑡) and unspiked 𝐼(0, 𝑡) for the case of full depolarization with V rising from -90mV to -8mV at t=10, holding on till t=150, and dropping back to -90mV. The inset plot is a blow-up of ON-current to visualize the difference of 𝐼(0, 𝑡) and unspiked 23 𝐼(0, 𝑡) more clearly. (b) Time courses of 𝑄(cid:2869), 𝑄(cid:2871), ∫ (𝑐(cid:3015)(cid:3028) − 𝑐(cid:3004)(cid:3039)) (cid:3013)(cid:3267) (cid:2868) 𝑑𝑧, and ∫ (cid:2870)(cid:3013)(cid:3267)(cid:2878)(cid:3013) (cid:3013)(cid:3267)(cid:2878)(cid:3013) (𝑐(cid:3015)(cid:3028) − 𝑐(cid:3004)(cid:3039)) 𝑑𝑧 under the same depolarization scenario as (a). Figure 8. (a) and (c) are time courses of amount of arginines moved to extracellular vestibule. (b) and (d) are center-of-mass trajectories of individual arginines and S4. (a) and (b) are the case of large depolarization with V rising from -90mV to -8mV at t=10, holding on till t=150, and dropping back to -90mV. (c) and (d) are the case of partial depolarization with V rising from -90mV to -50mV at t=10, holding on till t=150, and dropping back to -90mV. Figure 9. (a) Time courses of gating current with voltage rising from -90mV to V mV at t=10, holds on till t=150, and drops back to -90mV, where V=-62, -50, … -8 mV. (b) 𝜏(cid:2870) versus V compared with experiment [7]. 24 Figure 10. Effect of voltage pulse duration: (a) V increases from -90mV to -35mV at t=10 and drops back to -90mV at various times, (b) V increases from -90mV to 0mV at t=10 and drops back to -90mV at various times. 25
1705.00025	1	1705	2017-04-28T18:18:14	Frustration induced phases in migrating cell clusters	[ "physics.bio-ph" ]	Collective motion of cells is common in many physiological processes, including tissue development, repair, and tumor formation. Recent experiments have shown that certain malignant cancer cells form clusters in a chemoattractant gradient, which display three different phases of motion: translational, rotational, and random. Intriguingly, all three phases are observed simultaneously, with clusters spontaneously switching between these modes of motion. The origin of this behavior is not understood at present, especially the robust appearance of cluster rotations. Guided by experiments on the motion of two-dimensional clusters in-vitro, we developed an agent based model in which the cells form a cohesive cluster due to attractive and alignment interactions but with potentially different behaviors based on their local environment. We find that when cells at the cluster rim are more motile, all three phases of motion coexist, in excellent agreement with the observations. Using the model we can identify that the transitions between different phases are driven by a competition between an ordered rim and a disordered core accompanied by the creation and annihilation of topological defects in the velocity field. The model makes definite predictions regarding the dependence of the motility phase of the cluster on its size and external chemical gradient, which agree with our experimental data. Our results suggest that heterogeneous behavior of individuals, based on local environment, can lead to novel, experimentally observed phases of collective motion.	physics.bio-ph	physics	Frustration induced phases in migrating cell clusters Katherine Copenhagena, Gema Malet-Engrab,c, Weimiao Yud, Giorgio Scitab,c, Nir Gove, and Ajay Gopinathana,1 aDepartment of Physics, University of California Merced, Merced CA 95343 USA bUniversity of Milan, School of Medicine, Department of Oncology and Hemato-Oncology-DIPO cIFOM Foundation, Institute FIRC of Molecular Oncology, Milan 20139, Italy dInstitute of Molecular and Cell Biology, National University of Singapore eDepartment of Chemical Physics, Weizmann Institute of Science, Rehovot, Israel November 10, 2018 Abstract Collective motion of cells is common in many physiological processes, including tissue development, repair, and tumor formation. Recent experiments have shown that certain malignant cancer cells form clusters in a chemoattractant gradient, which display three different phases of motion: translational, rotational, and random. Intriguingly, all three phases are observed simultaneously, with clusters spon- taneously switching between these modes of motion. The origin of this behavior is not understood at present, especially the robust appearance of cluster rotations. Guided by experiments on the motion of two-dimensional clusters in-vitro, we developed an agent based model in which the cells form a cohesive cluster due to attractive and alignment interactions but with potentially different behaviors based on their local environment. We find that when cells at the cluster rim are more motile, all three phases of motion coexist, in excellent agreement with the observations. Using the model we can identify that the transitions between different phases are driven by a competition between an ordered rim and a disordered core accompanied by the creation and annihilation of topological defects in the velocity field. The model makes definite predictions regarding the dependence of the motility phase of the cluster on its size and external chemical gradient, which agree with our experimental data. Our results suggest that heteroge- neous behavior of individuals, based on local environment, can lead to novel, experimentally observed phases of collective motion. Introduction Collective motion is an emergent phenomenon in large groups of individuals where the motion can arise from purely local interactions. This phenomenon occurs across scales in systems ranging from bacteria to fish [1, 2]. Studies of such systems in the thermodynamic limit of infinite size have revealed a number of interesting features including long-range, scale-free correlations and a discontinuous phase transition [3]. These thermodynamic limit studies have spurred interest in hydrodynamic and mean field theories to describe such phenomena [4]. Finite groups display collective motion that closely model schools of fish or flock dynamics with boundaries. Their kinematics are characterized by unique behaviors including guidance by asymmetric boundaries [5] and the ability to simultaneously exhibit different phases of motion [6]. Collective motion of groups was found to exhibit three distinct phases: running, rotating and random [7, 8]. In the running phase, the individuals are all more or less aligned, leading to a large translational velocity of the cluster center of mass. In the random or disordered phase, individual velocities are uncorrelated and there is very little overall motion of the cluster. In the rotating phase on the other hand, the cluster rotates as a 1 whole around a common center. While the running and random phases have analogs in infinite systems, the mechanisms that can give rise to rotations are less clear. Through simulations, confinement has been shown, to be one mechanism that results in uniform popu- lations of self propelled particles exhibiting rotational modes [9, 10, 11, 12]. Simulations of large groups of unconfined agents can also display rotational phases or milling states where the group rotates in a donut shape under certain conditions [13, 14, 15, 16, 17]. However to achieve these rotational milling states, the agents interact over a range up to tens of times the size of an individual agent, and form a low density 'hole' at the center, where the defect in the velocity orientation field resides. Groups of cells have also been shown to display such rotations though it is unlikely that cells can interact much beyond their nearest neighbors [18, 19]. This rotational motion has been studied both experimentally and using simulations for small groups of cells confined to different geometries [20, 21, 22, 23]. Perhaps even more remarkably, uncon- fined cell clusters have also been observed experimentally to show transient rotational phases [24], and this behavior has been speculated to promote chemotaxis. In this paper, we use an agent-based swarming model, which only allows short-range, nearest-neighbor interactions and unconfined space, similar to previous models found in the literature [25, 26, 27], to address the phenomenon of transient rotations in unconfined clusters. We show that a possible mechanism for driving cluster rotations is density-dependent cell propulsion. This density-dependent propulsion may be caused by contact inhibition of locomotion (CIL), whereby cell protrusions are inhibited by the adhesions between cells [28, 29]. This causes cells at the cluster core, surrounded by other cells, to move slower than those at the edge of the clusters, which have a lower local cell density [30]. This results in an outer rim of cells that move faster than central core ones and display stronger alignment interactions. We also find that decoupling the motion of the rim and central cells suppresses any rotational motion, suggesting that it is the coupling of two systems with different motilities (rim and core) that leads to rotational phases. Specifically, rotations arise in this model when the internal noise is such that the rim cells are in an ordered state with respect to velocity alignment, while the core of the cluster is disordered. The coupling of these two systems (rim around core) results in a frustrated state of the ordered rim being pinned by the disordered core. The whole coupled system is then able to relieve this frustration most effectively by existing in a rotational phase where the more ordered rim is able to move around the disordered core pinning it in place. This model successfully captures the dynamics of transitions between the modes of motion and proportions of time spent in each phase observed experimentally [24]. In the experiments, when the cell clusters are subjected, to a chemical gradient [31, 32] there is an increase in the proportion of running phase, and a decrease in the rotational phase. This trend is also captured by our model when a chemical gradient is introduced. Furthermore, our model predicts an increase in the proportion of rotating phase with the size of the cluster, which we confirmed with experimental data. Taken together, our results suggest a novel form of frustrated interactions between behaviorally different parts of the same cluster that can lead to different collective dynamics -- a finding that may have applications beyond the context of cellular clusters. Model Cell clusters are modeled as groups of particles that move with overdamped dynamics in two-dimensional continuous space (see SI-1). Cells are initially arranged in a circular disk, with velocities pointing in random directions. Cell velocities are determined by their internal self propulsion (with magnitude pi), as well as physical interactions between cells, such as adhesions and collisions [26, 33]. All of these interactions assume that cells communicate with each other by contact within a distance small enough to only include nearest neighbors. The cell diameter is selected from a Gaussian distribution, as uniform cell sizes lead to crystal lattice effects which are unlikely to exist in the experimental cell system (see SI-2). Finally, the velocities of the cells are subject to some uniform and uncorrelated noise ((cid:126)η) due to random traction forces with the substrate and the random nature of the protrusions that cells use for propulsion. Cell positions are then updated according to their individually calculated velocities. (cid:126)xi(t + ∆t) = (cid:126)xi(t) + (cid:126)vi(t) ∗ ∆t (1) 2 To determine the velocity of individual cells, a couple of interactions are taken into account. First, cells propel themselves in a direction (n, eq. 2) determined by the memory of their own previous polarization and an alignment interaction with the mean orientation of neighboring cells, V , with interaction strength α. n = v(t − ∆t) + α ∗ V v(t − ∆t) + α ∗ V , V = (cid:80) (cid:80) n.n. (cid:126)vi n.n. (cid:126)vi (2) Cell velocities for each cell are calculated as arising from the forces described here and illustrated in Fig. 1(a). The self propulsion magnitude is set by p. Additionally cells experience volume exclusion and adhesion with neighboring cells, which are modeled as arising from a Lennard-Jones force ( (cid:126)LJ) and a spring- like interaction ((cid:126)S). The spring force is longer range and acts past their nearest-neighbors, as long as it is not interrupted by other cells (Fig. 1(a)). This adhesion force acts to maintain compact, cohesive, and roughly circular clusters. (cid:126)vi(t) = p ∗ n + ∗ (cid:126)LJ + k ∗ (cid:126)S + (cid:126)η (3) Phase characterization We identify the mode of motion of the cell cluster by measuring the polarization (O) and angular momentum (A) vi (4) N(cid:88) N(cid:88) i=0 i=0 1 N O = A = 1 N vi × ri We can also calculate the average expected velocity of the cells assuming perfect alignment, or perfect rotational motion (see SI-3). We define the running phase to occur when the group polarization (O) is greater than 0.5, and the average cell velocity is over half the expected velocity assuming perfect alignment. Similarly, the rotational phase occurs when the cluster angular momentum (A) is greater than 0.5, and the average cell velocity is over half the expected velocity assuming perfect rotation of the cluster. Finally, the random phase is defined for all other combinations of O,A, and average velocity. These definitions are made to ensure consistency with the definitions in the experimental analysis [24]. In the experiments, malignant B and T type lymphocytes were placed in a chemical gradient of CCL19, where they assemble into clusters and move towards higher CCL19 concentration. Automated analysis of video recordings of the cell clusters was utilized to extract velocity vectors of individual cells, which were then used to compute polarization and angular momentum as functions of time (see [24] for details). Using the criteria described above, we can then label the phase of motion of the cluster for each time point. We then calculate the proportion of time that the cluster spends in each of the three phases throughout a simulation in a manner that allows for experimental comparison. Results Uniform cell clusters In the case where all cells within the cluster behave identically, the cluster remains in a single phase through- out the simulation. The dashed line in Fig. 1 (b, bottom) shows a time trace for a cluster in the running phase where the group polarization remains high and angular momentum remains low throughout. When compared to a time trace of the same quantities measured in experimental cell clusters (Fig. 1 (b, top), the features of the time traces are very different. In the experiments, the group polarization and angular momentum fluctuate from high to low values corresponding to spontaneous transitions of the cluster between 3 various phases of motion. In the simulations, on the other hand, the cluster undergoes a transition from a running phase to a random phase only with increasing noise or decreasing propulsion. Fig. 2a shows the proportion of time spent by clusters in each of the three phases plotted against noise and propulsion. The diagonal line of the transition between running at low noise and random at high noise is the well known noise-driven transition seen in Vicsek swarming models [3, 34, 35]. The running and random phases seen here are similar to those seen in experiments; however, the transition between the running and random phases in phase space is very sharp and there is very little overlap or mixing of the phases. Experimentally, cell clusters are observed to spontaneously switch between running, rotating, and random phases which means that they coexist within this parameter space. Alternatively one could say that cells change their internal parameters, such as the propulsion, p, or noise, η, so that they cross over the transition between the phases. However, it is implausible for the entire cell cluster to change internal parameters in a coordinated way. Additionally, the uniform clusters show a very low level of cell rearrangement or fluidity within the cluster, whereas, in the experiments, cells are observed to move between the rim and the core of the cluster regularly. We therefore conclude that additional features of the real system must be incorporated into the model to recapitulate a rotational phase and transitions between phases within a single set of parameter values, as well as large scale cluster rearrangement. Introducing heterogeneous behavior An aspect of cellular behavior that is missing in this description is the possibility that cells may behave differently in different regions of the cluster, say the periphery or the interior. Rim cells have increased propulsion compared to inner-cluster (core) cells due to reduced CIL, which causes cells adhered to other cells to form fewer protrusions than cells which have more open space around them. We implement this effect by scaling the propulsion with the number of neighbors, increasing for cells with fewer neighbors pi = pcore + ∗ (pcore − prim) ∗ (ni − 6) 3 7 (5) Here, ni is the number of neighbors around cell i. prim and pcore are the propulsion of the rim cells (average of 3.67 neighbors) and core cells (average of 6 neighbors) respectively. A similar inverse relation between local density and propulsion (and therefore, alignment), was explored for a semi-infinite system in [36]. This variation of cell propulsion causes the rotating phase to emerge, and to coexist with the other phases, as seen in experiments. There is now a region in parameter space where there is a peak in the rotational phase at low values of pcore and intermediate noise (Fig. 2b). In this region, there are proportions of all three phases which are close to those seen in experiments. The point highlighted in Fig. 2b is an example of a location in parameter space where the simulations closely match the experiments. The time series for the group polarization and angular momentum at this point in parameter space is shown in Fig. 1 (b, bottom), and agrees well with the experimental time series. Furthermore, when the proportion of time spent in each of the three phases is compared to the experimentally measured values, they match very closely (Fig. 1c). We next investigate, more closely, the mechanism that drives the rotating phase. Cluster rim to core coupling The values of propulsion for individual cells in our simulation with heterogeneity are mostly close to pcore, except for those near the periphery where it rapidly climbs to an average of prim (see SI-4). This prompts us to consider whether the behavior of the system can be understood as arising from the coupling of two different systems -- a ring-like rim with a higher propulsion and a uniform core with lower propulsion. We first examine the rim cell system by confining a ring of cells to a circular shape and assigning them a fixed value of propulsion (= prim) independent of neighbor number. This results in the phase diagram shown in Fig. 3, where the contours are boundaries of regions where the proportion of time spent in the corresponding phase exceeds 30% and 50%. This phase space shares some characteristics with the uniform cluster phase diagram (Fig. 2a), such as the transition from a running to a random phase with increasing 4 noise or decreasing propulsion, and the lack of a rotational phase. Due to the lower number of neighbors in the rim case, however, the slope of the diagonal running-random transition line is smaller compared to the case with a uniform cluster. We next couple the rim cells to the core, resulting in a ring of cells confined to a circle with propulsion prim = 8, positioned around a core of cells with propulsion pcore. The black dashed line in Fig. 3 shows the noise value below which a ring of cells with prim = 8 would be ordered (or in the running phase more than 30% of the time); while above the black solid line we expect a cluster with an average uniform p given by eq. 6 to be disordered (random phase greater than 30%). paverage = prim ∗ Nrim + pcore ∗ Ncore N (6) We notice that there is a triangular region between the dashed and solid black lines at low pcore and intermediate noise where the rim should be in its ordered state while the core would be in the disordered state. This suggests the possibility that when the rim is pinned in place by the random phase of the core cells it could relieve the frustration and maximize order by existing in a rotational phase moving around the core. Indeed, when we couple the ordered rim to the random phase core in this parameter regime, we find a peak in the rotating phase (Fig. 3, solid contours). The fact that the peak in the rotational phase does not exist for the rim or core alone but emerges when the two systems are coupled suggests that the rotational phase is driven by the coupling of the ordered rim to the disordered core. There are some differences between the ring-disk confined system and the full unconfined model shown in Fig. 2b, but these are mainly due to affects arising from the confinement of the rim cells to a circle. Indeed, relaxing the confinement of the rim cells, while still treating the cluster as two different coupled systems of the rim cells with higher propulsion around core cells with lower propulsion recovers the original, fully heterogeneous model phase space (see SI-5). Taken together, these results suggest that it is the coupling between the disordered core and ordered rim that is the mechanism behind the cell cluster rotations seen experimentally. Transitions into and out of rotating While we have shown that the running, rotating, and random phases can coexist within our model with heterogeneous neighbor-dependent propulsion, we have not yet examined the dynamics of the transitions be- tween these phases. To do this, we quantify these transitions by monitoring changes in the overall topological properties of the phases, which are easier to track. In particular, note that in condensed matter systems, including active matter [37, 38], phase transitions may be driven by the interactions and dynamics of topo- logical defects [39]. We first take a coupled rim-core cluster with rim cells confined to a circle and project the rim cell velocities onto the confining circle. We then identify a defect in this effectively one-dimensional velocity field as a point where the velocity projections switch directions. Note that the defects, as we have defined them, exist when there is no defect in the full velocity field of the entire cluster in its running phase and vanish when there in fact is a vortex in the cluster in its rotating phase (+/- 1 defect in the director field). Fig. 4a shows these defects for a cluster in the running phase and the rotating phase. In the running phase, the cluster has two defects of opposite signs at roughly opposite sides of the cluster. The formation and spreading apart of a defect pair coincides with the transition from rotating to running phase while the annihilation of the pair results in the running to rotating phase transition. By measuring the frequency of occurrence of any given number of defect pairs in each cluster phase, we investigate the correlation between the phase and number of defect pairs (Fig. 4b). We see a large peak in the rotating phase for zero defect pairs and a peak in the running phase for one defect pair. The random phase has a much broader peak around two or three defect pairs, suggesting that the random phase could occur when multiple defect pairs spontaneously form. To observe the effective interactions of defects with each other, we calculate the pair distribution function (g(r)) for the spacing between two individual defects when a single pair exists. Fig. 4c shows the pair 5 distribution function calculated over all time throughout a simulation, independent of what phase the cluster is in at any particular point in time. We see that for parameter values when the cluster is predominantly in the running phase, the pair distribution indicates that the two defects will repel and largely exist at maximum separation. When the cluster is mostly in the rotating or random phase (Fig. 4c inset), there is a small peak at zero separation, implying that there may be a small effective attractive force between the defects if they get within one cell diameter of each other (rc ∼ 30). At longer ranges, the interaction is repulsive, though the slope is much smaller for both of these cases than in the running case, so the defects only repel weakly, increasing their chance of annihilating and transitioning out of the running phase. Thus, system-wide parameters have a significant influence on the interactions between topological defects, which in turn controls the dynamics of the defects, the formation or annihilation of which are correlated with cluster phase transitions. Cluster size dependence We next examine the effect of cluster size on the phase diagram. Fig 5a shows regions of the parameter space with a proportion of rotating phase greater than 30% for different cluster sizes. Compared to the predictions of the simulations regarding the proportion of the rotating phase, the predictions regarding the running and random phases have a very large spread across the parameters tested. Thus, we focus on the more significant rotational phase shown here (see SI-6 for other phases). Again, the dashed black line shows the noise value below which the rim alone would be in an ordered phase, while the colored dashed lines show the transition between the running and random phases for a uniform system of average p (eq. 6), for each different system size. Our results indicate that larger clusters have a higher proportion of rotational phase while smaller clusters are less likely to rotate. This is consistent with the idea that the coupling-induced rotational phase only exists in the area of phase space where the rim propulsion would result in an ordered state (running phase greater than 30% of the time; below the dashed black line) while the average p of the whole cluster would lead to a disordered state (random phase greater than 30% of the time; above the colored dashed lines for each size). Larger systems have a higher proportion of core cells and therefore will have a larger proportion of the phase space where the average p results in a disordered cluster (because pcore is always lower than prim) while the rim remains ordered, leading to a larger overlap of the two and a more stable rotational phase. Fig. 5(a inset) shows the comparison between experimental and simulation values, where the blue points are the experimentally measured proportions of rotating phase against cluster size, and the red shaded region shows the size dependence of the simulations with the parameter values marked by black crosses in Fig. 5a. Although these parameter values were chosen based solely on the proportions of all phases exhibited by a cluster of size N = 37, the dependence on system size seen in the simulations is very similar to that of the experiments, further supporting the idea that the rim/core coupling is a likely mechanism for experimental cell cluster rotations. Cluster fluidity Exchanges between the periphery and the interior of the cell clusters were proposed to have an important functional role in exposing 'fresh' cells with unsaturated receptors to the chemical gradient [24]. To examine this feature of our clusters, we look at the fluidity of a cluster as measured by the rate of exchange between core and rim cells. A rim cell is defined as any cell with an exposed edge larger than a single cell diameter. We then average the number of cells which switch between the rim and the core in each time step to get a measurement for the cluster fludity. We measure this exchange rate for the original, fully heterogeneous neighbor-number-dependent propul- sion model (phase space in Fig. 2b). Fig. 5(b inset) shows contours for the fluidity across parameter space. It turns out that the fluidity of the cluster is significantly higher when the cluster shows a majority of running or rotating phase compared to the random phase. This is presumably due to the fact that the large scale rearrangements or rim/core cell exchanges happen when the rim cells move past the slow moving core cells and then mix back into the cluster. This is why the fluidity drops for high pcore values, approaching prim. 6 Fig. 5b shows the dependence of the rim/core exchange on cluster size for the simulations, as well as the experiments (cross-hatched bars). The trend of decreasing rim/core exchange with increasing cluster size is more dramatic in the simulations, but is maintained between both experiments and simulation, further supporting a rim/core coupling as the mechanism for rotational phases in cell clusters. It should be noted that the simulations slightly overestimate the proportion of the running phase in smaller clusters at the expense of the random phase. We speculate that this might be a signature of cellular clusters maintaining a roughly constant effective prim across cluster sizes, perhaps by sensing curvature. In this case, our simulations are essentially overestimating the effective prim value because the rim cells in smaller clusters have fewer neighbors, and p increases linearly with decreasing number of neighbors. This would result in a slight overestimation of the running phase with decreasing cluster size at the expense of the random phase seen in the experiments (SI-7). Similarly, for large cluster sizes, the simulations underestimate the effective prim, leading to a higher proportion of the random phase at the expense of the running phase, as well as an underestimation of the rim/core exchange in simulations due to the fact that about 60% of the exchanges that take place occur during the running phase in both experiments and simulations. Cell clusters in a chemical gradient Cell clusters can chemotax robustly up a chemical gradient [24] and it has been shown that such a collective chemotactic motion can be obtained by cells at the cluster rim having a propulsive force normal to the surface of the cluster with a magnitude that depends on the local concentration of the chemokine [24, 40, 41]. We are interested in how such a gradient would affect the proportion of time the clusters spend in each of the different phases. To implement a chemical gradient into our model, we introduce an additional term into the calculation for the cell propulsion direction, n, by replacing α ∗ V by α ∗ V + (cid:126)g in eq. 2), where p.a.n.(cid:88) (cid:126)gi = gc(cid:48)y (cid:126)fj (7) j and the sum j is over each distinct pair of adjacent neighbors of cell i. (cid:126)fj is a vector pointing in the direction bisecting the angle subtended by the centers of the cells of the neighbor pair at the center of cell i, with a magnitude equal to the arc length between the two neighbors (see SI-1.2). Here, g reflects the strength of the influence on propulsion direction from the chemokine gradient per unit distance, of exposed cell edge arclength, c(cid:48) is the change in chemokine concentration per unit distance and y is the distance (in microns) from the 0 ng/ml concentration point. This results in a gradient force in the direction of the most vacant region around a cell, with a magnitude proportional to the size of the vacancy, causing a large outward force on rim cells and negligible force on core cells, resulting in an overall upward drift due to the unbalanced forces [24]. We introduce such a chemical gradient to a system with running, rotating and random proportions close to those measured experimentally in the absence of a chemical gradient. We find that the gradient leads to cluster motion up the gradient as anticipated, and observed experimentally. We then measure the changes in the proportion of phases as a function of the gradient (shown in Fig. 5c). With increasing gradient, we find an increase in the proportion of the running phase and a decrease in the amount of the rotating phase, similar to what is seen experimentally, (Fig. 5c inset) while the random phase proportion stays essentially unchanged. Discussion Cell clusters exhibit running, rotating, and random phases in experiments. We have identified, using a theoretical model, the possible cause for the coexistence of these phases, in a cluster that has only short range cohesive and alignment interactions. It is based on the tendency of cells at the cluster rim to have increased propulsion due to less contact inhibition, and therefore stronger alignment interactions, compared 7 to cells at the core of the cluster. We have identified a likely mechanism by which this increased propulsion can lead to rotational phases in cell clusters. This effect involves the effective existence of two different systems within the cell cluster -- a high propul- sion, ordered rim system and a low propulsion, disordered core. When these two systems are disconnected, there is no significant rotational phase present in either of them. However, when they are coupled together the rotational phase appears robustly, indicating that it is the coupling of these two systems that leads to the observed rotations. We find that the ordered rim is capable of dragging the disordered core with it, resulting in a solid-body-like rotation of the entire cluster (see SI-8). This behavior, whereby the ordered phase induces large-scale coherence in the adjacent disordered phase, is somewhat reminiscent of the coupling between super-conducting and normal metals (proximity effect) [42]. Our simulations exhibit multiple features that are seen experimentally, including spontaneous transitions between the different phases, the cluster size dependence of the proportions of phases and the fluidity, as well as the response to chemical gradients. Our results indicate that larger clusters show an increased proportion of the rotational phase. An increase in rotational motion with cluster size has also been observed in simulations of flocks and experiments with fish schools [7], suggesting that this mechanism for rotations may extend to systems other than just cell clusters. With increasing chemokine concentration gradient, the cell cluster spends an increased proportion of time in the running phase. An interesting consequence of this is that since the majority of rim/core exchanges take place while the cluster is in the running phase, we would expect the exchange to increase with increasing chemical gradient as well. This is indeed what we see in the simulations (see SI-9) with a 50% increase in exchange over a four-fold increase in gradient. These results are supportive of the conjectured functional benefit of exchanges of rim and core cells in maintaining robust chemotaxis [24]. An increase in rim/core exchange allows for the cells on the rim in high concentration gradients to shuffle back into the center of the cluster in order to replenish their chemical receptors, which become saturated while they are on the exposed rim of the cluster. At the same time, this brings core cells with unsaturated receptors to the rim allowing for more chemokine sensitivity. Thus, the cluster can utilize its collective dynamics to ensure a more robust response to gradients compared to individual cells. The emergence of exchanges between the periphery and interior, especially with increasing directional input at the periphery, might be of importance to flocks and swarms where sharing the inherent advantages/disadvantages of being at the core/rim (like temperature extremes in penguin colonies or the threat of predation in fish schools) is beneficial to the group as a whole. Taken together, our results show that the rotations induced by rim-core coupling hold across a range of system sizes, propulsion strengths, noise values, and even in the presence of directional forcing. They may even extend into three-dimensional rotations [43, 44, 45, 46], suggesting that the coupling between two swarming systems which are in different ordered phases can lead to interesting behaviors not seen in either system alone. Heterogeneous behavior within a single group is a robust mechanism that cells or other types of swarming organisms may use to enhance rotating phases or other phases that would be unlikely or impossible to achieve otherwise. [47] Acknowledgements AG and KC were partially supported by National Science Foundation NSF grant EF-1038697 and NSF grant DMS-1616926, a James S. McDonnell Foundation Award and in part by the NSF-CREST: Center for Cellular and Bio-molecular Machines at UC Merced (NSF-HRD-1547848). NSG gratefully acknowledges funding from the ISF (Grant No. 580/12). Work in GS lab was partially supported by the Associazione Italiana per la Ri cerca sul Cancro (AI RC 10168); the Worldwide Cancer Research (AICR- 14-0335); from the European Research Council (Advanced-ERC-268836) 8 References [1] Iain D Couzin and Jens Krause. Self-Organization and Collective Behavior in Vertebrates. Advances in the study of behavior, 32:1 -- 75, 2003. [2] Tamas Vicsek and Anna Zafeiris. Collective motion. Physics Reports, 517(3-4):71 -- 140, 2012. [3] Gregoire Guillaume and Hugues Chate. Onset of Collective and Cohesive Motion. Physical Review Letters, 92(2), 2004. [4] John Toner, Yuhai Tu, and Sriram Ramaswamy. Hydrodynamics and phases of flocks. Annals of Physics, 318(1):170 -- 244, jul 2005. [5] M. B. Wan, C. J. Olson Reichhardt, Z. Nussinov, and C. Reichhardt. Rectification of swimming bacteria and self-driven particle systems by arrays of asymmetric barriers. Physical Review Letters, 101(1):1 -- 4, 2008. [6] Katherine Copenhagen and Ajay Gopinathan. Active matter clusters at interfaces. Frontiers in mate- rials, 3(March):1 -- 8, 2016. [7] Kolbjorn Tunstrom, Yael Katz, Christos C. Ioannou, Cristian Huepe, Matthew J. Lutz, and Iain D. Couzin. Collective States, Multistability and Transitional Behavior in Schooling Fish. PLoS Computa- tional Biology, 9(2), 2013. [8] Zhao Cheng, Zhiyong Chen, Tamas Vicsek, Duxin Chen, and Hai-Tao Zhang. Pattern phase transitions of self-propelled particles: gases, crystals, liquids, and mills. New Journal of Physics, 18(10):103005, 2016. [9] B. Szabo, G.J. Szollosi, B. Gonci, Zs. Juranyi, D. Selmeczi, and Tamas Vicsek. Phase transition in the collective migration of tissue cells: Experiment and model. Physical Review E, 74(6):1 -- 5, 2006. [10] Fong Yew Leong. Physical explanation of coupled cell-cell rotational behavior and interfacial morphol- ogy: A particle dynamics model. Biophysical Journal, 105(10):2301 -- 2311, 2013. [11] Brian A. Camley, Yunsong Zhang, Yanxiang Zhao, Bo Li, Eshel Ben-Jacob, Herbert Levine, and Wouter- Jan Rappel. Polarity mechanisms such as contact inhibition of locomotion regulate persistent rota- tional motion of mammalian cells on micropatterns. Proceedings of the National Academy of Sciences, 111(41):14770 -- 14775, 2014. [12] Jakob Lober, Falko Ziebert, and Igor S Aranson. Collisions of deformable cells lead to collective migra- tion. Scientific reports, 5:9172, 2015. [13] Herbert Levine, Wouter-Jan Rappel, and Inon Cohen. Self-organization in systems of self-propelled particles. Physical Review E, 63(1):017101, dec 2000. [14] Udo Erdmann, Werner Ebeling, and Alexander S. Mikhailov. Noise-induced transition from translational to rotational motion of swarms. Physical Review E, 71(5):051904, 2005. [15] Iain D Couzin, Jens Krause, Richard James, Graeme D Ruxton, and Nigel R Franks. Collective Memory and Spatial Sorting in Animal Groups. Journal of theoretical Biology, 218:1 -- 11, 2002. [16] Yao-li Chuang, Maria R. D'Orsogna, Daniel Marthaler, Andrea L. Bertozzi, and Lincoln S. Chayes. State transitions and the continuum limit for a 2d interacting, self-propelled particle system. 2008. [17] J A Carrillo, A Klar, and A Roth. Single to Double Mill Small Noise Transitions via Semi-Lagrangian Finite Volume Methods. M:1 -- 25, 2014. 9 [18] Elod Mehes and Tamas Vicsek. Collective motion of cells : from experiments to models. Integrative Biology, 6:831 -- 854, 2014. [19] Kandice Tanner, Hidetoshi Mori, Rana Mroue, Alexandre Bruni-Cardoso, and Mina J Bissell. Coherent angular motion in the establishment of multicellular architecture of glandular tissues. Proceedings of the National Academy of Sciences of the United States of America, 109(6):1973 -- 8, 2012. [20] Clifford Brangwynne, Kevin Kit Parker, Sui Huang, and Donald E. Ingber. Symmetry breaking in cultured mammalian cells. In Vitro Cell. Dev. Biol., 36, 2000. [21] Kevin Doxzen, Sri Ram Krishna Vedula, Man Chun Leong, Hiroaki Hirata, Nir S. Gov, Alexandre J. Kabla, Benoit Ladoux, and Chwee Teck Lim. Guidance of collective cell migration by substrate geom- etry. Integrative Biology, 5(8):1026, 2013. [22] Bo Li and Sean X. Sun. Coherent motions in confluent cell monolayer sheets. Biophysical Journal, 107(7):1532 -- 1541, 2014. [23] Felix J. Segerer, Florian Thuroff, Alicia Piera Alberola, Erwin Frey, and Joachim O. Radler. Emergence and persistence of collective cell migration on small circular micropatterns. Physical Review Letters, 114(22):1 -- 5, 2015. [24] Gema Malet-Engra, Weimiao Yu, Amanda Oldani, Nir Gov, Loic Dupre, and Giorgio Scita. Collective Cell Motility Promotes Chemotactic Prowess and Resistance to Chemorepulsion. Current biology : CB, 2014. [25] Iain D Couzin, Jens Krause, Nigel R Franks, and Simon a Levin. Effective leadership and decision- making in animal groups on the move. Nature, 433(7025):513 -- 6, February 2005. [26] Julio Belmonte, Gilberto Thomas, Leonardo Brunnet, Rita de Almeida, and Hugues Chate. Self- Propelled Particle Model for Cell-Sorting Phenomena. Physical Review Letters, 100(24):248702, jun 2008. [27] Katherine Copenhagen, David A Quint, and Ajay Gopinathan. Self-organized sorting limits behavioral variability in swarms. Scientific Reports, 6:31808, aug 2016. [28] Juliane Zimmermann, Brian A. Camley, Wouter-Jan Rappel, and Herbert Levine. Contact inhibition of locomotion determines cell-cell and cell-substrate forces in tissues. Proceedings of the National Academy of Sciences of the United States of America, 113(10):2660 -- 2665, 2016. [29] Brian A. Camley, Juliane Zimmermann, Herbert Levine, and Wouter Jan Rappel. Emergent Collective Chemotaxis without Single-Cell Gradient Sensing. Physical Review Letters, 116(9):1 -- 6, 2016. [30] Victoria Tarle, Andrea Ravasio, Vincent Hakim, and Nir S. Gov. Modeling the finger instability in an expanding cell monolayer. Integr. Biol., pages 1 -- 14, 2015. [31] Nikhil Mittal, Elena O Budrene, and Michael P Brenner. Motility of Escherichia coli cells in clusters formed by chemotactic aggregation. Proceedings of the National Academy of Sciences of the United States of America, 100(23):13259 -- 13263, 2003. [32] Wouter-Jan Rappel. Cellcell communication during collective migration. Proceedings of the National Academy of Sciences, 113(6):201524893, 2016. [33] Nestor Sepulveda, Laurence Petitjean, Olivier Cochet, Erwan Grasland-Mongrain, Pascal Silberzan, and Vincent Hakim. Collective cell motion in an epithelial sheet can be quantitatively described by a stochastic interacting particle model. PLoS computational biology, 9(3):e1002944, jan 2013. [34] Tamas Vicsek, A Czirok, E Ben-Jacob, I Cohen, and O Shochet. Novel type of phase transition in a system of self-driven particles. Physical Review Letters, 75(6):4 -- 7, 1995. 10 [35] F. Peruani, A. Deutsch, and M. Bar. A mean-field theory for self-propelled particles interacting by velocity alignment mechanisms. European Physical Journal: Special Topics, 157(1):111 -- 122, 2008. [36] Shradha Mishra, Kolbjorn Tunstrom, Iain D. Couzin, and Cristian Huepe. Collective dynamics of self- propelled particles with variable speed. Physical Review E - Statistical, Nonlinear, and Soft Matter Physics, 86(1):1 -- 11, 2012. [37] Christoph a. Weber, Christopher Bock, and Erwin Frey. Defect-mediated phase transitions in active soft matter. Physical Review Letters, 112(16):168301, 2014. [38] Luca Giomi, Mark J. Bowick, Xu Ma, and M. Cristina Marchetti. Defect annihilation and proliferation in active Nematics. Physical Review Letters, 110(22):1 -- 5, 2013. [39] G. Duclos, S. Garcia, H. G. Yevick, and P. Silberzan. Perfect nematic order in confined monolayers of spindle-shaped cells. Soft Matter, 10(14):2346 -- 2353, 2014. [40] Andre Levchenko and Pablo a Iglesias. Models of eukaryotic gradient sensing: application to chemotaxis of amoebae and neutrophils. Biophysical journal, 82(1 Pt 1):50 -- 63, 2002. [41] Brian A. Camley, Juliane Zimmermann, Herbert Levine, and Wouter-Jan Rappel. Emergent collective chemotaxis without single-cell gradient sensing. arXiv, page 20, 2015. [42] Michael Tinkham. Introduction to superconductivity. Dover Publications, Inc., 1996. [43] Pernille Rorth. Fellow travellers: emergent properties of collective cell migration. EMBO reports, 13(11):984 -- 991, 2012. [44] Mark J Miller, Sindy H Wei, Ian Parker, and Michael D Cahalan. Two-photon imaging of lymphocyte motility and antigen response in intact lymph node. Science (New York, N.Y.), 296(5574):1869 -- 1873, 2002. [45] David Bilder and Saori L. Haigo. Expanding the Morphogenetic Repertoire: Perspectives from the Drosophila Egg. Developmental Cell, 22(1):12 -- 23, 2012. [46] Danfeng Cai, Wei Dai, Mohit Prasad, Junjie Luo, Nir S. Gov, and Denise J. Montell. Modeling and analysis of collective cell migration in an in vivo three-dimensional environment. Proceedings of the National Academy of Sciences, 113(15), 2016. [47] Eliseo Ferrante, Ali Emre Turgut, Marco Dorigo, and Cristian Huepe. Elasticity-based mechanism for the collective motion of self-propelled particles with springlike interactions: A model system for natural and artificial swarms. Physical Review Letters, 111(26):1 -- 5, 2013. 11 Figure 1: (a) Schematic for the model. Green direction indicators show the direction of the neighbors of the gray cells, and the green indicator on the gray cell shows the alignment interaction ( V ). The orange arrows show the Lennard-Jones interaction with each neighboring cell and the red arrow is the total LJ interaction ( (cid:126)LJ) on the gray cell. Finally, the blue spring denotes the cell-cell adhesion interaction ((cid:126)S). Note that it only exists between the gray cell and its second nearest neighbors that do not have cells interrupting the path between them. (b) Time series of the group polarization and angular momentum of the cell cluster. The colors along the bottom axis show the phase of the system with time (running -- red, rotating -- blue, and random -- green) for experimental data (top) and simulations (bottom), for a uniform cluster (dashed) and a cluster with behavioral heterogeneity (solid, corresponding to the point marked in Fig. 2b). (c) The proportion of time that the cluster spends in each phase (simulations (plain) and experiments (cross- hatched)), along with a typical illustration of what each phase looks like in the simulations, with velocity vectors as black arrows. The cluster size for simulations is N = 37 cells, while experiments are for an average cluster size of 50. 12 (a)(b)(c)Figure 2: (a) Proportion of time spent by the cluster (N = 37 cells) in each of the three phases plotted against propusion p and noise (cid:126)η for a cluster where all cells behave identically. (b) Phase diagram of the proportion of time spent in each of the three phases for a system with neighbor-number-dependent propulsion where the rim cells (those with 3.67 neighbors) have a propulsion of prim = 8. pcore is the propulsion of core cells (those with 6 neighbors), and η is the magnitude of the noise. The black 'x' shows the point where the time series and phase proportions shown in Fig. 1 are taken. Figure 3: Proportion of time spent by the system in each of the three phases as a function of propulsion p and noise (cid:126)η for a ring of 18 cells confined to a circle with propulsion prim = p. Dashed contour lines indicate regions (shaded) where the proportion of time spent in the corresponding phase exceeds 30% (blue) and 50% (red). Solid contour lines show the same contours but for the rim confined to a circle with prim = 8 coupled with a core of cells with pcore = p, and a full cluster size of N = 37. Note that the rotational phase only has non-zero values for the coupled system. The horizontal dashed line marks the noise value below which the rim alone would be ordered (greater than 30% running phase) with prim = 8, and the diagonal solid line marks the region above which a core with an average propulsion set by eq. 6 (with pcore = p) is disordered (greater than 30% random phase). 13 UniformHeterogeneous(a)(b)Figure 4: (a) Velocities of the rim cells of a 37 cell cluster which are confined to a circular shape projected onto the circle. In the running phase (red), there are two defects of opposite signs in the velocity field, denoted by the orange and blue points. There are no defects in the rotating phase (blue). (b) The proportion of the number of defect pairs for each phase, with a peak at zero defect pairs for the rotating phase (blue), and one defect for the running phase (red). (c) The pair distribution function plotted against the separation between two defects when only one defect pair exists for parameters where the cluster primarily displays a running phase (note that g(r) is calculated over the whole simulation, independent of specific phases at any given point in time). (c - inset) The pair distribution function for points in parameter space dominated by rotating (blue) and random (green) phases. Figure 5: (a) Proportion of time spent in the rotating phase by a system with neighbor-number-dependent propulsion as a function of pcore and noise (cid:126)η for prim = 8. The black horizontal dashed line marks the noise value below which the rim alone would be ordered (corresponding to the dashed transition line in Fig. 3 with prim = 8). The diagonal lines show the noise value above which a uniform system with average propulsion p would be disordered (red: N = 19, blue: N = 37, and green: N = 61). The shaded regions are where the clusters spend at least 30% time in the rotational phase, with the same color scheme (blue: N = 37, and green: N = 61; note that there is no red shaded region). (a - inset) The dependence of the proportion of rotating phase on system size. The shaded red region shows the range of dependence for a spread of parameter values marked with black crosses in the main figure. The experimental measurements are shown as the blue points. (b) The fluidity of the cluster measured as the rate of exchange between the core and rim cells of the cluster, for several systems sizes, for both simulations (plain bars), and experimental data (cross-hatched bars). (b - inset) Contours for the fluidity of the cluster over the pcore-η parameter space. (c) Simulated proportion of each phase (see legend) plotted with increasing chemical gradient (gc(cid:48)r in eq. 7, where r is the cell diameter), along with experimental data in the inset. The concentration gradient of chemokine in the experiments is measured in (ng/ml)/mm and shown on the x-axis for the inset. 14 (c)(a)(b)RunningRotatingRandom(a)(b)(c)RunningRotatingRandom
1702.08257	2	1702	2017-06-09T13:49:25	Gain and loss in open quantum systems	[ "physics.bio-ph", "physics.chem-ph", "quant-ph" ]	Photosynthesis is the basic process used by plants to convert light energy in reaction centers into chemical energy. The high efficiency of this process is not yet understood today. Using the formalism for the description of open quantum systems by means of a non-Hermitian Hamilton operator, we consider initially the interplay of gain (acceptor) and loss (donor). Near singular points it causes fluctuations of the cross section which appear without any excitation of internal degrees of freedom of the system. This process occurs therefore very quickly and with high efficiency. We then consider the excitation of resonance states of the system by means of these fluctuations. This second step of the whole process takes place much slower than the first one, because it involves the excitation of internal degrees of freedom of the system. The two-step process as a whole is highly efficient and the decay is bi-exponential. We provide, if possible, the results of analytical studies, otherwise characteristic numerical results. The similarities of the obtained results to light harvesting in photosynthetic organisms are discussed.	physics.bio-ph	physics	Gain and loss in open quantum systems Hichem Eleuch1∗ and Ingrid Rotter2† 1 Institute for Quantum Science and Engineering, Texas A&M University, College Station, Texas 77843, USA and 2 Max Planck Institute for the Physics of Complex Systems, D-01187 Dresden, Germany (Dated: July 3, 2018) Abstract Photosynthesis is the basic process used by plants to convert light energy in reaction centers into chemical energy. The high efficiency of this process is not yet understood today. Using the formalism for the description of open quantum systems by means of a non-Hermitian Hamilton operator, we consider initially the interplay of gain (acceptor) and loss (donor). Near singular points it causes fluctuations of the cross section which appear without any excitation of internal degrees of freedom of the system. This process occurs therefore very quickly and with high efficiency. We then consider the excitation of resonance states of the system by means of these fluctuations. This second step of the whole process takes place much slower than the first one, because it involves the excitation of internal degrees of freedom of the system. The two-step process as a whole is highly efficient and the decay is bi-exponential. We provide, if possible, the results of analytical studies, otherwise characteristic numerical results. The similarities of the obtained results to light harvesting in photosynthetic organisms are discussed. 7 1 0 2 n u J 9 ] h p - o i b . s c i s y h p [ 2 v 7 5 2 8 0 . 2 0 7 1 : v i X r a ∗ email: [email protected] † email: [email protected], corresponding author 1 I. INTRODUCTION Photosynthetic organisms capture visible light in their light-harvesting complex and transfer the excitation energy to the reaction center which stores the energy from the pho- ton in chemical bonds. This process occurs with nearly perfect efficiency. The primary process occurring in the light-harvesting complex, is the exciton transfer between acceptor and donor, while the transfer of the energy to the reaction center appears as a secondary process. Both processes are nothing but two parts of the total light harvesting. A few years ago, evidence of coherent quantum energy transfer has been found experimen- tally [1, 2]. Recent experimental results [3] demonstrated that photosynthetic bio-complexes exhibit collective quantum coherence during primary exciton transfer processes that occur on the time scale of some hundreds of femtoseconds. Furthermore, the coherence in such a system exhibits a bi-exponential decay consisting of a slow component with a lifetime of hundreds of femtoseconds and a rapid component with a lifetime of tens of femtoseconds [4]. The long-lived components are correlated with intramolecular modes within the reaction center, as shown experimentally [5]. These results induced different theoretical considerations which are related to the role of quantum coherence in the photosynthesis. For example, the equivalence of quantum and classical coherence in electronic energy transfer is considered in [6]. In [7], the fundamen- tal role of noise-assisted transport is investigated. In [8], it is shown that the efficiency is increased by reducing radiative recombination due to quantum coherence. The Hamilto- nian of the system in these (and many other) papers is assumed to be Hermitian although photosynthesis is a process that occurs in an open quantum system. We mention here also the paper [9] on the dynamical theory of primary processes of charge separation in the photosynthetic reaction center. The emphasis in this paper is on the important role of the primary processes, in which light energy is converted into energy being necessary for the living organisms to work. The lifetime of the primarily excited state must be very short. Otherwise there is no chance for the reaction center to catch the energy received from the photosynthetic excitation which will change, instead, to heat and fluorescence (in the framework of Hermitian quantum physics). In the description of an open quantum system by means of a non-Hermitian Hamilton operator, the localized part of the system is embedded into an environment. Mostly, the 2 environment is the extended continuum of scattering wavefunctions, see e.g. the review [10]. Coherence is an important ingredient of this formalism. Meanwhile the non-Hermitian for- malism is applied successfully to the description of different realistic open quantum systems, see the recent review [11]. The paper [12] is one of the oldest references in which the resonance structure of the cross section in the regime of overlapping resonances is considered in the non-Hermitian formalism. In this paper, the resonance structure of the nuclear reaction 15N + p with two open decay channels is traced as a function of the degree of overlapping of the individual resonances by keeping constant the coupling strength between the localized part of the system and the environment of scattering wavefunctions. The distance between the energies of the individual resonance states is varied by hand. As a result, two short-lived states are formed at a critical value of the degree of overlapping. The widths of all the other states are reduced because(cid:80)N n=1 Γn has to be constant according to the constant coupling strength between system and environment. These states are called trapped states. In some following papers, this phenomenon is studied as a function of the coupling strength between system and environment and is called segregation of decay widths, see the recent review [13]. In these papers, the short-living states are called superradiant states which exist together with long-living subradiant states. This formalism is applied also to the problem of energy transfer in photosynthetic complexes [14, 15], see also [16]. In this formal- ism, the enhancement of the energy transfer is related to the existence of the superradiant state. In other papers, the resonance trapping phenomenon is related to singular points which exist in the formalism of non-Hermitian quantum physics, see the review [10] and the recent paper [17]. These singular points are known in mathematics since many years [18], and are called usually exceptional points (EPs). Most interesting new features caused by the EPs in the non-Hermitian physics of open quantum systems are, firstly, the non-rigid phases of the eigenfunctions and, secondly, the possibility of an external mixing (EM) of the states of the localized part of the system via the environment. Non-rigidity of the phases of the eigenfunctions of the Hamiltonian and an EM of the states are possible only in an open quantum system. They are not involved explicitly in any type of Hermitian quantum physics. Furthermore, superradiant and subradiant states do not appear in this formalism. Quite the contrary, phenomena that are related in, e.g., [13] to their existence, are an expression 3 for nothing but the nontrivial properties of the eigenfunctions of a non-Hermitian Hamilton operator, such as non-rigid phases and EM of the wavefunctions. In [19], the dynamics of the system and the efficiency of energy transfer are studied in a non-Hermitian formalism by taking into account noise acting between donor and acceptor, while in [20], the role of protein fluctuation correlations in the energy transfer is investigated and the spin-echo approach is extended to include bio-complexes for which the interaction with dynamical noise is strong. It is the aim of the present paper to provide the general formalism of non-Hermitian physics of open quantum systems [10, 17] by inclusion of gain which simulates the acceptor, as well as of loss which stands for the donor [21]. When additionally the coupling of the system to a sink is taken into account, this formalism can be applied to the description of light-harvesting of photosynthetic complexes. We underline that this formalism describes the process of photosynthesis as a whole, i.e. as a uniform process. While the first part occurs instantly, the second part of the process may last longer. The formalism is generic. In the future, it has to be applied to concrete systems with realistic parameters. In Sect. II, we sketch the formalism for the study of an open quantum system with gain and loss which is basic for the description of photosynthesis. In Sect. III, we include additionally a sink into the formalism simulated by coupling to a second environment. In both sections we provide analytical as well as numerical results. We discuss and summarize the results in Sect. IV and draw some conclusions in Sect. V. Before providing the formalism for the description of open quantum systems, it is neces- sary to clarify the meaning of some terms. We will use definitions similar to those used in nuclear physics. * In nuclear physics, channel denotes the coupling of a certain state of the nucleus A to its decay products after emission of particle a and leaving the residual nucleus (A − a) in a special state. The term channel is equivalent to embedding of a localized state of the system into an environment. The localized state in nuclear physics is the state of the nucleus A, while the environment is the the continuum of scattering wavefunctions of the particle a. * In difference to the definition of energy and width of a nuclear state in nuclear physics, we use the definition εk = ek+ i operator H. The widths of decaying states have thus a negative sign [22]. 2γk for the complex eigenvalues of the non-Hermitian Hamilton * The term internal mixing of the wavefunctions denotes the direct interaction between 4 two orthogonal eigenfunctions of a Hermitian Hamilton operator, (cid:104)ΦiV Φj(cid:54)=i(cid:105). In our cal- culations, it is supposed to be included in the energies ek and widths γk of the states that define the non-Hermitian Hamilton matrix, see e.g. Eq. (1). An external mixing of two eigenstates of a non-Hermitian Hamilton operator occurs via the environment and is thus a second-order process. It is defined only in an open system. * The singularity related to the coalescence [23] of two eigenvalues of a non-Hermitian Hamilton operator H is called, in recent literature, mostly exceptional point. In older papers, the equivalent expressions branch point in the complex plane or double pole of the S-matrix are mostly used. II. OPEN QUANTUM SYSTEMS WITH GAIN AND LOSS A. Hamiltonian We sketch the features characteristic of an open quantum system with gain and loss [21] by considering a localized 2-level system that is embedded in a common continuum of scattering wavefunctions. One of these two states gains particles from the environment by interacting with it, while the other one loses particles to the continuum by decay. For the description of the open quantum system, we use the non-Hermitian Hamilton operator [17] H(2,1) =  ε(1) 1 ≡ e(1) 1 + i ω(1) 2γ(1) 1 ω(1) 2 ≡ e2 + i ε(1) 2γ(1) 2 are the two complex eigenvalues of the basic non-Hermitian operator coupled to i Here, ε(1) the environment 1 (called also channel 1) [22]. The e(1) the γ(1) states (γ(1) i i are their widths. One of these eigenvalues describes loss characteristic of decaying 2 < 0) while the other one describes gain from the environment (γ(1) 1 > 0) [22]. are the energies of the states i and The ω(1) stand for the coupling matrix elements of the two states via the common envi- of H(2,1) of the states of the localized part of the system [22]. ronment 1. They are complex [10]. The complex eigenvalues E (1) give the energies E(1) and widths Γ(1) i ≡ E(1) i + 1 2Γ(1) i i i We will consider also the non-Hermitian Hamilton operator  ε(1) H(2,1) 0 = 1 ≡ e(1) 1 + i 0 2γ(1) 1 5 0 2 ≡ e(1) ε(1) 2 + i 2γ(1) 2  .  (1) (2) which describes the localized part of the open system without coupling of its states via the continuum (ω(1) = 0). The phases of the eigenfunctions Φ0 Hermitian quantum physics) when γ(1) 1 = −γ(1) 2 . i of H(2,1) 0 are rigid (like in B. Eigenvalues The eigenvalues of H(2,1) are, generally, complex and may be expressed as (cid:113) E (1) 1,2 ≡ E(1) 1,2 + i 2 Γ(1) 1,2 = ε(1) 1 + ε(1) 2 2 ± Z ; Z ≡ 1 2 1 − ε(1) (ε(1) 2 )2 + 4(ω(1))2 (3) i and Γ(1) where E(1) i here Γ(1) stand for the energy and width, respectively, of the eigenstate i. Also i ≥ 0 for gaining states [22]. The two states may repel each other in accordance with Re(Z), or they may undergo width bifurcation in i ≤ 0 for decaying states and Γ(1) accordance with Im(Z). When Z = 0 the two states cross each other at a point that is called usually exceptional point (EP) [18]. The EP is a singular point (branch point) in the complex plane where the S-matrix has a double pole [10]. According to its definition [18], the EP is meaningful in an open quantum system which is embedded in one common environment c = 1. Correspondingly, we denote e.g. the eigenvalues by E (1) . i We consider now the behavior of the eigenvalues when the parametrical detuning of the two eigenstates of H(2,1) is varied, bringing them towards coalescence [23]. According to (3), the condition for coalescence reads (cid:115) Z = 1 2 1 − e(1) (e(1) 2 )2 − 1 4 1 − γ(1) (γ(1) 2 )2 + i(e(1) 1 − e(1) 2 )(γ(1) 1 − γ(1) 2 ) + 4(ω(1))2 = 0 . We consider two cases that can be solved analytically. (i) When e(1) 1 = e(1) 2 , and ω(1) is real, it follows from (4) the condition 2 = ± 4 ω(1) 2 )2 = 4 (ω(1))2 → γ(1) 1 − γ(1) (γ(1) 1 − γ(1) 1 4 for the coalescence of the two eigenvalues, i.e. for an EP. It follows furthermore 1 − γ(1) (γ(1) 1 − γ(1) (γ(1) 2 )2 < 16 (ω(1))2 → Z ∈ (cid:60) 2 )2 > 16 (ω(1))2 → Z ∈ (cid:61) . (4) (5) (6) (7) Eq. (6) describes the behavior of the eigenvalues away from the EP, where the eigen- values E (1) differ from the original ones through only a contribution to the energy. k 6 The widths in contrast, remain unchanged, and this situation therefore corresponds to that of level repulsion. Eq. (7), in contrast, is relevant at the other side of the EP. Here, the resonance states undergo width bifurcation according to Im(Z) (cid:54)= 0. The bifurcation starts in the neighborhood of the EP. Physically, the bifurcation implies that different time scales may appear in the system, while the states are nearby in energy. (ii) When e(1) 1 = −e(1) 2 (cid:54)= 0, and ω(1) is imaginary, then the condition (2e(1))2 = 4 (ω(1))2 → 2e(1) = ± 2 ω(1) , (8) together with γ(1) Here 2e(1) ≡ e(1) 1 = γ(1) 1 − e(1) 2 . Instead of (6) and (7) we have 2 , follows for the coalescence of the two eigenvalues from (4). (2e)2 > 4 (ω(1))2 → Z ∈ (cid:60) (2e)2 < 4(ω(1))2 → Z ∈ (cid:61) . (9) (10) Thus, the EP causes width bifurcation also in this case. However, this case is realized only when γ(1) 2 = 0 at the EP, i.e. when gain and loss vanish at the EP. 1 = γ(1) C. Eigenfunctions The eigenfunctions of a non-Hermitian Hamilton operator are biorthogonal (for details see [10, 17]) HΦi(cid:105) = EiΦi(cid:105) (cid:104)ΨiH = Ei(cid:104)Ψi . In the case of the symmetric 2 × 2 Hamiltonian (1), it is Ψi = Φ∗ i and the eigenfunctions should be normalized according to (cid:104)Φ∗ iΦj(cid:105) = δij (11) (12) (13) in order to smoothly describe the transition from a closed system with discrete states to a weakly open one with narrow resonance states. As a consequence of (13), the values of the standard expressions are changed, (cid:104)ΦiΦi(cid:105) = Re ((cid:104)ΦiΦi(cid:105)) ; Ai ≡ (cid:104)ΦiΦi(cid:105) ≥ 1 (14) 7 (cid:104)ΦiΦj(cid:54)=i(cid:105) = i Im ((cid:104)ΦiΦj(cid:54)=i(cid:105)) = −(cid:104)Φj(cid:54)=iΦi(cid:105) Bj i ≡ (cid:104)ΦiΦj(cid:54)=i ≥ 0 . (15) Furthermore, the phase rigidity which is a quantitative measure for the biorthogonality of the eigenfunctions, rk ≡ (cid:104)Φ∗ kΦk(cid:105) (cid:104)ΦkΦk(cid:105) = A−1 k (16) is smaller than 1. Far from an EP, rk ≈ 1 while it approaches the value rk = 0 when an EP is approached. , The Hamiltonian (2) describes the system around the EP without any mixing of its states via the environment, since ω(1) = 0 corresponds to vanishing EM of the eigenstates. In order to determine quantitatively the strength of the EM, we present the eigenfunctions Φi of H(2,1) in the set of eigenfunctions {Φ0 , Φi = under the condition that the bij are normalized by(cid:80) (17) j(bij)2 = 1. The coefficients bij2 differ from the (bij)2. They contain the information on the strength of EM via the environment which is determined by the value of ω(1). bij = (cid:104)Φ0∗ j Φi(cid:105) , i} of H(2,1) (cid:88) bij Φ0 j ; 0 For illustration, we consider the EM of the wavefunctions Φ1 and Φ2 around an EP in the two cases discussed in Sect. II B. (i) e(1) (ii) e(1) 1 − γ(1) 2 , and ω(1) ∈ (cid:60): according to (5), the strength of the EM via the environment 2 of the widths (which both have different 1 = e(1) is determined by the differences γ(1) sign). It depends thus on the fluctuations of the γ(1) 1 = −e(1) differences e(1) however realized only when gain and loss vanish at the EP (i.e. γ(1) 2 , and ω(1) ∈ (cid:61): according to (8) the strength of the EM is related to the . This case is 2 of the energies, i.e. to the fluctuations of the e(1) 1 − e(1) 2 = 0 at the 1 = γ(1) . i i EP). At the EPs, the two corresponding eigenfunctions are not orthogonal. Instead 1 → ± i Φcr Φcr 2 ; 2 → ∓ i Φcr Φcr 1 (18) according to analytical and numerical results [24–28]. We underline once more that an EP is, according to its definition, related to the common environment in which the system is embedded. In other words, it is well defined under the condition that the system is embedded in only one continuum. 8 D. Schrodinger equation with source term The Schrodinger equation ( H(2,1) − E (1) i Φ(1) i (cid:105) = 0 may be rewritten into a Schrodinger equation with source term [10, 17], ( H(2,1) 0 − E (1) i ) Φ(1) i (cid:105) = −  0 ω(1) ω(1) 0 Φ(1) i (cid:105) . (19) In this representation, the coupling ω(1) of the states i and j (cid:54)= i of the localized system via the common environment of scattering wavefunctions (EM) is contained solely in the source term. The source term vanishes, when e(1) γ(1) 1 = γ(1) 1 = e(1) 2 according to (5), what is fulfilled when γ(1) 2 around the EP under the condition i=1,2 = 0 . Far from EPs, the coupling of the localized system to the environment influences the spectroscopic properties of the system, in general, only marginally [10, 17]. The influence is however non-vanishing also in this case, see e.g. the experimental results [29]. In the neighborhood of EPs, however, the coupling between system and environment and therewith the source term play an important role for the dynamics of the open quantum system. The reason is, according to mathematical studies, that the source term causes nonlinear effects in the Schrodinger equation (19) around an EP. For details see [10, 17]. E. Resonance structure of the S-matrix Let us consider the resonance part of the S matrix from which the resonance structure of the cross section can be calculated, σ(E) ∝ 1 − S(E)2 . (20) A unitary representation of the resonance part of the S matrix in the case of two resonance states coupled to a common continuum of scattering wavefunctions reads [30] S = (E − E1 − i (E − E1 + i 2Γ1) (E − E2 − i 2Γ1) (E − E2 + i 2Γ2) 2Γ2) . (21) Here, the influence of the EPs onto the cross section is contained in the eigenvalues Ei = Ei + i/2 Γi. The expression (21) allows therefore to receive reliable results also when the phase rigidity is reduced, rk < 1. 9 Let us assume real ω(1) and γ(1) 2 . First we consider the case corresponding to the condition (6), i.e. for large coupling strength ω(1) of the system to the environment of continuous scattering wavefunctions. In this case Γ1 = −Γ2 = 0 and, according to (6), 1 = −γ(1) S = (E − E1) (E − E2) (E − E1) (E − E2) = 1 . In the other case, (7), it is E1 = E2; Γ1 = −Γ2 (cid:54)= 0; and 2Γ1) (E − E1 + i 2Γ1) (E − E1 − i (E − E1 − i (E − E1 + i S = 2Γ1) 2Γ1) (22) (23) = 1 . In both cases, S = 1, i.e. σ(E) = 0 according to (20). This result corresponds to the well-known fact that EPs cannot be identified in the resonance structure of the S-matrix and therefore also not in the resonance structure of the cross section. Most important is however the result that no resonances will be excited due to γ(1) 1 = −γ(1) 2 . 1 = −γ(1) The result S = 1 is violated when the conditions ω(1) ∈ (cid:60) and γ(1) are not 2 exactly fulfilled. This may happen, e.g., under the influence of external random (stochastic) processes that cause fluctuations of the γ(1) In such a case, S < 1; and the energy (or . i information) will be transferred with an efficiency of nearly 100 % (because no resonances can be excited under this condition in the localized part of the system). Results for this case can be obtained only numerically. F. Numerical results: one-channel case 1. Merge of states with gain and loss Let us first consider the results that are obtained by using a parametric dependence of the energies e(1) i and widths γ(1) i of the states of the localized part of the system which is analog to that used in [17, 31] for a decaying system. The results are shown in Fig. 1. In this figure, the existence of an EP at the parameter value a = acr = 2/3 can clearly be seen. Here, the two states are exchanged (Figs. 1.a,b), the phase rigidity ri approaches the value 0 (Figs. 1.c, d) and the EM of the states via the continuum increases limitless (Fig. 1.e). The contour plot of the cross section (Fig. 1.f) shows the wavefunctions of the two states: while the eigenvalues of H(2,1) cannot be seen according to (22) and (23), the eigenfunctions show some fluctuating behavior around the positions of the eigenstates 10 FIG. 1: Eigenvalues E (1) i ≡ E(1) i + i 2 Γ(1) i (a,b) and eigenfunctions Φ(1) i (c,d,e) of the Hamiltonian H(2,1) as a function of a and contour plot (f) of the cross section; ω(1) = 0.5025. Parameters: e(1) 1 = 1 − a/2; e(1) 2 = a; γ(1) 1 /2 = 0.55; γ(1) 2 /2 = −0.5 (dashed lines in a, b). according to the finite (nonvanishing) range of their influence (see Figs. 1.c,d,e). These fluctuations of the eigenfunctions can be seen in the contour plot. Although they follow the positions of the eigenvalues, their nature is completely different from that of the eigenvalue trajectories. The eigenfunction trajectories in Fig. 1.f show the exchange of the two states at the EP. That means: the state with positive width turns into a state with negative width and vice versa. This underlines once more that the two trajectories shown in Fig. 1.f have really nothing in common with the eigenvalue trajectories of resonance states the widths of which are always negative (or zero at most). We underline once more that the results shown in Fig. 1 are formally similar to those obtained and discussed in [17, 31] for a decaying system. In the latter case, both states which are exchanged at the EP, are of the same type: they are resonance states with negative widths. It is interesting to see from the numerical results (Fig. 1), that the non-Hermitian 11 FIG. 2: The same as Fig. 1 but ω(1) = 0.45 + 0.05i and e(1) 1 = 0.5; e(1) 2 = 0.5 + 0.1a; γ(1) 1 /2 = 0.5; γ(1) 2 /2 = −0.4. Hamilton operator H can be used, indeed, for the description of these two different types of open quantum systems as suggested in Sect. II [21]. Additionally we show some numerical results for the case that, respectively, the distance differ in energy of the two states is smaller (Fig. 2) and the EM as well as the widths γ(1) more from one another (Fig. 3) than in Fig. 1. The eigenvalue and eigenfunction figures i (Figs. 1.a-e, 2.a-e, 3.a-e) are similar to one another and show clearly the signatures of an EP at a certain critical value of the parameter a = acr. The contour plots (Fig. 1.f, 2.f, 3.f) differ however from one another. When the states are nearer to one another in energy, the two states with negative and positive width merge (Fig. 2.f). Under the influence of stronger EM (stronger coupling strength ω(1) between system and environment) as well as of a larger difference between the , the extension of the region with non-vanishing cross section is enlarged in two values γ(1) relation to the energy (Fig. 3.f). In any case, the cross section vanishes around a = acr. i 12 FIG. 3: The same as Fig. 1 but ω(1) = 0.9 + 0.1i and e(1) −1.3. 1 = 0.5; e(1) 2 = 0.5 + 0.1a; γ(1) 1 /2 = 0.5; γ(1) 2 /2 = Resonance states are not excited. 2. Level repulsion of states with gain and loss More characteristic for an open quantum system with gain and loss than those in Sect. 2 , e(1) 1 = −γ(1) II F 1 are the analytical results given in Sect. section is zero when γ(1) 1 = e(1) which causes differences between the original spectroscopic values ε(1) eigenvalues E (1) 2Γ(1) i 1 = −Γ(1) least one of the conditions Γ(1) II E. According to these results, the cross 2 , and ω(1) ∈ (cid:60). Under the influence of an EP and the of H(2,1), a non-vanishing cross section is expected when at 2 , together with ω(1) ∈ (cid:60), is not fulfilled. In Fig. 4 we show the corresponding numerical results obtained for two neighboring i ≡ E(1) 2 , E(1) 1 = E(1) i ≡ e(1) i + i i + i 2γ(1) i states, e(1) widths γ(1) 2 , and ω(1) almost real. We fix the energies e(1) and vary parametrically the , see the dashed lines in Figs. 4.a,b. The results show an EP at a = −1.8 and i 1 ≈ e(1) i 13 FIG. 4: The same as Fig 1 but ω(1) = 0.9 + 0.1i and e(1) 1 = 0.5; e(1) 2 = 0.3; γ(1) 1 /2 = 0.5a; γ(1) 2 /2 = −0.5a. the hint to another EP at a = 1.8. At the EP, the phase rigidity approaches the value zero, ri → 0 (Fig. 4.c), and the EM of the states is extremely large, bij → ∞ (Fig. 4.e). Of special interest is the parameter range between the two EPs. Here, ri → 1 at a ≈ 0 (Figs. 4.c,d). At this parameter value, the level repulsion is maximum; and the two eigenfunctions of the non-Hermitian Hamiltonian H(2,1) are (almost) orthogonal. An analogous result is known from calculations for decaying systems, i.e. for systems with excitation of resonance states [10, 17]. In these calculations, the energies are varied parametrically and ω is almost imaginary (Fig. 1 in [17]). Therefore, the two eigenfunctions of the non-Hermitian Hamilton operator are (almost) orthogonal (ri → 1) at maximum width bifurcation (instead of at maximum level repulsion in Fig. 4). In any case, the two eigenstates of the non-Hermitian operator turn irreversibly into two states with rigid phases in spite of the non-Hermiticity of the Hamiltonian. This unexpected result occurs due to the evolution of the system to the point of, respectively, maximum level repulsion and maximum width bifurcation, which is driven exclusively by the nonlinear 14 source term of the Schrodinger equation, see Sect. II D. The eigenfunctions of these two states are mixed. In order to receive a better understanding of this result, we mention here another un- expected result of non-Hermitian quantum physics, namely the fact that a non-Hermitian Hamilton operator may have real eigenvalues [21]. This fact is very well known in literature for a long time, for references see the review [10]. The corresponding states are called usually bound states in the continuum. Most interesting for a physical system is the contour plot of the cross section (Fig. 4.f). According to the analytical results discussed in Sect. II E, the cross section vanishes far from the parameter range that is influenced by an EP. It does however not vanish completely in the parameter range between the two EPs. Around the EP at a = −1.8, the conditions for vanishing cross section are quite well fulfilled, while this is not the case around the other EP at a = 1.8. In approaching the two EPs by increasing and decreasing, respectively, the value of a the cross section vanishes around a = −1.8, and is non-vanishing around a = 1.8. The cross section vanishes also around the point of maximum level repulsion at which the two eigenfunctions of H(2,1) are orthogonal (and not biorthogonal). Additionally, we performed calculations (not shown in the paper) with the reduced value ω(1) red = 0.5(0.9 + 0.1i) in order to determine the role of EM in the cross section picture. The obtained results are similar to those shown in Figs. 4.a-f. The parameter range influenced by the two EPs is however smaller when ω(1) is reduced: it ranges from a ≈ −1 to a ≈ 1 when ω(1) = ω(1) red. Accordingly, the region of the non-vanishing cross section in the contour plot shrinks in relation to a, and also in relation to the energy. In calculations with vanishing external mixing (ω(1) = 0), the cross section vanishes everywhere. 15 III. OPEN QUANTUM SYSTEM WITH GAIN AND LOSS COUPLED TO TWO ENVIRONMENTS A. Hamiltonian for coupling to two environments Let us consider the 4 × 4 non-Hermitian matrix H(2,2) =  ε(1) 1 ω(1) 0 0 ω(1) ε(1) 2 0 0 0 0 ε(2) 1 ω(2)  . 0 0 ω(2) ε(2) 2 (24) i ≡ e(1) 2γ(1) i i + i Here, ε(1) are the complex eigenvalues of the basic non-Hermitian operator H(2,2) relative to channel c = 1 and c = 2, respectively [22]. The two channels (environments) are independent of and orthogonal to one another what is and ε(2) i + i 2γ(2) i ≡ e(2) i expressed by the zeros in the matrix (24). One of the channels may be related to gain and loss [21] (acceptor and donor) considered in the previous section II, while the other channel may simulate a sink. In this case, the two widths γ(1) the first channel. Relative to the second channel however, both γ(2) 2 have different sign relative to are negative according 1 and γ(1) i to a usual decay process of a resonance state. The ω(1) and ω(2) stand for the coupling matrix elements between the two states i = 1, 2 of the localized part of the open quantum system and the environment c = 1 and 2, respectively. In the case considered above, these two environments are completely different from one another and should never be related to one another. In more detail: an EM of the considered states may be caused only by ω(1) or by ω(2), and never by both values at the same time [32]. This is guaranteed when ω(2) = 0 what is fulfilled when there is only one state in the second channel. When there are more states, then ω(2) should be much smaller than ω(1) (here we point to the general result that the values ω are related to the widths γi of the states [10]). i and γ(2) The values γ(1) are independent of one another and express the different will be usually very large, time scales characteristic of the two channels. While the γ(1) are generally much smaller. Accordingly, the two-step process as a whole will the γ(2) show, altogether, a bi-exponential decay: first the decay occurs due to the exponential quick i i i process; somewhere at its tail it will however switch over into the exponential decay of the 16 slow process. The Hamiltonian which describes vanishing coupling of the states of the localized part of the open quantum system to both environments is (25)  H(2,2) 0 = ε(1) 1 0 0 0 0 ε(1) 2 0 0 0 0 ε(2) 1 0 0 0 0 ε(2) 2  by analogy to (2). It does not contain any EM via an environment. B. Eigenvalues and eigenfunctions of H(2,2) The eigenvalues E (c) i ≡ E(c) i + i 2Γ(c) i and eigenfunctions Φ(c) i of (24) are characterized by two numbers: the number i of the state (i = 1, 2) of the localized part of the system and the number c of the channel (c = 1, 2), called environment, in which the system is embedded. Generally, E(1) . Also the wave functions Φ(1) (cid:54)= E(2) (cid:54)= Γ(2) and Φ(2) and Γ(1) i differ from one i i i i i another due to the EM of the eigenstates via the environment c = 1 and c = 2, respectively. From a mathematical point of view, the system has therefore four states. An EP influences the dynamics of the open quantum system also in the two-channel case. Without an EP in the considered parameter range in relation to both channels, we have E(1) . Accordingly, one has to consider effectively and Φ(1) , Γ(1) i ≈ Φ(2) i ≈ Γ(2) i i i ≈ E(2) only two states i Under the influence of an EP relative to c = 1 (or/and relative to c = 2), the eigenvalues and eigenfunctions will be, however, different from one another, E(1) Φ(1) (cid:54)= Φ(2) in the corresponding parameter range. We have to consider therefore effectively i (cid:54)= E(2) i , Γ(1) i (cid:54)= Γ(2) i and i i four states in this case. According to [18] an EP is defined when the system is embedded in one common envi- ronment. Under this condition, it causes nonlinear processes in a physical system, which is the crucial factor for the dynamical properties of an open quantum system [10]. This is valid not only for systems all states of which decay (corresponding to some loss), but also for systems with loss and gain, as shown in [10]. Due to the nonlinear processes occurring near to an EP, it is difficult to receive analytical solutions for the eigenvalues and eigenfunctions of (24). We will provide the results of some 17 numerical simulations, above all with e(1) 2 , almost real ω(1) and almost imaginary ω(2), which is the most interesting and general case for a system with gain and loss that is coupled 2 and ω(1) ∈ (cid:60), Eq. (5), and the to a sink (see the analytical results obtained with e(1) 1 = e(1) 1 ≈ e(1) corresponding results for decaying systems in [17]). C. Schrodinger equation with source term and coupling to two environments Using (24), we can write down the Schrodinger equation with source term for the two- channel case in analogy to (19) for the one-channel case. The corresponding equation reads (H(2,2) 0 − E (c) i ) Φ(c) i (cid:105) = − (26)  0 ω(1) 0 0 ω(1) 0 0 0 0 0 0 0 ω(2) 0 ω(2) 0 Φ(c) i (cid:105) . The source term depends on the coupling of the system to both channels, i.e. on ω(1) and on ω(2). We will consider the general case with two channels (two environments) in which ω(2) (cid:28) ω(1). We repeat here that, according to their definition [18], EPs occur only in the one-channel case, i.e. only in the submatrices related either to channel 1 or to channel 2. They are not defined in the 4 × 4 matrix (24). However, each EP in one of the two submatrices in (24) influences the dynamics of the open two-channel system. This will be shown in the following section by means of numerical results for the case that there is an EP in the first channel which simulates acceptor and donor (gain and loss), while the second channel being of standard type with resonance states, may or may not have an EP. D. Numerical results: two-channel case We performed some calculations for the two-channel case by starting from the calculations for the one-channel case in, respectively, Fig. 2 and 4 and by adding a second channel that describes decaying states (corresponding to loss). There are, of course, very many possibilities for choosing the number of states as well as the parameters for the second channel. One possibility is to keep the parameters constant by varying the parameter a of the first channel. Another possibility is to relate them directly to the parameter a, or to 18 Eigenvalues E (1,2) ≡ E(1,2) i FIG. 5: (c,d,e) of the Hamiltonian H(2,2) and contour plot (f) of the cross section as function of a with two merging states in the first channel, 2 /2 = −0.05 (dashed lines in a, b). The 1 /2 = −0.05; γ(2) ω(2) = 0.005 + 0.045 i. e(2) 1 = 0.5; e(2) 2 = 0.3; γ(2) (a,b) and eigenfunctions Φ(1,2) i + i 2 Γ(1,2) i i parameters of the first channel are the same as those in Fig. 2. introduce another independent parameter b. The choice should correspond to the physical situation considered. The aim of our calculations is to illustrate the influence of a second channel onto the eigenvalues and eigenfunctions of H(2,2) and onto the contour plot of the cross section. We exemplify this by choosing parameter dependent values ε(1) in the first channel and parameter independent values ε(2) in the second channel. In the following, we show a few characteristic i i results. 19 FIG. 6: Left: Eigenvalues E (1,2) H(2,2) as function of a with one state in the second channel, ω(2) = 0. e(2) (c,d,e) of the Hamiltonian 1 /2 = −0.05 (dashed lines in a, b). The parameters of the first channel are the same as those in Fig. 4. Right: contour (a,b) and eigenfunctions Φ(1,2) 1 = 0.5; γ(2) ≡ E(1,2) 2 Γ(1,2) i + i i i i plots of the cross section with the same parameters of the first channel as in Fig. 4; ω(2) = 0; and (f) 1 = 0.5; γ(2) e(2) 1 /2 = −0.05; (g) e(2) 1 = 0.5; γ(2) 1 /2 = −0.5; (h) e(2) 1 = −0.5; γ(2) 1 /2 = −0.5. 1. Merge of states with gain and loss; second channel with two states We start these calculations with two channels by choosing two merged states with gain and loss according to Fig. 2. The second channel contains two resonance states i = 1, 2 with , see Fig. 5. The eigenvalue and eigenfunction pictures negative widths and γ(2) Fig. 5.a,b and c,d,e, respectively, show the eigenvalues and eigenfunctions of the first channel (cid:28) γ(1) i i 20 ≡ E(1,2) i i FIG. 7: Eigenvalues E (1,2) (c,d,e) of the Hamiltonian H(2,2) and contour plot (f) of the cross section as function of a with two states in the second channel, 2 /2 = −0.06 (dashed lines in a, b). The (a,b) and eigenfunctions Φ(1,2) 1 /2 = −0.06; γ(2) e(2) 1 = 0.4; e(2) 2 = 0.5; γ(2) ω(2) = 0.01 + 0.09 i. + i 2 Γ(1,2) i i parameters of the first channel are the same as those in Fig. 4. (Fig. 2.a-e) as well as the eigenvalues and eigenfunctions of the second channel. The last ones are constant as function of the parameter a what follows from the assumption of their parameter independence. The contour plots of the cross section (Figs. 2.f and 5.f) are different from one another. Common to both of them is that the cross section does not vanish in a finite range of the energy around E ≈ 0 for all a. Under the influence of the two states in the second channel which are exactly in this energy and parameter range, the cross section is somewhat reduced. It does however not vanish. These (and similar) simulations show clearly the following result. The fluctuations of the cross section which are caused by the merging of two states with gain and loss in the first channel, excite resonance states in the second channel. This happens although the nature of 21 both channels is completely different. In the first channel, internal degrees of freedom of the system are not excited, while the appearance of the resonance states in the second channel occurs via excitation of internal degrees of freedom of the system. 2. Level repulsion of states with gain and loss; second channel with one state In these calculations we start from the results shown in Fig. 4 for the first channel. The second channel contains only one state. The eigenvalue and eigenfunction pictures Fig. 6.a-e contain the eigenvalues and eigenfunction trajectories of Fig. 4.a-e as well as those of the second channel, e.g. the energy trajectories at Ei = 0.5 and width trajectories at Γi/2 = −0.05. The phase rigidity approaches the value 1 at a = 0 in both cases. The corresponding contour plot of the cross section is shown in Fig. 6.f. It is related to the contour plot of the first channel (Fig. 4.f); and shows additionally the parameter independent state of the second channel in the whole parameter range. We performed further calculations with parameters similar to those used in Fig. 6.a-e and show two of the corresponding contour plots in Figs. 6.g,h. Both contour plots are obtained with the comparably large width γ(2) different from one another. The difference between e(2) 1 /2 = −0.5; the two energies e(2) 1 = 0.5 and e(2) 1 are however 1 = −0.5 can clearly seen in the corresponding contour plots Figs. 6.g and h. The results of these simulations with level repulsion in the first channel and one state in the second channel show the same characteristic features as those discussed above for merging states. The fluctuations observed in the first channel are able to excite resonance states in the second channel. 3. Level repulsion of states with gain and loss; second channel with two states The situation with two states in the second channel is richer than that with only one state because the two states can mix via the common continuum. We show results for one special case in Fig. 7. As in Figs. 5 and 6, the eigenvalue figures 7.a,b contain the eigenvalue trajectories of both channels. The eigenfunction trajectories refer to the existence of an EP in the second channel: the phase rigidity of states related to the second channel is independent of the parameter a, as expected. It is however smaller than 1. Also the mixing bij of the 22 two wavefunctions via the common continuum in the second channel shows small deviations from the expectations. The relation of these results for the eigenfunctions of H(2,2) to an EP in the second channel is discussed in detail in appendix A. The contour plot Fig. 7.f shows the same characteristic features as Fig. 6.f. Both, its relation to the contour plot of the first channel (Fig. 4.f) as well to the parameter independent state of the second channel can clearly be seen. Thus, the fluctuations observed in the first channel excite resonance states in the second channel also in this case. Summarizing the numerical results shown in the three figures 5 to 7 we state the following. Gain and loss in the first channel and excitation of resonance states in the second channel are a uniform process. This process can be described as a whole in the formalism for the description of open quantum systems which is used in the present paper. IV. DISCUSSION AND SUMMARY OF THE RESULTS In our paper, we considered gain and loss in an open quantum system [21]. Of special interest is the interplay of these two opposed processes in the neighborhood of singular points where it causes fluctuations of the cross section. These fluctuations are observable and can excite resonance states in the system. The time scale of the fluctuations and that of the excitation of resonance states are very different. The fluctuations of the cross section occur quickly, without any excitation of internal degrees of freedom of the system. The excitation of resonance states is however much slower, and internal degrees of freedom of the system are involved. The results of our calculations meet therefore the condition that the lifetime of the pri- mary process in the photosynthetic reaction center has to be very short. Otherwise the energy received from the photosynthetic excitation, will change into heat and fluorescence [9]. This is a statement of Hermitian quantum physics, since the change of energy into heat and fluorescence is impossible in non-Hermitian quantum physics according to the re- sults of our calculations. Here the photosynthesis does not excite any eigenstate of the non-Hermitian Hamiltonian H. Instead, the primay process occurs due to fluctuations of the eigenfunctions of H around EPs. The mechanism of photosynthetic excitation in non- Hermitian quantum physics is therefore, as a matter of principle, different from that in Hermitian quantum physics. 23 We sketched first the formalism by means of which both the interplay between gain and loss in an open quantum system [21] and the excitation of resonance states can be described as a uniform process. In any case, the widths γi of the states are (generally) different from zero. In the first case we have two states with different sign of the widths (corresponding to gain and loss), while the states in the second case are standard decaying (resonance) states with negative sign [22]. Generally, the cross section related to the two states with gain and loss vanishes. Devi- ations from this rule appear around the position of an eigenstate and in the neighborhood of singular (exceptional) points, where they may cause non-vanishing fluctuations of the cross section. These fluctuations have nothing in common with resonances. Rather, they are merely deviations from the vanishing value of the cross section and are not related to the excitation of any internal degrees of freedom of the system. They occur therefore with an efficency of nearly 100 % at a very short time scale. The fluctuations caused by the interplay of the states with gain and loss, are observable and may excite resonance states of the system after a comparably long time (which corre- sponds to the widths Γi of these states). The whole process of gain and loss together with the excitation of resonance states is therefore characterized by two very different time scales: the quick process which creates the fluctuations, and the slow process which is related to the excitation of resonance states. The decay occurs thus bi-exponential. Initially, it is determined by the quick process of the interplay between gain and loss. Somewhere at its tail however, it will switch over to the slow process with excitation of internal degrees of freedom of the system. The states with gain and loss as well as the resonance states excited by the fluctuations, can each interact via a common environment into which the corresponding states are em- bedded. The two environments (channels) will never mix. This request is guaranteed in our formalism due to the very different time scales of the two processes. In any case, this so-called external mixing of the states occurs additionally to the direct so-called internal mixing of the states which is supposed in our calculations to be involved in the complex energies εi ≡ ei + iγi/2 of the states. According to the numerical results of our paper, the fluctuations are very robust. They appear in a relatively large finite parameter range around the positions of the eigenstates and around EPs. 24 Finally, we mention a few interesting results which are characteristic of open quantum systems including those considered in the present paper. – The states of an open quantum system may interact via a common environment into which the system is embedded. This mixing is called usually external interaction. – The states of an open quantum system may have positive or negative widths. The states with positive width [22] gain excitons (or information) from the environment while those with negative width lose excitons (or information) due to their coupling to the environment. As function of a parameter, gain may pass into loss and vice versa [10]. – The phases of the eigenfunctions of a non-Hermitian operator are, generally, not rigid, see Figs. 1 to 7 and [10, 17]. – Irreversible processes determine the evolution of an open quantum system up to the occurrence of orthogonal eigenstates at maximum width or level repulsion, see Figs. 4, 6, 7 and [10, 17]. We mention further that results similar to those discussed above for systems with two states, appear also in calculations for systems with more than two states, see [17, 33, 34]. This holds true also for systems with gain and loss. V. CONCLUSIONS In our paper, we provided some results for a two-step process which we obtained in the framework of the non-Hermitian formalism [10, 17] for the description of open quantum systems. The first step is the interplay between gain and loss of information (excitons) from an environment, while the second step is the excitation of a resonance state. The two steps are treated as two parts of the whole process. The total process might simulate photosynthesis: the first step is the capture of light in the light-harvesting complex while the second step is the transfer of the excitation energy to the reaction center which stores the energy from the photon in chemical bonds. That means: gain simulates the acceptor for light, and loss stands for the donor which excites a resonance state and simulates the coupling to the sink. Altogether, the energy of the light 25 is transferred to the reaction center of the light-harvesting complex. The obtained results are very robust, and fluctuations play an important role. The results show some characteristic features which correspond, indeed, to those discussed in the literature for the photosynthesis. Most interesting are the following results of our calculations: 1. the efficency of energy transfer is nearly 100 %; 2. the energy transfer takes place at a very short time scale; 3. the storage of the energy in the reaction center occurs at a much longer time scale; 4. according to points 2 and 3, the decay is bi-exponential. In future studies, the theoretical results have to be confirmed by application of the for- malism to the description of concrete systems in close cooperation between theory and experiment. Acknowledgment We are indebted to J.P. Bird for valuable discussions. Appendix A: Exceptional point in the second channel At an EP the two eigenfunctions of a non-Hermitian Hamilton operator H are exchanged according to (18). In more detail: tracing the eigenfunctions of H as function of a certain parameter b, the two eigenfunctions jump according to (18) at the critical parameter value b = bcr which defines the position of the EP. Some years ago, it has been shown [24, 30] that the influence of the EP is not restricted to the jump occurring at bcr. It appears rather in a finite parameter range of b in which the wavefunctions of the two states are mixed according to i = βkΦk ± iβlΦl Φch with k (cid:54)= l. The two wavefunctions Φch their components) everywhere but at b = bcr. Using the representation i vary smoothly (i.e. without any jump of the sign of (A1) i = Φch Φch i eiθi 26 (A2) θi depends on the parameter b. After removing a common phase factor, it follows θ(1) and ±3π/4, respectively, in approaching b = bcr; and θ(2) critical region around bcr. In between the values θ(1) i → π/4 i → 0 or π when b is far from the , the angle θi varies smoothly. i and θ(2) i Corresponding to the dependence of θi on the parameter b, also the phase rigidity (16) de- pends on this parameter in a certain finite parameter range. The phase rigidity is determined by the ratio ri ≡ Re(βk + βl)2 βk + βl2 (A3) which approaches 0 at the EP (due to βk + βl2 → ∞) and 1 far from the EP (because here the wavefunctions are almost real). The intermediate values of the phase rigidity ri are determined by the expression (A3) calculated with the actual values βk and βl. The results of these calculation give values for ri that are between the two limiting values 0 and 1. Fig. 7 shows numerical results for the eigenfunctions of H which are obtained for cal- culations with two channels and two states in the second channel. We see the values ri for the states of the first channel (see Figs. 2 and 4) as well as those for the two states of the second channel. They are independent of one another. The ri related to the second channel are constant in the whole parameter range a shown in the figures (which is defined for the first channel). They may be smaller than 1, see Fig. 7. The results obtained for the phase rigidity ri of the two states in the second channel, may be considered as a proof of the finite parameter range in which an EP influences the properties of the system. The value ri is nothing but an expression for the distance of the system from an EP in the second channel: the larger ri, the more distant is the EP, while ri → 0 indicates that the EP is approached. Thus, the value ri in the eigenfunction pictures of the two-channel system with two states in the second channel additionally to those of the one channel system (Figs. 2 and 4), allows us to determine the position of the EP in the second channel. [1] G. S. Engel, T. R. Calhoun, E. L. Read, T.-K. Ahn, T. Mancal, Y.-C. Cheng, R. E. Blanken- ship, and G. R. Fleming, Nature 446, 782 (2007) 27 [2] H. Lee, Y.-C. Cheng, and G. R. Fleming, Science 316, 1462 (2007) R.J. Sension, Nature 446, 740 (2007) [3] M. Mohseni, Y. Omar, G.S. Engel, and M.B. Plenio (eds.), Quantum Effects in Biology, Cambridge University Press, Cambridge, UK, 2014 [4] H. Dong and G. R. Fleming, J. Phys. Chem. B 118, 8956 (2014) [5] E. Romero, R. Augulis, V.I. Novoderezhkin, M. Ferretti, J. Thieme, D. Zigmantas, and R. van Grondelle, Nature Physics 10, 676 (2014) S.F. Huelga and M.B. Plenio, Nature Physics 10, 621 (2014) [6] J. S. Briggs and A. Eisfeld, Phys. Rev. E 83, 051911 (2011) [7] F. Caruso, A.W. Chin, A. Datta, S.F. Huelga, and M.B. Plenio, J. Phys. Chem. C 131, 105106 (2009) [8] M.O. Scully, Phys. Rev. Lett. 104, 207701 (2010) E. A. Sete, A. Svidzinsky, H. Eleuch, R. D. Nevels and M. O. Scully Journ. Mod. Opt. 57, 1311 (2010) A.A. Svidzinsky, K.E. Dorfman, and M.O. Scully, Phys. Rev. A 84, 053818 (2011) [9] V.D. Lakhno, Journ. Biological Physics 31, 145 (2005) [10] I. Rotter, J. Phys. A 42, 153001 (2009) [11] I. Rotter and J.P. Bird, Rep. Prog. Phys. 78, 114001 (2015) [12] P. Kleinwachter and I. Rotter, Phys. Rev. C 32, 1742 (1985) [13] N. Auerbach and V. Zelevinsky, Rep. Prog. Phys. 74, 106301 (2011) [14] G.L. Celardo, F. Borgonovi, M. Merkli, V.I. Tsifrinovich, and G.P. Berman, J. Phys. Chem. C 116, 22105 (2012) [15] D. Ferrari, G.L. Celardo, G.P. Berman, R.T. Sayre, and F. Borgonovi, J. Phys. Chem. C 118, 20 (2014) G.L. Celardo, G.G. Giusteri, and F. Borgonovi, Phys. Rev. B 90, 075113 (2014) G.G.Giusteri, G.L. Celardo and F. Borgonovi, Phys. Rev. E 93, 032136 (2016) [16] G.P. Berman, A.I. Nesterov, G.V. Lopez, and R.T. Sayre, J. Phys. Chem. C 119, 22289 (2015) [17] H. Eleuch and I. Rotter, Phys. Rev. A 95, 022117 (2017) [18] T. Kato, Perturbation Theory for Linear Operators, Springer, Berlin 1966 [19] A.I. Nesterov, G.P. Berman, and A.R. Bishop, Fortschr. Phys. 61, 95 (2013) A.I. Nesterov, G.P. Berman, J.M.S. Martinez and R.T. Sayre, J. Math. Chem. 51, 2514 (2013) 28 [20] A.I. Nesterov and G.P. Berman, Phys. Rev. E 91, 042702 (2015) A.I. Nesterov and G.P. Berman, Phys. Rev. E 91, 052702 (2015) [21] We underline that we consider open quantum systems with gain and loss. This should not be confused with the consideration of exactly balanced gain and loss in PT-symmetric systems which are neither open nor closed, but nonisolated according to the definition in, e.g., C.M. Bender, Journal of Physics: Conference Series 631, 012002 (2015). [22] In contrast to the definition that is used in, for example, nuclear physics, we define the complex energies before and after diagonalization of H by εk = ek+ i 2 Γk, respectively, with γk ≤ 0 and Γk ≤ 0 for decaying states. This definition is useful when discussing systems with gain (positive widths) and loss (negative widths). 2 γk and Ek = Ek+ i [23] The coalescence of two eigenvalues of a non-Hermitian operator should not be confused with the degeneration of two eigenstates of a Hermitian operator. The eigenfunctions of two degen- erate states are different and orthogonal while those of two coalescing states are biorthogonal and differ only by a phase, see Eq. (18). [24] I. Rotter, Phys. Rev. E 64, 036213 (2001) [25] A.I. Magunov, I. Rotter and S.I. Strakhova, J. Phys. B 34, 29 (2001) [26] U. Gunther, I. Rotter and B.F. Samsonov, J. Phys. A 40, 8815 (2007) [27] B. Wahlstrand, I.I. Yakimenko, and K.F. Berggren, Phys. Rev. E 89, 062910 (2014) [28] F. Tellander and K.F. Berggren, Phys. Rev. A 95, 042115 (2017) [29] J.B. Gros, U. Kuhl, O. Legrand, F. Mortessagne, E. Richalot, and D. Savin, Phys. Rev. Lett. 113, 224101 (2014) [30] I. Rotter, Phys. Rev. E 68, 016211 (2003) [31] H. Eleuch and I. Rotter, Eur. Phys. J. D 69, 229 (2015) [32] Numerical calculations with γ(2) i ≈ γ(1) i have shown that, in such a case, the states of the second channel mix also via the first channel. The obtained results are abstruse from the point of view of physics, although they are mathematically correct. [33] H. Eleuch and I. Rotter, Phys. Rev. A 93, 042116 (2016) [34] H. Eleuch and I. Rotter, Eur. Phys. J. D 69, 230 (2015) 29
1903.03325	1	1903	2019-03-08T08:47:24	Changing cell mechanics -- a precondition for malignant transformation of oral squamous carcinoma cells	[ "physics.bio-ph", "q-bio.TO" ]	Oral squamous cell carcinomas (OSCC) are the 6th most common cancer and the diagnosis is often belated for a curative treatment. The reliable and early differentiation between healthy and diseased cells is the main aim of this study in order to improve the quality of the treatment and to understand tumour pathogenesis. Here, the optical stretcher is used to analyse mechanical properties of cells and their potential to serve as a marker for malignancy. Stretching experiments revealed for the first time that cells of primary OSCCs were deformed by 2.9 % rendering them softer than cells of healthy mucosa which were deformed only by 1.9 %. Furthermore, the relaxation behaviour of the cells revealed that these malignant cells exhibit a faster contraction than their benign counterparts. This suggests that deformability as well as relaxation behaviour can be used as distinct parameters to evaluate emerging differences between these benign and malignant cells. Since many studies in cancer research are performed with cancer cell lines rather than primary cells, we have compared the deformability and relaxation of both types, showing that long time culturing leads to softening of cells. The higher degree of deformability and relaxation behaviour can enable cancer cells to traverse tissue emphasizing that changes in cell architecture may be a potential precondition for malignant transformation. Respecting the fact that even short culture times have an essential effect on the significance of the results, the use of primary cells for further research is recommended. The distinction between malignant and benign cells would enable an early confirmation of cancer diagnoses by testing cell samples of suspect oral lesions.	physics.bio-ph	physics	Changing cell mechanics - a precondition for malignant transformation of oral squamous carcinoma cells Felix Meinhövel1,2, Roland Stange4, Jörg Schnauss4,5, Michael Sauer6, Josef A. Käs4 and Torsten W. Remmerbach1,2,3 1 Section of Clinical & Experimental Oral Medicine, University Hospital, Leipzig University, Germany 2 Department of Oral, Maxillofacial and Facial Plastic Surgery, University Hospital, Leipzig University, Germany 3 Griffith Institute of Health, Griffith University, Queensland, Australia 4 Faculty of Physics and Earth Sciences, Peter Debye Institute, Leipzig University, Germany 5 Fraunhofer Institute for Cell Therapy and Immunology (IZI), DNA Nanodevices Group, Leipzig, Germany 6 Department of Oral Maxillofacial and Facial Plastic Surgery, SRH Zentralklinikum Suhl GmbH, Suhl, Germany Corresponding author: Prof. Dr. Josef A. Käs, Faculty of Physics and Earth Sciences, Peter Debye Institute, Leipzig University, Germany, Linnéstrasse 5, D-04103 Leipzig, Germany; Email: [email protected]; Phone: +493419732470; Fax: +493419732479 Keywords oral squamous carcinoma, keratinocyte, cell mechanics, optical stretcher, malignant transformation 1 Abstract Oral squamous cell carcinomas (OSCC) are the 6th most common cancer and the diagnosis is often belated for a curative treatment. The reliable and early differentiation between healthy and diseased cells is the main aim of this study in order to improve the quality of the treatment and to understand tumour pathogenesis. Here, the optical stretcher is used to analyse mechanical properties of cells and their potential to serve as a marker for malignancy. Stretching experiments revealed for the first time that cells of primary OSCCs were deformed by 2.9 % rendering them softer than cells of healthy mucosa which were deformed only by 1.9 %. Furthermore, the relaxation behaviour of the cells revealed that these malignant cells exhibit a faster contraction than their benign counterparts. This suggests that deformability as well as relaxation behaviour can be used as distinct parameters to evaluate emerging differences between these benign and malignant cells. Since many studies in cancer research are performed with cancer cell lines rather than primary cells, we have compared the deformability and relaxation of both types, showing that long time culturing leads to softening of cells. The higher degree of deformability and relaxation behaviour can enable cancer cells to traverse tissue emphasizing that changes in cell architecture may be a potential precondition for malignant transformation. Respecting the fact that even short culture times have an essential effect on the significance of the results, the use of primary cells for further research is recommended. The distinction between malignant and benign cells would enable an early confirmation of cancer diagnoses by testing cell samples of suspect oral lesions. Key words oral squamous carcinoma, keratinocyte, cell mechanics, optical stretcher, malignant transformation Abbreviations OSCC, oral squamous cell carcinoma; OPMD, oral potentially malignant disorders 2 Introduction Oral squamous cell carcinomas (OSCCs) are the most common cancers of the head and neck. Usually, the OSCC develops on the basis of oral potentially malignant disorders (OPMD) [1 -- 5]. In 50% - 60% of the cases the confirmation of diagnosis is belated, because the neoplasm has not been recognized or has been mistaken as harmless [6, 7]. Early diagnosis and treatment of this entity enables recovery. Consequently, an early and efficient clarification of doubtful lesions is requested to reduce the unacceptably high morbidity and mortality of oral squamous cell carcinomas [8]. Various in vitro studies based on cell lines of breast- and cervix carcinomas have shown that the viscoelastic properties of native and healthy cells compared to malignant cells differ significantly [9 -- 11]. The filaments of the cytoskeleton (actin microfilaments, intermediate filaments and microtubules) form a structure that determines the shape and the mechanical properties of the cell [12 -- 20]. In cancer cells, however, the neoplastic transformation is accompanied by a reduction of the cytoskeletal polymers, in particular actin-associated proteins that lead to a reduction of the structural integrity of the cell [21 -- 24]. Results of a pilot study of our group indicate that this phenomenon seems to occur in well-established cell lines of the oral cavity. The work of Remmerbach et al. yielded first insights into biomechanical changes correlated to tumour growth, but at that time the optical stretcher technology did not allow to measure statistically relevant cell numbers [25]. The continuation of the studies revealed that the deformability of the cells is significantly influenced by culturing conditions and cell treatment before measurement [26]. In addition, it became increasingly clear that the cell deformability is temperature-dependent [27 -- 29]. Consequently, we have used a well-defined culture protocol and an accurate temperature control during the measurements in order to acquire reproducible and statistically meaningful data. This study considers the heterogeneity of the samples especially when originating from primary tissue. Tissue of the oral cavity constitutes a special case compared to most other tissues previously used for biomechanical measurements. It contains a large variety of cell types and a fraction of not fully differentiated cells since the tissue is self-renewing frequently. The measurements with the optical stretcher were performed in a way that cells were not pre-selected according to their morphology to mirror the heterogeneous cell populations also in the mechanical data. 3 An additional experiment investigated the comparability of cell lines derived from OSCCs to the primary samples. The cell lines are widely used in cancer research as a model system for cancer. With respect to the mechanical aspects there is a large doubt that those properties are preserved in cells that have undergone many passaging cycles on non-physiological substrates. 4 Results To check the influence of cultivation on the primary cells, their stiffness was measured over a number of passage cycles. We found that cells with increasing passage number became increasingly softer (Figure 1A; significance tested with a t-test). After a few passages the cells stopped growing entirely. This effect could be determined for the primary samples of OSCC as well as for healthy mucosa. Since the growth duration of the samples was inconsistent, only the first six passages are shown (Figure 1A). Furthermore, the proliferation rate of the cells was studied by measuring their doubling time. Figure 1B shows that the proliferation rate increases with increasing passage number, before the growth was stalled globally since cells lost the ability to attach to the surface (and grow) possibly due to the repetitive treatment with trypsin. This phenomenon appeared almost consistently for all samples. The expectation that cancer cells proliferate faster than healthy oral mucosa cells could not be confirmed in this study [35 -- 37]. Doubling times of healthy oral mucosa cells are comparable or even shorter than the doubling time of oral squamous cell carcinoma. The period from initial cultivating to first passage has been excluded from analysis since the cells from different pieces of tissue require a varying time to adhere -- a phenomenon that is not understood yet. A defined amount of single cells was seeded after the first passage in order to provide comparable conditions. In our experiments, cells were grown to 60 - 70 % confluence before passaging, which was reached with the healthy control cells after seven to eighteen days whereas the cancer cells needed five to twenty days (Figure 1B). 5 Figure 1. Overview of all measurements of primary samples. (A) The boxplot shows the relative deformation at the end of the stretching phase with regard to the passage count, with the whiskers representing the error bars. With increasing passage number, the primary cells become softer. (B) Plotting the passage time vs. passage number reveals a increasing proliferation rate with increasing passage number. In contrast to the cells from commonly used cell lines (CAL 27, CAL 33, BHY), the primary cells (benign as well as cancerous) are very heterogeneous due to their varying sizes, different stages in the cell cycle and occasionally appearing microtentacles (Figure 2A). This already becomes apparent when comparing the distribution of the cell sizes. While healthy and cancerous primary cells have a comparable size, cells from cell lines are in average 2 µm smaller and sizes spread less compared to primary samples (Figure 2A). This is especially important since the cell size directly affects the deformability showing that larger cells are more deformed than smaller cells (Figure 2A). The different average cell size as well as the larger spread of sizes in primary tissue directly reflects the variety of cells from a heterogeneous tissue like the oral cavity and even more the OSCC. Only outliers with obvious measurement artefacts were excluded from data analysis of the deformability measurements to avoid bias and preselection. The differences in deformability are represented in Figure 2B as the relative change of ellipticity during optical stretcher measurements. On average, the healthy samples (n= 360) have a relative change in ellipticity and deformation of about 1.9 % compared to 2.9 % in primary cancer cells (n= 1465) at the end of the stretch. The statistical evaluation revealed a highly significant difference between both tissue types (p < 0.001) displaying that cell originating from cancerous tissue are softer than their healthy counterparts. Since their 6 sizes are comparable, the effect is not based only on geometrical arguments. Figure 2. (A) Representative examples for the diversity of cells. The cells in the upper line originate from cancer cell lines (CAL 27, CAL 33, BHY). The cells below originate from primary cancerous tissue (middle) and primary benign tissue (bottom). The boxplot illustrates that cell originating from primary tissue are slightly larger and show a broader size distribution compared to cell lines. Cells from primary tissue as well as from cell lines display that larger cells are more deformed than smaller ones. (B) Comparing the relative deformability of primary OSCC (red; Passage 0; n= 1465) and benign oral epithelial tissue (blue; P0; n= 360) reveals that malignant cells are significantly softer than their benign counterparts. At the end of the stretching (dashed line; t = 3 s) the counts of the relative deformations show the distribution of the maximum deformability within the distinct cell population in the according histograms. (C) The magnitude of relaxation for primary OSCC (red) and primary oral epithelial tissue (blue), derived from the relative change in ellipticity normalized to the end of stretching (t = 3 s), illustrates that malignant OSCC cells were still contracting after 7 2 s while malignant cells seem to reach a stable plateau. Representative data after a total observation time of t = 4.5 s are shown in the according histograms displaying the distribution of relaxation values. However, the maximum deformation at the end of the deformation period is not the only meaningful parameter which can be extracted from the measurements since the Optical Stretcher provides a very good time resolution of the whole deformation process. Evaluating the relaxation behaviour after deformation allows to study an important cytoskeletal property of cells - their viscoelastic behaviour triggered by external stresses -- showing that cells do not return to their original shape within the observation time. Investigating the relaxation of cells after the stretching procedure revealed further differences between malignant and benign tissue. Tumour cells do not only become softer (Figure 2B) but also return faster towards their original shape than healthy cells (Figure 2B & C). When normalizing the curves to the maximal deformation (Figure 2C), it becomes apparent that cancerous and healthy cells initially relax comparably for the first 0.4 s after the stretching phase. Subsequently, the relaxation of healthy cells is almost completely halted and plateaus at still high deformability values (compared to the maximal deformation) while cancer cells continued to relax towards their initial shape over the entire relaxation period (t = 3-5 s). Since the behaviours are comparable in the initial relaxation phase but diverge later, it can be expected that the difference is not a pure requisite of the magnitude of the deformation but rather a result of differing internal dynamics. Generally, the statistical evaluation revealed a highly significant difference between healthy probes and primary cancer samples (p < 0.001) implying that primary cancer cells differ not only in their maximum deformability from healthy probes but also in their relaxation behaviour. Deformability curves of the OSCC cell lines (CAL 27, CAL 33, BHY) are presented in Figure 3. They all exhibit a higher deformability than the primary samples, which at first sight appears counterintuitive since cells from cell lines are smaller than the primary cells (Figure 2A) and should deform less. However, they are significantly softer, which is in accordance with the observed effect that primary samples become softer with increasing passage number (Figure 1). In contrast, the relaxation behaviour is decreasing compared to the primary samples. 8 Figure 3. (A) A comparison of the relative deformability of primary OSCC cells (red; Passage 0; n= 1465) and OSCC cell lines (back; pooled, n= 1780) reveals a much higher deformability for cells from OSCC cell lines. Histograms show the distribution of the maximum deformability (dashed line; end of stretching at t = 3 s) within the cell population and (B) the magnitude of relaxation (dashed line; t = 4.5 s) for primary OSCC (red) and OSCC cell lines (black). Cell lines exhibit a smaller relaxation than primary OSCC cells, which is, however, still larger than for the relaxation of primary benign cells. Histograms show the distribution of relaxation for both populations at t = 4.5 s. 9 Discussion This is the first representative study that investigates the biomechanical properties of primary cells of the oral cavity measured with the optical stretcher without preselecting any homogeneous subpopulations of cells. Even after some time in culture, the primary cells exhibit a higher heterogeneity than the commonly used cell lines, which suggests that some of the differentiated properties of the tissue cells are preserved. The optical stretcher was used as a diagnostic tool to analyse mechanical properties of the cytoskeleton of single cells from primary OSCCs and healthy control tissue. Results of both tissue sample types were compared to measurements of the standard cell lines being used as model systems for OSCCs. The results clearly indicate that healthy oral mucosa cells exhibit significantly different mechanical properties compared to malignant cells. Benign cells are stiffer and therefore less deformable than cancerous primary tissue cells. In addition, it has been shown that the relaxation of the cells after active deformation is a second parameter which can be used to analyse mechanical differences: The evaluation of these data revealed that tumour cells have a greater ability to return to their initial state than healthy cells, which have a noticeable plastic deformation behaviour. Thus, while the deformation experiments show that tumour cells are softer with more pronounced viscous contributions and appear more fluid-like than healthy cell, they appear more elastic than healthy cells during the relaxation phase. They display a continuous return towards their initial shape - a characteristic of an elastic deformation -- while healthy cells quickly plateau at still high deformability values. Consequently, the question arises what kind of mechanisms are responsible for the different mechanical behaviour of cells in the tumour cluster and why these differences are essential for malignancy? It can be speculated that plasticity effects such as bond breaking during the deformation or active responses superimpose classical viscoelastic effects. Due to the multitude of variables in these systems, a comprehensive modelling of arising behaviours is not auxiliary to gain new insights, which unfortunately impedes a direct translation of optical deformability into the more conventional metrics of moduli. From the proliferation assays presented in this study, it can be excluded that malignant cells are softer solely due to a higher proliferation rate and therefore to a lack of time for cytoskeleton formation. Our results clearly show that the proliferation rates of primary benign and malignant cells do not significantly differ in 10 vitro. Currently, there is a controversial discussion whether tumour cells are already soft in the tumour or if softening can be interpreted as a progressive course of disease leading to metastatic tumours [38]. Due to the fact that the cells in our experiments were removed from solid primary tumours and not from metastases or already metastasised tumours (all tumours G2 (moderately differentiated) and M0 (no distant metastasis); see supplemental table 1), it is obvious that the primary tumour cells in tumour tissue already show changed mechanical properties such as deformability and relaxation. The property of stronger contractility combined with a more elastic behaviour could be essential for the malignancy of cells that have the potential to metastasise or to traverse tissue as a pathogenic factor. Therefore, we speculate that the different mechanical behaviour may be a result of the transformation from benign to malignant tissue. It can be also speculated that the softening of the cells is a necessary alteration or a precondition for tumour aggressiveness that may facilitate the traverse in cell clusters to be reshaped as soon as possible. The difference of mechanical properties concerning cell lines seems to have another origin. We have observed that primary cells show decreasing cell stiffness with increasing passage number. This trend matches with the findings from experiments with established cell lines, where a greater deformability and accordingly softer cells were detected. Considering that most cell lines, including the cell lines used in this study, were subject to numerous passages and are in cell culture for many years, comparable data between primary tissue cells and established cell lines cannot be expected. The effect of softer behaviour with increasing passage number might be induced by the unphysiological micro-environment of in vitro culture conditions. These cells do not grow in their natural environment with mediators etc., but on a rigid plastic surface with an artificial medium and repetitive enzymatic treatment. Thus, to establish a serious diagnostic tool for direct application in the medical practise, the use of primary samples without cultivation is strongly recommended. We would like to note that cell lines are in fact robust test systems for new diagnostic and/or therapeutic platforms, especially since the cells are more homogenous and comparable than primary cells. However, for further investigations of tumour related cell softening, we suggest to measure the malignant tissue as well as a benign tissue sample in pairs originating from the same patient and omit direct 11 comparisons of primary malignant cells with benign cell lines. In conclusion, we were able to show that healthy and malignant oral tissue can be significantly distinguished from one another. They can be discriminated by their deformation as well as their relaxation behaviour. Our results indicate that the change of the cytoskeletal properties is rather a precondition for malignant transformation than the outcome of the transformation from the healthy to the diseased cell. These findings strongly emphasize the need to study if the deformability can be applied as an area-wide diagnostic tool for an early diagnosis of OSCCs. 12 Materials and methods Acquisition of primary cells The ethics committee of the University of Leipzig has reviewed and approved the research protocol (Reg.- No. 193/2003) and all participants gave written informed consent according to the Declaration of Helsinki. The tissue samples of the OSCC were generated during tumour resection. A small part of the resected tumour was separated and put into a centrifuge tube, containing 5 ml of Keratinocyte Basal Medium (Clonetics, Cologne, Germany) with 100 units/ml penicillin, 100 μg/ml streptomycin and 10 μg/ml amphotericin. The specimens of healthy donors were acquired during the surgical removal of third molars and were treated in the same way as the tumour specimens. Cell lines The established and commercially available cell lines CAL 27, CAL 33 [30] and BHY [31] were used for this study. All cell lines were obtained directly from the cell bank DSMZ (Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, Braunschweig, Germany), who performed cell line characterizations. All cell lines were passaged in our laboratory for less than 6 months after resuscitation. Cultivation protocol The tissue samples were washed in Betaisodona solution (Mundipharma, Limburg/ Lahn, Germany) for five minutes for disinfection, rinsed in phosphate buffered saline, and cut into pieces of about 1 mm³. These pieces were placed in an empty cell culture flask. After about 30 minutes, 20 ml of Keratinocyte Basal Medium (Clonetics, Cologne, Germany) with 100 units/ml penicillin, 100 μg/ml streptomycin and 10 μg/ml amphotericin were added and incubated at 37 °C/ 5 % CO2. The primary cultivation time varied from 14 to 49 days until 60 - 70 % confluence was reached. Prior to the experiment or to passaging the cells were detached by incubation with 2 ml of 0.25 % trypsin/EDTA for five minutes at 37 °C/ 5 % CO2. The trypsin reaction was stopped by adding 2 ml of DMEM (10% serum) to the cell suspension. To remove cell conglomerates and dead cells the suspension was strained through a 40 µm cell strainer (Falcon™ Cell 13 Strainers No.: 352340) and centrifuged twice for 2 min at RCF = 200 g at room temperature. Subsequently, the pellet was resuspended in 1 ml Keratinocyte Basal Medium (Clonetics, Cologne, Germany) with 100 units/ml penicillin, 100 µg/ml streptomycin and 10 μg/ml amphotericin. Optical stretcher It has been shown in several experiments (carcinoma of cervix or breast [11]), that the optical stretcher produces relevant data to differentiate between cancerous and healthy tissue. With the two-beam optical stretcher single cells in suspension were initially trapped with a laser power of 100 mW using two facing and slightly divergent laser beams [9, 11, 20, 23, 32, 33]. The cell suspension was given into a microfluidic system to channel a single cell into the optical trap. Applying higher laser power of 800 mW for two seconds, the momentum transfer at the interface leads to visible deformation of the cell along the laser axis. After deformation of the cell, the ability of retraction was analysed for two seconds [9]. A series of images was recorded during each measurement for further analyses of cell deformation and retraction. Therefore, the images were analysed automatically with a custom-made image analysis software (MATLAB 7.11.1, The Mathworks, Natick, Massachusetts, USA) using edge detection processes to determine the cell shape. To quantify the deformation, the ellipticity of the cells was extracted by fitting an ellipse on the edge points. The relative change in ellipticity (relative deformability) allows to compare different measurements (e.g. tumour1 vs. benign control). Statistics Biological cells are well known for their adaptability and manifoldness that do not necessarily arise randomly. Thus the frequency distribution of the deformability is not Gaussian. Consequently, median values were calculated and plotted for statistical evaluation. To generate confidence intervals for the non-parametric distributions we used the bootstrapping method; all confidence intervals indicate the 95% range. The Wilcoxon Rank sum test was used to check whether the values originate from the same basic distribution [34] as a measure for statistical significance. 14 Acknowledgement: We would like to thank the pathologist Dr. Schütze for the histopathological classification of all cancer samples. We also thank our technician Mrs Tröger for her support concerning cell culture and Ms Milani for her critical remarks. We acknowledge the financial supported by an R&D grant (SAB, Project 9889/1519) from the European Fund for Regional Development (EFRE) 2000-2006 and the state of Saxony, as well as the European Research Council (ERC-741350). Compliance with Ethical Standards Funding: We acknowledge the financial supported by an R&D grant (SAB, Project 9889/1519) from the European Fund for Regional Development (EFRE) 2000-2006 and the state of Saxony, as well as the European Research Council (ERC-741350). Potential conflict of interest: One of the authors (JAK) holds a patent on the optical stretcher technique and consults on its potential applications. One author (TWR) is CEO of the DGOD Deutsche Gesellschaft für orale Diagnostika mbH, Leipzig, Germany (Dental healthcare supplier). The other authors declare no competing financial interests. Ethical approval: All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. (Reg.-No. 193/2003) Informed consent: Informed consent was obtained from all individual participants included in the study. 15 References [1] van Zyl A W and Marnewick J C 2012 Aetiology of oral cancer SADJ : journal of the South African Dental Association = tydskrif van die Suid-Afrikaanse Tandheelkundige Vereniging 67 554 -- 6 [2] Mishra R 2012 Biomarkers of oral premalignant epithelial lesions for clinical application Oral oncology 48 578 -- 84 [3] Warnakulasuriya S 2009 Global epidemiology of oral and oropharyngeal cancer Oral oncology 45 309 -- 16 [4] van der Waal I 2009 Potentially malignant disorders of the oral and oropharyngeal mucosa; terminology, classification and present concepts of management Oral oncology 45 317 -- 23 [5] Scully C and Bagan J 2009 Oral squamous cell carcinoma overview Oral oncology 45 301 -- 8 [6] Lingen M W, Kalmar J R, Karrison T and Speight P M 2008 Critical evaluation of diagnostic aids for the detection of oral cancer Oral oncology 44 10 -- 22 [7] Reichart P 1991 Früherkennung von Neubildungen im Kiefer-Gesichtsbereich durch den praktizierenden Zahnarzt (Bonn: Forum-Medizin Verlagsgesellschaft) [8] van Zyl A and Bunn B K 2012 Clinical features of oral cancer SADJ : journal of the South African Dental Association = tydskrif van die Suid-Afrikaanse Tandheelkundige Vereniging 67 566 -- 9 [9] Guck J et al 2005 Optical deformability as an inherent cell marker for testing malignant transformation and metastatic competence Biophysical journal 88 3689 -- 98 [10] Beil M et al 2003 Sphingosylphosphorylcholine regulates keratin network architecture and visco-elastic properties of human cancer cells Nature cell biology 5 803 -- 11 [11] Fritsch A, Höckel M, Kiessling T, Nnetu K D, Wetzel F, Zink M and Käs J A 2010 Are biomechanical changes necessary for tumour progression? Nature Phys 6 730 -- 2 [12] Golde T, Schuldt C, Schnauss J, Strehle D, Glaser M and Käs J 2013 Fluorescent beads disintegrate actin networks Physical review. E, Statistical, nonlinear, and soft matter physics 88 44601 [13] Schnauss J, Händler T and Käs J 2016 Semiflexible Biopolymers in Bundled Arrangements Polymers 8 274 [14] Oswald L, Grosser S, Smith D M and Käs J A 2017 Jamming transitions in cancer J. Phys. D: Appl. Phys. 50 483001 [15] Schnauss J, Golde T, Schuldt C, Schmidt B U S, Glaser M, Strehle D, Händler T, Heussinger C and Käs J A 2016 Transition from a Linear to a Harmonic Potential in Collective Dynamics of a Multifilament Actin Bundle Physical review letters 116 108102 [16] Strehle D, Schnauss J, Heussinger C, Alvarado J, Bathe M, Kas J and Gentry B 2011 Transiently crosslinked F- actin bundles European biophysics journal : EBJ 40 93 -- 101 [17] Strehle D, Mollenkopf P, Glaser M, Golde T, Schuldt C, Käs J A and Schnauss J 2017 Single Actin Bundle Rheology Molecules (Basel, Switzerland) 22 1804 [18] Seltmann K, Fritsch A W, Käs J A and Magin T M 2013 Keratins significantly contribute to cell stiffness and impact invasive behavior Proceedings of the National Academy of Sciences of the United States of America 110 18507 -- 12 [19] Kubitschke H, Schnauss J, Nnetu K D, Warmt E, Stange R and Kaes J 2017 Actin and microtubule networks contribute differently to cell response for small and large strains New J. Phys. 19 93003 [20] Huber F, Schnauss J, Ronicke S, Rauch P, Muller K, Futterer C and Kas J 2013 Emergent complexity of the cytoskeleton: from single filaments to tissue Advances in physics 62 1 -- 112 [21] Efremov Y M, Lomakina M E, Bagrov D V, Makhnovskiy P I, Alexandrova A Y, Kirpichnikov M P and Shaitan K V 2014 Mechanical properties of fibroblasts depend on level of cancer transformation Biochimica et biophysica acta 1843 1013 -- 9 [22] Hall A 2009 The cytoskeleton and cancer Cancer metastasis reviews 28 5 -- 14 [23] Ananthakrishnan R, Guck J, Wottawah F, Schinkinger S, Lincoln B, Romeyke M, Moon T and Käs J 2006 Quantifying the contribution of actin networks to the elastic strength of fibroblasts Journal of theoretical biology 242 502 -- 16 [24] Rao K M and Cohen H J 1991 Actin cytoskeletal network in aging and cancer Mutation research 256 139 -- 48 [25] Remmerbach T W, Wottawah F, Dietrich J, Lincoln B, Wittekind C and Guck J 2009 Oral cancer diagnosis by mechanical phenotyping Cancer research 69 1728 -- 32 [26] Runge J, Reichert T E, Fritsch A, Käs J, Bertolini J and Remmerbach T W 2014 Evaluation of single-cell biomechanics as potential marker for oral squamous cell carcinomas: a pilot study Oral diseases 20 e120-7 [27] Kiessling T R, Stange R, Käs J A and Fritsch A W 2013 Thermorheology of living cells -- impact of temperature variations on cell mechanics New J. Phys. 15 45026 [28] Warmt E, Kiessling T R, Stange R, Fritsch A W, Zink M and Käs J A 2014 Thermal instability of cell nuclei New J. Phys. 16 73009 [29] Schmidt B U S, Kiessling T R, Warmt E, Fritsch A W, Stange R and Käs J A 2015 Complex thermorheology of living cells New J. Phys. 17 73010 [30] Gioanni J, Fischel J L, Lambert J C, Demard F, Mazeau C, Zanghellini E, Ettore F, Formento P, Chauvel P and Lalanne C M 1988 Two new human tumor cell lines derived from squamous cell carcinomas of the tongue: establishment, characterization and response to cytotoxic treatment European journal of cancer & clinical 16 oncology 24 1445 -- 55 [31] Kawamata H, Nakashiro K, Uchida D, Harada K, Yoshida H and Sato M 1997 Possible contribution of active MMP2 to lymph-node metastasis and secreted cathepsin L to bone invasion of newly established human oral- squamous-cancer cell lines International journal of cancer 70 120 -- 7 [32] Morawetz E W, Stange R, Kiessling T R, Schnauss J and Käs J A 2017 Optical stretching in continuous flows Converg. Sci. Phys. Oncol. 3 24004 [33] Guck J, Ananthakrishnan R, Mahmood H, Moon T J, Cunningham C C and Käs J 2001 The Optical Stretcher: A Novel Laser Tool to Micromanipulate Cells Biophysical journal 81 767 -- 84 [34] Mann H B and Whitney D R 1947 On a Test of Whether one of Two Random Variables is Stochastically Larger than the Other Ann. Math. Statist. 18 50 -- 60 [35] Charbit A, Malaise E P and Tubiana M 1971 Relation between the pathological nature and the growth rate of human tumors European journal of cancer 7 307 -- 15 [36] Pich A, Chiusa L and Navone R 2004 Prognostic relevance of cell proliferation in head and neck tumors Annals of oncology : official journal of the European Society for Medical Oncology 15 1319 -- 29 [37] Tubiana M 1989 Tumor cell proliferation kinetics and tumor growth rate Acta oncologica (Stockholm, Sweden) 28 113 -- 21 [38] Jonietz E 2012 Mechanics: The forces of cancer Nature 491 S56-S57 17
1511.08392	1	1511	2015-11-26T14:10:07	Computational Investigations on Polymerase Actions in Gene Transcription and Replication Combining Physical Modeling and Atomistic Simulations	[ "physics.bio-ph", "q-bio.BM" ]	Polymerases are protein enzymes that move along nucleic acid chains and catalyze template-based polymerization reactions during gene transcription and replication. The polymerases also substantially improve transcription or replication fidelity through the non-equilibrium enzymatic cycles. We briefly review computational efforts that have been made toward understanding mechano-chemical coupling and fidelity control mechanisms of the polymerase elongation. The polymerases are regarded as molecular information motors during the elongation process. It requires a full spectrum of computational approaches from multiple time and length scales to understand the full polymerase functional cycle. We keep away from quantum mechanics based approaches to the polymerase catalysis due to abundant former surveys, while address only statistical physics modeling approach and all-atom molecular dynamics simulation approach. We organize this review around our own modeling and simulation practices on a single-subunit T7 RNA polymerase, and summarize commensurate studies on structurally similar DNA polymerases. For multi-subunit RNA polymerases that have been intensively studied in recent years, we leave detailed discussions on the simulation achievements to other computational chemical surveys, while only introduce very recently published representative studies, including our own preliminary work on structure-based modeling on yeast RNA polymerase II. In the end, we quickly go through kinetic modeling on elongation pauses and backtracking activities. We emphasize the fluctuation and control mechanisms of the polymerase actions, highlight the non-equilibrium physical nature of the system, and try to bring some perspectives toward understanding replication and transcription regulation from single molecular details to a genome-wide scale.	physics.bio-ph	physics	Computational Investigations on Polymerase Actions in Gene Transcription and Replication Combining Physical Modeling and Atomistic Simulations Jin Yu, Beijing Computational Science Research Center #10 West Dongbei-Wang Road, Hai-Dian District, Beijing, P. R. China, 100094 ABSTRACT Email: [email protected] Tel +86-10-56981807 Polymerases are protein enzymes that move along nucleic acid chains and catalyze template-‐based polymerization reactions during gene transcription and replication. The polymerases also substantially improve transcription or replication fidelity through the non-‐equilibrium enzymatic cycles. We briefly review computational efforts that have been made toward understanding mechano-‐chemical coupling and fidelity control mechanisms of the polymerase elongation. The polymerases are regarded as molecular information motors during the elongation process. It requires a full spectrum of computational approaches from multiple time and length scales to understand the full polymerase functional cycle. We keep away from quantum mechanics based approaches to the polymerase catalysis due to abundant former surveys, while address only statistical physics modeling approach and all-‐atom molecular dynamics simulation approach. We organize this review around our own modeling and simulation practices on a single-‐subunit T7 RNA polymerase, and summarize commensurate studies on structurally similar DNA polymerases. For multi-‐subunit RNA polymerases that have been intensively studied in recent years, we leave detailed discussions on the simulation achievements to other computational chemical surveys, while only introduce very recently published representative studies, including our own preliminary work on structure-‐based modeling on yeast RNA polymerase II. In the end, we quickly go through kinetic modeling on elongation pauses and backtracking activities. We emphasize the fluctuation and control mechanisms of the polymerase actions, highlight the non-‐equilibrium physical nature of the system, and try to bring some perspectives toward understanding replication and transcription regulation from single molecular details to a genome-‐wide scale. Keywords: polymerase, molecular dynamics simulation, kinetic modeling， mechno-‐chemistry, fidelity PACS: 87.15.A-‐, 87.15.ap, 87.15.kj, 87.15.rp 1 in technological advancements Polymerases are key protein enzymes that direct gene transcription and replication in the central dogma of molecular biology. They move along nucleic acid (NA) track as molecular motors 1 and catalyze RNA or DNA synthesis according to template NA strand. The chemical catalysis, mechanical performance, and fidelity control of polymerases are therefore critical for maintaining genetic health and the malfunctions leading to diverse genetic diseases 2, 3. The polymerase enzymes are widely utilized in synthetic gene expression systems 4-‐6 and in genomic technologies 7-‐9. Engineering and redesigning of these enzymes are highly concerned and desired for various implementations. With tracking and manipulating polymerase enzymes at single molecule level in recent years 10-‐13, fundamental mechanisms of individual polymerase actions become approachable, and that greatly improves our understandings and further implementations. To understand the underlying functional mechanisms of polymerase enzymes with structural, dynamical, and energetic detail, computational studies are indispensible. With rapid developments on high-‐performance computing using parallel supercomputer clusters 14-‐16, molecular dynamics (MD) simulations of protein enzymes demonstrate great potential in elucidating the mechanisms from “bottom up”, at a full-‐atom resolution 17. Along with improvements on atomistic force field 18, long time simulations approaching microseconds to milliseconds physiological time scale have been achieved 15, 16, 19. With improvements on sampling techniques and data analysis methods 20-‐24, simulations become much more efficient in evaluating energetics and other physiologically relevant observables. On the other hand, a “top down” modeling strategy toward solving specific problems at commensurable levels, as commonly practiced in physical and mathematical sciences, can effectively deal with interested properties and easily connect to experimentally measurements. In this article, we aim at providing a brief review, based on our own efforts and practices, combining both the “top down” and “bottom up” computational strategies on studying the polymerase functions. Without being able to survey a full spectrum of computational approaches on this topic, we focus only on stochastic or kinetic modeling studies and all-‐atom molecular simulations that reveal mechano-‐chemical coupling and fidelity control properties of the polymerases. We start with a retrospect on early and general modeling frameworks built for the polymerase action, then emphasize on studies of single subunit DNA polymerases (DNAP) and RNA polymerases (RNAP) that are relatively simple in structures. For structurally more complex multi-‐subunit RNAPs that are studied intensively in recent years, we refer readers to two wonderful reviews 25, 26, which present highly I. Introduction 2 II. General physical models on polymerases active and detailed simulation studies on these systems. We then only show a few representative works that will shed light on future studies of polymerase transcriptional or replication controls and regulations. An early review on the single nucleotide addition cycle (NAC) of transcription provides a nice thermodynamics framework on RNA synthesis 27. For each NAC, an incoming NTP is added to the existing RNA strand and the product pyrophosphate ion (PPi) is released: RNAi +NTP ⇔ RNAi+1+PPi, the RNAP also moves from position i to i+1. The corresponding Gibbs free energy change ΔG can be decomposed into three parts: a chemical part, an RNA transcript folding part, and an RNAP elongation complex part (including the double and single-‐stranded DNA in the transcription bubble). In particular, the chemical part ΔG i →i+1, chem = ΔG0 i →i+1, chem + kBT ln [PPi]/[NTP], with [PPi]eq/[NTP] eq ~ 30 to 100 at an equilibrium condition 27. The latter two parts take into account the sequence-‐dependent impacts from the DNA track, though on average their contributions are close to zero. This thermodynamic framework was later employed in building a sequence-‐dependent kinetic model of the transcription elongation 28, which made good agreements with transcription gels and single-‐molecule data. force-‐velocity Inspired by single molecule measurements on relationships of E. coli RNAP, an early mechanical model 29 treated the RNAP as a processive molecular motor capable of generating force of 25~ 30 pN. The model assumed a rate-‐limiting step on pyrophosphate ion (PPi) release, and suggested that NTP binding rectifies RNAP diffusion on the DNA track. Although the assumption was not confirmed by later studies, the Brownian ratchet nature of the RNAP was captured nicely in that model 29. Based on similar single molecule measurements 30, another stochastic model of RNAP was built 31 with a focus on explaining the stall force distribution detected in the experiments. The model predicted that the stall force experimentally detected would be significantly smaller than the thermodynamic stall force 31. In both models, DNA sequence-‐dependent effects had been introduced. These early modeling studies shed light on using chemical kinetics or stochastic methods to effectively describe the mechano-‐chemical coupling in RNAPs. There are a few more recent modeling approaches toward understanding general RNAP properties. For example, a ‘look-‐ahead’ model for the transcription elongation has been proposed, in which NTP binds reversibly to a DNA site a few bps (~4 bp) ahead before being incorporated covalently into the nascent RNA chain 32. The model does not concern the mechanistic nature but provides a chemical kinetic framework, in which transcription fidelity control through NTP selection is performed at several DNA template site load 3 III. Single subunit DNA and RNA polymerases simultaneously 32. In another example, a general kinetic model was developed for the whole transcription cycle, taking into account that after RNA synthesis, RNAP may diffuse along DNA, desorb, or return to the promoter site to restart transcription 33. Interestingly, the model can predict transcriptional bursts even in the absence of explicit regulation of the transcription by master proteins 33. In a third example, dwell-‐time distributions in a two-‐state motor model was derived first, and on top of that, RNAP traffic model was developed considering steric interactions among many RNAPs moving simultaneously on the same track 34. One more example we want to mention here is the development of a ‘modular’ scheme of the RNAP transcription kinetics 35, which considers alternative and off-‐pathway states (e.g. paused, backtracked, arrested, and terminated states) of the RNAP elongation complex. The framework can be extended to study DNA replication, repair, RNA translation etc. 35. Below we focus on a group of single subunit polymerases 36-‐38, which include both DNA and RNA polymerases for gene replication and transcription, respectively. These polymerases adopt similar hand-‐like structures and are connected evolutionarily. We studies examining mechano-‐chemical coupling properties of the system. These studies mainly rely on molecular modeling and simulation techniques. Then we address how fidelity control is achieved at substrate selection stage, which has been studied from both molecular simulation and non-‐equilibrium statistic physics perspectives. Since polymerases work as molecular motors, we concern about how chemical free energy is transformed into mechanical work during each NAC cycle. The chemical free energy (ΔG i →i+1, checm) basically supports the phosphoryl transfer reaction, that adds the NMP part of NTP to the existing RNA strand while dissociates the PPi part. At the same time during each NAC, the polymerase undergoes substantial conformational changes to allow NTP binding and insertion; then it recovers back to the initial conformation, during or after the PPi release and translocation. Correspondingly, the mechanical motions involve both the substantial conformational changes of the polymerase and the relative translocation between the polymerase and the NA track. As a molecular motor moving along the track, the most concerned mechano-‐chemical coupling feature is whether the polymerase translocation is directly coupled to chemical step during the enzymatic cycle, from NTP binding to PPi dissociation. Besides, the translocation can also III. 1 Mechano-‐chemical coupling in single subunit polymerases first go through 4 I I) (pol couple to part of the substantial conformational changes. Indeed, a previous high-‐resolution structural study on bacteriophage T7 RNAP suggested a power stroke mechanism 39, in which the PPi release is tightly coupled to the translocation through an O-‐helix or fingers domain opening motion. Under this mechanism, the PPi release energetically supports the translocation. An all-‐atom MD simulation study was conducted on T7 RNAP, examining the energetics of the translocation 40. The MD study indicated that without the fingers domain opening after the product release, the translocation is not preferred. Though large fluctuations of the RNA 3’-‐end were detected within the nanosecond simulations, large conformational changes and critical translocation could not be sampled. The translocation mechanism of a structurally from Bacillus similar DNA polymerase stearothermophilus was also studied by all-‐atom MD simulations, employing biased and targeted MD methods 41. The study demonstrated that the PPi release precedes the translocation and facilitates the finger domain opening transition, which is then followed by DNA displacements for the translocation. Both studies suggested that the translocation of the polymerase is coupled to the opening conformational transition. In our most recent MD studies (ms in preparation) on the PPi release of T7 RNAP, we constructed the Markov state model (MSM) using many nanoseconds simulations, as that performed for the multi-‐subunit RNAPs 42-‐44. In addition, we also conducted a few microsecond simulations to detect slow motions in the release process. Interestingly, it is found that the PPi release proceeds through a ‘jump-‐from-‐cavity’ process, assisted by a large swing of side chain Lys472 (see Fig 1). The related structural features seem to be conserved in a group of structurally similar polymerases, including both RNAPs and DNAPs, so the mechanism can be general. On the other hand, the activated PPi release does not appear to be tightly coupled to the opening transition of the polymerase in the microsecond MD simulations. Hence, the studies do not support the power stroke mechanism, but are consistent with a Brownian ratchet mechano-‐chemical model. In the Brownian ratchet case, the PPi release precedes the translocation without direct couplings, while the translocation happens in Brownian motions without a significant free energy bias 29, 45-‐49. Indeed, previous single molecule measurements on T7 RNAP only revealed a very small free energy bias toward the post-‐translocation state (~ 1 kBT) 50, 51. The measurements thus supported a dominant Brownian ratchet feature of T7 RNAP. Our kinetic model then suggested that the small post-‐translocation free energy bias could actually aid nucleotide selection in T7 RNAP 52. From the above analyses, one can see that the substantial conformational changes in regard to the fingers domain opening and closing are crucial for the functioning of the single-‐subunit polymerases. Recently, the domain opening process of DNA pol I has been directly simulated using microsecond 5 unbiased MDs at atomistic resolution 53. An ‘ajar’ (semi-‐open) intermediate conformation, which had been discovered from a mismatched nucleotide bound structure 54, was examined in the simulation as well, and four backbone dihedrals were identified as important for the opening process. Fig 1 Transcription elongation of T7 RNAP and mechano-‐chemical coupling. (Right) A kinetic scheme of T7 elongation used in a recent modeling work 52, from NTP binding (pre-‐insertion) and insertion (the fingers domain or O-‐helix closing) to the chemical reaction and product (PPi) release, followed by translocation. The translocation proceeds in Brownian movements, while the NTP binding serves for a pawl in the Brownian ratchet model to prevent backward movements. A small post-‐translocation free energy bias has been suggested to stabilize Y639 for incoming nucleotide selection 52. (Left) The closed product structure of single subunit T7 RNAP elongation complex (in a surface representation: protein, while; NA, orange; the O-‐helix on the fingers domain, cyan; PPi, red; K472, blue, and the linked loop, light blue). The PPi release is found to be a jump-‐from-‐cavity process that is assisted by K472 side chain swing (ms in preparation). The release does not appear to be tightly coupled to the O-‐helix or the fingers domain opening, thus, cannot drive the translocation. While the opening conformational transition after PPi release somehow couples to the translocation, the close transition after NTP binding accompanies the NTP insertion to the active site, which can be a rate limiting process in some of polymerases. The domain open/closed motion has been examined previously through elastic network models combing with normal mode analyses 55, 56. It was noticed that the open to closed transition could be well approximated by a small number of normal modes of the open form polymerase 55. Later, a network of residues spanning the flexible fingers domain and the stable palmdomain are found to be involved in the open-‐closed transition, and the conserved network of residues supports a common induced-‐fit mechanism in the polymerase families for the closed 6 III. 2 Fidelity control in single subunit polymerases structure formation 56. A comprehensive report was made recently toward understanding the pre-‐chemistry conformational changes in eukaryotic DNA polymerase β 57. The NTP substrate induced domain closing transition in particular assembles the polymerase active site prior to chemistry, contributing essentially to DNA synthesis as well as on fidelity 57. The potential of mean force for the pol β closing pathway prior to chemistry was demonstrated in the study without NTP, and in the presence of correct and incorrect NTPs 57. It is shown that while subdomain motions appear intrinsic (as for conformational selection), subtle side chain motions and their favored states are largely determined by the binding of the substrate (as for induced fit). Hence, a hybrid of the conformational selection and induced fit mechanisms seems to apply to DNA polymerases 57. It was generally assumed that the polymerase fidelity control is achieved through both NTP binding and chemical steps. Some of the single subunit DNAPs, such as T7 DNAP and eukaryotic DNA pol β, had been studied systematically. For example, relative stability of Watson-‐Crick and mismatched dNTP*template base pairs in the active site of T7 DNAP and human DNA pol β had been examined using MD simulations and linear-‐response analyses 58, 59. It was found that the NTP binding selectivity of T7 DNAP is largely determined by the template-‐NTP interaction, while the binding contribution toward the replication fidelity control is less significant in pol β than that in T7 DNAP. Further progress understanding the fidelity control of these two types of DNAPs can be found, for example, in 60-‐62. In 59, a variety of computational methods, including the free energy perturbation, the linear response approximation, and an empirical valence bond method were summarized in calculating the binding free energy contribution. In 61, the full fidelity control of T7 DNAP was studied for both the substrate binding and chemical step, by taking into account contributions from the binding, pKa shifts, PO bonding breaking and making. More recently, a binding free energy decomposition approach aiming at an accurate quantification of the pol β fidelity control was implemented 62, in which separate calculations on the were conducted. neutral base and charged phosphate part using different dielectric constants As a small eukaryotic enzyme being able to repair short single stranded DNA, pol β has been extensively studied on its fidelity control. Beside the binding free energy, the closed to open transition of pol β was examined by targeted MD simulations in the mismatched system, in order to explain experimental results regarding following mis-‐incorporation, or polymerase proofreading 63. Recently, the crystal structures of pol β bound with the mismatched NTPs have been reported 64, inefficient DNA extension 7 together with MD simulation elucidating the replication fidelity control at both open and closed conformations. The results clearly show different DNAP responses toward different mismatches. Simulation studies on fidelity control of other single subunit polymerases also emerge recently. For example, MD studies on similar viral RNA-‐dependent RNA polymerases (RdRp) reveal coevolution dynamics derived from conserved and correlated dynamics of fidelity control and structural elements 65. More recently, the RdRp from Poliovirus has been studied through MD simulations and using free energy calculation 66. Interestingly, dynamic correlation between two important motifs appears sensitive to the incoming NTP species; the accessibility of the active site by one of the motifs also depends on the base pairing strength between the incoming NTP and the template, so that it explains why the active-‐site closure can be triggered by a correct NTP 66. Furthermore, studies on HIV reverse transcriptase have ben conducted both experimentally and computationally 67. The studies showed that the initial steps of weak substrate binding and protein conformational transition significantly enrich the yield of a reaction of a correct substrate but diminish that for an incorrect one. Among those above studies, controversies arose, for example, on how much pre-‐chemistry and chemical steps contribute to the DNAP replication fidelity control 57, 68, 69. Recently, we put up a kinetic framework on analyzing the stepwise nucleotide selection in the polymerase elongation 70, which considers contributions to the fidelity control from each kinetic checkpoint. When the elongation kinetics is described well by a three-‐state model (consisting of NTP binding, catalysis and translocation), the nucleotide selection can happen at two checkpoints, i.e., upon NTP binding and during chemistry step, as pointed out early. When the polymerase elongation cycle is detected with more intermediate states, however, additional kinetic transitions and checkpoints should be included. For example, NTP binding/ pre-‐insertion can be followed by another pre-‐chemistry step, which then allows the NTP insertion along with an open to closed conformational transition (see Fig 1 right). In that case, the nucleotide selection can happen at four selection checkpoints (see Fig 2): upon NTP binding/pre-‐insertion (S1), from NTP pre-‐insertion to insertion (S2), upon NTP insertion (S3), and during catalysis (S4). At each checkpoint, the wrong/non-‐cognate substrate bound the polymerase is either ‘rejected’ back to the previous state (S1 and S3, as the wrong one faces with a lower backward barrier comparing to the right), or ‘inhibited’ forward toward the next state (S2 and S4, as the wrong one incurs a higher forward barrier comparing to the right). The framework allows a stepwise examination of the fidelity control in a multiple-‐state kinetic scheme, without missing or biasing on any potential contribution. Using the master equation approach, we demonstrated some interesting properties in the stepwise selection system. First, we notice that selection through the initial selection checkpoint (S1), i.e, rejecting wrong nucleotides 8 right upon binding/pre-‐insertion, keeps the elongation at a relative high speed, which would not be maintained if the initial screening is not conducted. Fig 2. The nucleotide selection scheme in T7 RNAP elongation. (Top) The free energy landscape for incorporating right (solid line) and wrong (dashed) nucleotides in the five-‐state kinetic scheme. Four selection checkpoints (S1 to S4) are labeled. Δ1 and Δ2 are differentiation free energies between right and wrong at first two checkpoints. δG is an overall free energy differentiation without polymerase. See 70 for detail. (Bottom) Comparing the active site configurations when the right and wrong NTP bind respectively to the pre-‐insertion site 71. Y639 (red) is located on the C-‐term end of the O-‐helix (cyan) to assist the nucleotide selection. Left: rATP (right) forms the Watson-‐Crick base pairing with the template. The recognition is assisted by water bridging HB interactions with Y639-‐OH and 2’-‐OH of rNTP. Right: dATP cannot base pair with the template due to the Y639 interference, which associates with dATP and stacks well with DNA-‐RNA hybrid end, under water collision 71. Next, we find that for a same amount of free energy differentiation Δ, a same error rate is achieved for neighboring rejection and inhibition (i.e., S1 and S2, or S3 and S4). Finally, we show that the error rate achieved under the early checkpoints (S1 and S2) is lower than that achieved later on the reaction pathway (S3 and S4), if a same amount of free energy differentiation applies. One then can systematically characterize the stepwise selection i.e., by calculating the differentiation free energy at each checkpoint. We are now performing the analyses on T7 RNAP, to see if the selection system is evolved sufficiently efficient, in the absence of proofreading. 9 Besides, we have performed MD simulations to T7 RNAP and found a critical residue Tyr639 that assists nucleotide selection from pre-‐insertion to insertion 71. This residue is marginally stabilized inside the active site in post-‐translocation, by stacking its side chain with the end bp of the DNA-‐RNA hybrid. A cognate rNTP (rATP, see Fig 2 bottom) at pre-‐insertion site would form the Watson-‐Crick (WC) base pairing with the template, without further stabilizing of Tyr639 so that it can be easily pushed away during the cognate rNTP insertion. In contrast, the non-‐cognate would stabilize Tyr639 in the active site, so that Tyr639 keeps occupying the active site without allowing the non-‐cognate NTP. In particular, a dNTP (dATP, see Fig 2 bottom) is selected against by enhancing Tyr639 stacking with the end bp, under water collision 71. Interestingly, a non-‐cognate rNTP at pre-‐insertion grabs directly on Tyr639 instead 71. We also studied a mutant polymerase Y639F that cannot differentiate well dNTP from rNTP, and provided molecular basis for previous experimental findings 72. Furthermore, one should bear in mind that the polymerase elongation is a non-‐equilibrium process, and there are theoretical and modeling efforts made on this direction. It is understood that the equilibrium free energy difference between the right and wrong nucleotide incorporation contributes to transcription or replication fidelity, but the contribution is too small to fidelity. The template-‐based non-‐equilibrium for the overall account copolymerization process has been analyzed focusing on interplay between information acquisition and for the thermodynamic driving force copolymerization or elongation 73, 74. It is clearly shown that the polymerase must operate far from equilibrium to achieve a high fidelity level. Interestingly, close to equilibrium, the polymerase growth or elongation can be essentially supported by configuration disorder or the incorporation of ‘errors’ 73. The open-‐system thermodynamics to achieve DNA polymerase fidelity is systematically analyzed in 75. In particular, the nucleotide insertion selection in the absence of the exo-‐nuclease proofreading had been considered. The study indicates that a sustained non-‐equilibrium steady state essentially drives the polymerization error rate to transit from a thermodynamically determined value to a kinetically determined one, i.e., the fidelity is achieved under the “flux-‐driven kinetic checkpoints” 75. The two discrimination mechanisms involving either energetic (different binding energies) or kinetic (different kinetic barriers) differentiation are also analyzed more recently in 76. It is shown that though the two mechanisms cannot be mixed in a single-‐step reaction to reduce errors, they can be combined in coping schemes with error correction through proofreading 76. 10 IV. Multi-subunit RNA polymerases IV. 1 Probing molecular details of mechano-‐chemical coupling and Multi-‐subunit RNA polymerases (RNAPs) are widely distributed from bacteria to higher organisms, and have been extensively studied 77-‐80. Besides those general models developed for RNAPs, early kinetic modeling and analyses were developed side by side with single molecule experiments 48, 81, 82. For example, through a combination of theoretical and experimental approaches, a sequence-‐dependent thermal ratchet model of the transcription elongation was built 83. The NTP-‐specific model parameters were obtained, in particular, according to the force-‐velocity measurements on E. coli RNAP 83. A continuum Fokker-‐Planck framework of the RNAP elongation was also developed 84. Using high-‐resolution single-‐molecule data 48 of E. coli RNAP near the equilibrium condition, a free energy profile of the polymerase translocation was obtained 84, which shows consistently the ratchet character of the RNAP elongation. Stationary distributions of the RNAP translocation at far-‐from-‐equilibrium condition (e.g. very high [NTP]) can be easily derived under this framework 84. For multi-‐subunit RNAPs, we also cover two types of computational work: One is on structure-‐based modeling and simulations concerning molecular details of internal coupling and control. The other is on kinetic modeling focusing on backtracking, pauses, and related proofreading activities. Though both high-‐resolution structural studies and single molecule force measurements had been extensively conducted on multi-‐subunit RNAPs, very detailed structural dynamics is still lack of. Nevertheless, the dynamical detail can be probed directly from ‘computational microscope’ at atomistic resolution. As mentioned early, systematical reviews on employing MD simulation methods to study the multi-‐subunit RNAPs can be found in 25, 26. Here we only introduce representative works published very recently, which have not been included in the above reviews. In regard to the mechano-‐chemical coupling of RNAP, a central concern is the translocation mechanism. The intensively by constructing the Markov state model (MSM) for yeast Pol II, based on a large number of short (nanoseconds) atomistic MD simulations 85. The simulation system of Pol II reaches close to a half million atoms in explicit solvent condition. It is a big challenge, therefore, to simulate a molecular machine like Pol II up to biologically relevant time scales, i.e., from micro to milliseconds. Launching many short simulations essentially improve the computational efficiency, while constructing the MSM essentially extract the kinetic information from the simulated data. The studies show that the Pol II translocation is driven by thermal motions 85. In particular, metastable intermediate states between the fidelity control translocation is studied 11 pre-‐ and post-‐translocation states have been identified. It is also found that fluctuations of a bridge helix between bent and straight conformations facilitate the translocation of the upstream RNA:DNA hybrid, which turns out to be a rate-‐limiting step of the translocation. The bridge helix fluctuations also facilitate the translocation of a ‘transition nucleotide’, which moves asynchronously from the rest of the upstream RNA and DNA in the hybrid region 85. According to the MSM, the overall translocation rate was estimated to be about tens of microseconds at least, which is very fast comparing to the rate-‐limiting step of the elongation cycle (tens of milliseconds). Including the full transcription bubble may slow down the translocation rate 86. One noted that the translocation was simulated with a trigger loop in an open conformation 85, which had been suggested to be a pre-‐requisite for the translocation to happen 87, 88. Interestingly, recent single molecule experiments on Pol II identified a slow force-‐dependent step in the Pol II elongation, aside from a rate-‐limiting force-‐independent transition 89. Considering that the TL opening motion can be slow and force-‐dependent, we built a structure-‐based kinetic model of Pol II elongation 90 (see Fig 3), attributing the force-‐dependent slow step to the TL opening transition prior to the translocation. On one hand, the model is made consistent with both structural dynamics studies and single molecule Figure 3. A proposed five-‐state Brownian ratchet model of the multi-‐subunit RNA polymerase II (Pol II) elongation, adopted from 90. The structure of Pol II is provided (upper left). Configurations of the trigger loop (TL, in purple) and bridge helix (BH, in green) around the active site are shown for five kinetic states (I to V) in five windows. The non-‐template DNA strand is shown in blue, the template DNA strand is shown in cyan, and the synthesizing RNA strand in red. The incoming NTP molecule is shown in orange. 12 measurements, keeping a basic non-‐branched Brownian ratchet scenario; on the other hand, the model predicts the rate-‐limiting force-‐independent step conditional on accurate measurement of the NTP dissociation constant 90: If the dissociation constant is low (high NTP affinity), then the rate-‐limiting step is the TL closing transition accompanying the NTP insertion; or else (low NTP affinity), the NTP incorporation transition has to be fast to avoid too much NTP dissociation, the rate-‐limiting step can only be the catalysis after the TL closing . The study provides a working model of the complete productive elongation cycle of Pol II, and links local structure dynamics through the non-‐equilibrium enzymatic cycling kinetics 90. In regard to the fidelity control, a systematical illustration of ‘five checkpoints’ mechanisms is presented by using MD simulations 91. In the multi-‐subunit RNAP, there is an entry site (E-‐site) for the NTP binding prior to the NTP insertion into the active site (A-‐site). Correspondingly, the five checkpoints follow the reaction path along the elongation cycle, as the initial NTP binding to the E-‐site, a transition or rotation of NTP from the E-‐site to the A-‐site (E-‐A rotation), TL closing, active site re-‐arrangement for catalysis, and finally, the backtracking 91. The first four checkpoints are indeed for NTP selection, while the last checkpoint induces proofreading. In particular, the umbrella sampling method was implemented to calculate the free energy against the mismatched NTP binding at the first checkpoint, when TL is still open 91. The studies also found that the most important checkpoint for deoxy-‐NTP discrimination happens when the mismatched NTP triggers conformational distortions in the active site to hinder the catalysis 91. The final checkpoint to trigger the backtracking is through distortions of the template DNA nucleotide and DNA-‐RNA hybrid base pair around the active site. The studies open the door for further studies on the proposed mechanisms. It is noted, as the authors pointed out, that the efficiencies of the fidelity checkpoints on discriminating against different non-‐cognate rNTPs and dNTPs are sequence dependent and vary for different RNAP species. In multi-‐subunit RNAPs, pauses are frequently present to play important regulation roles 92. The paused are often linked to backtracking behaviors of the polymerases 93. A statistical mechanics approach toward predicting backtracked in bacterial transcription elongation had been conducted 94. A pauses thermodynamic model of complex was built with sequence-‐dependent free energy variations from the translocational and size fluctuations of the transcription bubble, as well as from accompanied changes in the RNA-‐RNA hybrid and the RNA transcript. The model produced statistically significant results toward predicting ~ 100 elongation pause sites for E. coli RNAP on 10 DNA templates 94. The study also provided a kinetic model on pause recovery, assuming slow RNA unfolding and fast translocation. In both models, the elongation IV. 2 Kinetic modeling on backtracking pauses 13 the sequence-‐specific kinetic barriers due to RNA co-‐transcriptional folding turn out to be essential to strongly inhibit the backtracking. Another essential feature identified was an intermediate state separating the productive elongation with the backtracking in a further developed thermal ratchet model 95. Whether the backtracking causes a wide range of pauses, including both long and short ones, was investigated in a study later 93. By modeling the backtracking as force-‐biased diffusion in a periodic one-‐dimension free energy landscape, the study showed a single mechanism of random walk backtracking can generate both the long (diffusive) and short (ubiquitous) pauses. In particular for short pauses, sequence-‐induced variations on the backward rates can have a large impact on the lifetime of the backtracking pauses 93. Actually, when the backtracking pauses are included in a full transcription elongation model, a broad, heavy-‐tailed distribution of the elongation time has been obtained 96. Interestingly, the authors of the study suggested that the pauses could even lead to bursts of mRNA production and non-‐Poisson statistics of mRNA levels 96, thus, contribute significantly to noise productions on a cellular level. Using a similar kinetic model and the master equation approach, these researchers studied proofreading activities involving the backtracking and RNA cleavage 97. Backtracking by more than one nucleotide provides a multiple-‐checking reaction to probe the fidelity of newly generated nucleotides before further nucleotide addition. The study showed that the accuracy improves along with longer delay caused by the backtracking and cleavage. In an extreme case, the error fraction scales exponentially with the maximum backtracking distances 97. The model thus predicts a strong dependence of transcriptional fidelity on the backtracking rates or probabilities. Gene transcription and replication are directed by RNA and DNA polymerases through enzymatic cycles, hence, their elongation processes are maintained at non-‐equilibrium steady states (NESS) driven by the chemical potential 98. It is key to understand the NESS basis in order to understand the mechano-‐chemical coupling mechanisms and fidelity control features of the polymerases. The non-‐equilibrium statistical physics in regard to corresponding heat production, growth rate, internal entropy, and durability of the ‘self-‐replication’ process has been built up in 99. Nevertheless, close-‐to-‐equilibrium properties of each kinetic intermediate state in the elongation cycle can be well probed through regular MD simulations, so that local structural dynamics and energetics reveal with substantial detail. For rate-‐limiting transitions in the elongation cycle, however, commensurable simulations should take into account the NESS chemical potential by simulating sufficiently fast processes of substrate binding and product release. The NESS dynamics would then become more of a concern when micro to milliseconds MD simulations become routine for polymerase V. Summary and Perspectives 14 machinery. It is quite interesting to notice that polymerases have been largely identified to work under the loosely coupled Brownian ratchet scenario, no matter for the single or multi-‐subunit polymerases. In addition to the experimental evidence mentioned early, a very recent example is on translocation of replicative DNAP from bacteriophage phi29 100. Under the Brownian ratchet mechanism, the translocation of the polymerase spontaneously happens without being directly coupled to chemical transition such as the substrate binding or product release. However, the translocation can still couple to some essential conformational changes (such as the O-‐helix or TL opening in the single and multi-‐subunit polymerases, respectively), which may be facilitated by the chemical transition but not at the same time as during the coupling to the translocation. In contrast, the tightly coupled power stroke scenario requires simultaneous coupling between the translocation and the chemical transition, no matter other conformational changes involved or not. The power stroke scenario, however, had not been gained continuous experimental support. Besides for the polymerases, ribosomes, the most essential translation machinery, have also been consistently demonstrated to work under the Brownian ratchet scenario 101-‐103. Since both machineries appeared very early in the molecular evolution history, one would speculate that the Brownian ratchet requires no highly sophisticated internal coupling mechanisms, therefore, might be easily adopted into those ancient molecular machineries. The fidelity control of polymerase transcription and replication is achieved in general by combining nucleotide selection before the catalysis with proofreading cleavage after the catalysis. Both mechanisms work at non-‐equilibrium or driven conditions. The selection proceeds stepwise through each kinetic intermediate state, starting right after the nucleotide binding or pre-‐insertion, and working all the way until the end of the catalytic reaction. Substantial selection has been found to happen through the slow process of nucleotide insertion or catalysis 57, 69. We notice that early selections outperform the late ones on the reaction path in reducing the error rate while the initial selection or screening is particularly helpful to maintain the elongation speed high. Since the template-‐based polymerization relies primarily on the WC base pairing, the differentiation between incoming rNTP and dNTP becomes highly subtle, involving delicate residue coordination such as ‘steric gate’ or hydroxyl-‐water interaction etc. 71, 104. On the other hand, tolerance on template backbone sugar heterogeneity is revealed as well for Pol II 105. There has also been evidence that the WC hydrogen bonding is not highly crucial for the fidelity control of T7 RNAP while the steric effect can be significant 106. Remarkably, it is reported that transient WC-‐like mispairs (with probabilities 10-‐3-‐10-‐5) stereochemically mimic the WC geometry so that to evade fidelity checkpoints 107, which can play some universal role in gene mutation and molecular evolution. One may hypothesize that the transient WC-‐like mispairs set a limit on the polymerase fidelity control. Anyhow, it remains elusive how much the polymerases contribute energetically 15 to select cognates over non-‐cognates, especially, for variant nucleotide species. Hence, it is still hard to quantitatively test the above hypothesis. The original idea of kinetic proofreading traced back to work of Hopfield 108 and Ninio 109. The proofreading activities of RNA polymerases have been investigated in recent years 80. Interestingly, in a newly published modeling work it is found that the proofreading supported fidelity control strongly depends on sequence context such that it brings to accuracy variation to several orders of magnitude 110. Though experimentally measured free energies of dsDNA and RNA-‐DNA hybrid has been incorporated into the model, the polymerase contribution to the sequence-‐dependent accuracy variation has not been considered 110. Again, it is because that the polymerase contribution to the accuracy has not been systematically investigated. Hence, it becomes highly desirable if computational studies in the near future could provide quantify how much the polymerases energetically differentiate cognates vs. non-‐cognates in a sequence specific manner. The sequence specific characterization of polymerase actions is expected to develop side by side with technology advancements on targeted and genome wide sequencing 7, 111. Current work is supported by NSFC under the grant No. 11275022. 1. 2. 3. 4. 5. 6. 7. 8. H. Buc and T. Strick, (The Royal Society of Chemistry, Cambridge, UK, 2009). L. A. Loeb and R. J. Monnat Jr, Nat. Rev. Gene. 9, 594 (2008). J. W. Conaway and R. C. Conaway, Annu. Rev. Biochem. 68, 301-‐319 (1999). D. L. Shis and M. R. Bennett, Molecular Systems Biology 10 (7), n/a-‐n/a (2014). F. W. Studier and B. A. Moffatt, Journal of Molecular Biology 189 (1), 113-‐130 (1986). K. J. Livak and T. D. Schmittgen, METHODS 25, 402-‐408 (2001). M. J. R. Previte, C. Zhou, M. Kellinger, R. Pantoja, C.-‐Y. Chen, J. Shi, B. Wang, A. Kia, S. Etchin, J. Vieceli, A. Nikoomanzar, E. Bomati, C. Gloeckner, M. Ronaghi and M. M. He, Nature Communications 6, 5936 (2015). J. Eid, A. Fehr, J. Gray, K. Luong, J. Lyle, G. Otto, P. Peluso, D. Rank, P. Baybayan, B. Bettman, A. Bibillo, K. Bjornson, B. Chaudhuri, F. Christians, R. Cicero, S. Clark, R. Dalal, A. deWinter, J. Dixon, M. Foquet, A. Gaertner, P. Hardenbol, C. Heiner, K. Hester, D. Holden, G. Kearns, X. Kong, R. Kuse, Y. Lacroix, S. Lin, P. Lundquist, C. Ma, P. Marks, M. Maxham, D. Murphy, I. Park, T. Pham, M. Phillips, J. Roy, R. Sebra, G. Shen, J. Sorenson, A. Tomaney, K. Travers, M. Trulson, J. Vieceli, J. Wegener, D. Wu, A. Yang, D. Zaccarin, P. Zhao, F. Zhong, J. Korlach and S. Turner, Science 323 (5910), 133-‐138 (2009). M. Ronaghi, M. Uhlén and P. Nyrén, Science 281 (5375), 363-‐365 (1998). M. Dangkulwanich, T. Ishibashi, L. Bintu and C. Bustamante, Chemical Review 9. 10. Acknowledgements References 16 114, 3203-‐3223 (2014). J. Michaelis and B. Treutlein, Chemical Reviews 113 (11), 8377-‐8399 (2013). K. M. Herbert, W. J. Greenleaf and S. M. Block, Annual Review of Biochemistry 77, 149-‐176 (2008). J. P. Gill, J. Wang and D. P. Millar, Biochemical Society Transactions 39, 595-‐599 (2011). Y. Ohno, R. Yokota, H. Koyama, G. Morimoto, A. Hasegawa, G. Masumoto, N. Okimoto, Y. Hirano, H. Ibeid, T. Narumi and M. Taiji, Computer Physics Communications 185 (10), 2575-‐2585 (2014). J. L. Klepeis, K. Lindorff-‐Larsen, R. O. Dror and D. E. Shaw, Current Opinion in Structural Biology 19 (2), 120-‐127 (2009). K. Schulten, J. C. Phillips, L. V. Kale and A. Bhatele, in Petascale Computing: Algorithms and Applications, edited by D. Bader (Chapman and Hall/CRC Press, Taylor and Francis Group, New York, 2008), pp. 165-‐181. M. Karplus and J. A. McCammon, Nat Struct Mol Biol 9 (9), 646-‐652 (2002). O. Guvench and A. MacKerell, Jr., in Molecular Modeling of Proteins, edited by A. Kukol (Humana Press, 2008), Vol. 443, pp. 63-‐88. T. J. Lane, D. Shukla, K. A. Beauchamp and V. S. Pande, Current opinion in structural biology 23 (1), 58-‐65 (2013). C. Junghans, D. Perez and T. Vogel, Journal of Chemical Theory and Computation 10 (5), 1843-‐1847 (2014). J. D. Chodera and F. Noé, Current Opinion in Structural Biology 25 (0), 135-‐144 (2014). P. Tiwary and M. Parrinello, Physical Review Letters 111 (23), 230602 (2013). O. Hisashi, Advances in Natural Sciences: Nanoscience and Nanotechnology 1 (3), 033002 (2010). T. Schlick, F1000 Biology Reports 1, 51 (2009). B. Wang, M. Feig, R. I. Cukier and Z. F. Burton, Chemical Reviews 113, 8546-‐8566 (2013). F. PARDO-‐AVILA, L.-‐T. DA, Y. WANG and X. HUANG, Journal of Theoretical and Computational Chemistry 12 (08), 1341005 (2013). D. A. Erie, T. D. Yager and P. H. von Hippel, Annual Review of Biophysics and Biomolecular Structure 21, 379-‐415 (1992). L. Bai, A. Shundrovsky and M. D. Wang, Journal of Molecular Biology 344 (2), 335-‐349 (2004). H.-‐Y. Wang, T. Elston, A. Mogilner and G. Oster, Biophysical Journal 74 (3), 1186-‐1202 (1998). H. Yin, M. D. Wang, K. Svoboda, R. Landick, S. M. Block and J. Gelles, Science 270 (5242), 1653-‐1657 (1995). F. Julicher and R. Bruinsma, Biophysical Journal 74 (3), 1169-‐1185 (1998). Y. R. Yamada and C. S. Peskin, Biophysical Journal 96, 3015-‐3031 (2009). V. P. Zhdanov, Physical Review E 80, 051925 (2009). T. Tripathi, G. M. Schutz and D. Chowdhury, Journal of Statistical Mechanics: Theory and Experiment 8, P08018 (2009). 17 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. S. J. Greive, J. P. Goodarzi, S. E. Weitzel and P. H. von Hippel, Biophysical Journal 35. 101, 1155-‐1165 (2011). S. Wu, W. A. Beard, L. G. Pedersen and S. H. Wilson, Chemical Reviews 114 (5), 36. 2759-‐2774 (2013). N. Cermakian, T. M. Ikeda, P. Miramontes, B. F. Lang, M. W. Gray and R. 37. Cedergrent, J Mol Evol 45 (6), 671-‐681 (1997). T. A. Steitz, S. J. Smerdon, J. Jager and C. M. Joyce, Science 266 (5193), 2022-‐2025 38. (1994). Y. W. Yin and T. A. Steitz, Cell 116 (3), 393-‐404 (2004). 39. H.-‐J. Woo, Y. Liu and R. Sousa, Proteins: Structure, Function, and Bioinformatics 40. 73 (4), 1021-‐1036 (2008). A. Golosov, J. Warren, L. Beese and M. Karplus, Structure 18, 83-‐93 (2010). 41. L. Da, D. Wang and X. Huang, Journal of the American Chemical Society 134, 42. 2399-‐2406 (2011). L.-‐T. Da, A. Pardo, F., D. Wang and X. Huang, PloS Computational Biology 9, 43. e1003020 (2013). L.-‐T. Da, F. Sheong, D.-‐A. Silva and X. Huang, in Protein Conformational Dynamics, 44. edited by K.-‐l. Han, X. Zhang and M.-‐j. Yang (Springer International Publishing, 2014), Vol. 805, pp. 29-‐66. C. S. Peskin, G. M. Odell and G. F. Oster, Biophysical Journal 65 (1), 316-‐324 45. (1993). H. Wang and G. Oster, Applied Physics A 75, 315-‐323 (2002). 46. G. Bar-‐Nahum, V. Epshtein, A. E. Ruckenstein, R. Rafikov, A. Mustaev and E. 47. Nudler, Cell 120 (2), 183-‐193 (2005). E. A. Abbondanzieri, W. J. Greenleaf, J. W. Shaevitz, R. Landick and S. M. Block, 48. Nature 438 (7067), 460-‐465 (2005). Q. Guo and R. Sousa, Journal of Molecular Biology 358 (1), 241-‐254 (2006). 49. P. Thomen, P. J. Lopez, U. Bockelmann, J. Guillerez, M. Dreyfus and F. Heslot, 50. Biophysical Journal 95 (5), 2423-‐2433 (2008). P. Thomen, P. J. Lopez and F. Heslot, Physical Review Letters 94 (12), 128102 51. (2005). J. Yu and G. Oster, Biophysical Journal 102, 532-‐541 (2012). 52. B. R. Miller III, C. A. Parish and E. Y. Wu, PLos Computational Biolgoy 10 (12), 53. e1003961 (2014). L. S. Beese and E. Y. Wu, J. Biol. Chem 286, 19758-‐19767 (2011). 54. M. Delarue and Y. H. Sanejouand, Journal of Molecular Biology 320 (5), 55. 1011-‐1024 (2002). 56. W. Zheng, B. R. Brooks, S. Doniach and D. Thirumalai, Structure 13 (4), 565-‐577 (2005). T. Schlick, K. Arora, W. A. Beard and S. H. Wilson, Theoretcial Chemistry 57. Accounts 131, 1287 (2012). J. Florian, A. Warshel and M. F. Goodman, The Journal of Physical Chemistry B 58. 106, 5754-‐5760 (2002). J. Florian, M. F. Goodman and A. Warshel, The journal of Physical Chemistry B 59. 18 106, 5739-‐5753 (2002). J. Florián, M. F. Goodman and A. Warshel, Biopolymers 68, 286-‐299 (2003). J. Florián, M. F. Goodman and A. Warshel, Proceedings of the National Academy of Sciences of the United States of America 102 (19), 6819-‐6824 (2005). R. Rucker, P. Oelschlaeger and A. Warshel, Proteins 78, 671-‐680 (2010). L. Yang, W. A. Beard, S. H. Wilson, B. Roux, S. Broyde and T. Schlick, Journal of Molcular Biology (321), 459-‐478 (2002). M. Koag, K. Nam and S. Lee, Nucleic Acids Research 42, 11233-‐11245 (2014). I. M. Moustafa, H. Shen, B. Morton, C. M. Colina and C. E. Cameron, Journal of Molecular Biology 410, 159-‐181 (2011). H. Shen and G. Li, Journal of Chemical Theory and Computation 10, 5195-‐5202 (2014). S. Kirmizialtin, V. Nguyen, K. A. Johnson and R. Elber, Structure 20, 618-‐627 (2012). B. R. Prasad, S. C. L. Kamerlin, J. Florian and A. Warshel, Theoretical Chemistry Accounts 131, 1288 (2012). A. J. Mulholland, A. E. Roitberg and I. Tunon, Theoretical Chemistry Accounts 131, 1286 (2012). J. Yu, Molecular Based Mathmatical Biology 2, 141-‐160 (2014). B. Duan, S. Wu, L.-‐T. Da and J. Yu, Biophysical Journal 107, 2130-‐2140 (2014). R. Sousa and R. Padilla, The EMBO Journal 14, 4609-‐4621 (1995). D. Andrieux and P. Gaspard, Proceedings of the National Academy of Sciences of the USA 105, 9516-‐9521 (2008). C. Jarzynski, Proceedings of the National Academy of Sciences of the USA 105, 9451-‐9452 (2008). F. Cady and H. Qian, Physical Biology 6, 036011 (2009). P. Sartori and S. Pigolotti, Physical Review Letters 110, 188101 (2013). S. Borukhov and E. Nudler, Trends in Microbiology 16 (3), 126-‐134 (2008). E. Nudler, Annual Review of Biochemistry 78 (1), 335-‐361 (2009). F. Brueckner, J. Ortiz and P. Cramer, Current Opinion in Structural Biology 19, 294-‐299 (2009). J. F. Sydow and P. Cramer, Current Opinion in Structural Biology 19 (6), 732-‐739 (2009). M. D. Wang, M. J. Schnitzer, H. Yin, R. Landick, J. Gelles and S. M. Block, Science 282 (5390), 902-‐907 (1998). L. Bai, T. J. Santangelo and M. D. Wang, Annual Review of Biophysics and Biomolecular Structure 35 (1), 343-‐360 (2006). L. Bai, R. M. Fulbright and M. D. Wang, Physical Review Letters 98, 068103 (2007). H.-‐J. Woo, Physical Review E 74, 011907 (2006). D. A. Silva, D. R. Weiss, F. Pardo Avila, L. T. Da, M. Levitt, D. Wang and X. Huang, Proc Natl Acad Sci U S A 111 (21), 7665-‐7670 (2014). M. Imashimizu and M. Kashlev, Proc Natl Acad Sci U S A 111 (21), 7507-‐7508 (2014). 19 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71. 72. 73. 74. 75. 76. 77. 78. 79. 80. 81. 82. 83. 84. 85. 86. M. Feig and Z. F. Burton, Proteins 78, 434-‐446 (2010). 87. B. Feig and Z. F. Burton, Biophysical Journal 99, 2577-‐2586 (2010). 88. M. Dangkulwanich, T. Ishibashi, S. Liu, M. L. Kireeva, L. Lubkowska, M. Kashlev 89. and C. Bustamante, eLIFE 2, e00971 (2013). J. Yu, L.-‐T. Da and X. Huang, Physical Biology 12, 016004 (2015). 90. B. Wang, K. Opron, Z. F. Burton, R. I. Cukier and M. Feig, Nucleic Acids Res 43 (2), 91. 1133-‐1146 (2015). F. Gong and C. Yanofsky, The Journal of Bacteriology 185, 6472-‐6476 (2003). 92. M. Depken, E. A. Galburt and S. W. Grill, Biophysical Journal 96 (6), 2189-‐2193 93. (2009). V. R. Tadigotla, D. O. Maoileidigh, A. M. Sengupta, V. Epshtein, R. H. Ebright, E. 94. Nudler and A. E. Ruckenstein, Proceedings of the National Academy of Sciences 103, 4439-‐4444 (2006). D. O. Maoileidigh, V. R. Tadigotla, E. Nudler and A. E. Ruckenstein, Biophysical 95. Journal 100, 1157-‐1166 (2011). M. Voliotis, N. Cohen, C. Molina-‐ParÌs and T. B. Liverpool, Biophysical Journal 94 96. (2), 334-‐348 (2008). M. Voliotis, N. Cohen, C. Molina-‐Paris and T. B. Liverpool, Physical Review Letters 97. 102 (25), 258101 (2009). H. Qian, Physical Review E 69, 012901 (2004). 98. J. L. England, Journal of Chemical Physics 139, 121923 (2013). 99. J. A. Morin, F. J. Cao, J. M. Lázaro, J. R. Arias-‐Gonzalez, J. M. Valpuesta, J. L. 100. Carrascosa, M. Salas and B. Ibarra, Nucleic Acids Research (2015). 101. J. Frank and R. Gonzalez Jr, Annual review of biochemistry 79, 381-‐412 (2010). 102. T. Liu, A. Kaplan, L. Alexander, S. Yan, J.-‐D. Wen, L. Lancaster, C. E. Wickersham, K. Fredrick, H. Noller, I. Tinoco and C. J. Bustamante, Direct measurement of the mechanical work during translocation by the ribosome. (2014). 103. A. Dashti, P. Schwander, R. Langlois, R. Fung, W. Li, A. Hosseinizadeh, H. Y. Liao, J. Pallesen, G. Sharma, V. A. Stupina, A. E. Simon, J. D. Dinman, J. Frank and A. Ourmazd, Proceedings of the National Academy of Sciences 111 (49), 17492-‐17497 (2014). 104. A. M. Delucia, N. D. Grindely and C. M. Joyce, Nucleic Acids Research 31, 4129-‐4137 (2003). L. Xu, W. Wang, L. Zhang, J. Chong, X. Huang and D. Wang, Nucleic Acids Res 43 105. (4), 2232-‐2241 (2015). 106. S. Ulrich and E. T. Kool, Biochemistry 50, 10343-‐10349 (2011). I. J. Kimsey, K. Petzold, B. Sathyamoorthy, Z. W. Stein and H. M. Al-‐Hashimi, 107. Nature 519 (7543), 315-‐320 (2015). 108. J. J. Hopfield, Proceedings of the National Academy of Sciences of the United States of America 71, 4135-‐4139 (1974). 109. J. Ninio, Biochimie 57, 587-‐595 (1975). 110. H. Mellenius and M. Ehrenberg, PLoS One 10 (3), e0119588 (2015). 111. M. Imashimizu, T. Oshima, L. Lubkowska and M. Kashlev, Nucleic Acids Res 41 (19), 9090-‐9104 (2013). 20
1111.3627	1	1111	2011-11-15T20:27:29	Correlated interaction fluctuations in photosynthetic complexes	[ "physics.bio-ph", "physics.chem-ph", "quant-ph" ]	The functioning and efficiency of natural photosynthetic complexes is strongly influenced by their embedding in a noisy protein environment, which can even serve to enhance the transport efficiency. Interactions with the environment induce fluctuations of the transition energies of and interactions between the chlorophyll molecules, and due to the fact that different fluctuations will partially be caused by the same environmental factors, correlations between the various fluctuations will occur. We argue that fluctuations of the interactions should in general not be neglected, as these have a considerable impact on population transfer rates, decoherence rates and the efficiency of photosynthetic complexes. Furthermore, while correlations between transition energy fluctuations have been studied, we provide the first quantitative study of the effect of correlations between interaction fluctuations and transition energy fluctuations, and of correlations between the various interaction fluctuations. It is shown that these additional correlations typically lead to changes in interchromophore transfer rates, population oscillations and can lead to a limited enhancement of the light harvesting efficiency.	physics.bio-ph	physics	Correlated interaction fluctuations in photosynthetic complexes Sebastiaan M. Vlaming1, ∗ and Robert J. Silbey1, † 1Center for Excitonics and Department of Chemistry, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, United States (Dated: March 26, 2018) Abstract The functioning and efficiency of natural photosynthetic complexes is strongly influenced by their embedding in a noisy protein environment, which can even serve to enhance the transport efficiency. Interactions with the environment induce fluctuations of the transition energies of and interactions between the chlorophyll molecules, and due to the fact that different fluctuations will partially be caused by the same environmental factors, correlations between the various fluctuations will occur. We argue that fluctuations of the interactions should in general not be neglected, as these have a considerable impact on population transfer rates, decoherence rates and the efficiency of photosynthetic complexes. Furthermore, while correlations between transition energy fluctuations have been studied, we provide the first quantitative study of the effect of correlations between interaction fluctuations and transition energy fluctuations, and of correlations between the various interaction fluctuations. It is shown that these additional correlations typically lead to changes in interchromophore transfer rates, population oscillations and can lead to a limited enhancement of the light harvesting efficiency. 1 1 0 2 v o N 5 1 ] h p - o i b . s c i s y h p [ 1 v 7 2 6 3 . 1 1 1 1 : v i X r a ∗Electronic address: [email protected] †deceased 1 I. INTRODUCTION The harvesting of sunlight by organisms such as plants and bacteria plays a crucial role in life on earth. Nature has optimized the photosynthesis process to a great extent, leading to a high efficiency in both the capturing of light and the subsequent energy transport within the organism's light harvesting complexes.1–5 Pigment molecules in these light harvesting structures absorb the light, after which the excitation energy is eventually funneled towards the reaction center, where charge separation takes place and the energy can be utilized for biochemical purposes. A closer understanding of the functioning of these biological systems and what features are crucial to their high efficiency may provide further insight in how to optimize synthetic light harvesting structures. Recent experimental studies have shown that quantum coherence in light harvesting com- plexes persists on surprisingly long timescales.6–9 While excitation energy transport (EET) in such molecular systems has been a subject of study for decades,10–12 the aforementioned experimental observation of long-lived coherence despite its occurrence in noisy environ- ments at room temperature has spurred further theoretical research. Naively, one might think that coupling the electronic excitations to a dissipative environment will lead to rapid decoherence and a loss of excitation energy. However, it has been shown that environment- induced dephasing and noise may actually enhance the transport efficiency of excitations in such structures.13–18 The rationale behind this lies in the fact that interaction with degrees of freedom in the bath can induce transitions of the excitation that are directed towards the reaction center. Recent studies have shed further light on the mechanisms that allow for optimal efficiency of the excitation, suggesting an intricate functioning where aspects such as the non-Markovian nature of the bath interactions19–21 and correlations in the bath18,22–24 may be exploited to achieve further optimization of the light harvesting complex. The interaction of the chromophores with their local environments leads to fluctuations of the transition energies and interactions. More specifically, changes in the environment will induce fluctuating local electrical fields. The resultant Stark shifts in the transition energies and transition dipoles of the chromophores will thus also fluctuate; in addition, environment-induced changes in relative position or orientation of the chromophores will also cause corresponding fluctuations in the interactions.24 A number of studies have been performed where only the effect of transition energy fluctuations has been included, as 2 well as possible spatial and temporal correlations between these transition energy fluctua- tions.18,22–24 This is commonly done in the framework of the extended Haken-Strobl-Reineker model,25,26 which provides an attractive approach due to its simplicity and tractability. The original Haken-Strobl-Reineker model allows for a treatment of both transition energy and interaction fluctuations, and while it does not include correlations and inherently assumes a high temperature, the formalism can straightforwardly be generalized to remove the former drawback. Fluctuations in the interactions, in contrast to transition energy fluctuations, have typically not been accounted for in studies of photosynthetic complexes, despite the fact that these will generally occur as well. Fluctuations in transition energies and interactions are induced by variations in the local environments of the chromophores, and environmental variations around one chromophore will generate both fluctuations in its transition energy and in its interaction with the other chromophores. In addition, different chromophores may share part of the environment; both these arguments imply that the generated fluctuations will in general be correlated. While correlations between transition energy fluctuations have been studied,18,22–24 there should generally also be correlations of transition energy fluctuations and interaction fluctuations, and between different interaction fluctuations. In previous studies, such correlations have commonly been discarded, even though there is no a priori reason to do so. A number of recent studies have also suggested the occurrence of such fluctuations in photosynthetic complexes.27–29 Moreover, fluctuations of the various intermolecular interactions are usually not accounted for, although also these are typically induced by changes in the local en- vironment. We present a study of the effects of fluctuations of the interactions and their possible correlations with the various transition energy and other interaction fluctuations, and provide a numerical analysis of their measurable effects. When applied to natural pho- tosynthetic complexes, such as the Fenna-Matthews-Olson (FMO) complex,30–34 it is shown that interaction fluctuations can have a considerable impact on the excitation dynamics and the efficiency, and these effects should generally not be neglected. Furthermore, we show that additional correlations can lead to enhanced oscillations of the exciton populations and modifications of the transfer and dephasing rates, and we quantify the dependence of the trapping efficiency on the correlations and initial conditions. The paper is structured as follows. In Sec. II, we provide the theoretical background of our study: in Sec. II A, we introduce the Haken-Strobl-Reineker formalism that is used to 3 describe the effect of environmentally induced fluctuations and their possible correlations on the excitation dynamics, Sec. II B discusses the possible correlations, and in Sec. II C we introduce the Fenna-Matthews-Olson photosynthetic complex that we will apply our theory to. In Sec. III, we show the results of application of our theory to FMO: Sec. III A elucidates the role of uncorrelated interaction fluctuations, Sec. III B focuses on the effect of correlations between fluctuations in transition energies and interactions, while subsequently in Sec. III C we investigate the effect of correlated interaction fluctuations. Our conclusions are presented in Sec. IV. II. THEORY A. The Haken-Strobl-Reineker model Upon absorption of the solar light and energy transfer into the FMO complex, an elec- tronic excited state will be created in the bacteriochlorophylls (BChl's). Due to the strong interchromophore interactions, this excited state (exciton) will be delocalized over a num- ber of BChl's. To describe the excitation and its dynamics, we employ the Frenkel exciton Hamiltonian,35,36 H = HS + HR + HS−R = X n En ni hn + X n,m6=n Jnm ni hm + X q ωqb† qbq + HS−R, (1) where ni describes a state where chromophore n is in its excited state while all others are in their ground states, Jnm is the interaction between chromophores n and m, and we have a bath of harmonic modes labeled by q. The first two terms are the system Hamiltonian HS, the third term corresponds to the bath Hamiltonian HR, and the final term describes the coupling between the system and the bath HS−R, which we will keep unspecified for now. In our case, the system consists of the BChl's within the photosynthetic complex, while the environment includes any other degrees of freedom the excitations can couple to, such as the protein surroundings. To adequately describe the environmental effects, it is necessary to work in the density matrix formalism.37–39 The density matrix corresponding to the wave function ψ(t)i is given by ρ(t) = ψ(t)ihψ(t), and evolves according to the Liouville-von 4 Neumann equation ( has been set to unity), i ∂ ρ(t) ∂t = [H(t), ρ(t)] ≡ Lρ(t). (2) Here, we have defined the Liouville superoperator as L = [H, ...]. The above time evolution equation of the density matrix is equivalent to the time evolution of the wave function ψ(t)i as given by the Schrodinger equation. At this point, we switch to the interaction picture, where the time evolution induced by the Hamiltonian terms HS + HR, which are large compared to the system-bath interaction HS−R, is explicitly removed from the time evolution of all operators A(t), AI(t) = ei(HS+HR)tA(t)e−i(HS +HR)t. (3) We are primarily interested in the evolution of the system degrees of freedom, which we obtain by taking the trace over the bath degrees of freedom. This leaves us with the reduced density matrix ρ(t) in the interaction picture, which is the quantity we will consider from this point on when referring to the density matrix. The approach we take in treating interactions with the environment is based on a method first introduced by Haken, Strobl and Reineker.25,26 In this approach, one models the bath- induced fluctuations as classical Gaussian Markov processes. While the original Haken- Strobl-Reineker (HSR) model assumed uncorrelated site energy fluctuations, one can extend the methodology to allow for correlations between the various fluctuations; we refer to Ref. 41 for the details. The Hamiltonian can then be written as H = X n En nihn + X n,m6=n Jnm nihm + X n,m Vnm(t) ni hm . (4) The terms Vnm(t) are the environmentally induced fluctuations of the various system Hamil- tonian matrix elements. Note that the Hermiticity of the Hamiltonian implies Vnm(t) = V ∗ mn(t). The averages hVnm(t)i can be set to zero without loss of generality, and since the fluctuations are assumed to be Gaussian Markov processes, the problem is fully defined when the correlation functions hVnm(t)Vn′m′(t′)i are known. The effect of the environment is now fully encoded in the bath correlation functions Cnmn′m′(τ ) = hVnm(τ )Vn′m′(0)i. 5 B. Correlation functions To proceed, we make the white noise assumption, where it is assumed that the bath relaxes on a time scale that is short compared to the exciton dynamics. The time dependence of the bath correlation functions is taken as a δ-function, Cnmn′m′(τ ) ≡ hVnm(τ )Vn′m′(0)i = γnmn′m′δ(τ ). (5) Substitution into the time evolution equation of the density matrix, Eq. 2, yields dρnm(t) dt = −iLsysρnm(t)+X n′m′ [γnn′m′mρn′m′(t) + γm′mnn′ρn′m′(t) − γnn′n′m′ρm′m(t) − γn′mm′n′ρnm′(t)] . (6) At this stage, it is worthwhile to consider the correlation matrix γ in more detail. First of all, the elements γnnnn and γnmnm(n 6= m) are simply the variances of respectively the transition energy fluctuations and the interaction fluctuations. All other elements of γ correspond to correlations between different fluctuations. Secondly, the correlation matrix fulfills a number of symmetries, due to the requirement of a Hermitian Hamiltonian and a set of trivial index permutation invariances, γabcd = γcdab = γbacd = γabdc. (7) We observe that there are generally three types of correlations, namely between fluctuations in • excitation energies (a = b, c = d), • excitation energies and interactions (a = b, c 6= d or vice versa), • interactions (a 6= b, c 6= d). A number of studies have included the first type of correlation, i.e. spatially correlated transition energy fluctuations, which have been shown to lead to appreciable changes in the transfer rates and efficiency of light-harvesting systems.18,22–24 However, the second type of correlation has only been studied in the dimer,41 while the third type of correlation has to the best of our knowledge not been included at all in studies of EET in photosynthetic 6 complexes. The central concern of this study is to provide an understanding of the relevance of such correlations to the time evolution of the various density matrix elements and the overall quantum efficiency. Besides the above symmetry considerations, additional restrictions apply to the various correlations in order to describe a physical system.41 This becomes clear when one considers the expectation value of the difference between two fluctuation elements squared, which should obviously give a positive result, (cid:10)(Vnm − Vn′m′)2(cid:11) ∝ γnmnm + γn′m′n′m′ − 2γnmn′m′ ≥ 0. (8) By considering such inequalities for various values of the indices, one finds a number of consistency conditions that need to be fulfilled, γnmnm ≥ 0 2 γnmnm′ ≤ γnmnm + γnm′nm′, (9) (10) where the indices n,m and m′ are allowed to be equal. In addition, the magnitude of the cross-correlation between two fluctuating matrix elements is limited by the magnitude of the original fluctuations. It is straightforward to show that γnmn′m′ ≤ √γnmnmγn′m′n′m′, (11) where the equality holds when the fluctuations of the Hamiltonian matrix elements Vnm and Vn′m′ are fully (anti-)correlated. These limitations on the allowed fluctuation correlations should be kept in mind when considering the effect of including the various additional correlations. C. The Fenna-Matthews-Olson complex and light harvesting efficiency The model light harvesting system for which we want to quantitatively probe the ef- fects of correlations in the environmentally induced fluctuations is the Fenna-Matthews- Olson (FMO) complex. FMO is the photosynthetic complex of the green sulfur bac- terium Chlorobium Tepidum, consisting of three weakly coupled, identical subunits of seven bacteriochlorophyll-a (BChl) molecules each.6–9,24,30,31,42,43 Within such a subunit, after ab- sorption of the incoming light, the excitation energy is transferred to the reaction center 7 where charge separation can take place. Due to their proximity to the primary light har- vesting antennae, energy typically flows into the system at BChl's 1 and 6.30 We model a subunit of the FMO complex by the following experimental Hamiltonian in the site basis (in units of cm−1),31,43 H =   6 −8 −4 1 2 8 −5 28 6 13 0 −62 −1 −9 280 −106 −106 420 28 17 6 −62 175 −70 −19 −57 −1 −70 320 2 40 −2 13 −9 −19 40 32 360 1 17 −57 −2 8 −5 6 −8 −4 32 260   . (12) The corresponding exciton states, which we label in order of increasing energy, are given in Appendix A. Of particular relevance for our numerical results are the exciton states s = 3 and s = 7 with energies of respectively E3 = 224 cm−1 and E7 = 480 cm−1, which are the exciton states mostly associated with BChl's 1 and 2, and the lowest energy exciton state s = 1 with energy E1 = −24 cm−1 which is localized mostly on BChl 3, from where energy will be trapped to the reaction center. In order to describe the influence of the various correlations on the light harvesting efficiency, we introduce two competing decay channels.18,44,45 Exciton decay, leading to an irreversible loss of the absorbed energy, is described by adding a decay term L(dec) nm = (cid:0)k(d) n + k(d) m (cid:1) /2, (13) where k(d) n is the exciton decay rate at chromophore n. In the FMO complex, energy is transferred to the reaction center, and we model capture of the excitation by the trap through a trapping term L(trap) nm = (cid:0)k(t) n + k(t) m (cid:1) /2, (14) where k(t) n is the exciton trapping rate at chromophore n. In the FMO complex, the reaction center is located close to BChl 3, and we will only include trapping from that particular chromophore, k(t) n = k(t)δn,3. The efficiency of the process can be defined as the branching ratio between energy trapped 8 at the reaction center and energy lost through exciton decay,18,44,45 q = Pn k(t) n τn n τn + Pn k(d) n τn Pn k(t) , (15) where τn = R ∞ 0 dtρnn(t) is the mean residence time at site n. III. NUMERICAL RESULTS FOR THE FMO COMPLEX As stated before, the FMO complex consists of seven coupled BChl's. In the follow- ing sections, we will first discuss the effects of uncorrelated interaction fluctuations, and subsequently proceed with a study of the effect of correlations between the various possi- ble fluctuations. Since many of the interactions are already small to begin with, we only consider fluctuations in the strongest interactions. Likewise, we only consider correlations involving the strongest interactions, i.e. interactions between BChl's of a magnitude exceed- ing 25 cm−1, which are shown in Fig. 1. In addition, we only include correlations between site energy fluctuations of site n and interactions involving that same site, and similarly, correlations between interaction fluctuations that have one site in common. This choice is motivated by the fact that correlations between fluctuations are predominantly caused by shared local environments. The numerical calculations reported in the upcoming sections concern energy flowing into the FMO complex at BChl 1. As we label the exciton states s in order of increasing energy, this corresponds to initially exciting exciton states s = 3 and s = 7. One of our goals is to in- vestigate the effect of including energy-interaction fluctuation correlations on the experimen- tally observed oscillations of the exciton populations.7 Therefore, we choose an initial excita- tion where exciton coherence is already initially present, ψ(0)i = (cid:0)s = 3i + eiθ s = 7i(cid:1) /√2. Obviously, different values of the mixing angle θ correspond to different initial populations Pn of BChl's 1 and 2. In particular, θ = 0 implies P1 = 0.08 and P2 = 0.89, θ = π/2 yields P1 ≈ P2 = 0.49, and for θ = π we have P1 = 0.90 and P2 = 0.08. Additional calculations have been performed for energy flowing in from the other side, that is, from BChl 6. This corresponds (mostly) to initially exciting exciton states s = 5 and s = 6. The results do not change qualitatively, except for a considerable decrease in population oscillation frequency due to the smaller energy difference between exciton states s = 5 and s = 6 as compared to exciton states s = 3 and s = 7. For future reference, the 9 FIG. 1: A schematic view of the FMO complex, with the strongest inter-BChl interactions denoted by arrows. The excitation will typically enter the complex at either BChl 1 or BChl 6, and will flow towards the reaction center which is most closely associated with BChl 3. exciton that most strongly overlaps with the trap state BChl 3 is exciton state s = 1. The parameters considered in our simulations have previously been used in Ref. 18 to model the FMO system. We take all exciton decay rates equal at a value k(d) n = k(d) = 1 ns−1, while only trapping from BChl 3 occurs, with a trapping rate k(t) = 1 ps−1. Furthermore, we take γ0 = γnnnn = 94 cm−1 for the transition energy fluctuations; note that there is a factor of 2 difference in the definition of γnnnn as compared to the quantity Γ in Ref. 18. A. Uncorrelated interaction fluctuations The effect of interaction fluctuations can be investigated straightforwardly with the cur- rent formalism, by introducing nonzero values for the fluctuations of the off-diagonal Hamil- tonian matrix elements, γnmnm ≡ γ1. As stated in the previous section, only fluctuations of the strongest interactions, shown in Fig. 1, are included. Note that, since γnmnm is the expectation value of a squared quantity, it is necessarily positive. In the original works by Haken, Strobl and Reineker,25,26 interaction fluctuations were already considered, and 10 also a recent paper by Chen and Silbey41 showed that interaction fluctuations can have a pronounced effect on exciton dynamics in the dimer. This makes it all the more surprising that this effect is commonly neglected in recent studies concerning the FMO complex. The time evolution of the population of exciton state s = 3 and the (real part of the) coherence between exciton states s = 3 and s = 7 is shown in Fig. 2, clearly showing that their time evolution strongly depends on the presence of interaction fluctuations. ] ) t ( 7 3 ρ [ e R , ) t ( 3 3 ρ increasing γ 1 0.6 0.5 0.4 0.3 0.2 0.1 0 −0.1 0 0.1 0.2 0.3 0.4 0.5 t (ps) FIG. 2: Time evolution of the exciton population ρ33 (solid lines) and the real part of the exciton coherence ρ37 (dashed lines), with initial phase θ = 0, for γ0 = 94 cm−1 and γ1 = 0, 1, 2, 5, 10, 20 cm−1. Note the appreciable increase in both transfer rate and decoherence rate with γ1. Analysis of the temporal evolution of the exciton populations and coherences, in particular the ones shown in Fig. 2 and estimating the rates by assuming simple exponential decay and looking at times where the oscillations have died out, reveals that the population transfer rate and decoherence rate both increase approximately linearly with the interaction fluctuations γ1. This is to be expected from the general form of the time evolution equations, Eq. 6. While these full equations contain a very large amount of terms, all couplings between the density matrix elements are proportional to either γ0 or γ1. The population transfer rate, in particular, contains only a small contribution from γ0 terms for these parameters, and 11 therefore the population transfer is increased dramatically upon introduction of interaction fluctuations. Note that this behavior is consistent with previous results for the dimer,40,41 where the exciton population transfer rate for a strongly detuned dimer is to a very large extent dominated by the interaction fluctuations. This is analogous to the situation in the FMO complex, where the differences between the BChl transition energies are also typically large compared to the interactions. More specifically, for a dimer detuned by an energy difference ∆E and having an interaction strength J, the mixing angle θ′ defined by tan θ′ = 2J/∆E is small. It can analytically be shown40,41 that the population transfer rate Γ between the two exciton states is given by Γ = γ0 sin2 θ′ + 2γ1 cos2 θ′. However, since the first term is typically very small for strongly detuned dimers (i.e., ∆E ≫ J), the population transfer rate is fully dominated by the second term as soon as interaction fluctuations are present. An analysis of the population evolution shown in Fig. 2 confirms that the population transfer rates in FMO are indeed also proportional to γ1. Thus, interaction fluctuations lead to dramatic increases in population transfer rates, in turn enhancing the efficiency of the FMO complex. To further illustrate the increase in population transfer rates with increasing interaction fluctuations, Fig. 3 shows the harvesting time as a function of the interaction fluctuation amplitude γ1. In Fig. 3, we consider the energy fluctuation magnitude γ0 that optimizes the harvesting time, γ0 = 94 cm−1, and an increase respectively decrease of one order of magnitude. Note that increasing γ1 leads to more efficient light harvesting due to a corre- sponding increase in population transfer within the FMO complex; this holds for different values of the energy fluctuations γ0 and for different initial conditions. Additionally, while the optimal dephasing value γ0 does not change with γ1, the optimum becomes considerably less deep, making the FMO complex more robust to variations in the dephasing rate γ0. For very large, in fact unphysically large, interaction fluctuations, the efficiency of the light har- vesting complex becomes independent of the energy fluctuation magnitude. This situation corresponds to large population transfer rates, dominated by the interaction fluctuations γ1, and subsequent rapid redistribution of population over all BChl's. The harvesting time will in that case converge to τ = 7 ps, corresponding to a trapping rate kt = 1 ps−1 from BChl 3 where at all times 1/7th of the total untrapped population will reside. 12 ) s p ( τ 18 16 14 12 10 8 6 10−2 γ =9.4 cm−1 0 γ =94 cm−1 0 γ =940 cm−1 0 100 γ (cm−1) 1 102 FIG. 3: Dependence of the harvesting time on the interaction fluctuation magnitude γ1, for various values of the energy fluctuations γ0. Notice the decrease in harvesting time with γ1 as a result of increasing population transfer rates. B. Effect of correlated energy-interaction fluctuations 1. Parameter set 1 Here, we consider γ0 = γnnnn = 94 cm−1 which corresponds to parameters close to the optimal energy transport conditions reported in Ref. 18. The magnitude of the allowed correlations is limited by Eqs. 9; to allow for appreciable amounts of correlation, we take a large interaction fluctuation value γ1 = γnmnm = 40 cm−1 (n 6= m), with the same provisions as before. In Sec. III B 2, we discuss the effect of correlated energy and interaction fluctuations for a smaller, more realistic value of the interaction fluctuation magnitude γ1. As stated in Sec. III, we only include correlations that involve the strongest interactions (see Fig. 1) and we focus on correlations between interaction fluctuations and energy fluctuations of the BChl's involved in that particular interaction; γn′n′nm ∝ (δn′n + δn′m), n 6= m. First of all, taking all energy-interaction fluctuation correlations equal in sign and mag- 13 nitude leads to negligible effects. This is consistent with a previous study by Chen and Silbey, where it was shown that in a dimer the relevant quantity is the difference between the two transition energy-interaction fluctuation correlations.41 While the full time evolu- tion equations for the more complicated FMO system are more involved, a similar result is observed to hold. This is not surprising, as fluctuation correlations of equal sign would imply that a change of interaction would mean that both energies involved change in the same direction, leaving the energy difference unchanged. Therefore, we choose the various energy-interaction correlations of equal magnitude, but with alternating sign; that is, we take γ1112 = −γ2212 = γ2223 = −γ3323 ≡ γ2 and so on. There is an ambiguity near BChl 4, as it strongly interacts with three other BChl's and it is therefore necessary to choose two out of γ4434, γ4454 and γ4474 with equal sign. We take γ4434 = γ4474 = −γ4454 = −γ2; other choices result only in inconsequential numerical changes. A few observations can be made regarding these plots. First of all, energy-interaction fluctuation correlations affect both the energy transfer rates and the coherence decay rates. Depending on the sign of the correlations, both increases and decreases of the various rates can occur. This can be observed in Fig. 4, where the population of exciton state s = 3 shows appreciable differences in transfer rate to other exciton states for different correlation magnitudes. While this particular exciton state and initial condition shows an increase in the transfer rate upon increasing correlations, this is not general; a decrease in transfer rate may be obtained for different exciton states and initial conditions. Secondly, it is clear that these parameters lead to a quick redistribution of population over all chromophores, only showing oscillatory features on a short time scale. Therefore, in the next subsection, we consider the same quantities for different parameters, where the effect on population oscillations can be observed more clearly. It is also possible to investigate the dependence of the harvesting efficiency, here quantified through the average trapping time τ , on the magnitude of the energy-interaction fluctuation correlation γ2. This is shown in Fig. 5; the effects are rather limited, with changes of at most a few percent. Both the overall effect of accounting for these correlations and the dependence on the initial condition, defined by the mixing angle θ, is rather weak. This is related to the fact that we are close to the conditions that constitute optimal harvesting behavior (i.e., γ0 = 94 cm−1), which is corroborated by the harvesting times which are only slightly above τ = 7 ps. In other words, the bottleneck in the overall harvesting process is 14 0.5 0.4 0.3 0.2 increasing γ 2 ) t ( 3 3 ρ 0.1 0 0.1 0.2 0.3 0.4 0.5 t (ps) FIG. 4: Time evolution of the exciton population ρ33, θ = 0, with γ0 = 94 cm−1 and γ1 = 40 cm−1. The considered correlation values are γ2 = −40,−20, 0, 20, 40 cm−1. not population transfer between the BChl's, but the trapping of excitations from BChl 3 into the reaction center. Therefore, the effect of limited changes in the population transfer rates have relatively little effect on the overall efficiency. Note that negative values of γ2 may give a small enhancement of the harvesting time; depending on the initial conditions, a nonzero optimal choice for the correlations may be made such that light harvesting is at its fastest. Again, we note that the fluctuations considered here lead to a rapid decay of the co- herences. In the next section, we will consider parameters that lead to slower decoherence and longer lived population oscillations, in order to evaluate the effect of energy-interaction fluctuation correlations on population oscillations. 15 θ=0 θ=π/2 θ=π 7.3 7.25 7.2 7.15 7.1 ) s p ( τ 7.05 −40 −30 −20 −10 10 0 γ (cm−1) 2 20 30 40 FIG. 5: Dependence of the harvesting time on the magnitude of the energy-interaction fluctuation correlations, for various initial conditions and with γ0 = 94 cm−1 and γ1 = 40 cm−1. Note that negative values of γ2 are required for enhancement of the efficiency, and that the optimal correlation value depends on the initial condition. 2. Parameter set 2 In order to clearly observe the effect of additional energy-interaction fluctuation correla- tions on population oscillations, it is necessary to switch to parameters where the decoherence occurs on a slower timescale. Here, we consider γ0 = γnnnn = 10 cm−1, and γ1 = γnmnm = 2 cm−1 (n 6= m) for the interactions shown in Fig. 1. The same observations as before hold with regards to the signs of the correlations: alternating signs are required for observable effects and are used throughout this section, while a negative γ2 can produce enhancement of the transport efficiency. Again, as an example we consider the time evolution of the population of the exciton states s = 3 and initial phase θ = 0 for various values of γ2, shown in Fig. 6. It can now be clearly seen that the populations oscillate in time, an effect which is induced 16 increasing γ 2 0.5 0.45 0.4 0.35 ) t ( 3 3 ρ 0.3 0 0.1 0.2 0.3 t (ps) 0.4 0.5 0.6 FIG. 6: Time evolution of the exciton population ρ33, θ = 0 and with γ0 = 10 cm−1 and γ1 = 2 cm−1. We consider the values γ2 = −4,−2, 0, 2, 4 cm−1. by the coherence that is present between the two exciton states. Indeed, the period of the oscillations corresponds to the energy difference between the exciton states, and is identical to the coherence oscillation period, which corresponds to τosc = 130 fs. A clear augmentation of the amplitude of the oscillations can be observed when one increases the magnitude of γ2. The introduction of energy-interaction fluctuation correlations can thus enhance the population oscillations, which are driven by the presence of coherence. As before, we also observe a change in population transfer rates, with in particular a strong suppression of population transfer with increasing γ2. The dependence of the harvesting efficiency on the correlation magnitudes can be calcu- lated exactly as before, and is shown in Fig. 7. There is now a stronger dependence on the magnitude of the correlations, which is caused by the fact that this parameter set constitutes suboptimal conditions, so that in this case changes in the population transfer rates have a larger effect on the overall harvesting time. Also, the behavior is almost independent of the initial conditions, even more so than for the parameters in the previous section. Negative values for γ2 are required for enhancement, where the transport is now optimized by choos- ing γ2 as negative as possible. Positive values of γ2 lead to a strong increase in harvesting 17 time; this corresponds to the strong decrease in the relevant population transfer rates, as observed in for example Fig. 6. θ=0 θ=π/2 θ=π 22 20 18 16 14 12 10 ) s p ( τ 8 −4 −3 −2 −1 1 2 3 4 0 γ (cm−1) 2 FIG. 7: Dependence of the harvesting time on the magnitude of the energy-interaction fluctuation correlations, for various initial conditions and with γ0 = 10 cm−1 and γ1 = 2 cm−1. Note that negative values of γ2 are required for enhancement of the efficiency. C. Effect of correlated interaction-interaction fluctuations It is straightforward to include correlations between the various interaction fluctuations. As in the previous section, we anticipate that such correlations are strongest between in- teractions that have one molecule in common, and in addition we focus on the strongest interactions. First of all, it is important to note that while such correlations may very well be present, also these are naturally limited in magnitude in order to fulfill the consistency conditions Eqs. 9. In particular, we have (cid:10)(Vab − Vac)2(cid:11) ∝ γabab + γacac − 2γabac > 0, (16) 18 0.5 0.45 0.4 0.35 increasing γ 3 ) t ( 3 3 ρ 0.3 0.25 0.2 0 0.1 0.2 t (ps) 0.3 0.4 0.5 FIG. 8: Time evolution of the exciton population ρ33, θ = 0, γ0 = 94 cm−1, γ1 = 40 cm−1, and γ2 = 40 cm−1. The considered correlation values are γ3 = −30,−15, 0, 15, 30 cm−1. where a 6= b, a 6= c. From this, it follows that 1 2 γabac ≤ (γabab + γacac) . (17) Nevertheless, it is instructive to quantify the effects of these additional correlations. The conclusions are typically similar to those in the previous section, although the effects are typically less pronounced. For brevity, we will not show all the same figures as before, but we show a typical result in Fig. 8. Here, we again consider the parameters previously used in Sec. III B 1, with in addition an energy-interaction fluctuation correlation magnitude γ2 = 40 cm−1 with the sign conventions defined there. First of all, here too alternating signs are required to amplify the effect of interaction-interaction fluctuation correlations. We take the interaction-interaction fluctuation correlations of the form γ1223 = −γ2334 = γ3445 = γ3 et cetera. Again there is some ambiguity in the definition of the signs for the correlations around BChl 4, however, since the details of those choices hardly matter for the initial conditions we have chosen, we will not go in to this in more detail. We show time evolution plots only for θ = 0, behavior for other parameters is not significantly different. First of all, it is clear that interaction-interaction fluctuation correlations, too, lead to changes in the population transfer and dephasing rates, as exemplified by the population evolution of 19 θ=0 θ=π/2 θ=π ) s p ( τ 7.5 7.4 7.3 7.2 −30 −20 −10 10 20 30 0 γ (cm−1) 3 FIG. 9: Dependence of the harvesting time on the magnitude γ3 of the interaction-interaction fluctuation correlations, for various initial conditions. We have chosen γ0 = 94 cm−1, γ1 = 40 cm−1 and γ2=40 cm−1. The different offsets of the curves are consistent with the observed θ- dependence of the harvesting time in Fig. 5. exciton state 3 shown in Fig. 8. In addition, while there is a small change in the size of the populations, the amplitude of the oscillations is influenced by the interaction-interaction fluctuation correlation magnitude to only a small extent. These conclusions are observed to hold for different sets of parameters and initial conditions. Fig. 9 shows the dependence of the harvesting time on the magnitude of the interaction- interaction fluctuation correlations. Again, there is an optimum for some nonzero value of the correlation magnitude γ3; the exact value of the optimal magnitude for the correlations depends on the initial conditions and on the choice for the energy-interaction correlations. The behavior for other values of the initial phase factor θ is very similar, implying that the transfer rates and other relevant parameters for trapping will only depend weakly on possible interaction-interaction fluctuation correlations. While Fig. 9 shows that the effect of interaction-interaction fluctuation correlations is roughly of the same magnitude as the effect of transition energy-interaction fluctuation correlations, this is only the case if the former occurs in conjunction with the latter. The changes induced by interaction-interaction fluctuation correlations for the case of γ2 = 0 cm−1 are considerably smaller. Finally, as was previously the case in Sec. III B, a more suboptimal choice of parameters leads to a 20 larger effect of changes in transfer rates, and will thus also amplify the effect of including interaction-interaction fluctuation correlations to an extent. IV. CONCLUSIONS Since in principle environmental changes will induce fluctuations of both transition en- ergies and interactions that should be correlated to some extent, we have performed a study of the effect of such fluctuations and their possible correlations on the energy transfer and excitation dynamics in photosynthetic complexes, specifically focusing on the Fenna- Matthews-Olson complex as a model system. The inclusion of interaction fluctuations and correlations between transition energy fluctuations and interaction fluctuations, and between different interaction fluctuations, has been studied by a straightforward generalization of the Haken-Strobl-Reineker model. These additional correlations have been shown to be natu- rally limited in magnitude, in order to correspond to a physically consistent system. Interaction fluctuations on the excitation dynamics in photosynthetic complexes can in general not be neglected. Not only will these in general occur, but it has been shown that even small values of interaction fluctuations can already lead to an appreciable increase in transfer rates and a corresponding increase in efficiency of the FMO complex. In particu- lar, for suboptimal values of the energy fluctuations, interaction fluctuations can lead to a considerable optimization of the light harvesting process, even to such an extent that the transfer rates and the eventual efficiency is dominated by interaction fluctuations and not by energy fluctuations. The presence of interaction fluctuations may thus make the efficiency of the FMO complex more robust to changes in the dephasing rate. Generally, for various initial conditions and energy fluctuation magnitudes, an increase in the efficiency of the FMO complex is observed with increasing interaction fluctuation amplitude. When including correlated transition energy fluctuations and interaction fluctuations, it is first of all possible to observe somewhat enhanced population oscillations, driven by the coherence between the initially excited exciton states. Depending on the signs of the various correlations, a suppression of the oscillations can also occur. In addition, one sees that the transfer rates between exciton states is modified, which in turn leads to accompanying changes in the overall efficiency of the photosynthetic complex. The overall effects on the efficiency are limited in scope when one considers parameter values that approximately 21 correspond to an optimal functioning of the light harvesting complex. In that case, not the excitation transfer but the trapping into the reaction center is the bottleneck, so that changes in transfer rates will only change the harvesting time by up to a few percent. For less optimal conditions, the effects can be considerably larger, as the excitation transfer processes play a larger role in determining the overall harvesting time. Typically, a small increase in the light harvesting efficiency can be obtained by using a nonzero amount of correlation between the aforementioned fluctuations. By the same approach, one can quantify how correlations between the various interaction fluctuations modify previously obtained results. The effect of these additional correlations turns out to be qualitatively very similar to that of transition energy-interaction fluctuation correlations, and are quantitatively of a comparable magnitude. The net effect of such correlations is limited to small changes in the scattering rate and resultant changes in the overall efficiency. This study thus suggests that one should not neglect interaction fluctuations in models describing the energy transfer in photosynthetic complexes. Correlations may also play a relevant role in the excitation dynamics, and this may (but does not necessarily) translate into appreciable changes in the efficiency. Acknowledgments The authors thank Dr. Jianlan Wu, Dr. Xin Chen and Dr. Jianshu Cao for useful discussions. This work is supported by ARO under grant W911NF-09-0480, and DARPA grant N66001-10-1-4063. Appendix A: Exciton states and energies in the FMO complex The exciton states are found by diagonalization of the Hamiltonian matrix (12); their coefficients and energies are given in the table below. 22 BChl 1 BChl 2 BChl 3 BChl 4 BChl 5 BChl 6 BChl 7 Es (cm−1) s=1 0.046 0.076 -0.940 -0.321 -0.066 -0.032 -0.005 s=2 -0.066 -0.036 0.285 -0.786 -0.319 0.064 -0.435 s=3 s=4 s=5 0.877 0.461 0.089 -0.011 -0.090 0.043 -0.019 0.013 0.000 -0.138 0.340 0.254 0.268 -0.854 0.040 0.098 0.073 -0.265 0.712 -0.629 -0.103 s=6 -0.080 0.111 -0.034 0.304 -0.560 -0.710 -0.264 s=7 -0.465 0.871 0.043 -0.007 0.032 0.144 0.038 -24 139 224 276 311 408 480 1 H. van Amerongen, L. Valkunas, and R. van Grondelle, Photosynthetic Excitons, World Scien- tific: Singapore (2000). 2 V. Sundstrom, T. Pullerits, and R. van Grondelle, J. Phys. Chem. B 103, 2327 (1999). 3 H. Sumi, J. Phys. Chem. B 103, 252 (1999). 4 K. Mukai, S. Abe, and H. Sumi, J. Phys. Chem. B 103, 6096 (1999). 5 G. McDermott, S. M. Prince, A. A. Freer, A. M. Hawthornthwaite-Lawless, M. Z. Papiz, R. J. Cogdell, and N. W. Isaacs, Nature 374, 517 (1995). 6 T. Brixner, J. Stenger, H. M. Vaswani, M. Cho, R. E. Blankenship, and G. R. Fleming, Nature 434, 625 (2005). 7 G. S. Engel, T. R. Calhoun, E. L. Read, T.-K. Ahn, T. Mancal, Y.-C. Cheng, R. E. Blankenship, and G. R. Fleming, Nature 446, 782 (2007). 8 H. Lee, Y.-C. Cheng, and G. R. Fleming, Science 316, 1462 (2007). 9 G. Panitchayangkoon, D. Hayes, K. A. Fransted, J. R. Caram, E. Harel, J. Wen, R. E. Blanken- ship, and G. S. Engel, Proc. Natl. Acad. Sci. USA 107, 12766 (2010). 10 Th. Forster, Ann. Phys. 437, 55 (1948). 11 M. K. Grover and R. Silbey, J. Chem. Phys. 52, 2099 (1970); M. Grover and R. Silbey, J. Chem. Phys. 54, 4843 (1971). 12 V. M. Kenkre and R. S. Knox, Phys. Rev. B 9, 5279 (1974); Phys. Rev. Lett. 33, 803 (1974). 13 K. Gaab and C. Bardeen, J. Chem. Phys. 121, 7813 (2004). 14 M. Mohseni, P. Rebentrost, S. Lloyd, and A. Aspuru-Guzik, J. Chem. Phys. 129, 174106 (2008). 15 M. B. Plenio and S. F. Huelga, New J. Phys. 10, 113019 (2008). 23 16 A. Ishizaki and G. R. Fleming, Proc. Natl. Acad. Sci. USA 106, 107255 (2009). 17 P. Rebentrost, M. Mohseni, I. Kassal, S. Lloyd, and A. Aspuru-Guzik, New J. Phys. 11, 033003 (2009). 18 J. Wu, F. Liu, Y. Shen, J. Cao, and R. J. Silbey, New J. Phys. 12, 105012 (2010). 19 P. Rebentrost, R. Chakraborty, and A. Aspuru-Guzik, J. Chem. Phys. 131, 184102 (2009). 20 J. Roden, A. Eisfeld, W. Wolff, and W. Strunz, Phys. Rev. Lett. 103, 058301 (2009). 21 X. Chen and R. J. Silbey, J. Phys. Chem. B, in press, doi: 10.1021/jp111068w 22 J. Strumpfer and K. Schulten, J. Chem. Phys. 134, 095102 (2011). 23 F. Caruso, A. W. Chin, A. Datta, S. F. Huelga, and M. B. Plenio, J. Chem. Phys. 131, 105106 (2009). 24 J. Adolphs and T. Renger, Biophys. J. 91, 2778 (2006). 25 H. Haken and P. Reineker, Z. Phys. 249, 253 (1972). 26 H. Haken and G. Strobl, Z. Phys. 262, 135 (1973). 27 C. Olbrich, T. L. C. Jansen, J. Liebers, M. Aghtar, J. Strumpfer, K. Schulten, J. Knoester, and U. Kleinekathofer, J. Phys. Chem. B (2011). 28 S. Shim, P. Rebentrost, S. Valleau and A. Aspuru-Guzik, arXiv:1104.2943v1. 29 D.F. Coker, private communications. 30 Y.-F. Li, W. Zhou, R. E. Blankenship, and J. P. Allen, J. Mol Biol. 271, 456 (1997). 31 S. I. E. Vulto, M. A. de Baat, R. J. W. Louwe, H. P. Permentier, T. Neef, M. Miller, H. van Amerongen, and T. J. Aartsma, J. Phys. Chem. B 102, 9577 (1998). 32 R. E. Fenna and B. W. Matthews, Nature 258, 573 (1975). 33 J.M. Olson, Biochim. Biophys. Acta 594, 33 (1980). 34 R. E. Blankenship, D. C. Brune and B. P. Wittmershaus, Chlorosome antennas in green pho- tosynthetic bacteria. In: S.E. Stevens, Jr and D.A. Bryant, Editors, Light Energy Transduction in Photosynthesis: Higher Plant and Bacterial Models, Am. Soc. Plant Physiol, Rockville, MD (1988), pp. 32-46. 35 A. S. Davydov, Theory of Molecular Excitons, Plenum: New York (1971). 36 V. M. Agranovich and M. D. Galanin, in Electronic Excitation Energy Transfer in Condensed Matter, edited by V. M. Agranovich and A. A. Maradudin, North-Holland: Amsterdam (1982). 37 J. von Neumann, Gottinger Nachrichten, 245 (1927); Mathematical Foundations of Quantum Mechanics, Princeton University Press: Princeton, NJ (1955). 24 38 K. Blum, Density Matrix Theory and Applications, Plenum: New York (1981). 39 V. May and O. Kuhn, Charge and Energy Transfer Dynamics in Molecular Systems, Wiley- VCH: Berlin (2000). 40 R. Wertheimer and R. Silbey, Chem. Phys. Lett. 75, 243 (1980); J. Chem. Phys. 74, 686 (1981). 41 X. Chen and R. J. Silbey, J. Chem. Phys. 132, 204503 (2010). 42 U. Ermler, G. Fritzsch, S. K. Buchanan, and H. Michel, Structure 2, 925 (1994). 43 M. Cho, H. M. Vaswani, T. Brixner, J. Stenger, and G. R. Fleming, J. Phys. Chem. B 109, 10542 (2005). 44 T. Ritz, S. Park, and K. Schulten, J. Phys. Chem. B 105, 8259 (2001). 45 J. S. Cao and R. J. Silbey, J. Phys. Chem. B 112, 12867 (2008). 25
1610.07784	1	1610	2016-10-25T08:35:55	Robustness of flight leadership relations in pigeons	[ "physics.bio-ph", "q-bio.QM" ]	Collective animal movements produce spectacular natural phenomena that arise from simple local interactions among group members. Flocks of homing pigeons, Columba livia, provide a useful model for the study of collective motion and decision making. During homing flights, flock members are forced to resolve potentially divergent navigational preferences in order to stay together and benefit from flying in a group. Recent work has demonstrated that some individuals consistently contribute more to the movement decisions of the flock than others do, thereby generating stable hierarchical leader/follower networks. Yet, what attributes of a flying pigeon reliably predict leadership remains an open question. We examined the flexibility of an individual's hierarchical leadership rank (i.e. its ordinal position when flock members are ranked according to the average time differences with which they lead or follow others) as a function of changes in its navigational knowledge. We manipulated already established hierarchical networks in three different flocks, by providing certain individuals with additional homing experience. We found that such training did not consistently lead to an increase in birds' leadership ranks, and that, in general, the nature of leader/follower interactions between trained and untrained birds remained unaffected. Thus, leadership hierarchies in pigeon flocks appear resistant to changes in the navigational knowledge of a subset of their members, at least when these changes are relatively small. We discuss the implications of our results in light of the potential benefits of structural stability in decision-making networks.	physics.bio-ph	physics	Introduction A flock of birds circling over its roosting site is a magnificent aerial display. Theoretical work suggests that these highly synchronised and coordinated movements arise from simple interaction rules, without the need for centralised organisation (Vicsek et al. 1995; Couzin et al. 2002; Vicsek & Zafeiris 2012). Nonetheless, we are only just beginning to understand how rules implemented in models relate to those applied by animals. Progress in digital image processing and high temporal resolution tracking has allowed the inference of interaction rules in bird and fish species (e.g., Ballerini et al. 2008; Lukeman et al. 2010; Herbert-Read et al. 2011; Katz et al. 2011). Furthermore, in line with researchers' increasing interest in the role of inter-individual differences in shaping interactions (Conradt et al. 2009; Nakayama et al. 2012a), it has been found that homing pigeon flocks are hierarchically organised, where individuals contribute with different weights to the movement decisions of the flock (Nagy et al. 2010). Such hierarchical networks consist of transitive leader-follower relationships in which birds consistently copy the directional choices of individuals above them in the hierarchy, while being copied by those lower in rank. Little is known about what attributes of a flying pigeon can reliably predict leadership in flocks, although it has been suggested that leadership may be related to individual navigational efficiency (Nagy et al. 2010). Empirical studies have identified a variety of traits (e.g., age, experience, social rank, and motivation; McComb et al. 2011; Reebs 2000; King et al. 2008; Nakayama et al. 2012b) that can modify an individual's behaviour towards conspecifics. With respect to the context of collective motion, recent work has demonstrated that navigationally less experienced birds are likely to follow more experienced conspecifics (Flack et al. 2012). More specifically, the larger the difference in homing experience between two partners, the higher the likelihood that the more experienced bird will emerge as leader. Additionally, in highly experienced birds the accuracy with which individuals recapitulate previously established idiosyncratic routes when flying solo has been suggested to predict relative influence when flying in pairs (Freeman et al. 2011), suggesting that some aspect of navigational certainty (or perhaps inflexibility) may promote leadership. These findings raise new questions about how variations in navigational knowledge possessed by individual members influence group dynamics in pigeon flocks. If a bird's position in the hierarchy correlates positively with its own navigational experience, we should be able to manipulate the network by providing selected individuals with the opportunity to acquire additional spatial knowledge. Here we evaluate whether it is indeed possible to alter individuals' ranks attained during flock homing flights by providing them with additional homing experience before re-testing them with their group mates. Methods Subjects and experimental procedure We used 30 adult homing pigeons (Columba livia) bred at the Oxford University Field Station at Wytham (51°46'58.34''N, 1°19'02.40''W). They were kept in a social group of ca. 120 pigeons inside two lofts. Birds normally had free access to the outside, except on the days when the experiments were conducted. Food (a commercially available multigrain mixture), water, minerals and grit were provided ad libitum throughout the study. All experimental birds were between 4 and 8 years old, and had homing experience but had never visited the release site used in the current study. They were trained to carry miniature GPS logging devices (see below) attached to their back by a small Velcro strip glued to clipped feathers. All releases were performed from Radford (distance and direction to home: 15.7 km, 151°, respectively). The experiment had three phases. First, we trained three flocks of 10 birds (designated groups A, B and C), by releasing all 10 birds of a flock simultaneously at the release site (Phase I: group training). Each flock performed eight group training flights, with a maximum of two releases per day. We then calculated for each group a leadership hierarchy among flock members using the methods described in Nagy et al. (2010). In Phase II (solo training), we allowed three randomly chosen individuals from each flock to gain additional homing experience by performing 10 individual flights from the same site (one of these nine birds was lost during its 8th individual training flight, and therefore did not participate in the third phase for group C). Finally, in Phase III (group tests), we released each original flock six more times in order to evaluate any changes in the hierarchy's structure – in particular, whether the additional homing experience resulted in any changes in the ranks attained by the three individuals that had received additional solo training. Phase I was completed in 10 days, Phase II in 6 days, and Phase III in 3 days, with releases conducted on all consecutive days when weather conditions were favourable (dry and with winds<7 ms-1). GPS device and data handling The GPS device was based on a commercially available product (Gmsu1LP, from Global Top), weighed 13 g, and was capable of logging time-stamped longitude, latitude and altitude data at 10 Hz. The geodetic coordinates provided by the GPS were converted into x, y and z coordinates using the Flat Earth model. These coordinates were smoothed by a Gaussian filter (σ=0.2 s), and we used a cubic B-spline method to fit curves onto the points obtained with the 0.1 s sampling rate. Only the x and y coordinates were used for analysis (the number of data points recorded per bird is shown in Table S1 of the Supplementary Material). In independent tests, using the devices in fixed relative positions to each other, the deviation between real and measured distance was 0.00 ± 0.34 m (mean ± S.D.). This degree of accuracy is sufficient for calculating directional correlation delay functions that characterise relations among the birds' motion (see Supplementary Figure S1 and Supplementary Methods for further details). Data analysis To evaluate the effect of training on homing performance, we calculated homing efficiency and homing time for each flight. Efficiency was measured by dividing the straight-line distance between the release site and the loft with the actual distance travelled by the bird to reach home. Homing time was the length of time that elapsed between release and the bird reaching a radius of 250 m from the loft. It should of course be noted that the two measures are not independent of each other, although the relationship between them can vary to some extent as a function of the bird's speed. In addition, to measure the trained birds' change in homing performance, we calculated the difference in efficiency and homing time between the average of the first two and the average of the last two solo training flights in Phase II. To determine leader-follower relations inside the flock, we calculated the directional correlation delay for each pair of birds i and j (i ≠ j). The directional correlation delay of a pair is 𝐶𝑖𝑗(𝜏) = ⟨𝑣⃗𝑖(𝑡) ∙ 𝑣⃗𝑗(𝑡 + 𝜏)⟩𝑡, where 𝑣⃗𝑖(𝑡) is the normalised velocity of bird i at time t and 𝑣⃗𝑗(𝑡 + 𝜏) is the normalised velocity of bird j at time t + τ. Note that 𝐶𝑖𝑗(𝜏) = 𝐶𝑗𝑖(−𝜏). We then determined the maximum value of the 𝐶𝑖𝑗(𝜏) correlation function at 𝜏𝑖𝑗 ∗ , 𝐶𝑖𝑗(𝜏𝑖𝑗 ∗ ). We identified the corresponding 𝜏𝑖𝑗 ∗ as the directional correlation delay time. 𝜏𝑖𝑗 ∗ values focus on the relationship between specific pairings of individuals while ignoring hierarchy changes caused by other flock members. Note also that 𝜏𝑖𝑗 ∗ =−𝜏𝑗𝑖 ∗ . Negative 𝜏𝑖𝑗 ∗ values mean that the flight directional changes of bird i fall behind that of bird j, and can thus be interpreted as a case of j leading. In order to compare relationships among flock members before and after the solo training we focused on pairwise 𝜏𝑖𝑗 ∗ values, averaged across pre- and post-training separately. For every specific pair ij, we averaged those 𝜏𝑖𝑗 ∗ values that exhibited a 𝐶𝑖𝑗(𝜏𝑖𝑗 ∗ ) larger than 0.95. Because the relationships between specific pairings are non-independent data points, we used the number of individuals as our sample size for correlations between pre- and post- training 𝜏𝑖𝑗 ∗ values. Only edges with values higher than 0.02 were retained. For the calculation of the 𝐶𝑖𝑗(𝜏) correlation function, we included only those pairs of data points from birds i and j where the two birds were a maximum of 100 m apart (i.e. dij<100 m). We chose this threshold based on the distributions of inter-individual distances (see Fig. S2). A bird's closest neighbour was less than 10 m away in 71% of all recorded data points (see inset of Fig. S2). However, to be able to detect potential interactions between more distant flock members we used a threshold of 100 m, although only few data points fall into this bin category. By averaging the 𝜏𝑖𝑗 ∗ values of bird i and the rest of the flock, we obtained a second measure, denoted 𝜏𝑖. Because of full transitivity of each hierarchy, this measure allowed us to resolve fully the hierarchical order among all group members. On two occasions, the 𝜏𝑖-value was 3.4 times higher than the standard deviation of all values (see Fig. S2); in these cases we removed the two birds from these particular flock flights and re-ran the analyses without them (see Table S2 in the Supplementary Material for the results including the outliers). We calculated for each bird the average of the 𝜏𝑖 values for the flights before (Phase I, 8 flights; 𝑝𝑟𝑒 𝜏𝑖 ) and after (Phase III, 6 flights; 𝜏𝑖 𝑝𝑜𝑠𝑡 ) the individual training period. 𝜏𝑖 values have similar properties to linear ranks (positive and negative values correspond to leading and following behaviour, respectively). Results Following the group releases of Phase I, we identified fully transitive hierarchies in each of our three flocks (Fig. 1, top row). Besides confirming the findings of Nagy et al. (2010), this initial result also provided the necessary premise for Phases II and III. Figure 1 Pre- and post-training hierarchical networks of three flocks, generated using 𝜏𝑖𝑗 ∗ values. Rectangles correspond to individual birds; trained birds are shown with black borders. The three- digit alphanumeric codes indicate in which group the subject was tested (A, B or C) and its rank during the pre-training flights. Edges indicate leader-follower relations pointing from the leader to the follower (only edges where 𝜏𝑖𝑗 ∗ ≥0.02 are shown). Edges that have the same directionality in pre- and post-training networks are indicated as thick blue lines; those that undergo a change in direction between pre- and post-training are shown as red lines; those that appear in only one of the networks are shown as dotted green lines. Numbers on edges correspond to 𝜏𝑖𝑗 ∗ . A, B, C Pre-training (left column) and post-training (right column) hierarchies of groups A, B and C, respectively. First, we evaluated the effect of the training (flock and solo flights) on homing performance, by examining homing efficiency and homing time over the course of Phases I, II and III (Fig. 2A, B). Birds improved in both measures of homing performance during the flock releases of Phase I. Furthermore, during Phase II the solo trained birds increased their efficiency by an average of 0.13 (S.D.=0.06, difference between the average of the first two and the average of the last two solo training flights in Phase II, Fig. 2C) and decreased their homing time by -343.4 s (S.D.=209.0 s, Fig. 2D). Both these changes differed significantly from zero (one- sample t-tests, efficiency: t7=5.77, P<0.001; time: t7=4.65, P=0.002). Figure 2 (A, B) Homing efficiency (mean ± S.E.M., A) and homing time (mean ± S.E.M., B) as a function of release number. Data from all groups were averaged according to release number. Grey circles indicate Phases I (N=30) and III (N=29), orange circles indicate Phase II (N=8). (C, D) Changes in homing efficiency (C) and homing time (D) during solo flights by trained individuals. Black line corresponds to mean (± S.E.M.). We next used data from Phase III to measure the stability of the hierarchies by comparing the relative ranks of the untrained birds before and after solo training (𝜏𝑖 𝑝𝑜𝑠𝑡 vs. 𝜏𝑖 𝑝𝑟𝑒 ). We found a positive correlation between 𝜏𝑖 𝑝𝑟𝑒 and 𝜏𝑖 𝑝𝑜𝑠𝑡 (Pearson's correlation, group ABC together: r=0.72, N=21, P<0.001, Fig. 3A, group A: r=0.80, N=7, P=0.030, group B: r=0.87, N=7, P=0.011, group C: r=0.69, N=7, P=0.090), which indicates the persistence of a robust hierarchical order among untrained flock members. However, the ranks of the trained birds exhibited variability: we found no correlation between 𝜏𝑖 𝑝𝑟𝑒 and 𝜏𝑖 𝑝𝑜𝑠𝑡 (Pearson's correlation, r=- 0.08, N=8, P=0.846, Fig. 3A), with some birds experiencing a rise and others a drop in 𝜏𝑖. Also, the change in the birds' relative rank did not correlate with their changes in homing performance (Pearson's correlation, efficiency: r=0.247, N=8, P=0.556; time: r=0.072, N=8, P=0.866). Figure 3 Relationship between τ before and after individual training flights (mean ± S.E.M.) (A) 𝑝𝑜𝑠𝑡 𝜏𝑖 as a function of 𝜏𝑖 𝑝𝑟𝑒 for solo-trained (orange circles) and untrained individuals (light grey circles). (B-C) averaged 𝜏𝑖𝑗 ∗ after individual training as a function of averaged 𝜏𝑖𝑗 ∗ before individual training for untrained-untrained pairings (B) and trained-untrained pairings (C). We further investigated the changes the hierarchies underwent using pairwise directional correlation time (𝜏𝑖𝑗 ∗ ). Again, we observed a positive correlation between pre- and post-training for the untrained birds (Pearson's correlation, group ABC together: r=0.72, N=21, P<0.001, Fig.3B, group A: r=0.70, N=7, P=0.083, group B: r=0.92, N=7, P=0.004, group C: r=0.62, N=7, P=0.138), further confirming the stability of their relationships over time and over repeated interactions. Moreover, pairwise 𝜏𝑖𝑇𝑗𝑈 ∗ values enable us to compare the changes in pairs consisting of a trained (𝑖𝑇) and an untrained (𝑗𝑈) individual. The relationship between trained and untrained pigeons also remained stable, as evidenced by the positive correlation between their pre- and post-training 𝜏𝑖𝑇𝑗𝑈 ∗ (Pearson's correlation, r=0.42, N=29, P=0.023, Fig. 3C). Despite the extra experience gathered by certain flock members, their positions in the hierarchy relative to untrained birds showed, on average, no improvement: the difference in directional correlation delay times in trained-untrained pairs before and after the individual training was on average 0.00 s (S.E.M.=0.01 s) and did not differ significantly from zero (one-sample t-test, t55=0.005, P=0.996). Thus, although the overall hierarchical rank of the trained individuals changed slightly, the direction of these changes was not consistent. In addition, the changes were small enough that across the flock as a whole the position of the untrained birds in relation to trained flock members remained mostly unchanged. The extra training had an even smaller effect on the positions of the untrained birds relative to each other (Fig. 1). Separate examination of the three flocks showed that in group A two of the trained birds improved their relative ranks and one maintained its position (the average change in 𝜏𝑖𝑗 ∗ before and after training, A: ∆𝜏𝑖𝑗 ∗ =0.06s (S.E.M.=0.02s), one-sample t-test, t20=3.34, P=0.003, Fig. 1A). In group B, no clear change was found (B: ∆𝜏𝑖𝑗 ∗ =0.02 s (S.E.M.=0.01 s), one-sample t-test, t20=1.77, P=0.09, Fig. 1B), whereas in group C the trained birds decreased their relative ranks C: (∆𝜏𝑖𝑗 ∗ =-0.11 s (S.E.M.=0.02 s), one-sample t-test, t13=5.30, P<0.001, see Fig. 1C). An additional statistical analysis, making use of the full dataset rather than per-bird averages as above, further confirmed the robustness of the measured hierarchies (see Supplementary Materials). Discussion Previous research has shown that group decision-making in pigeon flocks is hierarchically organised, with certain individuals consistently contributing with relatively more weight to movement decisions than others (Nagy et al. 2010). Here, we re-confirmed the existence of such hierarchical flight dynamics, demonstrating distinct leadership hierarchies in three separate flocks during repeated homing flights. Moreover, we showed that additional solo training given to specific group members did not affect the overall hierarchy of the flock: although trained birds increased their navigational efficiency during these solo flights (thus suggesting that they had gained additional navigational knowledge), this increase in efficiency was not accompanied reliably by improvement in their hierarchical position. Overall, pairwise leader-follower relations between flock members remained stable. This implies that leadership ranks within flocks do not directly relate to individual navigational experience, but that some other intrinsic property, or a combination of several properties, defines the organisation of the hierarchy. Two possible mechanisms might allow the establishment and maintenance of robust flight hierarchies. The first requires individual recognition in the air. Flock members may have fixed leader-follower relationships that are based on individual identity and are consequently maintained across multiple flights. Such relationships may be similar to those based on dominance (King et al. 2008), familiarity (Flack et al. 2013) or individual affiliations (Jacobs et al. 2011). Alternatively, hierarchies might derive from individuals reacting in consistent ways to other group members' movements, without necessarily identifying them. Each individual may respond to flockmates in a way that is defined by its own specific features and the features it perceives in others. This would allow leadership to emerge passively as a consequence of simple interaction rules (Vicsek et al. 1995; Couzin et al. 2002). In other species these responses have been described to change with experience (Reebs 2000), motivation (Nakayama et al. 2012b), age (McComb et al. 2011), or sex (Ihl & Bowyer 2011). The fact that we found no consistent effect of the extra training on birds' leadership ranks is a somewhat surprising result, given previous suggestions of the effect of navigational experience and skill on leadership (Nagy et al. 2010; Freeman et al. 2011). The trained birds' increase in experience might not have been large enough to induce changes in the organisation of the flock. Prior to the solo training, each subject had already performed eight flock homing flights and reached high, asymptotic levels of homing efficiency (Meade et al. 2005). Even though solo training did improve birds' solo homing efficiency, their advantage over the rest of the flock remained small or was only temporal. This interpretation is in agreement with past results showing that birds with more experience will more clearly emerge as leaders when the difference in experience between them and their flight partners is large (Flack et al. 2012). Future research should focus on the effect of experience while birds are still far from asymptotic levels of efficiency (e.g. with tests run after fewer homing flights for the most inexperienced birds). Furthermore, a control group in which every flock member receives extra solo training flights in Phase II would useful as a baseline measure of how flock homing efficiency changes in response to training given equally to all group members. Flack et al. (2012) tested mixed-experience pairs of pigeons and found that navigational experience had an effect on leadership, with birds that had performed more training flights more likely to emerge as leaders. In the present study, using groups of ten birds, no such effect was detected, which may indicate that influencing flockmates' movements is easier in smaller groups. Also, theoretical work by Couzin et al. (2011) showed that the presence of uninformed individuals can inhibit decisions made by a knowledgeable minority and enable the numerical majority to control movements. Investigating the potential link between group size and group dynamics – both empirically and theoretically – is a promising avenue for future research. Although flock dynamics can be observed without hierarchical organisation (Xu et al. 2012), such structure might be beneficial for establishing a "flight routine" that demands less attention from group members. The fact that hierarchies seem resistant to small changes once they are established indicates that rather than benefitting from particular features of the leader (such as navigational experience) their advantage might lie in the stability of the structure itself. Recent theoretical work has found that underlying social structures can improve the navigational accuracy of large, leaderless groups (Bode et al. 2012). Furthermore, it is suggested that hierarchical group dynamics could be based purely on social preferences (Bode et al. 2011). Social relationships can be found between relatives, familiar conspecifics or individuals of similar attributes such as size, personality or sex. Hence, the stability in our hierarchical networks may arise from preferential attachments that may have developed during training and that may not be susceptible to changes in individuals' navigational experience. Acknowledgements A.F. was supported by Microsoft Research, Cambridge. This work was partly supported by the EU ERC COLLMOT project (grant No. 227878) and EU ESF TÁMOP-4.2.1/B-09/1/KMR. M.N. was partly supported by a Royal Society Newton International Fellowship and Somerville College, Oxford. D.B. was supported by a Royal Society University Research Fellowship. The authors would like to thank Benjamin Pettit for technical assistance with the GPS tests and statistical advice. The authors are also grateful to three anonymous referees for helpful comments on an earlier version of the manuscript. References Ballerini, M., Cabibbo, N., Candelier, R., Cavagna, A., Cisbani, E., Giardina, I., Lecomte, V., Orlandi, A., Parisi, G., Procaccini, A., Viale, M. & Zdravkovic, V. 2008. Interaction ruling animal collective behavior depends on topological rather than metric distance: Evidence from a field study. Proceedings of the National Academy of Sciences, 105, 1232–1237. Bode, N. W. F., Wood, A. J. & Franks, D. W. 2011. The impact of social networks on animal collective motion. Animal Behaviour, 82, 29–38. Bode, N. W. F., Wood, J. A. & Franks, D. W. 2012. Social networks improve leaderless group navigation by facilitating long-distance communication. Current Zoology, 58, 329–341. Conradt, L., Krause, J., Couzin, I. D. & Roper, T. J. 2009. "Leading according to need" in self‐ organizing groups. The American Naturalist, 173, 304–312. Couzin, I. D., Krause, J., James, R., Ruxton, G. D. & Franks, N. R. 2002. Collective memory and spatial sorting in animal groups. Journal of Theoretical Biology, 218, 1–11. Couzin, I. D., Ioannou, C. C., Demirel, G., Gross, T., Torney, C. J., Hartnett, A., Conradt, L., Levin, S. A. & Leonard, N. E. 2011. Uninformed Individuals Promote Democratic Consensus in Animal Groups. Science, 334, 1578–1580. Flack, A., Pettit, B., Freeman, R., Guilford, T. & Biro, D. 2012. What are leaders made of? The role of individual experience in determining leader–follower relations in homing pigeons. Animal Behaviour, 83, 703–709. Flack, A., Freeman, R., Guilford, T. & Biro, D. 2013. Pairs of pigeons act as behavioural units during route learning and co-navigational leadership conflicts. The Journal of Experimental Biology, 216, 1434–1438. Freeman, R., Mann, R., Guilford, T. & Biro, D. 2011. Group decisions and individual differences: route fidelity predicts flight leadership in homing pigeons (Columba Livia). Biology Letters, 7, 63–66. Herbert-Read, J. E., Perna, A., Mann, R. P., Schaerf, T. M., Sumpter, D. J. T. & Ward, A. J. W. 2011. Inferring the rules of interaction of shoaling fish. Proceedings of the National Academy of Sciences, 108, 18726–18731. Ihl, C. & Bowyer, R. T. 2011. Leadership in mixed-sex groups of muskoxen during the snow- free season. Journal of Mammalogy, 92, 819–827. Jacobs, A., Sueur, C., Deneubourg, J. L. & Petit, O. 2011. Social Network Influences Decision Making During Collective Movements in Brown Lemurs (Eulemur fulvus fulvus). International Journal of Primatology, 32, 721–736. Katz, Y., Tunstrøm, K., Ioannou, C. C., Huepe, C. & Couzin, I. D. 2011. Inferring the structure and dynamics of interactions in schooling fish. Proceedings of the National Academy of Sciences, 108, 18720–18725. King, A. J., Douglas, C. M. S., Huchard, E., Isaac, N. J. B. & Cowlishaw, G. 2008. Dominance and affiliation mediate despotism in a social primate. Current Biology, 18, 1833–1838. Lukeman, R., Li, Y.-X. & Edelstein-Keshet, L. 2010. Inferring individual rules from collective behavior. Proceedings of the National Academy of Sciences, 107, 12576–12580. McComb, K., Shannon, G., Durant, S. M., Sayialel, K., Slotow, R., Poole, J. & Moss, C. 2011. Leadership in elephants: the adaptive value of age. Proceedings of the Royal Society B: Biological Sciences, 278, 3270–3276. Meade, J., Biro, D. & Guilford, T. 2005. Homing pigeons develop local route stereotypy. Proceedings of the Royal Society of London. Series B: Biological Sciences, 272, 17–23. Nagy, M., Ákos, Z., Biro, D. & Vicsek, T. 2010. Hierarchical group dynamics in pigeon flocks. Nature, 464, 890–893. Nakayama, S., Harcourt, J. L., Johnstone, R. A. & Manica, A. 2012a. Initiative, personality and leadership in pairs of foraging fish. PLoS ONE, 7, e36606. Nakayama, S., Johnstone, R. A. & Manica, A. 2012b. Temperament and hunger interact to determine the emergence of leaders in pairs of foraging fish. PLoS ONE, 7, e43747. Reebs, S. G. 2000. Can a minority of informed leaders determine the foraging movements of a fish shoal? Animal Behaviour, 59, 403–409. Vicsek, T. & Zafeiris, A. 2012. Collective motion. Physics Reports, 517, 71–140. Vicsek, T., Czirók, A., Ben-Jacob, E., Cohen, I. & Shochet, O. 1995. Novel type of phase transition in a system of self-driven particles. Physical Review Letters, 75, 1226–1229. Xu, X.-K., Kattas, G. D. & Small, M. 2012. Reciprocal relationships in collective flights of homing pigeons. Physical Review E, 85, 026120. Supplementary Material for Flack et al., Robustness of flight leadership relations in pigeons Table S1 Mean ± S.D. number of data points per bird Group A B C Average no. of flock flight data points per bird 167 725 189 878 169 082 (S.D.=7831) (S.D.=17422) (S.D.=8266) Figure S1 Spatial and temporal error of the GPS trajectories and the directional correlation delay method for parallel (first column), globally fixed (second column) and perpendicular (third column) orientation tests. The pole (illustrated as coloured lines) was moved along a path (black line) in parallel (A), globally fixed (B) and perpendicular (C) orientation relative to the movement direction. (D-F) Relative position of each device in a pair relative to the direction of motion as a function of its actual position. D shows only one value of each pair (i<j). E and F show both values. (G-H) Probability density function (PDF) of the measured forward position of panel D-F. G shows the deviation between measured and actual position for each pair. (J-K) Forward ratio defined as the time ratio a device was detected to be at front relative to the motion direction. (M-O) Directional correlation function (Cij(τ)) between GPS 0 and all other devices. (P-Q) The directional correlation delay time (τij) of each pair. Figure S2 Histogram illustrating the frequency distribution of distances (bin=1 m) to the first nearest neighbour of all three groups and flock flights (before and after solo training) pooled. Inset shows probability density functions of distances (bin=0.25 m) to the first, second, third and fourth nearest neighbours in groups A, B and C in red solid, green dashed and blue dotted lines, respectively (data shown only up to the fourth nearest neighbours for better visibility). Table S2 Results of the 𝜏𝑖 𝑝𝑟𝑒 vs 𝜏𝑖 𝑝𝑜𝑠𝑡 correlation analysis with and without two outliers Correlation between 𝜏𝑖 𝑝𝑟𝑒 and 𝜏𝑖 𝑝𝑜𝑠𝑡 Without outliers With outliers untrained ABC (N=21) untrained group A (N=7) untrained group B (N=7) untrained group C (N=7) r=0.72, P<0.001 r=0.80,P=0.031 r=0.87, P=0.011 r=0.69, P=0.090 0.62, P=0.003 r=0.80, P=0.031 r=0.61, P=0.149 r=0.69, P=0.090 trained ABC (N=8) r=-0.08, P=0.846 r=-0.176, P=0.677 Figure S3 Scatter plot of the relationship between an individual's 𝐶𝑖𝑗(𝜏𝑖𝑗 ∗ ) value and τi for groups A, B and C (panels (A), (B), and (C), respectively) for each flight. Red circles in B indicate τi -outliers ∗ ), τi value pairs for group B with low correlation values. Panel (D) shows the re-calculated 𝐶𝑖𝑗(𝜏𝑖𝑗 after excluding those two outliers. Figure S4 Mean (± S.D.) speed as a function of training progression in Phases I, II and III. Data from all groups were averaged according to phase. Grey circles indicate Phases I (N=30) and III (N=29), orange circles indicate Phase II (N=8). Spatial and temporal error of the GPS devices and their impact on directional correlation delay analysis To test the spatial and temporal error originating from the GPS devices, we performed a variety of tests. 10 GPS devices (labelled 0 to 9) were attached to a rigid, 3 m long pole with an inter-device distance of 33 cm. We moved the pole along a free path in an open field using 3 different orientations: (1) with the pole's orientation parallel to the direction of motion (GPS 0 at the front and 9 at the back, Fig. S1A); (2) with the pole in a fixed orientation relative to the field (Fig. S1B); and (3) with the pole's orientation perpendicular to the direction of motion (Fig. S1C.) Each test lasted 10 minutes, and the pole moved between 1 and 3 ms-1 (typical flight speed of a pigeons is 18-22 ms-1). An important aspect of analysing flock flights is the relative position of each device within a pair in relation to the movement direction of the whole flock. This is why we measured the average forward position of each device (Fig. S1D-E). In both, the perpendicular and the globally-fixed orientation case, we expect an average forward position of zero. We show the probability density function of this measure in Figure S1G-H. We also measured the time a device was detected to be in front relative to the direction of motion, and calculated the time ratio for the 10-min test (Fig. S1J-K). We also performed directional correlation delay analyses for all devices (Fig. S1 M-R). The absolute error of the GPS device arises from the relative error of the velocity which decreases as speed increases. Hence, our tests give an upper approximation of the noise due to the fact that each test lasted only 10 minutes and the pole was moved at low speeds. Additional test of hierarchy robustness We used a linear mixed-effects model to test the robustness of the hierarchies, using as our dataset the τ values calculated for each individual in every flight. All data were analysed using R (R Development Core Team, 2009) and the R packages lme4 (Bates & Maechler 2009) and languageR (Baayen 2009; cf. Baayen 2008). We included Subject as a random effect. As fixed effects, we added Training Phase (Phase I, pre-training or Phase III, post-training) and Treatment Group training group (trained or untrained individuals) to the model, as well as the interaction term between them. We verified that the normality of error and homogeneity of variance assumptions of parametric analysis were statisfied by visual inspection of plots of residuals against fitted values. To assess the validity of the mixed effects analyses, we performed likelihood ratio tests comparing the models with fixed effects to the null models with only the random effect. The model that included fixed effects did not differ significantly from the null model (P=0.222). The given P-values were based on Markov-chain Monte Carlo sampling. The directional delay times did not differ from zero (PMCMC=0.159), and we found no significant differences between pre- and post-training τ values (PMCMC=0.656). The interaction between Training Phase and Treatment Group was not significant (PMCMC =0.321). We also found no difference between pre- and post-training when examining untrained and trained birds in separate models (trained: PMCMC =0.772; untrained: PMCMC=0.219). Together, these results further confirm that the solo training had no effect on the groups' hierarchies. References Bates, D.M. & Maechler, M. (2009). lme4: Linear mixed-effects models using S4 classes. R package version 0.999375-32. Baayen, R.H. (2008). Analyzing Linguistic Data: A Practical Introduction to Statistics Using R. Cambridge: Cambridge University Press. Baayen, R. H. (2009). languageR: Data sets and functions with "Analyzing Linguistic Data: A practical introduction to statistics". R package version 0.955.
1010.1984	1	1010	2010-10-11T01:40:46	Artificial photosynthetic reaction centers coupled to light-harvesting antennas	[ "physics.bio-ph", "physics.chem-ph" ]	We analyze a theoretical model for energy and electron transfer in an artificial photosynthetic system. The photosystem consists of a molecular triad (i.e., with a donor, a photosensitive unit, and an acceptor) coupled to four accessory light-harvesting antennas pigments. The excitation energy transfer from the antennas to the artificial reaction center (the molecular triad) is here described by the F\"{o}rster mechanism. We consider two different kinds of arrangements of the accessory light-harvesting pigments around the reaction center. The first arrangement allows direct excitation transfer to the reaction center from all the surrounding pigments. The second configuration transmits energy via a cascade mechanism along a chain of light-harvesting chromophores, where only one chromophore is connected to the reaction center. At first sight, it would appear that the star-shaped configuration, with all the antennas directly coupled to the photosensitive center, would be more efficient. However, we show that the artificial photosynthetic system using the cascade energy transfer absorbs photons in a broader wavelength range and converts their energy into electricity with a higher efficiency than the system based on direct couplings between all the antenna chromophores and the reaction center.	physics.bio-ph	physics	Artificial photosynthetic reaction centers coupled to light-harvesting antennas Pulak Kumar Ghosh1, Anatoly Yu. Smirnov1,2, and Franco Nori1,2 1 Advanced Science Institute, RIKEN, Wako-shi, Saitama, 351-0198, Japan 2 Physics Department, The University of Michigan, Ann Arbor, MI 48109-1040, USA (Dated: August 26, 2018) We analyze a theoretical model for energy and electron transfer in an artificial photosynthetic system. The photosystem consists of a molecular triad (i.e., with a donor, a photosensitive unit, and an acceptor) coupled to four accessory light-harvesting antennas pigments. The excitation energy transfer from the antennas to the artificial reaction center (the molecular triad) is here described by the Forster mechanism. We consider two different kinds of arrangements of the accessory light- harvesting pigments around the reaction center. The first arrangement allows direct excitation transfer to the reaction center from all the surrounding pigments. The second configuration transmits energy via a cascade mechanism along a chain of light-harvesting chromophores, where only one chromophore is connected to the reaction center. At first sight, it would appear that the star- shaped configuration, with all the antennas directly coupled to the photosensitive center, would be more efficient. However, we show that the artificial photosynthetic system using the cascade energy transfer absorbs photons in a broader wavelength range and converts their energy into electricity with a higher efficiency than the system based on direct couplings between all the antenna chromophores and the reaction center. PACS numbers: I. INTRODUCTION The reaction centers of natural photosystems are sur- rounded by a number of accessory light-harvesting com- plexes [1]. These light-harvesting antennas absorb sun- light photons and deliver their excitation energy to the reaction center, which creates a charge-separated state. The photosystem of green plants is made up of six pho- tosynthetic accessory pigments: carotene, xanthophyll, phaeophytin a, phaeophytin b, chlorophyll a, and chloro- phyll b [1]. Each pigment absorbs light in a differ- ent range of the solar spectrum. As a result, the an- tenna complex significantly increases the effective fre- quency range for the light absorption, resulting in a highly-efficient photocurrent generation. In the presence of excessive sunlight the antenna complex can reversibly switch to the photo-protected mode, where harmful light energy is dissipated. The efficient performance of natural photosystems mo- tivates researchers to mimic their functions by creating photosynthetic units, which are combined to antenna complexes with artificial reaction centers. For exam- ple, a light-harvesting array of metalated porphyrins has been developed in Ref. [2]. This array absorbs light and rapidly transfers the excitation energy to the reaction center, so that the porphyrin (P) -- fullerene (C60) charge- separated state, P+ -- C− 60, is formed with a quantum yield ∼ 70%. Mixed self-assembled monolayers of the ferrocene (Fc) -- porphyrin -- fullerene molecular triad and the boron dipyrrin (B) dye have been made in Ref. [3, 4] with the goal to examine both energy and electron transfers in the artificial reaction center (Fc -- P -- C60), coupled to the light-harvesting molecule B. A quantum yield of ∼ 50% for photocurrent production at the wavelength 510 nm and a quantum yield of ∼ 21% at the wavelength 430 nm have been reported [3]. Recently, a more efficient, sophisticated and rigid antenna-reaction system has been designed in Refs. [5, 6]. This system includes three kinds of light-absorbing chro- mophores: (i) bis (phenylethynyl)anthracene (BPEA), which absorbs around 450 nm wavelength (blue region); (ii) borondipyrromethene (BDPY), having a strong ab- sorption at 513 nm (green region); and (iii) zinc tetraaryl- porphyrin, which absorbs both at 418 nm and at 598 nm. This study reports ∼100% quantum yield for the excita- tion transfer and ∼95 % quantum yield for the generation of the charge-separated state P+ -- C− 60. Energy transfer mechanisms in the multi-chromophoric light-harvesting complexes of bacterial photosystems (e.g., excitation-transfer between chlorophyll molecules in the Fenna-Matthews-Olson (FMO) protein complex) have been studied elsewhere (see, e.g., Ref. [7 -- 10]). Those works [7 -- 10] have mainly focused on the quantum effects in excitation-transfer across the bacteriochloro- phyll units. Theoretical studies of antenna-reaction center com- plexes can be useful for a better understanding of, and for optimizing light-to-electricity conversion, as well as for developing new and efficient designs of solar cells. Recently we have theoretically analyzed [11] the light-to- electricity energy conversion in a molecular triad (Fc -- P -- C60) electronically coupled to conducting leads. It was shown that the Fc -- P -- C60 triad can transform light en- ergy into electricity with a power-conversion efficiency of order of 40%, provided that the connection of the triad to the leads is strong enough. It should be noted, however, that this prototype solar cell absorbs photons with ener- gies in close proximity to the resonant transition of the central porphyrin molecule. Therefore, a major fraction of the sunlight spectrum is not converted to the electrical form by this device. In this paper we examine a theoretical model for 0 1 0 2 t c O 1 1 ] h p - o i b . s c i s y h p [ 1 v 4 8 9 1 . 0 1 0 1 : v i X r a 2 the light-to-electricity energy conversion by a molecu- lar triad, which is surrounded by four additional light- harvesting antenna complexes. We show that this ar- tificial photosystem is able to generate a photocurrent with a quantum yield of the order of 90 % (when the re- organization energy for the Forster transfer is relatively high) absorbing photons in a wide range (420 -- 670 nm) of the solar spectrum. We consider two different con- figurations for the antenna complexes: (a) where each light-harvesting molecule is independently connected (by the Forster energy-transfer mechanism) to the central porphyrin (P) molecule of the triad (Fig. 1a); and (b) where the light-harvesting molecules are arranged in a line (Fig. 1b), with only one molecule directly coupled to the porphyrin and with other molecules forming a chain where the energy propagates in a cascade-manner. Let us consider a (D-P-A) reaction center surrounded by several (say, four) accessory light-harvesting anten- nas. Which is the best way to arrange in space these accessory antennas? In other words, which network or topology would provide more energy from the surround- ing accessory antennas to the central reaction center? A star-shaped configuration? (with each accessory antenna directly coupled to the central reaction center). Or in a somewhat opposite configuration: as a linear chain with nearest-neighboring couplings between the antennas and only one of these coupled to the reaction center? These two cases are shown in Fig. 1. In principle, very many possible topologies could be considered. However, to simplify this analysis, we will now focus on two extreme cases, with somewhat opposite topologies or networks: a well-connected reaction center (directly connected to all four accessory light harvesting antennas), and the opposite case where the central re- action center is coupled to only one antenna, which is now part of a linear chain. These two extreme-opposite topologies or networks can be denoted as star-shaped (shown in Fig. 1a) and linear-chain (Fig. 1b), respec- tively. At first sight, it would seem that the star-shaped con- figuration, with each antenna directly connected to the reaction center, can provide far more energy to the reac- tion center, due to the multiple connections between the central unit, and the surrounding antennas. However, an energy bottleneck can arise even in this apparently- optimal topological configuration. Each antenna is here assumed to operate optimally in a limited, and shifted, frequency range. In other words, we are not assuming all accessory light-harvesting antennas to be equal to each other. In general, various antennas can operate in dif- ferent frequency ranges, and this difference is crucial in our analysis. Recall that the photosystem of green plants is made of six light-harvesting accessory pigments: each pigment absorbing light in a different range of the solar spectrum. Thus, in the star-shaped topology shown in Fig. 1a, antennas with a large energy mismatch would not provide energy to the central reaction center. Only the surrounding accessory antennas that have an approx- FIG. 1: (Color online) Schematic diagram of an artificial pho- tosystem comprised of a molecular triad (D -- P -- A) and four additional light-harvesting complexes (An1, An2, An3, An4). Here, D = donor, A = acceptor, and P = photosensitive part (porphyrin). The molecular triad D -- P -- A is inserted between two electrodes (leads) L and R. Energy exchange processes are denoted by straight red arrows. The green bent curved arrows describe electron pathways, L → D → P → A → R, via the molecular triad. (a) The photosensitive part, P, of the molecular triad is surrounded by four accessory light- harvesting complexes An1, An2, An3, and An4. In this case the surrounding antenna complexes can transfer excitations to the reaction center directly. (b) Here the antenna com- plexes form a linear chain coupled to the reaction center via nearest-neighbor couplings. An4LRAn3An2An1PAD(a)RL(b)An4An3An2 An1PADLRimate energy match to the central reaction center would transfer energy. This energy bottleneck works against the star-shaped network shown in (a). Thus, the two issues considered here are: (1) how to physically arrange antennas around the central reaction center, and also (2) how to arrange these in energy-space, so to speak. The first issue is topological and focuses on the network connectivity in real space: for instance, how many accessory antennas are connected to a central reac- tion center. The second issue refers to the energy match (or mismatch) between neighboring antennas, and be- tween them and the central reaction center. A large en- ergy mismatch between any connected units in the chain would preclude energy transfer between them. This ap- proximate "energy matching" issue between neighboring units is equally important to keep in mind, not just the real-space topological arrangement of the units. As mentioned above, Fig. 1 shows the connectivity be- tween the different units: star-shaped topology in (a), and linear-chain configuration in (b). Moreover, the col- ors there represent, very schematically, the energy range where each unit operates optimally. The linear chain shown there has antennas arranged in a way that nearby units operate in approximately similar energy ranges. This energy-matching issue perhaps is not very clear in Fig. 1, even when seen in color. The energy scales are shown far more explicitly in Fig. 2. Figure 2b clearly shows that the linear chain model considered here op- erates via an energy cascade, or linear chain-reaction. Like a line of falling dominoes, one event triggering the next one, in a sequential manner, with small energy mis- matches between successive energy transfer events. As shown in Fig. 2b, the more energetic antenna is located far away from the reaction center, and it is coupled to an antenna with a slightly lower energy, which is coupled to another antenna with an even slightly lower energy, and so on, moving energetically "downhill" along the chain. Thus, the neighboring antennas must be so both in real space, and also in "energy space", to allow for the effi- cient transfer of energy between them. Thus, "proxim- ity" between units must be in two spaces: real space and energy space. This article is organized as follows: in Section II we outline a model for the artificial reaction center (molec- ular triad) coupled to the antenna complex. We briefly describe our mathematical methods in Sec. III. The pa- rameters are listed in Sec. IV. In Sec. V, we numerically solve the master equations and analyze the energy trans- fer process. Conclusions are presented in Sec. VI. The methods used are described in more details in the online supplementary material. II. MODEL We start with a schematic description of the energy and electron transfer processes in an artificial reaction center D -- P -- A (Donor -- Porphyrin -- Acceptor) combined 3 FIG. 2: (Color online) Energy diagram of the antenna com- plexes An1, . . ., An4, energetically coupled to the reaction center (RC), D -- P -- A, for the case (a) where each light- harvesting molecule is directly connected to the porphyrin molecule, P, of the RC; and (b) where the energy transfer occurs via a linear-chain of antenna with nearest-neighbor energy exchange. Note that the four antenna chromophores in (b) are arranged in a way to have a relatively small energy difference between neighboring units. PPAn1An1An2An2An4An4An3An3D A (a)EnergyRLPPAn3An1An2An2An4An4An1An3D A EnergyLR(b)with four antenna chromophores: An1, An2, An3, An4 (see Figs. 1a and 1b, showing two different configurations for these antennas: star-shaped configuration in 1a and chain in 1b). The photosensitive molecular triad, D -- P -- A, is inserted between two electron reservoirs (electrodes) L and R. The donor, D, is coupled to the left lead, L, and the acceptor, A, is connected to the right lead, R. As in Refs. [3, 11], the donor and acceptor molecules (e.g., fer- rocene and fullerene), are connected to each other via the photosensitive molecule (porphyrin, P). This molecule is surrounded by four Accessory Light-Harvesting Pigments (ALHP). Figure 1a corresponds to the situation where all pigments are directly coupled to the photosensitive part (P) of the molecular triad. In this "star-shaped" geometric arrangement, all the ALHP can directly trans- fer excitations to the photosensitive part of the molecu- lar triad. Figure 1b corresponds to the case where the light-harvesting pigments form a chain, which transfers energy where the excitation moves from one pigment to the next one via nearest-neighbor couplings: An4 → An3 → An2 → An1 → P. This cascade-like excitation trans- fer occurs in an energetically-downhill direction, akin a one-dimensional chain reaction or domino-effect. Figures 2a and 2b present the energy diagrams of the photosystems described in Figs. 1a and 1b, respectively. The electron transfer chain, L→D→P→ P∗ →A→R, is the same for both configurations (a) and (b), and both begin on the left lead, L. The electrochemical potentials of the left (L) and right (R) electron reservoirs are de- termined by the parameters µL and µR, with µR > µL. Since the energy level ED of the donor D is lower than the potential µL of the left lead, ED < µL, electrons can move from the L-reservoir to the level D and, after- wards, to the low-lying ground energy level, EP , of the porphyrin. When absorbing a photon, the electron in the porphyrin molecule jumps from its ground state P to its excited state P∗. A subsequent electron transfer from P∗ to the acceptor A is driven by a negative en- ergy gradient, (EA − EP ∗ ) < 0. In view of the relation: EA > µR, the electron in A is finally transferred to the right, R, electron reservoir. This is the light-induced elec- tron transition in the porphyrin molecule, which results in an energetically-uphill electron flow in both photosyn- thetic systems (a) and (b). Even though these systems (a) and (b) have the simi- lar electron-transport chains, their light-harvesting com- plexes are arranged quite differently. Each of these com- plexes, An = An1, . . ., An4, can be characterized by a ground, EAn, and an excited, EAn∗ , energy levels with an energy difference ωAn = EAn∗ − EAn. Hereafter, we assume that ¯h = 1 and kB = 1. For the light- harvesting complex (a) (see Figs. 1a, 2a) all frequencies ωAn1, . . . , ωAn4 should exceed the porphyrin transition frequency, ωP = EP ∗−EP . In this case the energy of pho- tons collected by each individual antenna can be trans- ferred directly to the photosensitive part of the artificial reaction center (porphyrin molecule). However, in the light-harvesting complex (b) (see Figs. 1b and 2b) only 4 the antenna An1 is coupled (by a Forster mechanism) to the porphyrin, whereas the other light-harvesting pig- ments form a linear chain that transfers energy down- hill along the chain: An4 → An3 → An2 → An1 → P. This energy transfer can be energetically-allowed pro- vided that ωAn4 > ωAn3 > ωAn2 > ωAn1 > ωP . In Sec. V we compare these two artificial photosystems and deter- mine which arrangement of the antenna complexes pro- vides more energy to the reaction center. III. METHODS The electron flow through a molecular triad coupled to two electron reservoirs can be described with meth- ods of quantum transport theory and the theory of open quantum systems [12 -- 15]. In addition to four sites (D, P, P∗, A), describing the molecular triad, we introduce four pairs (An1, An1∗, . . ., An4, An4∗), which character- ize the ground and excited states of the light-harvesting antennas. The whole system, the molecular triad plus four antenna complexes, can be analyzed within a math- ematical formalism presented in Supporting Information and also in Ref. [11]. The total Hamiltonian of the system includes the fol- lowing components: (i) energies of the electron sites and leads, as well as the Coulomb interactions between the electrons located on different sites of the triad; (ii) tunneling couplings between the electron sites on the triad and the electron reservoirs; (iii) electron tunneling between the electron sites belong- ing to the molecular triad; (iv) coupling of electron sites and antenna complexes to an environment; (v) interactions of the porphyrin molecule and antenna complexes with an external electromagnetic field (laser field) and with (vi) blackbody radiation and Ohmic bath, responsible for the quenching (energy loss) of the porphyrin and antenna excited states. (vii) In the case of the design in Fig. 1a the Hamiltonian includes the direct Forster coupling between the por- phyrin molecule, P, and the light-harvesting complexes An1, . . ., An4. For the linear-chain configuration shown in Fig. 1b, the Forster mechanism provides the energy transfer between the nearest-neighbors in the antenna chain, as well as between the complex An1 and the por- phyrin molecule. IV. PARAMETERS Energy levels and electrochemical potentials: The en- ergy levels of the Fc -- P -- C60 molecular triad are ED = −510 meV, EP = −1150 meV, EP ∗ = 750 meV and EA = 620 meV. These values are obtained by esti- mating the reduction potentials (using a reference elec- trode Ag/AgCl) of ferrocene (D), porphyrin (P, P∗) and fullerene (A) molecules [16]. For the electrochemical potentials of the left (µL) and the right (µR) leads, we choose the following values: µL = −410 meV and µR = 520 meV, with the electrochemical gradient ∆µ = µR − µL = 930 meV. Coulomb interactions: The spatial separations be- tween D -- P, P -- A and D -- A are of order of 1.62 nm, 1.8 nm and 3.42 nm, respectively [16]. The Coulomb energies uDP , uDA and uP A can be calculated with the formula uij = e2 4π0rij where, {ij} = {DP}, {DA}, {PA}, and 0 is the vac- uum dielectric constant. For ∼ 4.4, the Coulomb inter- action energies are uDP = 200 meV, uDA = 95 meV and uP A = 180 meV. The Forster coupling, VF , between the photosensitive molecules l and l(cid:48) is proportional to the product of the dipole moments of these molecules, erl and erl(cid:48), and in- versely proportional to the cubic power of the distance (Rll(cid:48)) between them [17]: e2 rlrl(cid:48) R3 VF = (1) For the case where rk ∼ rl ∼ 0.3 nm, R ∼1 nm, and at ∼ 4, the Forster coupling is about VF ∼ 65 meV. 2π0 assumed the Tunneling amplitudes: We have ferrocene -- porphyrin and porphyrin -- fullerene tunneling amplitudes are about ∼ 3 meV, so that ∆AP ∗ ∆DP ∗ ∆DP ∆AP = 4.5 ps−1. = ¯h ¯h = = ¯h ¯h For the tunneling rates between the left lead and fer- rocene (ΓL) and between the right lead and fullerene (ΓR) we choose the following values [11]: ΓL/¯h = 1800 µs−1 and ΓR/¯h = 180 µs−1. Radiation leakage and quenching rates: We take the following estimates for the radiation leakage time: τP ∗→P = τP ∗→D = τA→P ∼ 0.4 ns. Similar estimates have been used for the radiation leakage timescales of the antenna molecules. For the quenching (or energy- loss) time of the porphyrin excited state P∗ we use the value: τquen ∼ 0.1 ns. Reorganization energies for the electron and energy transfers: For the molecular triad analyzed in Ref. [11] we obtain the relatively high power-conversion efficiency, η ∼ 42%, provided that the donor-porphyrin and acceptor-porphyrin electron transfer reorganization en- ergies are about ΛDP ∼ 600 meV and ΛAP ∼ 100 − 400 meV. A much smaller value, ΛP P ∗ ∼ 100 meV, is as- sumed for the light-induced electron transitions between the ground (P) and excited (P∗) levels of the porphyrin molecule. The Forster energy transfer between the light- harvesting molecules and between these molecules and the porphyrin is also accompanied by an environment- reorganization process, which can be characterized by a 5 smaller energy scale, ΛF ≤ 100 meV (see, e.g., the energy transfer in the B850 complex [18], where ΛF is assumed to be about 30 meV). Hereafter, we assume that the external light source has a fixed intensity, I = 100 mW/cm2, and that the environment is kept at the room temperature, T = 298 K. We also assume that the reorganization energy for the Forster energy transfer, ΛF , is about 100 meV, unless otherwise specified. We analyze the Fc -- P -- C60 molecular triad, where the ferrocene molecule, Fc, is attached to the gold surface (left lead, L), and the fullerene, C60, is in contact with an electrolyte solution (right lead, R) filled with oxygen molecules, which are able to accept electrons from the C60 molecules. V. RESULTS AND DISCUSSION We derive and solve numerically a set of master equa- tions for the probabilities to find the system in a definite eigenstate of the basis Hamiltonian. This is explained in the online supplementary material. After that, we calcu- late: (i) the energetically-uphill electron current through the triad, (ii) the energy of the photons absorbed by the triad and by the light-harvesting molecules. This allows us to determine a quantum yield and power-conversion efficiency of the system (see all definitions in the support- ing information). A. Photocurrent through the molecular triad directly coupled to four porphyrin light-harvesting molecules Here we consider the situation where both the reaction center and the antenna complexes are made of porphyrin molecules with the geometrical arrangement shown in Fig. 1a. This arrangement allows direct energy transfer from each light-harvesting chromophore to the reaction center. In Fig. 3 we plot the photocurrent through the triad as a function of the wavelength of light for different values of the Forster coupling strength : VF = 0, 0.1, 1, 10 (in meV) and for the above-mentioned set of parameters of the system. It is apparent from Fig. 3 that the magnitude of the light-induced pumping current at λ = 620 nm is significantly enhanced (about 5 times larger when VF = 10 meV) by the antenna system. However, the spectral range of the light absorption remains the same as for the detached porphyrin reaction center (see Fig. 3, where the red curve with square symbols describes the photocur- rent through the molecular triad completely disconnected from the antenna chromophores, VF = 0). We also find that the quantum yield, Φ, taken in the middle of the res- onant peak (λ = 620 nm), non-monotonically depends on the Forster coupling strength VF measured here in 6 FIG. 3: (Color online) Photoinduced electron current Ie ( number of electrons, in one ms, pumped from the L to the R lead) versus the wavelength of the incident light for a pho- tosystem with four antenna complexes, which are made of porphyrin molecules. Both the antenna complexes and the reaction center absorb at the same wavelength. The whole complex now absorbs more photons than the single porphyrin molecule; thus, pumping many more electrons from the left (with µL = −410 meV) to the right (with µR = 520 meV) electron reservoir. The other parameters are listed in the text (see Sec. IV). The peak in the current increases for larger val- ues of the Forster coupling strength VF . We also studied (not shown here) higher values of VF , but these produced the same results as VF = 10 meV. Thus, this value of VF provides a saturation in the electron current. The resonant peak here is λ = 620 nm. meV: Φ(VF = 0) (cid:39) 0.85; Φ(0.1) (cid:39) 0.3; Φ(1) (cid:39) 0.75 and Φ(10) (cid:39) 0.82. B. Molecular triad connected to two BPEA and two BDPY chromophores FIG. 4: (Color online) Electron current Ie and quantum yield Φ as functions of the wavelength λ of the external radia- tion for the configuration shown in Fig. 1a, where two BPEA and two BDPY antenna chromophores are directly coupled to the centrally-located reaction center. The Forster cou- pling constant, VF , which is assumed to be the same for ev- ery chromophore-RC connection, takes four values (in meV): VF = 0, 1, 25, 50. For other parameters see Sec. IV. Note that the electron current and the quantum yield grow for in- creasing values of the Forster coupling energy strength VF . More importantly, the direct coupling (Fig. 1a) suppresses the peak at λ= 450 nm (Fig. 4a), which is present in the linear chain configuration (Fig. 1b), as shown in Fig. 5a Now we consider a different case: an antenna system comprised of two BDPY and two BPEA molecules. The BDPY molecule has the maximum absorbance in the green region (at 513 nm) of the solar spectrum, where neither BPEA (with maximum absorbance at 450 nm) nor the porphyrin, which absorbs at 620 nm, have max- ima of absorption spectra. It should be noted that a mul- tichromophoric hexad antenna system having three light- absorbing BDPY, BPEA and porphyrin has been devel- oped in Ref. [5]. The generation of the charge-separated state, P+ -- C− 60, with almost 95% quantum yield [5]. In our case, the porphyrin unit of the molecular triad is cou- pled to the four antenna chromophores (two BDPY and two BPEA). We consider two situations: (a) where the antenna chromophores are directly coupled to the porphyrin unit of the reaction center (Fig. 1a); and (b) where the an- tenna chromophores are arranged in line: BDPY → BDPY → BPEA → BPEA → RC, with the nearest- neighbor coupling between chromophores. See Fig. 1b. Thus, the configuration (a) might appear to be energet- ically more efficient than (b). However, our calculations below indicate that this is not the case. (a) For the case of direct connection between the four antennas chromophores and the triad (see Fig. 1a and Fig. 4) we calculate a photocurrent and a quantum yield, Φ, as functions of the wavelength of light, at ΛF = 100 meV, and at five values of the Forster cou- pling: VF = 0, 1, 25, 50 meV. The wavelength depen- dence of the current has two maxima centered at 513 nm and 620 nm. The BPEA molecules, which absorb at 450 nm, give a negligible contribution to the cur- 54058062066002060100140λ (nm)Current, Ie (ms-1) VF = 10 meVVF = 1 meVVF = 0.1 meVVF = 0 4505005506006500510152025λ (nm)Current, Ie (ms-1)45050055060065000.10.30.50.70.9λ (nm)Quantum yield, Φ VF = 50 meVVF = 25 meV VF = 1 meVVF = 0 (a)(b)7 FIG. 5: (Color online) Electron current Ie and a quantum yield Φ versus the wavelength of light, λ, for the configuration shown in Fig. 1b, where the excitation energy moves along the following chain of light-harvesting molecules: BPEA → BPEA → BDPY → BDPY → RC. The parameters used here are the same as in Fig. 4. However, the current peak at λ ∼ 450 nm is present in (a) here, but absent in Fig. 4(a), which used a direct-coupling configuration to the central reaction center. rent since their spectral maxima are too far from the absorbance maximum of the porphyrin spectrum. As a consequence, the BPEA-porphyrin energy transfer is sig- nificantly suppressed at moderate values of the Forster reorganization energy, ΛF ≤ 100 meV in the range of the coupling constants VF ≤ 50 meV. It follows from Fig. 4 (see a peak at λ = 513 nm) that the BDPY molecules start working as efficient light-harvesters only at suffi- ciently strong Forster coupling, VF ≥ 10 meV, to the porphyrin unit of the molecular triad. We also note that when λ ∼ 513 nm, both the photoinduced current and the quantum yield grow with increasing the Forster cou- pling strength, so that the quantum yield, Φ, can be around 48%. In the range of porphyrin absorption (at λ ∼ 620 nm and VF = 0) the quantum yield is of the order of 90%. (b) A much broader light spectrum can be converted into electrical current in the linear configuration in Fig. 1b, where the light-harvesting chromophores are arranged along a line: BPEA → BPEA → BDPY → BDPY → FIG. 6: (Color online) Quantum yield as a function of reor- ganization energy, ΛF , for the linear-chain nearest-neighbor coupling between chromophores, BPEA → BPEA → BDPY → BDPY → RC, for a the wavelength of light λ = 450 nm (blue peak in Fig. 5a). This figure is plotted for three different temperatures: (a) Very low T = 77 K, (b) room temperature T = 298 K and (c) very high T = 500 K as well as for three values of the Forster constant VF = 1, 25, and 50 meV. As shown in the figures (b) and (c), increasing the reorganization energy ΛF , can sharply increase the quantum yield. The ex- treme low-temperature case in (a) is just a limit case, shown for comparison with the higher-temperature cases in (b) and (c). RC, with the only one BDPY molecule directly coupled to the porphyrin unit of our artificial reaction center (RC) (see Fig. 1b and Fig. 5). This system is able to collect photons in the range of wavelength from 420 nm up to 650 nm covering a significant part of the visible sunlight spectrum. The chain of BPEA and BDPY molecules cre- ates an efficient channel, which gradually transmits en- ergy from the collectors of high-energy photons (BPEA molecules), via the intermediate BDPY antennas, to the molecular triad. In Figs. 5a, 5b we plot the photoinduced current and the quantum yield versus the wavelength of 4505005506006500510152025λ (nm)Current, Ie (ms-1)45050055060065000.10.30.50.70.9λ (nm)Quantum yield, Φ VF = 50 meVVF = 25 meVVF = 1 meVVF = 0 meV(b)(a)10020030000.20.40.60.81ΛF (meV)Quantum yield, Φ 10020030000.20.40.60.8ΛF (meV)Quantum yield, Φ10020030040000.20.40.60.8ΛF (meV)Quantum yield, ΦVF= 50 meVVF= 25 meVVF= 1 meVT = 298 Kλ = 450 nm(c)(b)(a)T = 77 Kλ = 450 nmT = 500 Kλ = 450 nm8 chromophores and the RC [Fig. 1a, design (a)]. In this star-shaped configuration, the BPEA molecules, which collects blue photons, are not able to transfer their en- ergy to the photosensitive unit of the triad due to a sig- nificant difference between the energies of the BPEA an- tennas (λ = 450 nm) and the porphyrin-based reaction center (λ = 620 nm). The antenna-RC energy transfer is facilitated by the strong coupling to the environment (when the energy difference is high), characterized by the reorganization energy ΛF , as well as by a tight Forster binding between chromophores, which is described by the constant VF . In Figs. 6 and 7 we plot the quantum yield, Φ, as a function of the reorganization energy, ΛF , for three different tem- peratures (in K): T = 77, 298, 500, and for three values of the coupling constant VF = 1, 25, and 50 meV. Figure 6 is related to the blue peak of the absorption spectrum (λ = 450 nm), whereas figures 7 describes the behavior of the green peak (λ = 513 nm). The peak centered at λ = 620 nm is produced by the porphyrin molecule, be- longing to the triad, and, therefore, shows no dependence on ΛF and VF . The Marcus rates, describing the energy transmission between the photosensitive elements of the system, de- pend (i) on the energy difference, ∆E, between the pho- tosensitive units, (ii) on the Forster coupling VF , and (iii) on the Forster reorganization energy ΛF . The large energy separation, ∆E, of the energies of the nearby pho- tosensitive molecules leads to a decrease of the Marcus rates and, thus, to the suppression of the energy transfer. In our case, the energy distance between the BPEA (λ = 450 nm) and BDPY (λ = 513 nm) molecules is about 340 meV, whereas the energy separation of the BDPY chromophore and porphyrin (λ = 620 nm) is of order of 415 meV. This energy gap can be partially compensated by the large reorganization energy, ΛF , which reflects the significant fluctuations of the relative positions of the en- ergy levels. Here, the environment plays a positive role assisting the efficient and fast energy transfer between chromophores (see also Ref. [19, 20]). The timescales for the energy and electron transfers should be shorter than the radiation leakage time and the quenching time, oth- erwise the energy of the photons absorbed by the system will be lost. At very low temperatures (e.g., liquid nitrogen, T = 77 K), the fluctuations in the positions of the energy levels of the chromophores are frozen, so that the light- to-electricity conversion requires sufficiently large values of the reorganization energy, ΛF > 180 meV (see Figs. 6a and 7a). Note that for VF > 25 meV the linear-chain arrangement system has an optimal performance at ΛF (cid:39) 225 meV for both frequency ranges. This extreme low temperature case is only shown for comparison with the higher temperatures cases. At room temperature (T = 298 K) and at strong enough Forster coupling, VF ≥ 25 meV, the blue and green spectral peaks demonstrate similar behaviors as functions of ΛF (see Figs. 6b and 7b). Here, the quantum FIG. 7: (Color online) Quantum yield versus reorganization energy, ΛF , for the green peak (λ = 513 nm) of the spec- trum (see Fig. 5a) for three values of the Forster constant, VF = 1, 25, 50 meV, and at three different temperatures: (a) T = 77 K, (b) T = 298 K and (c) T = 500 K. The other parameters are listed in Sec. IV. Figures 5, 6, and 7 focus on the linear chain configuration (Fig. 1b) with nearest-neighbor couplings between chromophores. Figures 6 and 7 show the same quantities, but centered at different peaks (λ ∼ 450 nm versus λ ∼ 513 nm). the external radiation, at ΛF = 100 meV, and at four val- ues of the Forster constants (in meV) VF = 0, 1, 25, 50. For large value of the Forster coupling strength VF the quantum yield Φ reaches ∼ 48% in Fig.4(a) around λ = 513 nm [for the star-shaped topology in Fig. 1a] and ∼ 30% in Fig. 5(b) [for the linear chain case in Fig 1b]. The quantum yield in Fig. 5b is lower than the one in Fig. 4a, but extends over a wider range of wavelengths including the peaks around λ ∼ 450 nm and λ ∼ 513 nm. It follows from Fig. 4 and Fig. 5, that the configu- ration using the chain-like nearest-neighbor coupling be- tween light-harvesting chromophores [Fig. 1b, design (b)] converts much more blue (λ = 450 nm) light into elec- tricity with a higher quantum yield than the star-shaped configuration with direct coupling between the antenna 5010015020025030035000.20.40.60.81ΛF (meV)Quantum yield, Φ5010015020025030035000.20.40.60.8ΛF (meV)Quantum yield, Φ10020030040000.20.40.60.8ΛF (meV)Quantum yield, Φ VF = 50 meVVF = 25 meVVF = 1 meV(a)(c)(b)T = 500 Kλ = 513 nmT = 298 Kλ = 513 nmT = 77 Kλ = 513 nmyield begins to grow when the reorganization energy ex- ceeds ∼ 80 meV, reaching finally 90% at ΛF > 150 meV. These numbers are determined by the parameters of the antenna-triad complex, and, especially, by the radiation leakage time τrad, of the excited porphyrin state P∗, es- timated above as τrad ∼ 0.4 ns. At high temperatures (see Figs. 6c and 7c, plotted for T = 500 K) the facilitating effect of the environment increases, and the efficient energy transfer starts at the lower reorganization energies, ΛF ≥ 75 meV. This value of ΛF is comparable with the reorganization energy for the energy transfer in the B850 light-harvesting complex, where ΛF (cid:39) 27 meV [18]. At the smaller value of the Forster coupling, VF = 1 meV, the conversion of the blue light (λ = 450 nm) to electricity is significantly suppressed (see Fig. 6), whereas for green light (λ = 513 nm) the dependence of the quan- tum yield on ΛF are shifted to higher reorganization ener- gies (Fig. 7), compared to the case of the larger couplings, VF = 25, 50 meV. It should be noted that, to cover a broader range of the spectrum of light with a fixed number of antenna chro- mophores, the resonance energies of the light-harvesting complexes should be very well separated. However, in this case the energy transfer between the antenna chro- mophores would be quite slow, since this transfer is gov- erned by the rates corresponding to the inverted regions of the Marcus parabola. Then, the dissipation comes into play, and the energy of the absorbed photons is lost on its way from the antennas to the reaction center. The en- ergy transfer rates and, thus, the efficiency of the system can be maximized in the case when the energy distance, ∆Ei = Ei+1 − Ei, between the nearby light-harvesting complexes (labeled by indices i + 1 and i, with energies Ei+1 and Ei) is equal to the corresponding reorganization energy, Λi F . That is, ∆Ei = Ei+1 − Ei = Λi F . VI. CONCLUSIONS In this paper we have studied theoretical aspects of the operation of an artificial reaction center (a ferrocene -- porphyrin -- fullerene molecular triad) coupled to the com- plex of four light-harvesting molecules. We have ana- lyzed two configurations of the antenna complex: (a) a star-shaped configuration, where each light-harvesting molecule is able to transfer energy directly to the centrally-located reaction center, and (b) a case where the antenna molecules form a linear chain, which gradu- ally transfers excitations from the high-energy antenna located in the far end, to the antenna chromophore with the lowest energy. The last antenna chromophore in the chain is energetically connected to the reaction center (RC). To be specific, we have considered the case when the antenna complex is comprised of two molecules of bis(phenylethynyl)anthracene (BPEA), ab- sorbing blue photons (λ = 450 nm), and two molecules of borondipyrromethene (BDPY), having an absorption 9 maximum in the green region (λ = 513 nm). We have shown that the configuration with a linear arrangement of the antenna chromophores (configuration (b)) is able to convert blue and green photons to electricity with a quantum yield of order of ∼30% (over a wide range of wavelengths), whereas the energy of the red photons, ab- sorbed by the molecular triad itself (λ = 620 nm), is converted to a current with a quantum yield reaching the value of 90%. We have investigated dependencies of the quantum yield on the Forster reorganization energy as well as on the Forster coupling constants between chro- mophores and have shown that the environment plays a significant role in facilitating the antenna-RC energy transfer, thus, improving the light-harvesting function of the system. Overall, the configuration (b) is more effi- cient than (a) in transferring energy to the reaction cen- ter. We emphasize that the artificial photosystem an- alyzed in this work can be implemented with real light-harvesting components, such as porphyrin and BPEA/BDPY molecules. The excitonic (Forster) cou- pling strongly depends on the mutual distances and the orientations of the chromophores. Similar to the wheel- shaped antenna-reaction center complex implemented in Ref. [5], the components of the photosystem can be placed at distances of the order 10 A, which allows for a sufficiently strong Forster coupling between the antenna chromophores and the reaction center. At the same time, the chromophores comprising the light-harvesting com- plex retain their individual molecular features. The re- organization energy, another controlling parameter for energy transfer, is varied for the system under study. Namely, we numerically calculate both the light-induced electron current and the quantum yield as functions of the reorganization energy. This allows us to determine the value of the reorganization energy at which the sys- tem works with maximum optimal efficiency. Acknowledgements. FN acknowledges partial sup- port from the Laboratory of Physical Sciences, National Security Agency, Army Research Office, DARPA, Na- tional Science Foundation grant No. 0726909, JSPS- RFBR contract No. 09-02-92114, Grant-in-Aid for Sci- entific Research (S), MEXT Kakenhi on Quantum Cy- bernetics, and Funding Program for Innovative Research and Development on Science and Technology (FIRST). Appendix A: Hamiltonian Here we describe the methods used in our work. We characterize the electrons in the states i (= D, P, P∗, An1, An1∗, An2, An2∗, An3, An3∗, An4, An4∗, A) by the Fermi operators a+ i and ai with the electron population operator ni = a+ i ai. Each electron state can be occupied by a single electron, as the spin degrees of freedom are ne- glected. Electrons in the leads (electrodes) are described by the Fermi operators d+ kα, dk,α, where α = L, R; and k is an additional parameter which has the meaning of a (cid:80) wave vector in condensed matter physics. The number of electrons in the leads is determined by the operator kαdkα. The total Hamiltonian of the system is complicated. It includes the terms de- scribed below. k Nkα, with Nkα = d+ 1. Eigenenergies and Coulomb interactions (cid:88) This part of the Hamiltonian involved the eigenener- gies of the electron states (i = D, P, P∗, An1, An1∗, An2, An2∗, An3, An3∗, An4, An4∗, A) and the Coulomb interactions between the electron states. H0 = Eini + uP nP nP ∗ + uDP (1 − nD) (1 − nP− i nP ∗ ) − uDA(1 − nD)nA − uP A(1 − nP − nP∗)nA. (A1) The symbols uP , uDP , uDA, uP A represent the electro- static interactions between the electron sites. We have assumed that the empty donor state D (with nD = 0) as well as the empty photosensitive group (nP + nP ∗ = 0) have positive charges. Therefore, UDP > 0 because both D and P are positively charged and thus repulsive. The acceptor state A becomes negatively charged when it is occupied by an electron and thus −UDA < 0 and −UP A < 0. This attraction occurs when the acceptor A is occupied (nA = 1) and the D and P states are both empty. Also, the acceptor state A is neutral when it is empty. 2. Forster couplings (cid:88) both given by the Hamiltonian, Htr = −(cid:88) D ckL −(cid:88) TkL a+ k k 10 TkR c+ kR aA + H.c. , (A4) where c+ kα, ckα are the electron creation and annihilation operators, and α is the index for the leads. The Hamil- tonian of the leads is given by (cid:88) HLR = εαnα with nα = c+ kα ckα . α k 4. Thermal tunneling Htun = −(cid:88) Activated by thermal fluctuations, electrons can tun- nel between the electron sites. Here, Htun, given by the following expression, ∆l,l(cid:48) a+ l al(cid:48) + H.c. , (A5) l which accounts for the thermal tunneling effects. Here ∆l,l(cid:48) is the strength of the tunnelling coupling and the {l, l(cid:48)} indices refer to the pairs: {D, P}, {D, P∗}, {A, P}, {A, P∗}. 5. Light-induced excitations This part of the Hamiltonian accounts for the inter- action of light with the molecular triad and the antenna complexes. Under the rotating-wave approximation, the light-induced excitation processes can be described as HLight = − (cid:88) F eiω0t a+ k ak∗ + H.c. (A6) HForster = −(cid:88) We consider the energy transfer between the reaction center (P) and the antenna complexes, and also among the antenna complexes by introducing Forster coupling terms, k where k = P, An1, An2, An3, An4 and the field amplitude F = Eext dkk∗ , Vkl a+ l al∗ a+ k∗ ak + H.c. , (A2) where, dkk∗ is the dipole moment. kl where, the pair {k, l} = {P, An1}, {P, An2}, {P, An3}, {P, An4}, {An1, An2}, {An2, An3}, {An3, An4}. Here, VF determines the strength of the Forster coupling and is proportional to the product of the dipole moments, erl and erk, and inversely proportional to the cubic power of the separating distance, R, between the chromophores, VF = e2 2π0 rkrl R3 . (A3) 3. Tunneling couplings to the leads The electron tunnelling from the left lead to the donor state and from the acceptor state to the right lead are 6. Coupling to a radiation heat bath and an Ohmic bath HQ = −(cid:88) Coupling the system to a radiation heat bath causes radiation leakage from the excited states. The following Hamiltonian accounts for this radiation leakage Qσσ(cid:48)a+ σ aσ(cid:48) + H.c. , (A7) σ where, {σ, σ(cid:48)} denotes the pairs of sites {D, P∗}, {A, P}, {P, P∗}, {An1,A n1∗}, {An2, An2∗}, {An3, An3∗}, {An4, An4∗}. The operators for the radiation bath, Qσσ(cid:48) = e xσσ(cid:48) × Erad (A8) are proportional to the projection of the fluctuating elec- tromagnetic field, Erad, along the direction of the corre- sponding dipole moment, dσσ(cid:48) = e xσσ(cid:48). The excited state of the photosensitive part of the molecular triad can be quenched by the electrode. Namely, lose the excitation energy when interacting with the electrodes. We introduce Hquench to account for this energy-loss or quenching processes., Hquench = −Ql a+ (A9) where, Ql is the variable of the Ohmic bath and {l, l(cid:48)} = {P, P∗}. l al(cid:48) + H.c. , 7. Interaction with the environment We have taken into account the effects of a dissipative environment by the well-known system-reservoir model [11, 13, 15, 17, 21, 22].  p2 j 2mj (cid:88) j (cid:32) + mjω2 j 2 (cid:88) 1 2 Henv = xj + xji ni (cid:33)2 , i (A10) where xj, pj are the position and momentum of the jth oscillator with effective masses mj and frequencies ωj. Here, xji is a measure of the strength of the coupling be- tween the electron subsystem and the environment. We characterize the phonon modes of the bath by the spec- tral functions Jii(cid:48)(ω), defined by mjω3 j (xji − xji(cid:48))2 δ(ω − ωj) . (A11) Jii(cid:48)(ω) = j 2 (cid:88) The spectral function Jii(cid:48) is related to the reorganiza- tion energy Λii(cid:48) for the i → i(cid:48) transition, by the following equation: Λii(cid:48) = dω ω Jii(cid:48)(ω) = mjω2 j (xji − xji(cid:48))2 . (A12) j 2 (cid:90) ∞ 0 (cid:88) 8. Unitary transformation To remove the electron population operators of the electron subsystem from Henv, we use the unitary trans- foration U =(cid:81) Ui, where i  i 2 (cid:88) j , Ui = exp pjxjini (A13) The results of this unitary transformation are: (i) When U is operated on an arbitrary function Φ(xj), a shift of the oscillator positions xj is produced. (cid:32) (cid:88) i 1 2 (cid:33) U†Φ(xj) U = Φ xj + xjini . (A14) 11 (ii) Another result of this unitary transformation is that all the transition operators acquire fluctuating factors. † U†a i ai(cid:48) U = eiξii(cid:48) a † i ai(cid:48). (A15) The stochastic phase operators ξii(cid:48) are given by ξii(cid:48) = ξi − ξi(cid:48) with ξi = 1 ¯h pjxji . (A16) (cid:88) j Appendix B: Master equations (cid:88) (cid:88) The system under study can be characterized by the 256 eigenstates of H0. We expressed all the operators described in previous section in the terms of the density operators ρµν ≡ µ(cid:105)(cid:104)ν. To derive the time evolution of the diagonal elements ρµµ ≡ ρµ of the density matrix (ρµν), we write the Heisenberg equation for the operators, with the subsequent averaging (cid:104)ρµ(cid:105) over the environment fluctuations. Finally, we obtain the master equation for the density matrix of the system [11, 13, 15, 17, 22], (cid:104) ρµ(cid:105) + γνµ(cid:104)ρµ(cid:105) = γµν(cid:104)ρν(cid:105), (B1) ν ν where, γµν is the total relaxation matrix, which is the sum of six types of relaxation rates: γµν = γtr µν + kForster µν + ktun µν + klight µν + krad µν + kquench µν .(B2) These relaxation rates will be described right below. 1. Electron tunneling rates between the leads and the molecular triad The first term of Eq.(B2), γtr µν, represents the relax- ation rates due the couplings of the triad to the L and R electron reservoirs: (cid:8)aD;µν2[1 − fL(ωνµ)] + aD;νµ2fL(ωµν)(cid:9) (cid:8)aA;µν2[1 − fR(ωνµ)] + aA;νµ2fR(ωµν)(cid:9) ,(B3) γtr µν = ΓL +ΓR where the Γα (α = L, R) are the resonant tunneling rates. Here, the electron reservoirs have been characterized by the Fermi distributions fα(ω), (cid:20) (cid:18) EkB α − µα (cid:19) T (cid:21)−1 fα(Ekα) = exp + 1 . (B4) with the temperature T (kB = 1, ¯h = 1). The electro- chemical potentials µL and µR are the controlling fac- tors of the electron transition rates from the left lead to the donor state, and from the acceptor state to the right lead. 2. Forster relaxation rates 5. Relaxation rates due to radiation leakage 12 µν Here kForster accounts for the excitation transfer rates from the antenna complexes to the reaction center, and also among the antenna complexes. The excitation tran- sition rates via the Forster mechanism is given by k∗ ak)µν2 (cid:88) l al∗ a+ κForster µν = (cid:114) π Vkl2 (cid:2)(a+ (cid:20) k∗ ak)νµ2(cid:3) exp ΛklT l al∗ a+ kl + (a+ − (ωµν + Λkl)2 4ΛklT (cid:21) , (B5) where Vkl is the resonant Forster relaxation rate and Λkl stands for reorganization energy. We denote, Vkl = VF and Λkl = ΛF for any combinations of k and l. The non- resonant exponential term of the above expression arises due to the different energy gaps of the reaction center and the accessory antenna complexes. Moreover, the non- resonant exponential terms depend on the reorganization energy. 3. Thermal tunneling rates The matrix element ktun µν of the Eq. (B2) are respon- sible for the relaxation processes arising from thermal tunneling. These are given by (cid:114) π (cid:88) σ aσ(cid:48))νµ2(cid:3) exp Λσσ(cid:48)T ∆σσ(cid:48)2 (cid:2)(a+ (cid:20) σσ(cid:48) + (a+ σ aσ(cid:48))µν2 (cid:21) − (ωµν + Λσσ(cid:48))2 4Λσσ(cid:48)T , (B6) κtun µν = where ∆ is the resonant tunnelling rate, ωµν is the en- ergy difference between the states µ and ν (acting as a thermodynamic gradient), and the reorganization energy Λ are the main guiding factors of the thermal tunneling rates κtun µν . Neglecting the effects of the environment on the radi- ation transitions, krad κrad µν = (cid:88) (cid:17)3(cid:104) × (cid:16) ωµν 2n 3 σσ(cid:48) c µν is given by dσσ(cid:48)2 [(a+ (cid:16) ωµν 2T (cid:17) − 1 (cid:105) σ aσ(cid:48))µν2 + (a+ σ aσ(cid:48))νµ2] coth , (B8) where, n and dσσ(cid:48) stand for the refraction index and the dipole moment, respectively. 6. Lead-induced quenching rates of the excited states The last term of Eq. (B2), kquench , describes the en- ergy loss due to the quenching of the excited state of the photosensitive part of the triad. µν κquench µν = αp[(a+ × ωµν (cid:17) − 1 (cid:105) (cid:16) ωµν P aP ∗ )µν2 + (a+ P aP ∗ )νµ2] . coth (cid:104) 2T (B9) Appendix C: Current and efficiency 1. Electron current For weak couplings, the electron flowing between the leads and the molecular triad is given by: (cid:104)c+ kαckα(cid:105) , Ie = (cid:18) d (cid:19)(cid:88) dt k We derive the equation of the current in terms of the density matrix elements. Ie = ΓR − ΓR aA;µν2[1 − fR(ωνµ)](cid:104)ρν(cid:105) aA;µν2fR(ωνµ)(cid:104)ρµ(cid:105) . (C1) (cid:88) (cid:88) µν 4. Light-induced excitation rates µν The contribution klight to the total relaxation matrix µν (cid:114) π (cid:88) due to light-induced excitation processes is σ aσ∗ )µν2 (cid:21) κlight µν = Λσσ∗ T F2 (cid:20) k × exp − (ωµν + ω0 + Λσσ∗ )2 (cid:8)(a+ (cid:20) − (ωµν − ω0 + Λσσ∗ )2 + (a+ σ aσ∗ )νµ2 exp 4Λσσ∗ T 4Λσσ∗ T (cid:21)(cid:27) (B7) This rate includes contributions from the following tran- sitions, P→P∗, An1 → An1∗, An2 → An2∗, An3 → An3∗, and An4 → An4∗. 2. Absorbed energy ω0F2 (cid:88) (cid:18) The total amount of energy absorbed per unit time Ephoton by the molecular triad and antenna chromophores is Ephoton = (cid:114) π (cid:88) (a+ σ aσ)µν2 (cid:104)ρµ − ρν(cid:105) (cid:20) (cid:21) − (ωµν − Λσσ∗ + ω0)2 (cid:21)(cid:19) (cid:20) − (ωµν − Λσσ∗ − ω0)2 − exp 4Λσσ∗ T Λσσ∗ T (C2) exp × µν σ , 4Λσσ∗ T where σ = P, An1, An2, An3, An4. 3. Power-conversion efficiency The quantum yield is defined as The power-conversion efficiency of the system is the the ratio of the output Eoutput and the input Einput energies, η = Eoutput Einput = Epump Ephoton IR(µR − µL) Ephoton . = (C3) Φ = npump Nphoton = η ω0 V . 13 (C4) [1] Alberts B, Johnson A, Lewis J, Raff M, Roberts K, and Walter P 2002 Molecular Biology of the Cell (Garland Science, New York), Chap. 14. [2] Kuciauskas D, Liddell P A, Lin S, Johnson T E, Weghorn S J, Lindsey J S, Moore A L, Moore T A, and Gust D 1999 J. Am. Chem. Soc. 121 8604 -- 8614. [3] Imahori H 2004 J. Phys. Chem. B 108 6130 -- 6143. [4] Imahori H 2007 Bull. Chem. Soc. Jpn. 80 621-636. [5] Kodis G, Terazono Y, Liddell P A, Andr´easson J, Garg V, Hambourger M, Moore T A, Moore A L, and Gust D 2006 J. Am. Chem. Soc. 128 1818 -- 1827. [12] Wingreen N S, Jauho A P, and Meir Y 1993 Phys. Rev. B 48 8487. [13] Marcus R A, and Sutin N, 1985 Biochim. Biophys. Acta. 811 265 -- 322. [14] Smirnov A Yu, Mourokh L G, and Nori F, 2009 J. Chem. Phys. 130 235105. [15] Ghosh P K, Smirnov A Yu, and Nori F 2009 J. Chem. Phys. 131 035102. [16] Imahori H, Yamada H, Nishimura Y, Yamazaki I, Sakata Y 2000 J. Phys. Chem. B 104 2099 -- 2108. [17] Smirnov A Yu, Mourokh L G, and Nori F 2008 [6] Gust D, Moore T A and Moore A L 2009 Acc. Chem. Res. Phys. Rev. E 77 011919. 42 1890 -- 1898. [7] Engel G S, Calhoun T R, Read E L, Ahn T K, Mancal T, Cheng Y C, Blankenship R E and Fleming G R 2007 Nature 446 782 -- 786. [18] Urboniene V, Vrublevskaja O, Trinkunas G, Gall A, Robert B, and Valkunas L 2007 Biophys. J. 93 2188. [19] Caruso F, Chin A W, Datta A, Huelga S F, Plenio M B 2009 J. Chem. Phys. 131 105106. [8] Gilmore J and McKenzie R H 2008 J. Phys. Chem. A [20] Mohseni M, Rebentrost P, Lloyd S, and Aspuru-Guzik A 112 2162 -- 2176. 2008 J. Chem. Phys. 129 174106. [9] Rebentrost P, Mohseni M, Kassal I, Lloyd S and Aspuru- [21] Garg A, Onuchic J N, and Ambegaokar V 1985 Guzik A 2009 New J. Phys. 11 033003. J. Chem. Phys. 83 4491. [10] Yang S, Xu D Z, Song Z, and Sun C P 2010 J. Chem. [22] Smirnov A Yu, Savel'ev S, Mourokh L G, and Nori F Phys. 132 234501. 2008 Phys. Rev. E 78 031921. [11] Smirnov A Yu, Mourokh L G, Ghosh P K and Nori F 2009 J. Phys. Chem. C 113 21218 -- 21224.
1704.00118	2	1704	2017-09-08T08:02:37	Collision of two action potentials in a single excitable cell	[ "physics.bio-ph" ]	It is a common incident in nature, that two waves or pulses run into each other head-on. The outcome of such an event is of special interest, because it allows conclusions about the underlying physical nature of the pulses. The present experimental study dealt with the head-on meeting of two action potentials (AP) in a single excitable plant cell (Chara braunii internode). The membrane potential was monitored at the two extremal regions of an excitable cell. In control experiments, an AP was excited electrically at either end of the cell cylinder. Subsequently, stimuli were applied simultaneously at both ends of the cell in order to generate two APs that met each other head-on. When two action potentials propagated into each other, the pulses did not penetrate but annihilated (N=26 experiments in n=10 cells). APs in excitable plant cells did not penetrate upon meeting head-on. In the classical electrical model, this behavior is specifically attributed to relaxation of ion channel proteins. From an acoustic point of view, annihilation can be viewed as a result of nonlinear material properties the entire system. The present results suggest that APs in excitable animal and plant cells belong to a similar class of nonlinear phenomena. Intriguingly, other excitation waves in biology (intracellular waves, cortical spreading depression, etc.) also annihilate upon collision and are thus expected to follow the same underlying principles as the observed action potentials.	physics.bio-ph	physics	Collision of two action potentials in a single excitable cell Christian FILLAFER1, Anne PAEGER1, Matthias F. SCHNEIDER1,* 1Medical and Biological Physics Faculty of Physics Technical University Dortmund Otto-Hahn-Str. 4 44227 Dortmund Germany Address correspondence to: Matthias F. Schneider; Physics - Medical and Biological Physics Technical University Dortmund; Otto-Hahn-Str. 4; 44227 Dortmund (Germany); tel: +49-231-755-4139 email: [email protected] Abstract Background. It is a common incident in nature, that two waves or pulses run into each other head-on. The outcome of such an event is of special interest, because it allows conclusions about the underlying physical nature of the pulses. The present experimental study dealt with the head-on meeting of two action potentials (AP) in a single excitable plant cell (Chara braunii internode). Methods. The membrane potential was monitored at the two extremal regions of an excitable cell. In control experiments, an AP was excited electrically at either end of the cell cylinder. Subsequently, stimuli were applied simultaneously at both ends of the cell in order to generate two APs that met each other head-on. Results. When two action potentials propagated into each other, the pulses did not penetrate but annihilated (N=26 experiments in n=10 cells). Conclusions. APs in excitable plant cells did not penetrate upon meeting head-on. In the classical electrical model, this behavior is specifically attributed to relaxation of ion channel proteins. From an acoustic point of view, annihilation can be viewed as a result of nonlinear material properties the entire system. General significance. The present results suggest that APs in excitable animal and plant cells belong to a similar class of nonlinear phenomena. Intriguingly, other excitation waves in biology (intracellular waves, cortical spreading depression, etc.) also annihilate upon collision and are thus expected to follow the same underlying principles as the observed action potentials. Graphical abstract Highlights - When two pulses meet, they reveal information about their physical nature. - Upon running into each other, two action potentials in an excitable plant cell annihilate - Action potentials in plant cells and nerves are similar nonlinear phenomena. Keywords action potential; collision; annihilation; Chara; nonlinear wave; extinguish 1. Introduction Action potentials (AP) are propagating all-or-none pulses that have been observed in excitable cells such as neurons, myocytes, plant cells (e.g. Charophytes), etc. These cells can assume different morphologies which commonly contain elongated cylindrical elements. APs can be stimulated at either end of such a cylinder, i.e. propagation in the antero- as well as retrograde direction is possible (in the neurophysiology literature the terms ortho- (in "natural" direction of fibre) and antidromic (against "natural" direction of fibre) are most often used). Thus, two pulses may run into each other. This leads to several interesting questions, for instance, do two APs superimpose and penetrate or do they partially/fully annihilate? The outcome of such an experiment is an important criterion, since it provides information about the physical nature of an action potential. Moreover, it has been suggested that a reduction of pulse collision events may drive the formation of directional transmission in nerve cell networks (K. Kaufmann, personal communication). What is the present state of evidence? In neurophysiology, annihilation of two APs upon collision – due to the trailing refractory zone – is considered to be an established fact. This assertion provides the rationale for several experimental techniques. One of those is the "collision test", which is used, amongst others, to determine the level of natural (i.e. orthodromic) activity in a nerve trunk [1,2]. In a collision test, antidromic APs are stimulated artificially. Only those pulses which propagate in axons with no orthodromic activity will be detectable at the opposite end of the nerve, whereas the others will be extinguished by collision. Based on these ideas, a technique called "collision block" has been conceived [3]. The goal of the latter is to avoid overstimulation of an end organ as a result of high nerve activity. For this purpose, antidromic APs are stimulated artificially to eliminate orthodromic pulses. The results of collision test as well as collision block experiments provide evidence that APs annihilate upon meeting head on. However, since such studies assume the premise that collision leads to annihilation to be true they are not ideal to investigate it. Otherwise, to the best of our knowledge, only few systematic studies exist. In 1949, Tasaki demonstrated that two APs extinguish upon meeting head-on in single myelinated nerve fibres [4]. These results were corroborated in ventral nerve cord of earthworm (Lumbricus hortensis) [5], the stomatogastric nervous system of the crab (Cancer borealis) [5], as well as close to bifurcations in lobster axons [6]. In contradiction to these findings, Gonzalez- Perez et al. reported penetration of APs in ventral nerve cord from Lumbricus terrestris as well as Homarus americanus [7]. The potential importance of pulse collision for nervous network formation, the scarcity of research of the phenomenon and the possibility of different outcomes have motivated us to conduct the present study. So far, most experiments in the literature have been carried out with nerve trunks. The latter preparations are not ideally suited to obtain clear results. A nerve consists of several fibres with varying diameters which are usually excited and recorded by external electrodes. To be able to properly set up and interpret such experiments requires extensive experience. It has to be ascertained, for example, that the component fibre in which the pulses are excited is known. Otherwise, ambiguities remain (e.g. two different fibres may be stimulated and thus the pulses never come to meet each other which will be misinterpreted as penetration). In any case, it is difficult to judge from extracellular recordings from a nerve trunk whether annihilation in a component fibre is complete or partial. Thus, it seemed expedient to study the head-on meeting of two APs in a preparation that allows for unambiguous conclusions. Ideally, such a system is simple to set up so that the results can be easily reproduced/falsified by others. Giant internodes from Charophytes are well suited for this purpose. These cylindrical plant cells are one of the most extensively studied systems in excitable cell physiology [8–10]. Their large diameter (0.5-1 mm) and cell length (up to 15 cm) dramatically facilitate experimental work. Thus, we use Chara braunii internodes herein to study what happens when two APs run into each other. 2. Materials and Methods 2.1 Materials. All reagents used were purchased from Sigma-Aldrich (St. Louis, MO, USA) and were of analytical purity (≥99%). 2.2 Cell cultivation and storage. Chara braunii cells were cultivated in glass aquariums filled with a layer of soil (~2 cm), quartz sand (~1cm) and deionized water. The cells were grown under illumination from an aquarium light (14W, Flora Sun Max Plant Growth, Zoo Med Laboratories Inc., San Luis Obispo, CA, USA) at a 14:10 light:dark cycle at room temperature (~20°C). Prior to use, single internodal cells were stored for a minimum of 12 h in a solution containing 0.1 mM NaCl, 0.1 mM KCl and 0.1 mM CaCl2 [8]. 2.3 Experimental setup. A single internodal cell (6-12 cm long) was placed on a plexiglass chamber into which compartments (~5 x 5 x 10 mm; h x w x l) had been milled. Small extracellular sections (length ~5 mm) of the cell were electrically isolated against each other with vacuum grease (Dow Corning Corporation, Midland, MI, USA). The K+-anesthesia technique [8] in combination with extracellular Ag/AgCl-wire electrodes was used for monitoring the cell membrane potential. The extracellular solutions contained 110 mM KCl in the outmost compartments and artificial pond water (APW) in all other compartments (1 mM KCl, 1 mM CaCl2, 5 mM TRIS and 190 mM D-sorbitol; pH set to 7.0 with HCl). The potential between the virtual intracellular electrodes (KCl-compartments) and extracellular electrodes was recorded with voltage sensors (PS-2132; 100Hz sample rate; PASCO scientific, Roseville, CA, USA). All experiments were conducted at room temperature (20±2 °C). 2.4 Controls and collision experiment. Upon equilibration of the cell in the measurement chamber, an action potential was excited at either end of the cell by application of a rectangular current pulse across the outer compartments (Fig. 1A; 9V applied across a 1 MΩ resistor; stimulus duration: 50-100 msec; stimulus frequency: ~1AP per 15 min). This procedure ascertained that the internode was fully excitable and that the triggered AP propagated along the entire cell length. In order to generate a collision event, stimuli were applied simultaneously at both ends of the cell. In total, N=26 experiments were performed in which two APs met each other head-on (n=10 cells). All electrophysiological data for control and collision experiments are publicly available [11]. 3. Results and Discussion 3.1 Control experiments. An action potential was excited in a Chara cell by application of a superthreshold electrical shock. Subsequently, the pulse propagated along the cell cylinder. Upon stimulation at the left end of the cell, the AP passed by the two potential sensors as expected from their geometrical setup (Fig. 1B and 2B). It was experimentally confirmed that stimulation at the right end of the cell led to propagation along the reversed path (Fig. 1C and 2C). These control experiments ascertained that both ends of the cell are electrically excitable and they established the characteristic appearance of an AP at the sensor locations. Typically, the duration of an action potential was in the range of ~10 s and the conduction velocity was ~5 mm s-1 from which it is concluded that the pulse has a spatial extension or wavelength of ~5 cm. The pulse amplitude, duration as well as propagation velocity found herein are in accordance with the ones reported in the literature [8,9]. The pulse shape in Characean cells can be subject to slight variations (e.g. along the cell cylinder or with duration of an experiment). This is usually more pronounced in some cells than in others. Thus, additional control experiments were intermittently performed in course of an experiment in order to identify changes in pulse shape that are not due to pulse collision. These experiments revealed that broadening of an AP or small distortions can occur in absence of collision (see examples in [11]). Therefore, caution must be taken in order not to erroneously attribute such changes in pulse shape during a collision experiment to pulse penetration, etc. Figure 1. Collision of action potentials (AP) in Chara braunii cells. (A) The membrane potential was monitored at the two extremal regions (1) (solid line) and (2) (dotted line) of a Chara internode. By application of a superthreshold electrical stimulus, an AP was excited at the left (L) and right end (R) of the cell cylinder. (B) A single excitation at L led to the opposite propagation path as compared to a single excitation at R (C). (D, E) Simultaneous excitation at L and R led to the head-on meeting of two APs in the central region of the cell. At each sensor only one AP was recorded which indicates that the pulses did not penetrate. 3.2 Collision of two action potentials. When two APs are excited simultaneously at both ends of a cylindrical cell, they propagate towards the center and meet head-on. Prior to a discussion of the results, it is instructive to consider the possible outcomes of such an experiment. If the two APs superimpose and penetrate or reflect off of each other, one expects the appearance of two pulses at each sensor. The two pulses could manifest either as a fused event (the membrane potential deflection may assume, for example, an "M"-like shape) or as discrete peaks. Based on the spatial extension of an AP in Chara internodes as calculated above (~5 cm), two discrete pulse events should be observable at the extremal sensors if the pulses simply penetrate and if the cell length is larger than ~8 cm. In case of shorter cells, varying degrees of pulse overlap are expected. In contrast, none of these possibilities will be realized if the two pulses annihilate. In this case, only one discrete, undistorted AP should appear at each potential sensor. Which of these outcomes is realized in Chara internodes? To investigate this question, two APs were triggered simultaneously at both ends of the cell cylinder (Fig. 1D). In none of the experiments conducted (N=26 in n=10 cells; data for all control and collision experiments are publicly available [11]), an additional discrete AP or "M"-shaped pulse appeared at the extremal sensors. In one of the cells, one external sensor recorded a broadened AP. However, in a subsequent control experiment in the same cell a similar broadening was observed, which suggest that the reason for broadening is not a penetrated pulse. The lack of amplitude doubling in the central region of the cell (Fig. 2D and data in [11]), where collision occurs, provided further evidence that two APs do not superimpose and penetrate. Interestingly, the multisensor arrangement also allowed to extract the pulse propagation velocities for the control and collision experiments. In 19 of 25 measurements the velocity was increased (relative change on average ~10%) when two pulses were excited simultaneously. This essentially confirms an observation of Tasaki who reported an increase in conduction velocity when two nerve pulses approach each other head-on [4]. Taken together, these experiments with multiple sensors along the cell underline that two action potentials that meet head-on – within the conditions tested – neither summate and penetrate nor reflect off of each other. Figure 2. Collision of two action potentials monitored with multiple sensors along the cell. (A) The membrane potential was recorded at the two extremal regions (solid line (1) and dotted line (4)) and in the central parts (solid (2) and dotted (3)) of a Chara internode. (B) A single excitation at L led to the opposite propagation path as compared to a single excitation at R (C). (D) Simultaneous excitation at L and R led to the head-on meeting of two APs in the central region of the cell. Note: At all sensors only one, undistorted AP was recorded which indicates that the pulses neither superimposed nor penetrated. Annihilation is most evident in the absence of additional pulses in the extremal sensors (grey region). This conclusion is in agreement with previous findings in myelinated nerve fibers from frog [4], lobster axons [6], as well as ventral nerve cords of earthworms and the stomatogastric nervous system of the crab [5]. To the best of our knowledge, the only work which disagrees is the report of pulse penetration by Gonzalez-Perez et al. [7]. It will be valuable to address in future studies, if the latter findings can be replicated by other investigators. Until then, the majority of results suggest that annihilation of two APs is a general characteristic in excitable plant cells as well as nerve fibers. Albeit penetration of two APs in Chara can be excluded, it is of interest whether annihilation of the two pulses is complete or partial. In 1 out of 26 collision experiments, an additional low-amplitude * while the present work was under review, a debate of the study has taken place [36,37]. deflection of the membrane potential record was observed, which points to a residual pulse component (Fig. 1E). The existence of such a residual may be related to Tasaki's observation of record baseline fluctuations after collision of two APs in myelinated frog fibers [4]. Future studies could address if partial annihilation is generally a very rare event or if its likelihood increases for certain states of the excitable cell, for instance, when the refractory period becomes short (e.g. at elevated temperature). 3.3 Pulse collision and the physical nature of an action potential. At present, APs are considered to be electrical phenomena and their theoretical description is based on the formalism established by Hodgkin and Huxley [12]. The latter model, despite its widespread acceptance, has come under criticism. From a physical perspective, the severest objections are the existence of reversible mechanical, optical, thermal, etc. pulse components, which are not contained in a dissipative electrical framework. These discrepancies will not be discussed herein since they were laid out in detail by others (see e.g. [7,13–15]). The shortcomings of the classical electrical model have motivated the development of alternative theories of the action potential [7,13–15]. Kaufmann, for example, proposed that an AP is an acoustic† pulse in the lipid bilayer [14]. Recently, evidence in favor of this proposition has been obtained. In lipid monolayers, linear [16,17] as well as nonlinear waves [18] have been excited and detected by a variety of means (e.g. chemically, mechanically, electrically, etc.). The solitary waves in particular share several key characteristics with APs (threshold, all-or-none behavior, etc.) [18,19]. These pulses also propagate over extended distances without major changes to their shape. This behavior is usually attributed to particular nonlinear material properties which counteract dispersion of the wave packet. Solitary waves have been observed in many different fields such as hydrodynamics [20], optics [21], Bose-Einstein condensates [22] and during catalytic surface reactions [23]. Which predictions emerge from linear and nonlinear acoustics as pertaining to the head-on meeting of two pulses? In a recent work, it was suggested that the only conceivable outcome is superposition and penetration [5]. This is indeed expected in the realm of linear acoustics. In excitable cells, however, a linear relationship between stimulus and response is only observed in certain regimes. In postsynaptic membrane regions of neurons, for example, typical excitatory and inhibitory inputs summate [24]. This is reminiscent of wave superposition. The membrane of axons also responds linearly to hyperpolarizing stimuli as well as depolarizing stimuli with small amplitude [24] and such membrane potential displacements should superimpose. However, as the amplitude of depolarization of an axon is increased, the stimulus-response curve becomes nonlinear and an action potential is induced. Taken together, these observations demonstrate an important point, namely that linear as well as nonlinear perturbations can be excited in the same system. APs are phenomena in the nonlinear regime. There, the superposition principle does not apply and an extended spectrum of outcomes including complete and partial annihilation may be realized when two pulses meet. Some possibilities shall be discussed. † It has to be emphasized that the term "acoustic" does not imply that an AP is a purely mechanical process. Rather, a propagating perturbation must manifest in local oscillations of all thermodynamic observables of the excitable medium (area density, charge density, internal energy density, etc.). It has been suggested that APs are a class of nonlinear waves, so called solitons [7,15,25]. Upon meeting head-on, two solitons – similar to linear waves – penetrate and continue to propagate with widely preserved shape and speed [7,23,25]. This behavior is observed, for instance, for two such waves meeting on a water channel surface [20]. It was predicted that two solitons propagating in a lipid bilayer membrane should also penetrate [25]. The present results and the majority of reports in the literature [1,2,4–6] indicate, however, that pulse penetration is not observed in excitable membranes (note [7] as an exception). Considering this evidence, it is unlikely that APs are correctly described as solitons. It must be emphasized, however, that this does not falsify the proposition that an action potential is a nonlinear acoustic phenomenon. Other solitary waves annihilate upon meeting head-on, for example those in Bose Einstein condensates [22] as well as during catalytic surface reactions [23]. Interestingly, however, annihilation seems to be a comparatively rare scenario. One reason may be that annihilation requires a special type of nonlinearity in the system. There are indications, for example, that excitable cells reside close to a phase transition [13,26]. If the latter constitutes the nonlinearity that allows for the formation of an AP, the pulse could entail a phase change of the excitable medium. This transient realization of a different material phase may lead to pulse annihilation when two APs meet. In order to be able to penetrate, the pulses would have to propagate through this new phase. The latter, however, lacks the nonlinearity of the resting state and thus does not support this type of wave. This leads us to the traditional explanation in the electrophysiological framework [5]. There, it is held firmly that two APs annihilate upon collision because of a refractory zone that trails the pulse [1–3,5]. In the electrical framework, the latter acquires its impenetrability from the properties of single molecules (i.e. ion channel proteins). Shortly after an AP, these proteins are believed to be in an inactive state [5]. Only after relaxation to their resting state, a subsequent excitation of the membrane will be possible. In this explanation, the excitable membrane is separated into active (ion channel proteins) and passive parts (lipid bilayer). This opens up a possibility to test the different theories of the AP. In acoustics, the propagation medium is not subdivided. The latter theory will be falsified if it is shown that the lipid bilayer remains in a constant thermodynamic state during an AP. This has often been presumed to be the case, because early measurements indicated that the electrical capacity of the excitable membrane does not change during activity [27]. Later experiments demonstrated, however, that this assumption was incorrect and that capacity shifts by ~25% during activity in squid axons [28]. Thus, it will be of interest to re-investigate this central issue by monitoring the thermodynamic state of an excitable membrane during an AP. In addition, it will be important to study if solitary waves in lipid monolayers annihilate upon collision. This will provide a direct test of the hypothesis that specific membrane proteins are required for pulse annihilation in a lipid membrane interface. Another question that remains is what happens at the site of collision. The simplest possibility is that the local temperature increases. This should, in principle, be observable with very sensitive thermal sensors [29]. However, in particular in biological materials, which have many degrees of freedom, it is conceivable that non-thermal transformations take place (budding/fusion of vesicles, changes of pH, transformation into chemical energy, etc.). Finally, we would like to point out that there are several examples of other wave phenomena that annihilate upon meeting head-on. The latter was reported, for instance, in the Belousov-Zhabotinsky reaction system [30]. Waves of NADH and proton concentration which occur during glycolytic reactions in organelle-free yeast extracts also annihilate [31]. The same behavior was observed for intracellular waves in myocytes [32] and oocytes [33]. Further examples exist in multicellular systems. Mechanical stimulation of the nerve net of a sea pen is associated with luminescence waves and the latter extinguish upon running into each other [34]. Castro and Martins-Ferreira demonstrated in remarkable experiments with chick retina that this also holds true for two spreading depression waves in cortical tissue [35]. Taken together, these findings suggest (i) that specific molecules are unlikely to be the cause of annihilation because the phenomenon occurs in media with different composition and (ii) the intriguing possibility that different biological interfaces (plasma membrane, cytoplasm, cerebral cortical tissue, etc.) have been optimized for propagation of similar nonlinear waves. 4. Conclusions When two action potentials were excited simultaneously at the ends of an excitable plant cell cylinder (Chara internode), they propagated towards each other and met head-on. The pulses annihilated in all experiments conducted. The majority of collision experiments carried out under standard conditions in nerve fibers came to the same conclusion. Thus, APs in excitable plant cells and axons share this as a common feature. It will be of interest to study if nonlinear solitary waves in lipid monolayers, which resemble action potentials, also extinguish upon collision. Further experiments should aim at understanding the mechanism that underlies annihilation. Concerning the latter, different theories make different predictions: While the relaxation of single molecules is crucial in the electrophysiological framework, an acoustic theory emphasizes the role of nonlinear material properties of the excitable medium. Acknowledgements We thank K. Kaufmann for stimulating us to work on this problem and for numerous lectures and discussion sessions. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. References [1] J. Lipski, Antidromic activation of neurones as an analytic tool in the study of the central nervous system, J. Neurosci. Methods. 4 (1981) 1–32. [2] W. Douglas, J. Ritchie, A technique for recording functional activity in specific groups of [3] [4] [5] [6] [7] medullated and non-medullated fibres in whole nerve trunks, J. Physiol. 138 (1957) 19–30. C. van den Honert, J.T. Mortimer, Generation of unidirectionally propagated action potentials in a peripheral nerve by brief stimuli, Science. 206 (1979) 1311–2. I. Tasaki, Collision of two nerve impulses in the nerve fibre, Biochim. Biophys. Acta. 3 (1949) 494–497. R. Follmann, E. Rosa, W. Stein, Dynamics of signal propagation and collision in axons, Phys. Rev. E - Stat. Nonlinear, Soft Matter Phys. 92 (2015) 32707. Y. Grossman, I. Parnas, M.E. Spira, Differential conduction block in branches of a bifurcating axon., J. Physiol. 295 (1979) 283–305. A. Gonzalez-Perez, R. Budvytyte, L. Mosgaard, S. Nissen, T. Heimburg, Penetration of action potentials during collision in the medial giant axon of the earthworm, Phys. Rev. X. 4 (2014) 31047. H. Lühring, Algen unter Strom: Das Experiment, Biol. Unserer Zeit. 36 (2006) 313–321. R. Wayne, Excitability in Plant Cells, Am. Sci. 81 (1993) 140–151. [8] [9] [10] M.J. Beilby, Action potential in Charophytes, Int. Rev. Cytol. 257 (2007) 43–82. [11] C. Fillafer, A. Paeger, M. Schneider, Dataset "Collision of two action potentials in a single excitable cell," (2017). doi:10.17632/nmpxg8z5mb.1. [12] A.L. Hodgkin, A.F. Huxley, A quantitative description of membrane current and its application to conduction and excitation in nerve, J. Physiol. 117 (1952) 500–544. I. Tasaki, Evidence for phase transition in nerve fibers, cells and synapses, Ferroelectrics. 220 (1999) 305–316. [14] K. Kaufmann, Action potentials and electromechanical coupling in the macroscopic chiral [13] phospholipid bilayer, Caruaru, Brazil, 1989. [15] T. Heimburg, A.D. Jackson, On the action potential as a propagating density pulse and the role of anesthetics, Biophys. Rev. Lett. 2 (2007) 57–78. J. Griesbauer, S. Bössinger, A. Wixforth, M. Schneider, Propagation of 2D pressure pulses in lipid monolayers and its possible implications for biology, Phys. Rev. Lett. 108 (2012) 198103. [17] B. Fichtl, S. Shrivastava, M. Schneider, Protons at the speed of sound: Predicting specific [16] biological signaling from physics, Sci. Rep. 6 (2016) 22874. [18] S. Shrivastava, M. Schneider, Evidence for two-dimensional solitary sound waves in a lipid controlled interface and its implications for biological signalling, J. R. Soc. Interface. 11 (2014) 20140098. [19] S. Shrivastava, K. Kang, M. Schneider, Solitary shock waves and adiabatic phase transition in lipid interfaces and nerves, Phys. Rev. E. 91 (2015) 12715. [20] Y.S. Chen, H. Yeh, Laboratory experiments on counter-propagating collisions of solitary waves. Part 1. Wave interactions, J. Fluid Mech. 749 (2014) 577–596. [21] Z. Chen, M. Segev, D.N. Christodoulides, Optical spatial solitons: historical overview and recent advances, Reports Prog. Phys. 75 (2012) 86401. J.H. V. Nguyen, P. Dyke, D. Luo, B.A. Malomed, R.G. Hulet, Collisions of matter-wave solitons, Nat. Phys. 10 (2014) 918–922. [23] H.H. Rotermund, S. Jakubith, A. Von Oertzen, G. Ertl, Solitons in a surface reaction, Phys. [22] [24] D. Aidley, The physiology of excitable cells, 4th ed., Cambridge University Press, Cambridge, Rev. Lett. 66 (1991) 3083–3086. UK, 1998. [25] B. Lautrup, R. Appali, A. Jackson, T. Heimburg, The stability of solitons in biomembranes and nerves., Eur. Phys. J. E. Soft Matter. 34 (2011) 1–9. [26] T. Ueda, M. Muratsugu, I. Inoue, Y. Kobatake, Structural changes of excitable membrane formed on the surface of protoplasmic drops isolated from Nitella., J. Membr. Biol. 18 (1974) 177–186. [27] A.L. Hodgkin, A.F. Huxley, Resting and action potentials in single nerve fibres, J. Physiol. 104 (1945) 176–195. [28] S. Takashima, Admittance change of squid axon during action potentials, Biophys. J. 26 (1979) [29] 133–142. I. Tasaki, Rapid structural changes in nerve fibers and cells associated with their excitation processes, Jpn. J. Physiol. 49 (1999) 125–138. [30] P.M. Wood, J. Ross, A quantitative study of chemical waves in the Belousov–Zhabotinsky [31] T. Mair, S.C. Müller, Traveling NADH and proton waves during oscillatory glycolysis in vitro, reaction, J. Chem. Phys. 82 (1985) 1924. J. Biol. Chem. 271 (1996) 627–630. [32] N. Ishide, T. Urayama, K. Inoue, T. Komaru, T. Takishima, Propagation and collision characteristics of calcium waves in rat myocytes, Am. J. Physiol. Hear. Circ. Physiol. 259 (1990) H940–H950. J. Lechleiter, S. Girard, E. Peralta, D. Clapham, Spiral calcium wave propagation and annihilation in Xenopus laevis oocytes, Science. 252 (1991) 123–126. [33] [34] A. Moore, On the nature of inhibition in Pennatula, Am. J. Physiol. 76 (1926) 112–115. [35] G.O. Castro, H. Martins-Ferreira, Deformations and thickness variations accompanying spreading depression in the retina, J. Neurophysiol. 33 (1970) 891–900. [36] R. Berg, M. Tving Stauning, J. Balslev Sorensen, H. Jahnsen, Comment on "Penetration of action potentials during collision in the median and lateral giant axons of invertebrates," Phys. Rev. X. 7 (2017) 28001. [37] T. Wang, A. Gonzalez-Perez, R. Budvytyte, A.D. Jackson, T. Heimburg, Reply to "Comment on 'Penetration of action potentials during collision in the median and lateral giant axons of invertebrates,'" Phys. Rev. X. 7 (2017) 28002.
1811.12076	1	1811	2018-11-29T11:32:09	Emergence of active nematics in bacterial biofilms	[ "physics.bio-ph", "cond-mat.soft" ]	Growing tissue and bacterial colonies are active matter systems where cell divisions and cellular motion generate active stress. Although they operate in the non-equilibrium regime, these biological systems can form large-scale ordered structures such as nematically aligned cells, topological defects, and fingerings. Mechanical instabilities also play an essential role during growth by generating large structural folding. How active matter dynamics and mechanical instabilities together develop large-scale order in growing tissue is not well understood. Here, we use chain forming Bacillus subtilis, also known as a biofilm, to study the direct relation between active stress and nematic ordering. We find that a bacterial biofilm has intrinsic length scales above which series of mechanical instabilities occur. Localized stress and friction control both linear buckling and edge instabilities. Remarkably, these instabilities develop nematically aligned cellular structures and create pairs of motile and stationary topological defects. We also observe that stress distribution across the biofilm strongly depends on the defect dynamics which can further initiate the formation of sporulation sites by creating three-dimensional structures. By investigating the development of bacterial biofilms and their mechanical instabilities we are proposing a new type of active matter system which provides a unique platform to study the essential roles of nematics in growing biological tissue.	physics.bio-ph	physics	Emergence of Active Nematics in Bacterial Biofilms Yusuf Ilker Yaman1, Esin Demir2, Roman Vetter3, Askin Kocabas1,4∗ 1Department of Physics, Koç University, 34450 Sarıyer, Istanbul, Turkey 2Bio-Medical Sciences and Engineering Program, Koç University, 34450 Sarıyer, Istanbul, Turkey 3Department of Biosystems Science and Engineering, ETH Zurich, 4058 Basel, Switzerland 4Koç University Surface Science and Technology Center, Koç University, 34450 Sarıyer, Istanbul, Turkey Abstract Growing tissue and bacterial colonies are active matter systems where cell divisions and cellular motion generate active stress. Although they operate in the non-equilibrium regime, these biological systems can form large-scale ordered structures such as nematically aligned cells, topological defects, and fingerings. Mechanical instabilities also play an essential role during growth by generating large structural folding. How active matter dynamics and mechanical instabilities together develop large-scale order in growing tissue is not well understood. Here, we use chain forming Bacillus subtilis, also known as a biofilm, to study the direct relation between active stress and nematic ordering. We find that a bacterial biofilm has intrinsic length scales above which series of mechanical instabilities occur. Localized stress and friction control both linear buckling and edge instabilities. Remarkably, these instabilities develop nematically aligned cellular structures and create pairs of motile and stationary topological defects. We also observe that stress distribution across the biofilm strongly depends on the defect dynamics which can further initiate the formation of sporulation sites by creating three-dimensional structures. By investigating the development of bacterial biofilms and their mechanical instabilities we are proposing a new type of active matter system which provides a unique platform to study the essential roles of nematics in growing biological tissue. Main Biofilm formation is a collective response of bacteria1−6. Depending on the availability of food and environmental conditions7, B. subtilis produce matrix proteins and initiate the formation of a biofilm2. During biofilm development, motile bac- teria differentiate into an aligned chain of cells. Growing chains further develop fibers and bundles which shape the overall biofilm morphology8−13. These distinct aligned structures promote sliding of a colony on a solid surface where the swimming be- havior is not efficient12,14. Aligned cellular structures are also observed in a variety of biological phenomena. During wound healing, migrating cells align and form fingering structures at the leading edge of the tissue15. Sim- ilarly, cultured cells16−18 and isolated bacterial colonies19−23 can form nematic alignment and mod- ulate the cellular density and active stress. Recent studies have shown that liquid crystal the- ory can provide a suitable framework to study the dynamics of growing tissue as an active nematic sys- ∗Corresponding author: [email protected] 1 tem. Mainly, the dynamics of cellular alignment, topological defects, and edge instabilities have been explored16−20. Here, by investigating the formation of a bacterial biofilm starting from a single bac- terium, we have studied the detailed mechanical in- stabilities driving the dynamics of nematic active matter. We revealed the direct relation between lo- cal stress and localized buckling. This system pro- vides not only a new platform to observe the essen- tial dynamics of active nematics in a growing tis- sue but also the mechanical insights to explore the physics of nonbiological active nematics (AN)24−28. Results We first used time-lapse fluorescence microscopy to observe the temporal evolution of a biofilm forma- tion. Figure 1 shows the snapshots from the growth of a chaining B. subtilis biofilm, laboratory strain 168 (Fig. 1). This strain apparently forms a biofilm on a solid agar surface. Initially, divided cells give rise to a long and straight chain of attached bacte- ria. After several cell divisions, the first mechanical instability occurs (Supplementary Movie 1). Figure 1: Emergence of active nematics in a bacterial biofilm. Snapshots of a growing biofilm. a The bacterium initially elongates and forms a bacterial chain through proliferation (t = 2.5h). Scale bar, 30µm. b Due to local stress, the chain of attached bacteria buckles (t = 7h). Scale bar, 30µm. c Multilayered and circular structures appear at the region where the localized buckling takes place (t = 11h). Scale bar, 30µm. d The biofilm formation (t = 28h). Scale bar, 150µm. e Kink bands appear and separate different domains across the biofilm. Scale bar, 30µm. Scanning electron microscopy images of f the biofilm formed by GFP labeled laboratory strain 168 (BAK47) and g the isolated bacterial colony formed by the non-chaining strain. All colonies were grown at T = 21 oC. Interestingly, unlike Euler instability, this initial buckling is very localized. As the bacterial chain grows, the buckled region forms a crumpled struc- ture (Fig. 1c). Some parts of this structure become multilayered and circular. Similarly, these multilay- ered structures continue to grow radially, and they split into two (Supplementary Figure 1). Moreover, sharp walls appear across the biofilm, and these bands are observed as dark lines separating differ- ent domains in fluorescence images (Fig. 1d, e). All the basic dynamics and shapes described above were observed repeatedly across the biofilm and generate perfectly aligned cellular structures (Fig. 1f). This aligned structure and its dynamics strongly resem- ble the nematic active matter systems, particularly microtubule-based AN27,28 (Supplementary Movie 2, 3, 4). To clarify the structural differences between ne- matic biofilm and isolated bacterial colony, we grew a similar but non-chaining B. subtilis strain (BAK51, a derivative of 3610) under the same en- vironmental conditions. This wild isolate failed to form chaining on an agar surface but can form a biofilm at a later stage stochastically (Supple- mentary Movie 5). In our bacterial strains, the flagella producing gene (hag) was also mutated to eliminate the swimming induced motion. Detailed scanning electron microscopy images show tightly packed cells with fairly smooth colony edges (Fig. 1g, Supplementary Figure 2). Clear nematic alignment of the biofilm required optimization of several parameters. First, we re- duced the growth rate by decreasing the ambient temperature to eliminate the twisting and break- age of a bacterial chain (Supplementary Figure 3). Second, GFP labeling is essential to observe finely 2 Figure 2: Localized buckling of a bacterial chain. a A snapshot of a bacterial chain with a length just above the critical buckling length. The buckled region is highly localized due to b localization of stress generated by the uniform friction force. The stress accumulates from the tip to the center of the chain due to outward axial velocity. The uniform friction force creates an almost linear stress distribution along the chain. c Plot of critical buckling length against friction, simulation results. Inlet shows a snapshot of the simulation just after the buckling and the color code shows the tension at pre-buckling stage. d Successive localized buckling instabilities on 1.5 %, 2.5 %, and 3.5 %, respectively. After the first buckling, the subsequent buckling occurs at the middle of the straight part of the chain, which implies that the stress is symmetric around the center, and the connection of the former buckling region with the straight part is effectively a free end. Therefore, the separation between buckled regions is equal to half of the critical buckling length. e Experimental results of the critical buckling length for various agarose concentrations. The data is collected from more than 100 chains. f Schematic of a simplified regulatory network that controls the biofilm formation. The protein sinI regulates the chaining state through double-negative feedback. g Bacterial chain grown in liquid LB. The chaining formation is initiated by IPTG induction. In the later stages, the colony forms a biofilm via supercoiling. Scale bar, 50µm (top) and 200µm (bottom). aligned structures, but low power fluorescence ex- citation is also necessary to eliminate light toxicity during biofilm development. Localized buckling appears to be the first build- ing block of the nematic biofilm (Fig. 2a). To ex- plore the physical mechanism underlying localiza- tion, we quantified the buckling condition. The lo- calized buckling occurs just after reaching the crit- ical chain length. As the chain continues to grow, extended arms follow the same localized buckling scenario. The elasticity theory explaining the localization of the buckling is surprisingly complex. Both linear and nonlinear elastic properties of soft materials can contribute to the localization and formation of the post-buckling shape29−34. However, the theoreti- cal framework for growth-induced localized buck- ling is not well-studied. We have noticed that rail- road thermal buckling35 and growing plant roots36 show a similar spatial localization profile. There- fore, we used the notation and the framework de- veloped for these systems and followed the slender- body approximation to simulate the dynamics of a growing bacterial chain. We ignored the twisting dynamics, and we modeled the bacterial chain as an incompressible flexible elastic rod. Since the system is at low Reynolds number regime, we neglected the inertial forces and used the equations for mechanical equilibrium. According to the slender-body theory, in mechanical equilibrium, the following equation is satisfied. = −K(s) (1) dF(s) ds Where F(s) is the internal force, and K(s) is the 3 external force acting on a unit length of the rod. We are interested in the pre-buckling tension profile. Therefore, we assumed that the bacterial chain is a straight rod with total length ' lying on x-axis. Hence the tension can be written as: T(x) = f(abs(x) − '/2) (2) where f is the uniform friction force acting on a unit length of the rod (see Supplementary Information 1 for a detailed derivation). T is linear and always negative implying that the rod is compressed. Ten- sion profile locally exceeds the critical buckling lim- its when the total length reaches the threshold (Fig. 2b). This buckling condition sets a length scale for a bacterial chain. To further quantify the buckling conditions, we performed FEM simulations based on recently developed algorithms used to study growing elastic structures37−39. Using realistic bacterial pa- rameters, we found that an increase in friction force reduces the critical length (Fig. 2c). To experimen- tally verify this effect, we tuned the agarose con- centration which serves as a flat surface and imaged the periodically buckled regions of a single bacterial chain. Our experiments revealed that the critical length changes with the agarose concentration (Fig. 2d, e). Typically, the bacterial chain becomes unsta- ble above 120µm. Interestingly, high agarose con- centration reduces the friction force and increases the critical length. The exact mechanism behind the friction40 is not clear to us; we speculate that the soft agar surface allows more local deformations and increases the friction force. Further, we tested the local buckling in a liq- uid environment which eliminates the friction force. In its wild-type form, B. subtilis suppress biofilm formation in liquid LB culture. Detailed molecu- lar genetics studies have identified that protein sinI initiates and maintains the chaining state through double-negative feedback1,3,4 (Fig. 2f). To trig- ger biofilm formation in liquid, protein sinI should be overexpressed. To overcome this limitation, we used a genetically modified strain to drive sinI us- ing IPTG based induction (the background strains TMN1152 were received from R. Losick Lab). When we grew the biofilm in liquid culture with IPTG, lo- cal buckling disappeared, but the supercoiling pro- cess dominated the formation of the biofilm (Fig. 2g. Supplementary Movie 6, 7). We further clari- fied the structure of supercoiled bundles using flu- orescence time-lapse microscopy and SEM imaging (Supplementary Figure 4). Altogether, we showed that friction force between the agar surface and the bacterial chain controls the mechanical instability. Localization of stress further generates spatially lo- calized buckling and defines the critical length scale for a biofilm. Figure 3: Edge instabilities of a growing biofilm. a Snapshots of growing multilayered circles. The radius reaches the critical radius and splits into two. Scale bar, 30µm b Snapshots from the simulation of a droplet-like structure with a long tail. Color code shows the stress. While the droplet splits into two circles, the long tail locally buckles since its length is above the critical buckling length. c Multilayered growing circles are simulated under various friction forces. The plot of critical radius against friction, simulation results. Units are arbitrary. All snapshots were taken as the sec- ondary edge instability occurred. d Experimental data of the critical radius. The colonies were grown on 1.5 % agarose concentration. As a second step, we focused on the edge insta- bilities of a growing biofilm. The growing multi- layered circles are the most distinct geometric fea- tures observed around the leading edge (Fig. 3a). At first glance, the dynamics of these structures re- semble the fingering instabilities of Hele-Shaw cell41 and smectic-A liquid crystal filaments42,43. These circles are connected to the film with a tail, and they resemble growing droplets. As they grow out- 4 ward, instability occurs, and the droplets split into two. As gleaned from previous studies, we tested whether these structures have a characteristic crit- ical radius which triggers new mechanical instabili- ties. We performed FEM simulations to observe the splitting dynamics. First, we simulated the grow- ing multilayered circular structure with a tail (Fig. 3b, Supplementary Movie 8). We found that these structures indeed have a critical radius. Above this radius, circular structures are unstable and split into smaller but stably growing ones. As we observed in previous experiments, the straight tail also locally buckles in the direction perpendicular to growing axis after reaching the critical length. Our simula- tion verified that edge instability occurs when the circumference of the circular structure exceeds the linear critical length 2πRc = 'c (Fig. 3c, d). We confirmed this relationship experimentally. Simi- larly, both a critical radius and a critical length are controlled by friction. We also observed that, just after the splitting, an additional buckling might occur around the junction point (connection between the droplet and the tail) where the chain has the highest curvature (green ar- row in Fig. 3b). This buckling further makes the divided droplet fall back on the biofilm edge asym- metrically (Supplementary Figure 5). We then ex- tended our simulations to observe large-scale biofilm growth starting from many concentric circular rods (Supplementary Movie 9). As expected, friction shapes the overall biofilm morphology by defining the maximum radius of the curvature around the leading edges (Fig. 3d). Altogether our results show that a growing biofilm has a critical radius which controls the dynamics of edge instabilities. One of the characteristics of active nematics is the formation of topological defects16−18,27,28,44. To better understand the active matter nature of a biofilm, we further focused on how mechanical insta- bilities generate topological defects. In microtubule- based active matter systems, topological defects (+1/2 or -1/2) (Fig. 4a) are spontaneously created and annihilated. The exact mechanism behind the creation of these defects remains unknown45. Re- cent studies elegantly demonstrate that the con- finement strongly controls defects formation45,46. In contrast, our biofilm system does not require any physical confinement which enables the edge to freely move and allows observation of the evolution of the defect formation. Analyzing the development 5 progress, we found two main defect forming mech- anisms; the large-scale biofilm folding and edge in- stabilities. Using time-lapse microscopy, we recorded the first defect generation process. As it grows, the entire biofilm reaches the critical length scale and triggers large structural folding (Fig. 4b, c). This fold- ing clearly forms +1/2 defects. Fig. 4 shows the temporal evolution of this defect formation. As ob- served, the biofilm sequentially folds in orthogonal directions and generates several motile +1/2 defects (Supplementary Figure 6). Clearly, in this stage the total number of defects increases. Due to the small size, all defects move outward. As the size of a biofilm increases, the critical length drops below the biofilm size, and this type of folding appears locally (Supplementary Movie 2, 3). Notably, some of the propagating defects leave king bands as a clear trace (Fig. 4f). These bands originate from the differences in the radius of multi- layered structures which eventually define the total growth rate of the circumference. To further examine defect formations experimen- tally, we measured the speed distributions of de- fects within a biofilm (Fig. 4i). We found that +1/2 defects are more motile around the edge and We also found that the edge instabilities is the second type of defect forming mechanism. Growing droplet geometries around the edge simply behave like +1/2 topological defects (Fig. 4d). These struc- tures are very motile because the unbalanced force generated by the growing side chains can push the defect forward. In contrast to the structural fold- ing described above, this growing droplet splits into two and generates paired +1/2 defects, and it leaves one -1/2 defect around the junction point (Fig. 4e). This mechanism keeps the total topological charge constant . At later stages foldings and splittings become strongly coupled. Thus, the dynamics of defect formation become very chaotic. Our simulations verified the dynamics of defect formation around the edge . We observed that above the critical radius both outward and inward prop- agating bucklings appear. Accumulated localized stress gradually slow down the speed of the cen- tral region and then starts migrating inward. These inward propagating chains might collide with the junction point and then become a stationary -1/2 defect by leaving two motile paired +1/2 (Supple- mentary Movie 9). Figure 4: Dynamics of topological defects and vertical lift-off. a Comet-like motile (+1/2) and trefoil-like (-1/2) topological defects. b, c, d, e Snapshots of a growing biofilm. b Biofilm elongates and reaches the critical buckling length. Scale bar, 100µm. c As a result, biofilm shows a structural folding and creates (+1/2) topological defects. d, e (+1/2) topological defects at the edge move outward, split into two (+1/2) defects and create a (-1/2) defect at the junction. f Snapshot of a growing biofilm. g The director field of the biofilm (f). h Trajectories of both (+1/2) and (-1/2) defects. Red lines indicate the trajectories of (+1/2) defects which are motile, whereas blue lines represent the trajectories of (-1/2) defects, which are stationary. i Experimental data of velocity distribution of (+1/2) and (-1/2) defects. j, k, l Simulation results of a growing multilayered circular colony. j After reaching the critical radius, the colony folds and creates topological defects. k Inward propagating defects collide and create a (-1/2) defect. The stress is accumulated at the collision point, namely (-1/2) topological defect. l The accumulated stress results in vertical lifting around the (-1/2) topological defect. m, n, o SEM images of a biofilm in late stage. Red arrows indicate the propagation directions of defects. m Vertically lifted domains are in the vicinity of the center where the stress accumulates. n Motile (+1/2) defects collide and form a stationary (-1/2) defect. o Accumulation of stress causes vertical lift-off. 6 -1/2 defects are stationary. We also noticed that as the biofilm develops the central region becomes more stationary. The most parsimonious hypothesis based on these observations is that stress accumu- lates around the center and slows down the dynam- ics. Topological defects could also change the physi- ology of the biological tissues18,47−49. It was shown that defect-induced stress triggers mechanotrans- duction and alters the genes expressions16. Stress also modulates the cellular density of cultured cells. Topological defects and active stress could also play a biophysical role during the development of a biofilm. With this motivation, we further per- formed stress analysis of topological defects. Our simulations revealed that the stress accumulates as stationary -1/2 defect occurs (Fig. 4j, k). Around these defects, the unbalanced force generated by the growing chain can be balanced by counter propagat- ing ones. Further, colliding bent chains significantly increase the total stress. When we performed our simulation in the three-dimensional domain (so far we performed all simulation in a 2D confined envi- ronment), we observed vertical lifting of the elastic rods around the vicinity of -1/2 defect (Fig. 4l, Sup- plementary Movie 10). Finally, we asked whether we could experimen- tally observe the effects of stress accumulation in a biofilm. To do so, we analyzed the 3D surface topology after getting a stable biofilm (35 h after seeding). SEM images revealed vertical structures around the -1/2 defects sites (Fig. 4m, n, o). A close investigation clarified that these structures are actually the vertically lifted parts of the defects. These structures are more visible around the center of the biofilm. We grew the non-chaining bacteria in the same conditions and we did not observe any of these vertical structures (Supplementary Figure 7). The results are remarkably consistent with our FEM simulations. These structures have already been identified as aerosol or fruiting bodies of the biofilm which eventually produce bacterial spores2. We do not exclude other structural instabilities or the contribution of twisting process. Our results support the idea that topological defects are one of the possible origins of these vertical ordered geome- tries. Discussion The theory of active nematics provides a robust framework to explain the coordinated motion of cells and the emergence of large-scale order in bi- ological systems. Our study throws light upon dif- ferent aspects of biological active matter systems: First, mechanical instabilities can be simply char- acterized by a critical length scale. Different types of length scales have previously been defined for various AN systems24,26. These quantities mainly describe the activity of the system. In contrast, our length scale reflects the elastic properties of the biofilm. Second, the leading edge of the biofilm is not stable. Edge instabilities15,44,50−52 have re- ceived significant attention due to fingering forma- tion during wound healing15. Our results revealed the importance of cellular alignment and localized stress driving instabilities. Finally, we have shown the details of defect-forming mechanisms and their effect on the stress distribution across the biofilm. In growing tissue or bacterial colonies, mechanical stress can easily result in physiological stress. Re- cent studies have linked the physical forces with the cellular response16, and our system could fur- ther expand the understanding of stress manage- ment across growing tissue53. Another benefit of using B. subtilis is the availability of sophisticated genetic toolbox54. Genetic manipulations and con- trol could reveal unexplored dynamics of active mat- ter systems. We believe, our biofilm platform offers new opportunities to study active nematics in living systems. Methods Sample preparation and growth conditions. Bacterial cultures (BAK47 and BAK51) were grown in Luria-Bertani (LB) broth at 37oC on a shaker. An overnight culture was diluted 100x and grown 8 h. The culture was diluted 10000x and 50µ' of culture was seeded on an LB agarose plate. These isolated bacteria on plates were grown at 21oC for 12 h then imaged. Cover glass was not used. IPTG (isopropyl β-D- 1-thiogalactopyranoside) inducible (BAK50) were grown in LB broth at 37oC without IPTG induction to ensure that chaining does not occur. An overnight culture was diluted 10000x and supplemented with 100µM IPTG. A droplet of IPTG inducable bacterial suspension is placed between two glass slides with 100µm separation, then imaged at 21oC. Microscopy Imaging. Fluorescence time-lapse imaging was performed using a Nikon SMZ18 stereo microscope and im- ages were obtained using a Thor Labs DCC1545M CMOS camera. Time intervals between successive images are 15 minutes. To obtain the best image, the light exposure is controlled adaptively for different colony sizes by a custom- 7 written program in LabVIEW. A typical colony growth ex- periment was run for 1 day. SEM imaging. Before seeding the bacteria on LB agar sur- face, sterile filter paper with pore size 0.2 µm is placed on the surface. After seeding the bacteria on the filter paper, bacteria were grown on the paper for 24 h at 21oC. Then the filter paper was peeled off from the surface and the colonies were fixed using paraformaldehyde and left to drying. Fixed colonies were imaged using Zeiss Ultra Plus Field Emission Electron Microscope. Strains and labeling. Parent 168 Strain BAK47 Operation Transformed with plasmid ECE321 from Bacillus Genetic Stock Center BAK50 TMN1152 Transformed with plasmid ECE321 from Bacillus Genetic Stock Center BAK51 TMN1138 Transformed with plasmid ECE321 from Bacillus Genetic Stock Center Genotype amyE::Pveg-sfGFP (Spc) amyE::Pveg-sfGFP (Spc) ywrK::Pspank-sinI (Spc) amyE::Pveg-sfGFP (Spc) sacA::Phag-mKate2L (Kan) hagA233V (Phleo) Detection of topological defects. Images were smoothed using Bandpass filter in ImageJ, and Coherence-Enhanced Diffusion Filter was applied to images in MATLAB. The Im- ageJ plugin OrientationJ was used to find the nematic di- rector field of the colony. We followed the defect detection algorithm27 using a custom-written MATLAB code. Computer Simulations. For the computer simulations in- volving large growth, we employed a dynamic finite element program37−39. The filament was modelled as an isotropic, linearly elastic continuum with three-dimensional beam the- ory using cubic Hermite shape functions and a corotational quaternion formulation for geometric nonlinearity. The fil- ament was assumed to have a circular cross section with constant uniform radius r = 0.7µm, a Young's modulus of E = 5300P a (corresponding to a bending stiffness of Eπr4/4 = 10−21N m2), a Poisson ratio of 0.33 and a mass density of 1g/cm3. The elastic energy of the filament con- sisted of the sum of bending, torsion and axial compression energies. Hertzian repelling forces were exchanged between contacting filament elements in normal direction, whereas the tangential contact forces were realized with a Coulomb slip-stick friction model. Newton's equations of motion were integrated in time with a second-order predictor-corrector scheme. For numerical robustness and to keep the filament near static equilibrium during growth, subcritical damping forces were added. The initial configuration was assumed to be stress-free, and the filament was grown exponentially in length over time, according to '(t) = '0eσt where σ is the exponential growth rate. This was realized by continuously increasing the equilibrium length of each rod element. The substrate was modeled as an ideal elastic horizontal plane onto which the filament was placed, and a gravitational force was applied perpendicular to the plane. Acknowledgements This work was supported by an EMBO installation Grant (IG 3275, A.K.) and BAGEP young investigator award (A.K.). Author Contributions Y.I.Y and A.K. designed and performed experiments, ana- lyzed data and developed the imaging systems. Y.I.Y and E.D. A.K. performed the molecular biology experiments. R.V. designed and developed the FEM simulation toolbox and optimized the codes with realistic bacterial parameters. 8 Y.I.Y performed the simulations. Y.I.Y and A.K. prepared the draft, and all authors contributed to the writing of the manuscript. Competing Interests Authors declare no competing interests. References 1. Chai, Y. R., Norman, T., Kolter, R. & Losick, R. An epigenetic switch governing daughter cell separation in Bacillus subtilis. Gene Dev 24, 754-765 (2010). 2. Branda, S. S., Gonzalez-Pastor, J. E., Ben-Yehuda, S., Losick, R. & Kolter, R. Fruiting body formation by Bacillus subtilis. P Natl Acad Sci USA 98, 11621-11626 (2001). 3. Norman, T. M., Lord, N. D., Paulsson, J. & Losick, R. Memory and modularity in cell-fate decision making. Nature 503, 481-+ (2013). 4. Chai, Y. R., Kolter, R. & Losick, R. Reversal of an epi- genetic switch governing cell chaining in Bacillus sub- tilis by protein instability. Mol Microbiol 78, 218-229 (2010). 5. Vlamakis, H., Chai, Y. R., Beauregard, P., Losick, R. & Kolter, R. Sticking together: building a biofilm the Bacillus subtilis way. Nature Reviews Microbiology 11, 157-168 (2013). 6. Liu, J. T. et al. Coupling between distant biofilms and emergence of nutrient time-sharing. Science 356, 638- 641 (2017). 7. Liu, J. T. et al. Metabolic co-dependence gives rise to collective oscillations within biofilms. Nature 523, 550- + (2015). 8. Mamou, G., Mohan, G. B. M., Rouvinski, A., Rosen- berg, A. & Ben-Yehuda, S. Early Developmental Pro- gram Shapes Colony Morphology in Bacteria. Cell Re- ports 14, 1850-1857 (2016). 9. Honda, R., Wakita, J. I. & Katori, M. Self-Elongation with Sequential Folding of a Filament of Bacterial Cells. Journal of the Physical Society of Japan 84 (2015). 10. Klapper, I. Biological applications of the dynamics of J Comput Phys 125, 325-337 twisted elastic rods. (1996). 11. Wolgemuth, C. W., Goldstein, R. E. & Powers, T. R. Dynamic supercoiling bifurcations of growing elastic fil- aments. Physica D 190, 266-289 (2004). 12. Mendelson, N. H. Bacillus subtilis macrofibres, colonies and bioconvection patterns use different strategies to achieve multicellular organization. Environ Microbiol 1, 471-477 (1999). 13. Mendelson, N. H., Thwaites, J. J., Kessler, J. O. & Li, C. Mechanics of Bacterial Macrofiber Initiation. J Bac- teriol 177, 7060-7069 (1995). 14. van Gestel, J., Vlamakis, H. & Kolter, R. From Cell Differentiation to Cell Collectives: Bacillus subtilis Uses Division of Labor to Migrate. Plos Biol 13 (2015). 15. Basan, M., Elgeti, J., Hannezo, E., Rappel, W. J. & Levine, H. Alignment of cellular motility forces with tis- sue flow as a mechanism for efficient wound healing. P Natl Acad Sci USA 110, 2452-2459 (2013). 16. Saw, T. B. et al. Topological defects in epithelia govern cell death and extrusion. Nature 544, 212-+ (2017). 17. Duclos, G., Erlenkamper, C., Joanny, J. F. & Silberzan, P. Topological defects in confined populations of spindle- shaped cells. Nature Physics 13, 58-62 (2017). 18. Kawaguchi, K., Kageyama, R. & Sano, M. Topological defects control collective dynamics in neural progenitor cell cultures. Nature 545, 327-+ (2017). 19. Doostmohammadi, A., Thampi, S. P. & Yeomans, J. M. Defect-Mediated Morphologies in Growing Cell Colonies. Phys Rev Lett 117 (2016). 20. Dell'Arciprete, D. et al. A growing bacterial colony in two dimensions as an active nematic. Nat Commun 9 (2018). 21. Wensink, H. H. et al. Meso-scale turbulence in living fluids. P Natl Acad Sci USA b, 14308-14313 (2012). 22. Beroz, F. et al. Verticalization of bacterial biofilms. Na- ture Physics 14, 954-+, doi:10.1038/s41567-018-0170-4 (2018). 23. Hartmann, R. et al. Emergence of three-dimensional order and structure in growing biofilms. Nature Physics, doi:10.1038/s41567-018-0356-9 (2018). 24. Guillamat, P., Ignes-Mullol, J. & Sagues, F. Control of active liquid crystals with a magnetic field. P Natl Acad Sci USA 113, 5498-5502 (2016). 25. Guillamat, P., Ignes-Mullol, J., Shankar, S., Marchetti, M. C. & Sagues, F. Probing the shear viscosity of an active nematic film. Phys Rev E 94 (2016). 26. Guillamat, P., Ignes-Mullol, J. & Sagues, F. Taming active turbulence with patterned soft interfaces. Nat Commun 8 (2017). 27. DeCamp, S. J., Redner, G. S., Baskaran, A., Hagan, M. F. & Dogic, Z. Orientational order of motile defects in active nematics. Nat Mater 14, 1110-1115 (2015). 28. Sanchez, T., Chen, D. T. N., DeCamp, S. J., Heymann, M. & Dogic, Z. Spontaneous motion in hierarchically assembled active matter. Nature 491, 431-+ (2012). 29. Hunt, G. W., Wadee, M. K. & Shiacolas, N. Localized Elasticae for the Strut on the Linear Foundation. J Appl Mech-T Asme 60, 1033-1038 (1993). 30. Shan, W. L. et al. Attenuated short wavelength buck- ling and force propagation in a biopolymer-reinforced rod. Soft Matter 9, 194-199 (2013). 31. Diamant, H. & Witten, T. A. Compression Induced Folding of a Sheet: An Integrable System. Phys Rev Lett 107 (2011). 32. Diamant, H. & Witten, T. A. Shape and symmetry of a fluid-supported elastic sheet. Phys Rev E 88 (2013). 33. Oshri, O., Brau, F. & Diamant, H. Wrinkles and folds in a fluid-supported sheet of finite size. Phys Rev E 91 (2015). 34. Thompson, J. M. T. & Champneys, A. R. From helix to localized writhing in the torsional post-buckling of elas- tic rods. P Roy Soc a-Math Phy 452, 117-138 (1996). 35. Tvergaard, V. & Needleman, A. On Localized Thermal Track Buckling. Int J Mech Sci 23, 577-587 (1981). 36. Silverberg, J. L. et al. 3D imaging and mechanical mod- eling of helical buckling in Medicago truncatula plant roots. P Natl Acad Sci USA 109, 16794-16799 (2012). 37. Vetter, R., Wittel, F. K., Stoop, N. & Herrmann, H. J. Finite element simulation of dense wire packings. Eur J Mech a-Solid 37, 160-171 (2013). 38. Vetter, R., Wittel, F. K. & Herrmann, H. J. Morphogen- esis of filaments growing in flexible confinements. Nat Commun 5 (2014). 39. Vetter, R. Growth, Interaction and Packing of Thin Ob- jects, ETH Zurich, (2015). 40. Grant, M. A. A., Waclaw, B., Allen, R. J. & Cicuta, P. The role of mechanical forces in the planar-to-bulk transition in growing Escherichia coli microcolonies. J R Soc Interface 11 (2014). 41. Paterson, L. Radial Fingering in a Hele Shaw Cell. J Fluid Mech 113, 513-529 (1981). 42. Naito, H., Okuda, M. & Zhongcan, O. Y. Pattern forma- tion and instability of smectic-A filaments grown from an isotropic phase. Phys Rev E 55, 1655-1659 (1997). 43. Shelley, M. J. & Ueda, T. The Stokesian hydrodynamics of flexing, stretching filaments. Physica D 146, 221-245 (2000). 44. Blow, M. L., Thampi, S. P. & Yeomans, J. M. Biphasic, Lyotropic, Active Nematics. Phys Rev Lett 113 (2014). 45. Opathalage, A., Norton, M. M., Juniper, M. P. N., Aghvami, S. A., Langeslay, B., Fraden, S. & Dogic, Z. (arXiv:1810.09032, 2018). 46. Norton, M. M. et al. Insensitivity of active nematic liquid crystal dynamics to topological constraints. Phys Rev E 97 (2018). 47. Hawkins, R. J. & April, E. W. Liquid-Crystals in Living Tissues. Adv Liq Cryst 6, 243-264 (1983). 48. Bonhoeffer, T. & Grinvald, A. Iso-Orientation Domains in Cat Visual-Cortex Are Arranged in Pinwheel-Like Patterns. Nature 353, 429-431 (1991). 49. Gruler, H., Dewald, U. & Eberhardt, M. Nematic liquid crystals formed by living amoeboid cells. Eur Phys J B 11, 187-192 (1999). 50. Nesbitt, D., Pruessner, G. & Lee, C. F. Edge instability in incompressible planar active fluids. Phys Rev E 96 (2017). 51. Zimmermann, J., Basan, M. & Levine, H. An instability at the edge of a tissue of collectively migrating cells can lead to finger formation during wound healing. Eur Phys J-Spec Top 223, 1259-1264 (2014). 52. Doostmohammadi, A. et al. Celebrating Soft Mat- ter's 10th Anniversary: Cell division: a source of active stress in cellular monolayers. Soft Matter 11, 7328-7336 (2015). 53. Martinez-Corral, R., Liu, J. T., Suel, G. M. & Garcia- Ojalvo, J. Bistable emergence of oscillations in growing Bacillus subtilis biofilms. P Natl Acad Sci USA 115, E8333-E8340 (2018). 54. Guiziou, S. et al. A part toolbox to tune genetic expres- sion in Bacillus subtilis. Nucleic Acids Res 44, 7495- 7508 (2016). 9
1904.11453	1	1904	2019-04-25T16:51:20	A polymer model of bacterial supercoiled DNA including structural transitions of the double helix	[ "physics.bio-ph", "q-bio.BM" ]	DNA supercoiling, the under or overwinding of DNA, is a key physical mechanism both participating to compaction of bacterial genomes and making genomic sequences adopt various structural forms. DNA supercoiling may lead to the formation of braided superstructures (plectonemes), or it may locally destabilize canonical B-DNA to generate denaturation bubbles, left-handed Z-DNA and other functional alternative forms. Prediction of the relative fraction of these structures has been limited because of a lack of predictive polymer models that can capture the multiscale properties of long DNA molecules. In this work, we address this issue by extending the self-avoiding rod-like chain model of DNA so that every site of the chain is allocated with an additional structural degree of freedom reflecting variations of DNA forms. Efficient simulations of the model reveal its relevancy to capture multiscale properties of long chains (here up to 21 kb) as reported in magnetic tweezers experiments. Well-controlled approximations further lead to accurate analytical estimations of thermodynamic properties in the high force regime, providing, in combination with experiments, a simple, yet powerful framework to infer physical parameters describing alternative forms. In this regard, using simulated data, we find that extension curves at forces above 2 pN may lead, alone, to erroneous parameter estimations as a consequence of an underdetermination problem. We thus revisit published data in light of these findings and discuss the relevancy of previously proposed sets of parameters for both denatured and left-handed DNA forms. Altogether, our work paves the way for a scalable quantitative model of bacterial DNA.	physics.bio-ph	physics	A polymer model of bacterial supercoiled DNA including structural transitions of the double helix Thibaut Lepage, Ivan Junier CNRS, TIMC-IMAG, F-38000 Grenoble, France Univ. Grenoble Alpes, TIMC-IMAG, F-38000 Grenoble, France Abstract DNA supercoiling, the under or overwinding of DNA, is a key physical mecha- nism both participating to compaction of bacterial genomes and making genomic sequences adopt various structural forms. DNA supercoiling may lead to the formation of braided superstructures (plectonemes), or it may locally destabi- lize canonical B-DNA to generate denaturation bubbles, left-handed Z-DNA and other functional alternative forms. Prediction of the relative fraction of these structures has been limited because of a lack of predictive polymer models that can capture the multiscale properties of long DNA molecules. In this work, we address this issue by extending the self-avoiding rod-like chain model of DNA so that every site of the chain is allocated with an additional structural de- gree of freedom reflecting variations of DNA forms. Efficient simulations of the model reveal its relevancy to capture multiscale properties of long chains (here up to 21 kb) as reported in magnetic tweezers experiments. Well-controlled ap- proximations further lead to accurate analytical estimations of thermodynamic properties in the high force regime, providing, in combination with experiments, a simple, yet powerful framework to infer physical parameters describing alter- native forms. In this regard, using simulated data, we find that extension curves at forces above 2 pN may lead, alone, to erroneous parameter estimations as a consequence of an underdetermination problem. We thus revisit published data in light of these findings and discuss the relevancy of previously proposed sets of parameters for both denatured and left-handed DNA forms. Altogether, our work paves the way for a scalable quantitative model of bacterial DNA. Keywords: DNA polymer modeling, DNA supercoiling, Multiscale polymer models, DNA structural transitions 1. Introduction In bacteria, and probably in numerous eukaryotes, genome-wide coordination of transcription primarily relies on the physics of DNA in interaction with RNA polymerases [1, 2]. Understanding bacterial transcription coordination there- fore requires understanding the physics of DNA, more specifically the physics Preprint submitted to Physica A April 26, 2019 of under and over-wound DNA [3, 4, 5, 6, 7, 8], also known as DNA supercoil- ing. DNA is indeed continually processed in cells by topoisomerases [9], whose activity allows relaxing the transient constraints generated by replication and by transcription itself, both processes tending to overwind DNA downstream and to underwind it upstream [10, 11]. As a result, bacterial DNA is gener- ally underwound, which is commonly referred to as the negative supercoiling of bacterial genomes. In model organisms, and probably in most bacteria [12], this negative supercoiling is further partitioned into approximately 10 kb long topologically independent domains [13, 14]. Although supercoiling is produced at specific sites (e.g. at the target sites of gyrases), it may impact DNA globally as a consequence of the conservation of the linking number. This linking number, Lk, is equal to the sum of the twist (Tw), the cumulative helicity of the molecule, plus the writhe (Wr), the global intricacy of the molecule [15]. Thus, for topologically constrained DNA as in the case of circular molecules (e.g. plasmids), of topologically constrained linear domains [10], or when DNA molecules are manipulated by magnetic tweezers [16, 17, 18], any local variation of the twist results in a global variation of the writhe, and reciprocally. As a consequence, negative DNA supercoiling leads to both superstructuring participating to the global compaction of bacterial chromosomes (through e.g. the formation of so-called plectonemes) and to local modifications of the double helix structure. The latter can induce the formation of functionally important DNA forms different from B-DNA [19, 20], including denaturation bubbles [21], cruciforms [22], left-handed Z-DNA [23] and left- handed L-DNA [24, 25, 26, 27]. One of the most important challenges raised by the biophysical and func- tional characterizations of bacterial genomes thus consists in predicting the mul- tiscale distribution of supercoiling constraints. Such prediction nevertheless re- mains challenging, with often no other solution than to resort to numerical simulations [28]. In this regard, two main types of DNA polymer models have been proposed. On one hand, explicit polymer models have allowed investi- gating supercoiling-induced phenomena on the basis of numerical simulations of single DNA chains [29, 30, 31, 32]. On the other hand, phenomenological models [33, 34] have led to bona fide analytical solutions of thermodynamic properties. These have been particularly useful to infer microscopic parameters of alternative DNA forms [25, 34], whose knowledge is crucial to both address the exact nature of these forms and parametrize explicit polymer models. Explicit polymer models can be further divided into two types. Depending on the addressed question, structural details can indeed be coarse-grained at the scale of a single nucleotide [32, 35] or at a larger scale of a few tens base pairs [29, 30, 31, 36]. Single nucleotide resolution models have thus been useful to investigate properties of small (i.e. a few hundreds base pairs) DNA molecules as provided by cryo-electron microscopy [28, 37] and cyclization data [20], with the possibility to investigate sequence effects in detail (see e.g. [38] and references therein). Tens base pairs resolution models have instead been useful to address both mechanical and conformational properties of long (i.e. a few kb or tens kb) B-DNA molecules, taking advantage of much less time-consuming simulations 2 and of the possibility to use parsimonious sets of parameters that can capture bending, torsional and self-avoidance properties of molecules [15, 20]. Along this line, the self-avoiding rod-like chain (sRLC) model [29] has been paramount to analyze folding properties of positively supercoiled molecules (see below for details of the model). These include molecular extensions [39], torques [40, 41, 42] and conformation details of superstructures [29, 43, 42] as measured by magnetic tweezers and cryo-electron microscopy, as well as dynamical properties (see [44] and references therein) and sequence-dependent phenomena [45, 46]. The sRLC model has thus provided a solid framework to infer physical pa- rameters of B-DNA [15, 31]. To the best of our knowledge, it has however not yet been adapted to the co-existence of multiple DNA forms, precluding its use to investigate functional properties of bacterial genomes in vivo and leav- ing open several important biophysical questions. These include the balance between the global and local relaxations of supercoiling constraints, the exact nature of alternative forms (see e.g. [27] for a recent discussion about the exper- imental identification and characterization of denatured DNA (D-DNA) with respect to L-DNA) and, related to this, the inference of associated mechanical parameters. Namely, while some of the parameters, such as the free energy formation of left-handed DNA forms and of their associated junctions with B- DNA, have been estimated early on in bulk studies [22, 23, 47, 48, 49, 50, 51], the estimation of, e.g., the torsional and bending moduli of D-DNA and left- handed forms, has thus far relied exclusively on phenomenological models in the context of single molecule studies [33, 25, 26, 52, 34]. These models, which are based on bona fide free energy landscape descriptions of the co-existence of multiple DNA forms, have however never been quantitatively assessed, which may explain the existence of strong discrepancies in the prediction of some of the parameters [25, 34]. In this work, we aim at filling these gaps by extending the sRLC model so that it includes the possibility to have multiple DNA forms. To this end, we follow an approach initially proposed to tackle the problem of the large flexibility of small DNA molecules [53]. Namely, as proposed in [54, 55, 56], the softening of sharply bent DNA can be rationalized by considering a structural degree of freedom allowing the appearance of alternative forms more flexible than B-DNA, with a free energy cost reflecting the destabilization of the double helix. More elaborated models including torsional energies, but still neglecting DNA self-avoidance, have then been proposed to investigate the statistics of DNA denaturation [57] and the behavior of stretched DNA molecules [58] under torsional constraints. Models based on torsional energies alone have also been studied in the context of the statistics of DNA denaturation [59, 60] as well as to model properties of specific DNA sequences [26]. Following these studies, here we consider a version of the sRLC model, here- after referred to as the 2sRLC model, which includes an additional structural degree of freedom. Using numerical simulations, we first demonstrate the abil- ity of the 2sRLC model to reproduce magnetic tweezers experiments of long molecules (up to 21 kb) for a wide range of stretching forces and supercoiling levels typical of bacterial genomes in vivo. We next tackle the inference prob- 3 lem of parameters associated with alternative DNA forms. To this end, we circumvent the use of time-consuming simulations by deriving semi-analytical solutions of the thermodynamics when the writhe is negligible and we show that these can be used for any value of supercoiling level and stretching force where plectonemic superstructures are absent. As an application, we discuss the validity of the commonly used phenomenological model proposed by John Marko [33] and highlight a previously overlooked underdetermination problem related to the question of parameter inference, which is immanent to the high force analysis of single molecule extension curves. We provide, in this context, our best estimates of parameters associated with D-DNA and L-DNA. 2. Theory 2.1. The discrete sRLC model In the sRLC framework, a double-stranded DNA molecule, usually in the form of B-DNA, is modeled as a continuous self-avoiding chain characterized by five fundamental parameters: i) a bending modulus (K) defining an associated persistence length ((cid:96)p = K/kBT ), ii) a torsional modulus (C) measured in units of twist persistence length, iii) a winding angle at rest (ψ) of the implicitly embedded helix, iv) the distance (a) between any two consecutive base pairs, thus defining the number of base pairs per (cid:96)p, and v) an effective radius (re) reflecting the hard-core simplification of the electrostatic repulsion of the DNA backbone [61]. For B-DNA in physiological conditions ([NaCl] ∼ 100 mM), (cid:96)p typically lies in [40− 60 nm] [62, 63] (in the following, we use 50 nm), C is on the order of 100 nm [64, 24], while a = 0.34 nm, ψ = 0.6 rad/bp and re ≈ 2 nm [61]. To simulate the folding of a DNA molecule, a discrete version of the model is used, for which the continuous formulation is a good approximation if the dis- cretization is sufficiently fine (see hereafter for further details). Specifically, the chain is divided into identical impenetrable cylinders and an additional param- eter, n, specifies the number of base pairs per cylinder used in the simulation such that the length of every cylinder is equal to na (Fig. 1A) -- in this work, we use n ≤ 10 such that one cylinder corresponds at most to one B-DNA helix. The energy of any conformation C of a self-avoiding linear chain in the presence of a stretching force, f , then reads: EsRLC(C) = EB(C) + ET (C) − f z, (1) nC i=1 kB T 2 (cid:80)N i=1 (cid:96)p na θ2 2 (cid:80)N i and ET (C) = kB T where z is the extension of the chain along the axis of the force, while EB(C) = a (φi − ψ)2 respectively stand for the bending and torsional energies. In these equalities, the index i indicates a site around which two consecutive cylinders are articulated, while θi is the bending angle associated with the n base pairs of the site and φi is the winding angle per base pair (Fig. 1A); this means that nφi is the torsional counterpart of θi for the site i. Note that at the end of the molecule, two additional sites (at i = 0 and i = N + 1) set the boundary conditions, which here correspond to a situation where the orientation of the external cylinders is kept fixed along 4 Figure 1: A. Discrete version of the sRLC: two complete cylinders surrounding a site i are represented along with their local frames ((cid:126)t, (cid:126)u, (cid:126)v) used to compute the bending angle θi and the schematically depicted twist angle per bp φi. B. Extension of the model by allocating a variable s specifying the DNA form of every site. In this case, for symmetry reasons, the length of a cylinder located between sites k − 1 and k is equal to n(ask−1 + ask )/2, meaning that cylinders may have different lengths because of the different values of as characterizing the different forms. Here, we depict the case where sites i − 1 and i are under the B form (chain in blue), while sites i + 1 and i + 2 are denatured (chain in red). We also indicate by a gray disk the domain wall penalty, J, associated with the structural cost for transiting from one form to the next. the force axis. Altogether, a linear chain made of Nbp base pairs thus contains N + 1 sites and N segments (N = Nbp/n = L/na where L is the contour length of the molecule). A circular chain (e.g. a plasmid) would contain N sites and N segments, the first site joining the last and first segments together. (2π)−1(cid:80)N Importantly, just as θi, φi can be explicitly written as a function of the local frames associated with the surrounding cylinders [65] (Fig. 1A). Using Tw ≡ i=1 nφi, this ensures that the linking number, Lk = Tw + Wr, remains strictly constant for any deformation of a circular chain, provided the chain never crosses itself. Lk then reflects the topological status of the DNA molecule, for instance a supercoiling constraint if it is different from the corresponding value at rest, Lk0. For a linear chain, conservation of the linking number further requires the ends of the chain to be attached to two impenetrable walls [39, 66]. These respectively play the role of the fixed surface and of the magnetic bead in single-molecule experiments. 2.2. The discrete 2sRLC model: including multiple DNA forms In order to include varying DNA forms at every site i of the sRLC, we follow the lines of denaturation studies [59, 54, 55, 56, 57, 58] and associate a state variable s with every site (s = B, D, L or Z if one considers for instance B-DNA, 5 ABD-DNA, L-DNA and Z-DNA) specifying the associated mechanical parameters: Ks (persistence length (cid:96)s), Cs, as and ψs -- for simplicity, here we consider a form-independent electrostatic radius re. In this context, the length of cylinders may vary as a result of both a varying value for as and a fixed number of base pairs per cylinder (n); here, for symmetry reasons, we consider the length of a cylinder surrounded by two sites with forms s and s(cid:48) to be equal to n(as +as(cid:48))/2 (Fig. 1B). In addition to form-dependent mechanical parameters, alternative DNA forms come along with free energy formation costs on the order of one kBT per bp [67, 68], reflecting the internal deformation of base pairing and base pair stacking. Here, we denote this cost by γ0 B = 0 (reference form). Finally, a domain wall penalty, J, must be considered between any two sites having different states. This penalty accounts for the internal free energy cost to transit from one DNA form to another one [69, 57, 52] (Fig. 1B) -- in effect, it constrains alternative forms to gather in as few domains as possible. This wall penalty is in principle associated with two consecutive base pairs and therefore includes an entropic contribution on the order of ln(n) coming from the n possi- ble choices of these base pairs. For simplicity, here we consider this contribution as already included in the values of J. s and set γ0 Altogether, the energy of a self-avoiding 2sRLC conformation C having an arbitrary sequence {si}i=1..N of DNA forms reads: E2sRLC(C) = Eγ(C) + EB(C) + ET (C) + EJ (C) − f z, (2) i=1 γ0 where Eγ(C) = n(cid:80)N alternative forms, EB(C) = kB T 2 (cid:80)N nCsi asi kB T i=1 si is the total free energy formation costs coming from is the bending energy, ET (C) = i=1 (1 − (φi − ψsi)2 is the torsional energy and EJ (C) = J(cid:80)N−1 2 (cid:80)N (cid:96)si nasi θ2 i i=1 δsi,si+1) is the total wall penalty, with δsi,si+1 = 1 if si = si+1, 0 otherwise. 3. Results and discussion 3.1. Effective 2sRLC model Solving the thermodynamics of the 2sRLC model is challenging because of the self-avoidance constraints. Yet, following upon previous studies on single DNA forms [70] and on the statistics of DNA denaturation [59, 60], it is possible to integrate out the torsional degrees of freedom (the φi's) under the constraint of a fixed Lk. This leads to an equivalent effective model that has the benefit to offer much better simulation performances [39, 42] and further analytic treat- ment (see below). In the case of the 2sRLC model, the resulting conformational energy can be decomposed as in Eq. 2, but with an effective free energy of forma- tion and an effective torsional energy that respectively read (see Supplementary 6 methods inSupplementary material for details): Eef f γ (C) = n γsi where γsi = γ0 si − 1 2n ln asi Csi (3) N(cid:88)i=1 2 (cid:34) Nbp kBT Eef f T (C) = Tw(C) + ln(χ (C))(cid:35) Nbp − ψ(C)(cid:19)2 χ(C)(cid:18)2π Csi(cid:14)N are conformation-dependent average quan- (4) asi where, again, Nbp = nN is the length of the molecule in base pairs, ψ(C) ≡ i=1 ψsi(cid:14)N and χ(C) ≡(cid:80)N (cid:80)N i=1 tities, namely, the average twist angle per bp at rest and the average torsional susceptibility per bp. Eq. 4 thus generalizes to the case of any number of different DNA forms the possibility, for topologically constrained DNA molecules, to obtain an equivalent effective model with a torsional energy quadratic in Tw. Note, nevertheless, that here the associated parameters depend on the composition of the molecule in these DNA forms. In the particular case of a single (B-DNA) form, one for instance recovers ψ(C) = ψB and χ(C) = aB/CB for all conformations, such that Eef f as demonstrated in [70]. T (C) = kB T CB Nbp 2aB (cid:16)2π Tw(C) Nbp − ψB(cid:17)2 3.2. Capturing the phenomenology of negatively supercoiled DNA To demonstrate the ability of the 2sRLC model to capture the phenomenol- ogy of negatively supercoiled DNA, we performed Monte-Carlo simulations of long chains using the above effective energies. To this end, we have adapted the standard Monte-Carlo procedure used to simulate the equilibrium folding of the sRLC model [31] in order to include the possibility for mechanical parameters to change at any site of the chain, at any iteration of the simulation (see [66] for details). In practice, it involves a new transition type so that for any site i independently of the other sites, the value of the variable si may change under the constraint of detailed balance. Corresponding variations of cylinder lengths (Fig. 1B) are thus managed as other elementary Monte-Carlo moves. In partic- ular, if they lead to a collision, the transition is rejected. More generally, length variation of a cylinder always requires an adjustment of the positions and orien- tations of the surrounding cylinders to preserve the continuity of the chain. As explained in [66], this is realised by performing specific random rotations that are constrained by the geometry of the chain. Using this Monte-Carlo method, we could quantitatively reproduce exten- sions of several kilo base pairs long molecules as obtained in magnetic tweezers experiments for a wide range of forces and supercoiling densities σ = (Lk − Lk0)/Lk0. As an example, we report in Fig. 2 both our simulation results and the experimental results of Cees Dekker's lab [27] for a 21 kb molecule, for σ ∈ [−0.05, 0.05], f = 0.5 pN and f = 4.5 pN, providing as a bonus a predic- tion of the equilibrium fraction of alternative forms present along the molecule (top panel). DNA denaturation being expected to be the most likely alternative form at these forces and supercoiling densities [26, 27], here we considered the 7 Figure 2: Extension of a 21 kb molecule as a function of σ, for different forces, obtained experimentally (in red and blue, data from [27]) and numerically using a 2sRLC model that includes the possibility to have B-DNA plus an alternative DNA form along the chain (see main text for values of the microscopic parameters). Error bars and colored areas represent one standard deviation due to thermal fluctuations. At low forces, the curves are symmetrical with respect to σ, with the formation of branched plectonemic structures at large values of σ as indicated by the typical conformations observed in the simulations (leftmost and rightmost panels). At high forces, the buckling transition toward the plectonemic regime occurs only for positive supercoiling. For negative supercoiling, denaturation bubbles appear instead (upper left panel, in red). Upper panel: corresponding estimation of the equilibrium fraction λ∗ of the alternative form as a function of σ. situation where only B-DNA and D-DNA could form, with the following pa- rameters leading to good agreement with experimental curves: aD = 0.45 nm, (cid:96)D = 15 nm, CD = 10 nm and γD = 2kBT (see below for discussion of these val- ues) -- we intentionally used a (cid:96)D much larger than the 4 nm diameter of the chain because smaller values would lead, in any case, to larger effective persistence lengths [71]. We used J = 10kBT , in accord with previous analyses [69, 26, 52], and considered a discretization level of n = 10, leading to (cid:39) 3.3 cylinders per (cid:96)D and 15 cylinders per (cid:96)B(= 50 nm), a value that both offers reasonable computa- tional times and ensures properties of B-DNA superstructuring to be insensitive to discretization procedures [42, 66]. Finally, the quantitative reproduction of the buckling transition at σ > 0 (right part of the curves in Fig. 2) required us to use C = 80 nm at f = 4.5 pN and C = 100 nm at f = 0.5 pN (Fig. S1). This is in accord with the observed systematic discrepancies occurring at low forces between sRLC and experimental extensions [72], a consequence of a coupling between torsional and bending deformations stemming from the asymmetry of the DNA helix [73]. 8 Our simulations corroborate well-known results of magnetic tweezers exper- iments, with notably the presence of two regimes (Fig. 2): i) a low force regime (f (cid:46) 0.5 pN) that is symmetric under the transformation σ → −σ and where torsional stress is partially relaxed through the formation of plectonemes, some of them being branched, and ii) a σ-asymmetric high force regime (f (cid:38) 2 pN) where negative torsional stress is partially relaxed by forming denaturation bub- bles. For intermediate forces at σ < 0, both conformations with denaturation bubbles and plectonemic conformations become locally stable, being separated by high free energy barriers (Fig. S2). This is in accord with the experimen- tal findings, in this regime, of large equilibrium fluctuations of the extension [74, 34], which have been shown, in the case of a 2.4 kb long molecule, to reflect a two-state behavior between plectonemic conformations and conforma- tions with D-DNA bubbles [34]. In our simulations of comparable 2.4 kb long molecules, this manifests through an hysteresis cycle in the force-extension di- agram (Fig. S2), with the drawback that specific numerical methods, such as metadynamics [75], must be developed in this regime to get access to thermo- dynamic properties, including the height of the free energy barriers. Our simulations also corroborate the existence of conformations where de- naturation bubbles locate at the apex of plectonemes (Fig. S3). These have been predicted to occur using a polymer model of DNA coarse-grained at the nucleotide level [76]. Here, we observe that these tip-bubble conformations, as coined in [76], are all the more favored that the system is close to the "spin- odal" point associated with the plectonemic state -- the force at which the free energy barrier separating pure plectonemic states from states with denaturation bubbles becomes on the order of kBT . Most importantly, while trying to reproduce previously published experi- mental extensions, we have found that different sets of parameters could lead to similar results. For instance, the D-DNA persistence length used in Fig. 2 is typically 5 times larger than that previously estimated [33, 25], while previous works have led to values of the torsional modulus that may differ by more than 20-fold [25, 34]. Overall, this raises the question of the extent to which it is pos- sible to infer mechanical parameters of alternative forms using extension curves alone as obtained in magnetic tweezers experiments, what we discuss now in detail. 3.2.1. Zero writhe approximation in the high force regime One could in principle use the above numerical simulations to search for sets of parameters that best reproduce extension curves. However, independently of the efficiency of simulations, an exhaustive exploration of the 5 parameters ((cid:96)D, CD, aD, ψD and γ0 D) and the domain wall penalty J is extremely challenging. Moreover, the high free energy barriers separating plectonemic conformations from conformations with D-DNA bubbles precludes obtaining the relative frac- tion of each state at equilibrium using standard Monte-Carlo methods. Curves at intermediate forces between 0.5 pN and 2 pN cannot thus be currently ex- ploited to further constrain parameters as previously done in the context of phenomenological models [34]. 9 To partially circumvent these problems, we focus on the regime of high forces, for which the thermodynamics can be solved in the approximation of both a neg- ligible writhe, i.e. when Lk = Tw, and negligible self-avoidance properties (see below for the range of forces where this approximation holds). In this approx- imation, it is indeed possible to obtain an explicit formula for the free energy profile of the system, Ff,σ({λi}), as a function of the fraction of each of the possible alternative forms at a given stretching force f and supercoiling density σ. In the following, for simplicity we discuss the case of two forms, with in mind B-DNA plus a certain fraction λ of an alternative form, hereafter denoted as X-DNA -- the generalization to more than two forms is straightforward (see Supplementary methods). Calling J (λ) the free energy associated with the multiple possibilities to distribute the λN sites of the alternative form into distinct domains (see SI for its derivation), gs,f (Ls) the free energy of the WLC at force f associated with the form s with contour length Ls, and using the notations χs = as/Cs and χ/ψ(λ) = (1 − λ)χB/ψB + λχs/ψs, we obtain (Supplementary methods): + gX,f (λNbpaX ) + gB,f ((1 − λ)NbpaB) (cid:125) Ff,σ(λ) = λNbpγX + J (λ) free energy formation of alternative domains bending+stretching (cid:124) (5) kBT + (cid:125) (cid:123)(cid:122) 2 (cid:34) Nbp (cid:124) χ(λ)(cid:18)1 + σ − (cid:123)(cid:122) torsion (cid:124) ψ(λ) ψB (cid:19)2 (cid:123)(cid:122) + ln (χ(λ))(cid:35) (cid:125) Using this free energy profile, equilibrium properties of the 2sRLC model in the zero writhe approximation can be computed by performing numerical integration over λ of Boltzmann-weighted observables (see Eq. 10 in Supple- mentary methods and subsequent explanations) with, here, a particular in- terest in equilibrium values of the extension, z∗ = −(cid:104)∂f Ff,σ(λ)(cid:105)λ, the torque Γ∗ = (2πTw0)−1(cid:104)∂σFf,σ(λ)(cid:105)λ (see Supplementary methods for derivation and further details), and the fraction of alternative form, λ∗ = (cid:104)λ(cid:105)λ, where (cid:104)•(cid:105)λ ≡ (cid:82) dλ • exp [−Ff,σ(λ)/kBT ](cid:14)(cid:82) dλ exp [−Ff,σ(λ)/kBT ]. We also compute the equilibrium number of domains, X∗, using a free energy surface in the (λ, X) space (Eq. 6 in Supplementary methods). 3.2.2. Validity domain of the zero writhe approximation To assess the validity domain of the zero-writhe approximation, we compare the above Boltzmann-weighted equilibrium values to those obtained in Monte- Carlo simulations of the corresponding 2sRLC model. As a case study, we investigate a 1 kb long molecule with an alternative form characterized by (cid:96)X = 15 nm, aX = 0.54 nm, ψX = 0, CX = 10 nm and γX = 1.6kT , having in mind a form close to D-DNA; just as in the analysis of Fig. 2, the relatively high value of (cid:96)X compared to previous estimations for D-DNA (3− 4 nm [33, 25]) was chosen for practical purposes, that is, to avoid potential artifacts that may occur in the simulations when the persistence length is smaller or on the order of re [66] and to avoid considering (longer) effective persistence lengths in our analytical 10 Figure 3: Comparison between numerical simulations, the zero writhe approximation and Marko's phenomenological approach for the equilibrium value of different quantities of inter- est as a function of the negative supercoiling: the fraction λ∗ of denatured alternative form, the number X∗ of alternative domains, the torque Γ∗ and the extension of the molecule (relative to its contour length under the single B form). Error bars represent one standard deviation due to thermal fluctuations; in particular, this reveals that torque fluctuations are large. For readability, the green points (2 pN) and red points (10 pN) have respectively been slightly shifted left and right with respect to the exact values of σ: 0, −0.1, −0.2, −0.3, −0.4, −0.5 (−0.015, −0.02, −0.025, −0.03 in the inset of X∗). The transition between fully stretched con- formations and plectonemic conformations (where the approximation does not hold anymore) occurs between 1 and 2 pN. For all panels: the apparent absence of error bars implies that they are smaller than symbols. approximations [71]. We also used a fine discretization level of n = 4 to get a large number of cylinders per (cid:96)D ((cid:39) 7) so that to prevent any discrepancy coming from the discretization procedure of the rod-like chain. Regarding λ∗ and the extension, we observe an excellent agreement between simulation results and the zero-writhe approximation for any force above 2 pN (left panels in Fig. 3), with a relative error lower than 5% (Fig. S4). For the number of domains the error is on the order of 20% (a higher error is expected as a consequence of the discrete nature of X), and up to 30% for the torque (side effect of the error in X∗). Most remarkably, this agreement is observed for a wide range of supercoiling densities, from the transition point at physio- logical σ (cid:39) −0.025 associated with the onset of the first denaturation bubble (see inset in upper right panel of Fig. 3) to large non-physiological negative values of σ = −0.5, and for forces down to the onset of plectonemic superstruc- tures. The strong deviation observed at 1 pN is indeed the result of a buckling 11 transition that occurs in the simulations, which is not captured by the zero writhe approximation. In other words, the zero writhe approximation provides a good description of the system whenever plectonemic superstructures are ab- sent, paving the way for an efficient method for the quantitative inference of physical properties of alternative DNA forms in the high force regime (see be- low). In this regard, we note that the inclusion of finite size corrections [77] for computing the WLC terms gX,f and gB,f leads to very similar results (Fig. S5), suggesting the possibility to use the classical infinite length approximation of the WLC [78], although the fraction of alternative form may be large. We finally used the same simulations to assess the validity of the commonly used phenomenological approach proposed by John Marko a decade ago [33]. In a nutshell, Marko proposed a model with two separated phases (B-DNA and X-DNA), each of them being allocated with a certain supercoiling density and a free energy that is quadratic in the latter; the two supercoiling densities are coupled because of the conservation of the linking number. Then, supposing a linear relationship between the fraction λ of X-DNA and the total supercoiling density (σ) and considering the torques to be equal in both phases, Marko used a free energy minimization procedure to compute the supercoiling densities specific to each phase, leading back to the computation of the free energy as a function of σ. Here, our analysis shows that this phenomenological approach provides as well a remarkable description of the 2sRLC equilibrium properties (Fig. 3), excluding the number of domains, X, which is not a variable of the model [33]. 3.3. Inferring mechanical parameters of alternative forms from extension curves Alternative DNA forms mostly occur in presence of B-DNA. Estimation of their parameters, which ultimately reflect statistical properties of the dynam- ics of strands, has thus been a matter of debate. Both bulk [67] and single- molecule [68] measurements have nevertheless converged to free energy forma- tions of D-DNA on the order of one kBT per bp. Cyclization measurements of small circular DNA molecules have also revealed an independence of D-DNA strands, suggesting to use φD = 0 [79]. A zero-writhe phenomenological mod- eling of these experiments have then led to aD = 0.54 nm, (cid:96)D (cid:39) 2 to 3 nm and CD = 1 nm [79]. In contrast, investigation of the co-existence of D-DNA with B-DNA at forces (cid:46) 1 pN and σ (cid:46) 0.06, using a phenomenological model of the coexistence of twisted B-DNA, plectonemic B-DNA, and D-DNA, has led to CD = 28 nm [34]. Regarding left-handed DNA forms, static and dynamic laser light scattering experiments of Z-DNA solutions led to (cid:96)Z (cid:39) 200 nm [80], whereas a best fitting procedure using Marko's phenomenological model revealed a flexible L-DNA form, with aL = 0.48 nm and (cid:96)L = 3 nm [25], and a torsional modulus CL lying between 10 and 20 nm. Similar values of CL were found by investigating the collapse of pure L-DNA molecules [26]. Finally, location of the buckling transition of such "pure" L-DNA molecules at high forces led to ψL = −0.5 rad/bp [26, 27]. Here, we aim at using the predictive power of our zero writhe approxima- tion of the 2sRLC model (Fig. 3) in order to infer some of the parameters of 12 alternative forms. Specifically, we discuss the possibility, as proposed in [25], to estimate parameters of D-DNA and L-DNA using σ-extension curves, alone, in the high force regime. To this end, we first test our capacity to recover original parameters using simulated data and, then, discuss estimations from real data. 3.3.1. Simulated data: assessing parameter inference in an underdetermination context To test our capacity to recover original parameters (hereafter indicated by a hat, e.g., aX ) from simulated σ-extension data, we use as a generative model the 2sRLC model under the zero writhe approximation (Eq. 5) and question the possibility to recover original parameters using the same generative model. Regarding the inference method, we follow the procedure of [25], which consists in determining optimal parameters such that extensions of the model best fit experimental extensions over a wide range of forces and supercoiling densities (Supplementary methods). First, compared to the other parameters, we find that aX and (cid:96)X (or equiv- alently KX ) have a more systematic impact on σ-extension curves. More pre- cisely, using a large random set of parameters, the various σ-extension curves appear to be sorted according to the values of both aX and (cid:96)X , whereas they appear to be randomized with respect to the values of the other parameters (Fig. 4A, Fig. S6) -- we nevertheless note that best fits correspond to values of ψX that are distant from ψB (Fig. S6). The goodness of fit can be further quantified by computing the root-mean-square deviation (RMSD) of each tested curve with respect to the original curve. Reporting these RMSDs in the plane (aX , (cid:96)X ) at a given force then reveals the existence of a well-defined continu- ous set of parameters with similar locally optimal solutions and going through (aX , (cid:96)X ), hereafter called a crest of solutions (Fig. 4B) -- note that these so- lutions are always suboptimal with respect to the original set of parameters. The overall shape of this crest reflects the existence of two types of solutions: 1) solutions corresponding to almost straight pieces of X-DNA (large (cid:96)X , the exact value being irrelevant); 2) solutions corresponding to more disordered do- mains (small (cid:96)X ) with longer double helices (larger aX ). Most importantly, an analysis of solutions in the vicinity of (aX , (cid:96)X ) corroborates the existence of a large spectrum of possible values for J, γX , ψX and CX (Fig. S8), meaning that these parameters cannot be estimated using a single extension curve, i.e., a single force. Moreover, the presence of the crest implies the existence of some degeneracy that precludes the unambiguous determination of aX and (cid:96)X , too. In principle, the value of CX could be addressed using torque measure- ments [26], yet only if one has access to λ(σ) (see Supplementary methods). For aX and (cid:96)X , one might expect that the underdetermination problem akin to a single force could be resolved by considering multiple forces. However, we find that, without fixing any other parameters, the crest gets blurrier as the force increases (blue points in Fig. 4C). To understand this intriguing result, we performed the same analyses but by setting one arbitrary parameter to its right value. We found that fixing either J or γX leads to the same underdeter- 13 Figure 4: Test of the possibility to recover parameters of an alternative X-DNA co-existing with B-DNA, using simulated extension curves (red curves in panel A). To this end, we first generated σ-extension curves using the zero writhe approximation model of Eq. 5 with a specifically predefined X-DNA (see main text for parameters). We next performed an inference procedure using the same generative model by producing 10 000 different curves corresponding to a random combination of the 6 parameters of X-DNA. A: σ-extension curves at f = 2 pN where the color of each curve is given by the value of one of the parameters: aX (left), (cid:96)X (center) and γX (right). See Fig. S6 for the other parameters. We can see that, compared to γX , aX and (cid:96)X have a well-defined, strong influence on extension curves. In these cases, the curves indeed appear to be sorted as a function of the values of the parameters. B: Each point corresponds to a random set of parameter values, with x- and y-axes corresponding to the values of aX and (cid:96)X , respectively. The opacity is inversely correlated (using an arbitrarily scale) to the root-mean-square deviation (RMSD) of the curves in A with respect to the red one, so that only the best fits are visible. We observe a crest of small RMSD values that go through the original parameters (aX , (cid:96)X ) (black cross). C: At a higher force (12 pN), the front is much less visible and the original set of solutions becomes isolated among a broad cloud of RMSD values (in blue). The crest becomes clearer again by setting ψX = ψX (= 0) and CX = CX (= 10 nm) (in red). The intersection between crests obtained at different forces then allows for an accurate estimation of aX and (cid:96)X (Fig. S7). 14 mination problem. In contrast, fixing either ψX or CX makes a crest of locally suboptimal solutions re-appear. However, setting ψX to the right value ψX = 0 results in a crest that is off the right solution (aX , (cid:96)X ) (Fig. S9), a shift that disappears only by setting CX = CX (red points in Fig. 4C). Overall, these results show that inference of aX and (cid:96)X from extension curves requires to have some prior knowledge about the nature of the alternative form, more particularly about its torsional parameters ( ψX and CX ). This may be explained by the fact that the (effective) torsional energy is quadratic in σ, such that any parameter that affects this energy is expected to have a large impact on any curve plotted as a function of σ. When the values of ψX and CX are known, a minimization over all forces then lead to a fair estimation of both aX and (cid:96)X which is all the more accurate that the range of available forces is large (the crossing of the different crests of solutions becoming well-defined as shown in Fig. S7). 3.3.2. Experimental data: estimation of D-DNA and L-DNA parameters To estimate a and (cid:96)p for both D-DNA and L-DNA, we use the data of Sheinin et al. [25] obtained for 2.2 kb long molecules, which consist of extension curves measured over a wide range of values of σ (from -2.3 to +0.1) and f (from 2.5 to 36 pN). In accord with the findings of [27], we further consider that L-DNA occurs for σ (cid:46) −1 and, hence, estimate D-DNA and L-DNA parameters using an optimization procedure of the extension curves for σ lying in [−0.6,−0.1] and in [−1.8,−1.3], respectively. As discussed above, we lift the underdetermination problem by setting, on one hand, ψD = 0 and ψL = −0.5 rad/bp and we test, on the other hand, CD = 1 nm as discussed in [25], CD = 28 nm as proposed in [34] and CL = 10 nm in accord with previous estimations [25, 26]. Similar to the case of simulated data, optimizations over forces from 2.5 to 12 pN reveals the existence of small islands of optimal solutions, which we consider as our best estimations of parameters (Fig. 5). For D-DNA (Fig. 5A), CD = 1 nm leads to aD > 0.6 nm and (cid:96)D close to 1 nm, which are different, although being qualitatively similar, from previous estimations according to which aD = 0.54 nm and (cid:96)D ∈ [2−3 nm] (see [25] and references therein). When CD = 28 nm, we find aD ∈ [0.45 − 0.5 nm] and (cid:96)D ∈ [2 − 5 nm]. Note, here, that we used the finest level of chain discretization (n = 1) and we checked that our results remain identical for rougher coarse-graining up to n = 4 (Fig. S10). The effect of n is indeed expected to affect the exact value of the domain wall penalty J (see above) and the entropic contribution to the free energy associated with the multiple possible positions of the denaturation bubbles (the term J (λ) in Eq. 5, see Supplementary methods). However, as discussed using simulated data, the former is expected to play only a marginal role in the determination of aD and (cid:96)D, while in the case of a fixed, small number of denaturation bubbles as here (typically one), the latter scales as ln(n). For L-DNA (Fig. 5B), we find aL ∈ [0.5 − 0.6 nm], which is slightly larger than previous estimations [25], and (cid:96)L ∈ [1− 2 nm], which is slightly smaller. In all cases, the small value of (cid:96)L corroborates that L-DNA is not a pure left-handed structural form but made of denatured DNA [25, 26]. 15 Figure 5: A: Inference of D-DNA parameters using experimental data from [25]. Each point represents the cumulative (over several forces ) RMSD resulting from a best fit procedure of experimental extension curves using numerical curves of the 2sRLC model generated under the zero-writhe approximation (see Eq. 5) and for a given set of parameters. Points are all the darker that the cumulative RMSD is small (arbitrary scale). For each set of parameters, the cumulative RMSD was computed for σ varying in [−0.6, −0.1] and over 6 different forces ranging from 2.5 to 12 pN. To mitigate underdetermination problems, we set ψD = 0 and tested two values of CD, which are representative of values discussed in the literature [25, 34]. For CD = 1 nm (red points), we find an island of optimal solutions characterized by aD > 0.6 nm and (cid:96)D close to 1 nm. For CD = 28 nm, we find aD ∈ [0.45−0.5 nm] and (cid:96)D ∈ [2−5 nm]. The dashed horizontal line in the figure indicates (cid:96)D = 2 nm. B: Same analysis but for L-DNA. Here, we set ψL = −0.5 rad/bp and test CL = 10 nm [25, 26]. We find aL ∈ [0.5 − 0.6 nm] and (cid:96)L ∈ [1 − 2 nm]. 16 To summarize, our results question the compatibility of previously used aD = 0.54 nm with reported torsional parameters. They also show the importance of having precise knowledge of a certain number of parameters to be able to fully exploit magnetic tweezers experiments in the high force regime. In absence of such prior information (as it is usually the case for real molecules), several strategies can be contemplated. For instance, working with an optimization procedure that is constrained by both torque curves and extension curves should narrow down the set of parameters leading to good fits. Even in this case, yet, having access to the fraction of the alternative form seems mandatory (see Supplementary methods). Potential development along this line might come from the use of DNA binding proteins sensitive to alternative forms. Additional constraints on model parameters could also be imposed using cyclization data and conformation statistics from cryo-electron microscopy experiments. 3.3.3. Extension of the model by including sequence effects In effect, our 2sRLC model bridges "kinkable" worm-like chain models and self-avoiding rod like chain models, which have been developed to respectively address the high flexibility of sharply bent DNA [53] and the superstructuring properties of supercoiled B-DNA molecules [15, 20]. By doing so, it opens novel perspectives in the field of supercoiled DNA, not only to better rationalize mag- netic tweezers experiments but also to predict behaviors of bacterial genomes in vivo. In this regard, including sequence effects, such as the tendency of AT- tracks and CpG-tracks to respectively favor DNA denaturation and left-handed forms [81], can be easily realized by using sequence dependent free energy forma- tion costs of alternative forms (the γsi in Eq. 3). As an example, we simulated the folding of a 1 kb long molecule where one single site (iAT ) had a much lower γ0 D than the other sites, thus mimicking the presence of an AT-track. Just as previously found in a more detailed, base resolution polymer model of DNA [76], we find that for a force f = 1 pN and a supercoiling density σ = −0.06, the site iAT is systematically denatured and serves as both nucleation and anchor points for a plectoneme (Fig. S11). 4. Conclusion More elaborate 2sRLC models can be contemplated by e.g. including se- quence effects at the level of DNA bending [82, 38]. In particular, by properly distinguishing intrinsic curvature from stiffness [38], one can expect to quanti- tatively address the structuring properties of bacterial sequences, including the possible pinning of pure B-DNA plectonemes [83]. In all cases, our ability to develop truly predictive models of supercoiled DNA depends on our capacity to precisely infer mechanical parameters of alternative forms, which itself relies on the derivation of analytical expressions of thermodynamic quantities as a func- tion of system parameters, at least for some range of forces and supercoiling densities [25, 26, 34]. Here, we have derived simple analytical expressions that accurately describe the free energy landscape of the 2sRLC model in the approx- imation of zero writhe, providing a powerful alternative to phenomenological 17 approaches. It would then be interesting to compare our analytical results to other exact analytical results obtained in models devoid of self-avoidance [58]. It would also be interesting to further expand the free energy landscape in power series of the writhe. This would allow including additional constraints coming from data at lower forces where alternative forms co-exist with super- structures [34]. This latter possibility appears to be crucial for unambiguously estimating parameters of alternative DNA forms. We have indeed shown that the proper estimation of a specific parameter may critically depend on the knowledge of other parameters or, equivalently, on their independent estimation -- see e.g. the impact of the torsional modulus on the estimation of both the distance between consecutive base pairs and the persistence length in Fig. 5B. A systematic par- allel analysis of extension curves together with torque curves (which is par- ticularly adapted to highlight torsional properties) seems therefore necessary. In addition, sequences of interest, like GC or AAT repeats as implemented in [26, 52], could be systematically used to channel the formation of alternative forms, which would have two main advantages. On one hand, this would allow "controlling" the fraction of the alternative form (λ in Eq. 5), thus lifting an important unknown of the problem. On the other hand, this would sharpen sig- nals because of stronger cooperative transitions (see e.g. [52]) and, hence, should constrain further model predictions -- just as for the co-existence of plectonemic conformations and strecthed conformations with alternative forms (Figs. 2 and S2), this may also require specific numerical methods to efficiently sample the different states at equilibrium. For more general sequences, DNA binding pro- teins sensitive to alternative forms, as used e.g. in [27], might offer solutions to compute the fraction λ. Acknowledgements We thank Ruggero Cortini and Marc Joyeux for their early input in this project, Bahram Houchmandzadeh for useful suggestions, Daniel Jost for crit- ical reading of the manuscript and Olivier Rivoire for helpful comments. We also thank Cees Dekker and John van Noort for providing us with their experi- mental data. I.J. is supported by an ATIP-Avenir grant (Centre National de la Recherche Scientifique). Appendix A. Supplementary Material The supplementary material is a single pdf file consisting of two sections: I) the "Supplementary methods" section provides details related to the analytics and simulation of the 2sRLC model and II) the "Supplementary figures" section contains 11 figures. 18 Author Contributions I.J and T.L. performed research; T.L. implemented simulation tools; I.J. and T.L. analyzed data; I.J. wrote the article. References References [1] I. Junier, O. Rivoire, Conserved Units of Co-Expression in Bacterial Genomes: An Evolutionary Insight into Transcriptional Regulation., PLoS ONE 11 (5) (2016) e0155740. [2] I. Junier, E. B. Unal, E. Yus, V. Llor´ens-Rico, L. Serrano, Insights into the Mechanisms of Basal Coordination of Transcription Using a Genome- Reduced Bacterium., Cell Systems 2 (6) (2016) 391 -- 401. [3] C. J. Dorman, 1995 Flemming Lecture. DNA topology and the global con- trol of bacterial gene expression: implications for the regulation of virulence gene expression., Microbiology (Reading, England) 141 ( Pt 6) (6) (1995) 1271 -- 1280. [4] G. W. Hatfield, C. J. Benham, DNA topology-mediated control of global gene expression in Escherichia coli., Annual review of genetics 36 (2002) 175 -- 203. [5] A. Travers, G. Muskhelishvili, DNA supercoiling - a global transcriptional regulator for enterobacterial growth?, Nature Reviews Microbiology 3 (2) (2005) 157 -- 169. [6] A. Valenti, G. Perugino, M. Rossi, M. Ciaramella, Positive supercoiling in thermophiles and mesophiles: of the good and evil, Biochemical Society Transactions 39 (1) (2011) 58 -- 63. [7] J. Ma, M. Wang, Interplay between DNA supercoiling and transcription elongation., Transcription 5 (3) (2014) e28636. [8] M. C. Lagomarsino, O. Espeli, I. Junier, From structure to function of bacterial chromosomes: Evolutionary perspectives and ideas for new ex- periments., FEBS Letters 589 (20 Pt A) (2015) 2996 -- 3004. [9] J. C. Wang, DNA topoisomerases: why so many?, The Journal of biological chemistry 266 (11) (1991) 6659 -- 6662. [10] L. F. Liu, J. C. Wang, Supercoiling of the DNA template during transcrip- tion., Proc Natl Acad Sci U S A 84 (20) (1987) 7024 -- 7027. [11] L. Postow, N. Crisona, B. Peter, C. Hardy, N. Cozzarelli, Topological chal- lenges to DNA replication: conformations at the fork, Proceedings of the National Academy of Sciences 98 (15) (2001) 8219 -- 8226. 19 [12] I. Junier, P. Fr´emont, O. Rivoire, Universal and idiosyncratic characteristic lengths in bacterial genomes., Physical biology 15 (3) (2018) 035001. [13] L. Postow, C. D. Hardy, J. Arsuaga, N. R. Cozzarelli, Topological domain structure of the Escherichia coli chromosome, Genes & Development 18 (14) (2004) 1766 -- 1779. [14] S. Deng, R. A. Stein, N. P. Higgins, Organization of supercoil domains and their reorganization by transcription, Molecular Microbiology 57 (6) (2005) 1511 -- 1521. [15] R. Strick, M.-N. Dessinges, G. Charvin, N. H. Dekker, J.-F. Allemand, D. Bensimon, V. Croquette, Stretching of macromolecules and proteins, Reports on Progress in Physics 66 (2003) 1 -- 45. [16] Z. Bryant, F. C. Oberstrass, A. Basu, Recent developments in single- molecule DNA mechanics., Current opinion in structural biology 22 (3) (2012) 304 -- 312. [17] T. Lionnet, J.-F. Allemand, A. Revyakin, T. R. Strick, O. A. Saleh, D. Ben- simon, V. Croquette, Single-molecule studies using magnetic traps., Cold Spring Harbor protocols 2012 (1) (2012) 34 -- 49. [18] F. Kriegel, N. Ermann, J. Lipfert, Probing the mechanical properties, con- formational changes, and interactions of nucleic acids with magnetic tweez- ers., Journal of Structural Biology. [19] X. Du, D. Wojtowicz, A. A. Bowers, D. Levens, C. J. Benham, T. M. Przytycka, The genome-wide distribution of non-B DNA motifs is shaped by operon structure and suggests the transcriptional importance of non-B DNA structures in Escherichia coli., Nucleic Acids Research 41 (12) (2013) 5965 -- 5977. [20] A. Vologodskii, Biophysics of DNA, Cambridge University Press, 2015. [21] C. J. Benham, Torsional stress and local denaturation in supercoiled DNA, Proceedings of the National Academy of Sciences 76 (8) (1979) 3870 -- 3874. [22] C. J. Benham, Stable cruciform formation at inverted repeat sequences in supercoiled DNA., Biopolymers 21 (3) (1982) 679 -- 696. [23] A. Rich, A. Nordheim, A. H. Wang, The chemistry and biology of left- handed Z-DNA., Annual review of biochemistry 53 (1) (1984) 791 -- 846. [24] Z. Bryant, M. D. Stone, J. Gore, S. B. Smith, N. R. Cozzarelli, C. Busta- mante, Structural transitions and elasticity from torque measurements on DNA., Nature 424 (6946) (2003) 338 -- 341. URL http://dx.doi.org/10.1038/nature01810 20 [25] M. Y. Sheinin, S. Forth, J. F. Marko, M. D. Wang, Underwound DNA under Tension: Structure, Elasticity, and Sequence-Dependent Behaviors, Physical Review Letters 107 (1) (2011) 108102. [26] F. C. Oberstrass, L. E. Fernandes, Z. Bryant, Torque measurements reveal sequence-specific cooperative transitions in supercoiled DNA., Proceedings of the National Academy of Sciences 109 (16) (2012) 6106 -- 6111. [27] R. Vlijm, A. Mashaghi, S. Bernard, M. Modesti, C. Dekker, Experimen- tal phase diagram of negatively supercoiled DNA measured by magnetic tweezers and fluorescence., Nanoscale 7 (7) (2015) 3205 -- 3216. [28] R. N. Irobalieva, J. M. Fogg, D. J. Catanese, T. Sutthibutpong, M. Chen, A. K. Barker, S. J. Ludtke, S. A. Harris, M. F. Schmid, W. Chiu, L. Zechiedrich, Structural diversity of supercoiled DNA., Nature commu- nications 6 (2015) 8440. [29] A. V. Vologodskii, S. D. Levene, K. V. Klenin, M. Frank-Kamenetskii, N. R. Cozzarelli, Conformational and thermodynamic properties of supercoiled DNA, Journal of Molecular Biology 227 (4) (1992) 1224 -- 1243. [30] K. Klenin, H. Merlitz, J. Langowski, A Brownian dynamics program for the simulation of linear and circular DNA and other wormlike chain poly- electrolytes, Biophysical Journal 74 (2 Pt 1) (1998) 780 -- 788. [31] A. Vologodskii, Simulation of equilibrium and dynamic properties of large DNA molecules, Computational Studies of DNA and RNA, Springer, Berlin (2006) 579 -- 604. [32] T. E. Ouldridge, A. A. Louis, J. P. K. Doye, Structural, mechanical, and thermodynamic properties of a coarse-grained DNA model., Journal of Chemical Physics 134 (8) (2011) 085101. [33] J. F. Marko, Torque and dynamics of linking number relaxation in stretched supercoiled DNA, Physical Review E 76 (2007) 21926. [34] H. Meng, J. Bosman, T. vanderHeijden, J. vanNoort, Coexistence of Twisted, Plectonemic, and Melted DNA in Small Topological Domains, Biophysical Journal 106 (5) (2014) 1174 -- 1181. [35] M. Manghi, N. Destainville, Physics of base-pairing dynamics in DNA, Physics Reports 631 (2016) 1 -- 41. [36] B. A. Krajina, A. J. Spakowitz, Large-Scale Conformational Transitions in Supercoiled DNA Revealed by Coarse-Grained Simulation., Biophysical Journal 111 (7) (2016) 1339 -- 1349. [37] Q. Wang, R. N. Irobalieva, W. Chiu, M. F. Schmid, J. M. Fogg, L. Zechiedrich, B. M. Pettitt, Influence of DNA sequence on the struc- ture of minicircles under torsional stress., Nucleic Acids Research 45 (13) (2017) 7633 -- 7642. 21 [38] J. S. Mitchell, J. Glowacki, A. E. Grandchamp, R. S. Manning, J. H. Mad- docks, Sequence-Dependent Persistence Lengths of DNA., Journal of chem- ical theory and computation 13 (4) (2017) 1539 -- 1555. [39] A. V. Vologodskii, J. F. Marko, Extension of torsionally stressed DNA by external force, Biophysical Journal 73 (1) (1997) 123 -- 132. [40] J. Lipfert, J. W. J. Kerssemakers, T. Jager, N. H. Dekker, Magnetic torque tweezers: measuring torsional stiffness in DNA and RecA-DNA filaments., Nat Methods 7 (12) (2010) 977 -- 980. [41] R. Schopflin, H. Brutzer, O. Muller, R. Seidel, G. Wedemann, Probing the Elasticity of DNA on Short Length Scales by Modeling Supercoiling under Tension, Biophysical Journal 103 (2) (2012) 323 -- 330. [42] T. Lepage, F. K´ep`es, I. Junier, Thermodynamics of Long Supercoiled Molecules: Insights from Highly Efficient Monte Carlo Simulations., Bio- physical Journal 109 (1) (2015) 135 -- 143. [43] J. Bednar, P. Furrer, A. Stasiak, J. Dubochet, E. H. Egelman, A. D. Bates, The twist, writhe and overall shape of supercoiled DNA change during counterion-induced transition from a loosely to a tightly interwound super- helix. Possible implications for DNA structure in vivo., Journal of molecular biology 235 (3) (1994) 825 -- 847. [44] I. D. Ivenso, T. D. Lillian, Simulation of DNA Supercoil Relaxation, Bio- physical Journal 110 (10) (2016) 2176 -- 2184. [45] K. V. Klenin, M. D. Frank-Kamenetskii, J. Langowski, Modulation of intramolecular interactions in superhelical DNA by curved sequences: a Monte Carlo simulation study., Biophysical Journal 68 (1) (1995) 81 -- 88. [46] G. Chirico, J. Langowski, Brownian dynamics simulations of supercoiled DNA with bent sequences., Biophysical Journal 71 (2) (1996) 955 -- 971. [47] C. K. Singleton, J. Klysik, S. M. Stirdivant, R. D. Wells, Left-handed Z- DNA is induced by supercoiling in physiological ionic conditions., Nature 299 (5881) (1982) 312 -- 316. [48] L. J. Peck, J. C. Wang, Energetics of B-to-Z transition in DNA., Proceed- ings of the National Academy of Sciences of the United States of America 80 (20) (1983) 6206 -- 6210. [49] D. B. Haniford, D. E. Pulleyblank, Facile transition of poly[d(TG) x d(CA)] into a left-handed helix in physiological conditions., Nature 302 (5909) (1983) 632 -- 634. [50] A. V. Vologodskii, M. D. Frank-Kamenetskii, Left-handed z form in su- perhelical dna: A theoretical study, Journal of Biomolecular Structure and Dynamics 1 (6) (1984) 1325 -- 1333. 22 [51] W. R. Bauer, C. J. Benham, The free energy, enthalpy and entropy of na- tive and of partially denatured closed circular DNA., Journal of molecular biology 234 (4) (1993) 1184 -- 1196. [52] F. C. Oberstrass, L. E. Fernandes, P. Lebel, Z. Bryant, Torque Spectroscopy of DNA: Base-Pair Stability, Boundary Effects, Backbending, and Breath- ing Dynamics, Physical Review Letters. [53] P. Cong, J. Yan, Recent progress on the mechanics of sharply bent DNA, Science China Physics 59 (8) (2016) 99. [54] J. Yan, J. F. Marko, Localized single-stranded bubble mechanism for cy- clization of short double helix DNA., Physical Review Letters 93 (10) (2004) 108108. [55] P. A. Wiggins, R. Phillips, P. C. Nelson, Exact theory of kinkable elas- tic polymers., Physical review. E, Statistical, nonlinear, and soft matter physics 71 (2 Pt 1) (2005) 021909. [56] P. Ranjith, P. B. S. Kumar, G. I. Menon, Distribution functions, loop formation probabilities, and force-extension relations in a model for short double-stranded DNA molecules., Physical Review Letters 94 (13) (2005) 138102. [57] M. Manghi, J. Palmeri, N. Destainville, Coupling between denaturation and chain conformations in DNA: stretching, bending, torsion and finite size effects., Journal of physics. Condensed matter : an Institute of Physics journal 21 (3) (2009) 034104. [58] A. K. Efremov, R. S. Winardhi, J. Yan, Transfer-matrix calculations of DNA polymer micromechanics under tension and torque constraints., Phys- ical Review E 94 (3-1) (2016) 032404. [59] R. M. Fye, C. J. Benham, Exact method for numerically analyzing a model of local denaturation in superhelically stressed DNA, Physical Review E. [60] D. Jost, A. Zubair, R. Everaers, Bubble statistics and positioning in super- helically stressed DNA., Physical Review E 84 (3 Pt 1) (2011) 031912. [61] V. V. Rybenkov, A. V. Vologodskii, N. R. Cozzarelli, The effect of ionic conditions on DNA helical repeat, effective diameter and free energy of supercoiling., Nucleic Acids Research 25 (7) (1997) 1412 -- 1418. [62] C. G. Baumann, S. B. Smith, V. A. Bloomfield, C. Bustamante, Ionic effects on the elasticity of single DNA molecules, Proceedings of the National Academy of Sciences 94 (12) (1997) 6185 -- 6190. [63] J. R. Wenner, M. C. Williams, I. Rouzina, V. A. Bloomfield, Salt De- pendence of the Elasticity and Overstretching Transition of Single DNA Molecules, Biophys J 82 (6) (2002) 3160 -- 3169. 23 [64] T. Strick, D. Bensimon, V. Croquette, Micro-mechanical measure- ment of the torsional modulus of DNA, Genetica 106 (1999) 57 -- 62, 10.1023/A:1003772626927. [65] P. Carrivain, M. Barbi, J.-M. Victor, In Silico Single-Molecule Manipula- tion of DNA with Rigid Body Dynamics, PLoS Comput Biol 10 (2) (2014) e1003456. [66] T. Lepage, I. Junier, Modeling Bacterial DNA: Simulation of Self-Avoiding Supercoiled Worm-Like Chains Including Structural Transitions of the He- lix., Methods in molecular biology (Clifton, N.J.) 1624 (1) (2017) 323 -- 337. [67] J. SantaLucia, A unified view of polymer, dumbbell, and oligonucleotide DNA nearest-neighbor thermodynamics., Proceedings of the National Academy of Sciences of the United States of America 95 (4) (1998) 1460 -- 1465. [68] J. M. Huguet, C. V. Bizarro, N. Forns, S. B. Smith, C. Bustamante, F. Ri- tort, Single-molecule derivation of salt dependent base-pair free energies in DNA., Proceedings of the National Academy of Sciences 107 (35) (2010) 15431 -- 15436. [69] L. J. Peck, J. C. Wang, Energetics of B-to-Z transition in DNA., Proceed- ings of the National Academy of Sciences of the United States of America 80 (20) (1983) 6206 -- 6210. [70] J. A. Gebe, S. A. Allison, J. B. Clendenning, J. M. Schurr, Monte Carlo simulations of supercoiling free energies for unknotted and trefoil knotted DNAs., Biophysical Journal 68 (2) (1995) 619 -- 633. [71] J. L. Barrat, J. F. Joanny, Persistence length of polyelectrolyte chains, EPL (Europhysics Letters) 24 (5) (1993) 333 -- 338. [72] S. K. Nomidis, F. Kriegel, W. Vanderlinden, J. Lipfert, E. Carlon, Twist- Bend Coupling and the Torsional Response of Double-Stranded DNA., Physical Review Letters 118 (21) (2017) 217801. [73] J. F. Marko, E. D. Siggia, Bending and twisting elasticity of dna, Macro- molecules 27 (4) (1994) 981 -- 988. doi:10.1021/ma00082a015. [74] D. Salerno, A. Tempestini, I. Mai, D. Brogioli, R. Ziano, V. Cassina, F. Mantegazza, Single-Molecule Study of the DNA Denaturation Phase Transition in the Force-Torsion Space, Phys. Rev. Lett. 109 (2012) 118303. [75] A. Laio, F. L. Gervasio, Metadynamics: a method to simulate rare events and reconstruct the free energy in biophysics, chemistry and material sci- ence, Reports on Progress in Physics 71 (1) (2008) 126601. [76] C. Matek, T. E. Ouldridge, J. P. K. Doye, A. A. Louis, Plectoneme tip bub- bles: Coupled denaturation and writhing in supercoiled DNA., Scientific Reports 5 (2015) 7655. 24 [77] Y. Seol, J. Li, P. Nelson, T. Perkins, M. Betterton, Elasticity of Short DNA Molecules: Theory and Experiment for Contour Lengths of 0.6-7µm, Biophysical Journal 93 (12) (2007) 4360 -- 4373. [78] C. Bouchiat, M. D. Wang, J. Allemand, T. Strick, S. M. Block, V. Cro- quette, Estimating the persistence length of a worm-like chain molecule from force-extension measurements., Biophys J 76 (1 Pt 1) (1999) 409 -- 413. [79] J. D. Kahn, E. Yun, D. M. Crothers, Detection of localized DNA flexibility., Nature 368 (6467) (1994) 163 -- 166. [80] T. J. Thomas, V. A. Bloomfield, Chain flexibility and hydrodynamics of the B and Z forms of poly(dG-dC).poly(dG-dC)., Nucleic Acids Research 11 (6) (1983) 1919 -- 1930. [81] S. M. Mirkin, Discovery of alternative DNA structures: a heroic decade (1979-1989)., Frontiers in bioscience : a journal and virtual library 13 (2008) 1064 -- 1071. [82] S. Geggier, A. Vologodskii, Sequence dependence of DNA bending rigidity., Proceedings of the National Academy of Sciences 107 (35) (2010) 15421 -- 15426. [83] S. H. Kim, M. Ganji, J. van der Torre, E. Abbondanzieri, C. Dekker, DNA sequence encodes the position of DNA supercoils, biorxiv.orgdoi:http: //dx.doi.org/10.1101/180414. 25 Supplementalmaterial:ApolymermodelofbacterialsupercoiledDNAincludingstructuraltransitionsofthedoublehelixThibautLepageandIvanJunierCNRS,TIMC-IMAG,F-38000Grenoble,FranceUniv.GrenobleAlpes,TIMC-IMAG,F-38000Grenoble,France1CONTENTSI.Supplementarymethods2A.Derivationoftheeffectivemodelwithaglobaltorsionalenergy2B.Freeenergyprofilesinthezerowritheapproximation4C.Generalizationtoanarbitrarilynumberofalternativeforms5D.Torques61.Measurementinnumericalsimulations62.Expressioninthezerowritheapproximationandaccesstothetorsionalmodulus6E.Inferenceofparametersbytestingalargenumberofparameters7References7II.Supplementaryfigures9I.SUPPLEMENTARYMETHODSInthefollowing,parametersareuniformalongtheDNAmolecule.Sequenceeffectswillbetreatedinsubsequentworks.A.DerivationoftheeffectivemodelwithaglobaltorsionalenergyWestartbywritingdowntheequilibriumprobabilityofaconfigurationCwithDNA-formprofile{si}i=1..NinthecontextofanimposedlinkingnumberLk.Tothisend,wedividetheconformationalenergyintotwoterms:thetorsionalenergyET(C)=kBT2PNi=1nCsiasi(φi−ψsi)2thatdependsonthetorsionalangles{φi}i=1..Nandtherest,ET(C),whichdoesnot.Inthiscontext,denotingβ=1/kBT,wehave:P(C)=Z−1exp"−βET(C)−XinCsi2asi(φi−ψsi)2#δ(Tw(C)−TwLk(C))(1)withZthepartitionfunctionsuchthatPCP(C)=1.Theδ(x)function,whichisequalto1ifx=0and0otherwise,imposestheconservationofLk,thatis,itonlyretainsconformationsforwhichthetotaltwistofthemolecule,Tw(C),isequaltoTwLk(C)≡Lk−Wr(C).2UsingTw(C)=(2π)−1Pinφianddefiningt(C)=TwLk(C)/Nbp=TwLk(C)/nN,Eq.1canberewrittenas:P(C)=Z−1exp"−βET(C)−XinCsi2asi(φi−ψsi)2#δ Xiφi−2πtN!(2)Theintegrationovertheφi'sthenresultsinevaluatingthefollowingterm(forclarity,wedroptheC's):I=ZYkdφkexp"−XinCsi2asi(φi−ψsi)2#δ Xiφi−2πtN!=(usingtheFouriertransformofδ)ZdxYkZdφkexp(cid:20)−nCsk2ask(φk−ψsk)2−2πix(2πt−φk)(cid:21)=ZdxYkZdφkexp"−nCsk2ask(cid:18)φk−ψsk−2πiasknCskx(cid:19)2#exp(cid:20)−2π2asknCskx2−2πix(2πt−ψsk)(cid:21)="YkZdφkexp"−nCsk2ask(cid:18)φk−ψsk−2πiasknCskx(cid:19)2##Zdxexp"−Xj(cid:18)2π2askjnCsjx2+2πix(2πt−ψsj)(cid:19)#(3)Definingtheconformation-dependentaveragetorsionalsusceptibilityperbasepairχ(C)≡PNi=1asiCsi(cid:14)N,theconformation-dependentaveragetwistangleatrestψ(C)≡PNi=1ψsi(cid:14)N,weobtain(stilldroppingtheC's):I=(2π)N/2sYkasknCskZdxexp(cid:20)−2π2Nχn(cid:16)x2+inπχ(2πt−ψ)x(cid:17)(cid:21)=(2π)N/2sYkasknCskZdxexp(cid:20)−2π2Nχn(cid:18)(cid:16)x+in2πχ(2πt−ψ)(cid:17)2+n24π2χ2(2πt−ψ)2(cid:19)(cid:21)=(2π)N/2sYkasknCskexp(cid:20)−nN2χ(2πt−ψ)2(cid:21)Zdxexp(cid:20)−2π2Nχn(cid:16)x+in2πχ(2πt−ψ)(cid:17)2(cid:21)=(2π)(N−1)/2N1/2sYkaskχCskexp(cid:20)−nN2χ(2πt−ψ)2(cid:21)=(2π)(N−1)/2N1/2exp"−12 Nbp2χ(2πt−ψ)2+ln(χ)−Xiln(cid:18)asiCsi(cid:19)!#,(4)which,afterredistributionofthetermsln(χ)andln(cid:16)asiCsi(cid:17)inthetorsionalenergyandthefreeenergyofformation,respectively,leadstotheeffectiveenergiesofthemaintext(Eqs.3and4).NotealsothatinEq.4ofthemaintext,wehaveusedthefactthatTw(C)=TwLk(C)inthesimulatedpolymermodel.3B.FreeenergyprofilesinthezerowritheapproximationInthecaseoftheco-existenceoftwoforms,hereB-DNAandX-DNA(anyalternativeDNAform),usingEqs.2,3and4ofthemaintext,theeffectiveenergy,U(λ,X),ofacon-formationwhereX-DNAoccupiesλNsites(correspondingtoλNbpbasepairs)distributedinXdomainsreads:U(λ,X)=λNbpγX+2XJ+EXB−fzX+EBB−fzB+kBT2"Nbpχ(λ)(cid:18)2πTwNbp−ψ(λ)(cid:19)2+ln(χ(λ))#(5)whereEsBindicatesthetotalbendingenergyoftheformsandwherewehaveusedthenotationsχs=as/Csandχ/ψ(λ)=(1−λ)χB/ψB+λχs/ψs -- notethatψ(λ)representstheaveragetwistangleatrestofthemoleculewithafractionλofalternativeforms.Next,con-sideringthatthesupercoilingdensity,σ,istotallyconvertedintotwistintheapproximationofzerowrithe,wecanreplace2πTwNbpby(1+σ)ψB.Finally,integratingthecorrespondingpartitionfunctionoverthebendinganglesofeachsiteleadstoafreeenergywherethetermsEsB−fzsarereplacedbygs,f(X),thefreeenergyassociatedwiththeWLCatforcefwithbendingmodulusKsandcontourlengthX.Theresultingbi-dimensionalfreeenergysurface,Sf,σ(λ,X)thereforereads:Sf,σ(λ,X)=λNbpγX+2XJ+gX,f(λNbpaX)+gB,f((1−λ)NbpaB)+kBT2"ψ2BNbpχ(λ)(cid:18)1+σ−ψ(λ)ψB(cid:19)2+ln(χ(λ))#(6)TheunidimensionalfreeenergyprofileFf,σ(λ)(Eq.5inthemaintext)canbeobtainedbysumming(withBoltzmannweights)overthevaluesofX,leadingto:Ff,σ(λ)=λNbpγX+J(λ)+gX,f(λNbpaX)+gB,f((1−λ)NbpaB)+kBT2"ψ2BNbpχ(λ)(cid:18)1+σ−ψ(λ)ψB(cid:19)2+ln(χ(λ))#(7)with:J(λ)=−kBTln XXAN,XλNexp[−2XJ/kBT]!(8)wherewehavedefinedAN,XY=(cid:18)Y−1X−1(cid:19)(cid:18)N−Y+1X(cid:19),(9)4thenumberofwaystodistributeYsitesinXdifferentdomainsamongNpossiblesites.UsingEq.6,onecanthencomputetheequilibriumnumberofdomainsinthezerowritheapproximationbynumericallydetermining:X∗=ZdXdλXexp[−Sf,σ(λ,X)/kBT](cid:14)ZdXdλexp[−Sf,σ(λ,X)/kBT](10)Inaddition,usingEq.7andh•iλ≡Rdλ•exp[−Ff,σ(λ)/kBT](cid:14)Rdλexp[−Ff,σ(λ)/kBT],onecanderivetheotherequilibriumquantitiesasfollows(seebelowfortorques):λ∗=hλiλ,(11)z∗=−h∂fFf,σ(λ)iλ=λ∗X+(1−λ∗)B(12)withstheWLCextensionassociatedwiththeformsofDNA,whichdependsonlyonasand's[1].NotefinallythatnumericalintegrationisperformedbycircumventingnumericalproblemscomingfromthegenerationoflargecombinatorialquantitieswhenλNbecomeslarge(asinthecasee.g.ofEq.9).Recallingthatλ=K/NwhereKisthenumberofsitesoccupiedbythealternativeform,theintegrationthusconsistsofadiscretesumoverrestrictedvaluesofKandXforwhichBoltzmannweightsaremaximal.Tothisend,wefirstlocalisethesaddle-pointofSf,σ(λ,X)andthenperformthesumovervaluesofKandXlyingintherectanglethatcontainsthesaddle-pointandoutsideofwhichrelativeBoltzmannweightswithrespecttothesaddle-pointaresmallerthan10−4.Wefurthercheckedthatourresultdidnotdependonthechoiceofthiscut-off.C.GeneralizationtoanarbitrarilynumberofalternativeformsInthesituationofkpossibleDNAforms(oneB-DNAplusk−1alternativeforms),denotingthevectoroffractionofeachoftheseformsbyλ≡{λi}i=1..k,withPki=1λi=1,thefreeenergylandscapebecomes:Ff,σ(λ)=NbpkXi=1λiγi+I(λ)+kXi=1gi,f(λiNbpai)+kBT2ψ2BNbpPki=1λiχi 1+σ−kXi=1λiψiψB!2+ln kXi=1λiχi!(13)5whereheretheindexialwaysindicatesthedifferentformsandwhereI(λ)indicatesthefreeenergyassociatedwiththemultiplepossibilitiestodistributethedifferentalternativesitesintodistinctdomains.D.Torques1.MeasurementinnumericalsimulationsTomeasurethetorqueexertedbyamoleculeinoursimulations,weusethesamemethodasinourpreviouswork[2]whereweadaptedacommonlyusedexperimentalmethod[3,4]tothecaseofrod-likechainmodelswithglobaltorsionalenergy.Itconsistsinattachingtheendsofthemoleculetoareservoiroftwistswithwhichthiscanexchangehelixturns.Thereservoir,alsocalledamagnetictrap,ischaracterizedbyastiffnesskRsuchthatanyvariationofthetwistwithinit,δTwR,withrespecttoareferencetwist(seebelow)isassociatedwithanenergycost2π2kBTkRδTw2R.Similarlytothemeasurementsinmagnetictorquetweezers,theequilibriumtorqueexertedbythemoleculeisthengivenbyΓ∗=−2πkR(Tw∗−Tw∗0),whereTw∗andTw∗0arerespectivelytheequilibriumtwistsofthesupercoiledmoleculeandofthecorrespondingrelaxedmolecule,bothmeasuredinthepresenceofthemagnetictrap.Notehere,thatthelinkingnumberinthereservoirisfurthercalibratedsuchthatTw∗0isequaltothenumberofhelicesofthemoleculeatrest,Tw0.2.ExpressioninthezerowritheapproximationandaccesstothetorsionalmodulusTheequilibriumtorqueexertedbythemoleculeisdefinedbyΓ∗=h∂ΘFf,σ(λ)iλwhereΘistherotationangleofthetipofthemoleculearoundthestretchingdirection(thedirectionperpendiculartothemagneticfieldusedtoaddhelicalturnstothemolecule).Inmagnetictweezersexperiments,Θisacontrolparameterrelatedtothelinkingnumber:Θ=2πLk.Asaconsequence,Γ∗=(2π)−1h∂LkFf,σ(λ)iλ,whichbecomesΓ∗=(2πTw0)−1h∂σFf,σ(λ)iλintheapproximationofzerowrithe,whereLk=Twandσ=(Tw−Tw0)/Tw0.UsingEq.7,wethusobtainforthecaseoftheco-existenceofX-andB-DNA:Γ∗=kBTψB*1+σ−ψ(λ)ψBχ(λ)+λ(14)6WethennotethatwheneverthefractionλofX-DNAisfixed(andknown),Eq.14canbeusedtoestimateCX(appearinginχ(λ))bycomputingtheslopeoftheσ-torquecurvesinceinthiscaseΓ∗=Cte+kBTψBχ(λ)σ.ThishasbeenusedbyBryant'sgroupinthecontextofsmall,specificallydesignedbubbleofalternativeforms[5,6].E.InferenceofparametersbytestingalargenumberofparametersIntheabsenceofaprioriknowledgeaboutthemechanicalparametersofalternativeforms,weuseda"blind"approachconsistingini)drawingalargenumberofpoints(typi-callythousands)inthespaceofparametersaccordingtoauniformdistribution,ii)comput-ingσ-extensioncurvesusingthezero-writheapproximationandiii)comparingthegeneratedcurvestotheoriginaldata(eithersimulateddataasinFigure4ofthemaintextorexperi-mentaldataasinFigure5ofthemaintext).Tolimitthespaceofparameters,wemadeavarybetween0.34nm(B-DNA)and0.7nm(approximatelythelargestvaluemeasuredfordenaturedDNAbeforebreaking[7]),'varybetween1nm(lowestestimatefordenaturedDNA[8])and15nm(atleast5timeslargerthananypreviousestimationsforeitherD-DNAorL-DNA),Cvarybetween1nm(lowestestimateofbothD-DNAandL-DNA[8])and100nm(typicalvalueforB-DNA),ψvarybetween-0.6and0.6rad/bp(helicityofB-DNA),γvarybetween1kBTand3kBT(theaveragefreeenergyformationcostofal-ternativeformsisestimatedtobearound2kBT[9])andJvarybetween2kBT(leadingtoseveraldomainsinoursimulations)and20kBT(singledomain).[1]R.Strick,M.-N.Dessinges,G.Charvin,N.H.Dekker,J.-F.Allemand,D.Bensimon,andV.Croquette,"Stretchingofmacromoleculesandproteins,"ReportsonProgressinPhysics66,1 -- 45(2003).[2]ThibautLepage,Fran¸coisK´ep`es,andIvanJunier,"ThermodynamicsofLongSupercoiledMolecules:InsightsfromHighlyEfficientMonteCarloSimulations."BiophysicalJournal109,135 -- 143(2015).[3]L´aszl´oOroszi,P´eterGalajda,HubaKirei,S´andorBottka,andP´alOrmos,"Directmeasure-mentoftorqueinanopticaltrapanditsapplicationtodouble-strandDNA."PhysicalReviewLetters97,058301(2006).7[4]JanLipfert,JacobWJ.Kerssemakers,TessaJager,andNynkeH.Dekker,"Magnetictorquetweezers:measuringtorsionalstiffnessinDNAandRecA-DNAfilaments."NatMethods7,977 -- 980(2010).[5]FlorianCOberstrass,LouisEFernandes,andZevBryant,"Torquemeasurementsrevealsequence-specificcooperativetransitionsinsupercoiledDNA."ProceedingsoftheNationalAcademyofSciences109,6106 -- 6111(2012).[6]FCOberstrass,LEFernandes,PLebel,andZBryant,"TorqueSpectroscopyofDNA:Base-PairStability,BoundaryEffects,Backbending,andBreathingDynamics,"PhysicalReviewLetters(2013).[7]D.Bensimon,A.J.Simon,V.Croquette,andA.Bensimon,"StretchingDNAwithaRecedingMeniscus:ExperimentsandModels,"Phys.Rev.Lett.74,4754 -- 4757(1995).[8]MaximYSheinin,ScottForth,JohnFMarko,andMichelleDWang,"UnderwoundDNAunderTension:Structure,Elasticity,andSequence-DependentBehaviors,"PhysicalReviewLetters107,108102(2011).[9]JSantaLucia,"Aunifiedviewofpolymer,dumbbell,andoligonucleotideDNAnearest-neighborthermodynamics."ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica95,1460 -- 1465(1998).8II.SUPPLEMENTARYFIGURESFIG.S1.EffectofConthesimulatedextensioncurves.UsingC=100nmfitsbettertheexperimentaldataforf=0.5pNwhileC=80nmgivesabetteragreementforf=4.5pN.9FIG.S2.Hysteresiscyclesaroundtheplectonemic-denaturedtransition.Simulationsofa2.4kbmoleculewereperformed(σ=−0.05)duringwhichtheforcewascontinuouslyincreased(asindi-catedbythelowerleftarrow)anddecreased(upperrightarrowarrow)atconstantspeedfrom0.3to1.8pN.Foreachofthethreespeedsreportedhere(fromthefastesttotheslowest:red,greenandbluecurves),12cycleswereperformed,revealingineachcaseastronghysteresispatternintheforce-extensiondiagram.FIG.S3.Snapshotofasimulationclosetotheplectoneme-denaturationtransition(inthefigureS2).Adenaturationbubble(red)appearsattheapexofaplectoneme.10FIG.S4.Relativedifferencebetweenthezero-writheapproximationandoursimulationsandbetweenMarko'smodelandoursimulations,correspondingtoFig.3inthemaintext.FIG.S5.FInite-sizeeffectsonthezero-writheapproximation:extensionasafunctionofσfor2differentforces,includingfinite-sizeeffects(squares)ornot(circles).Thepointsareactuallyalmostexactlysuperimposed;theywereslightlyoffsetalongthex-axisforreadability.11FIG.S6.RainbowplotsoftheextensionwithrespecttothevaryingparametersnotshowninFig.3ofthemaintext,namely,CX,ψXandJ.CXandJhavenoeffectontheextension.ForψX,valuesawayfromψB=0.6rad/bptendtoyieldbetterfits.FIG.S7.Inasituationwhereψ=ψXandC=CX,combiningRMSDsformultipleforces(2,2.5,3,3.5,6,8.5and12pN),thebestsolutionsresultingfromtheintersectionofthecrestsshowninFig.2ofthemaintextarelocatedaround(aX,'X)(blackcross).12FIG.S8.Typicalvaluesofparametersforbestfitsolutions.WithinthesliceaX∈[0.52,0.56nm]centeredaroundaX(seeFig.4inthemainarticle),weselectedthe100bestfitstothefakedataandplottedthehistogramofvaluesfortheotherparameters(dashedlinesrepresent'X,CX,JandγX).Contrarytotheotherparametersthatspreadacrossalltheirallowedrange,thevaluesof'Xareconcentratedaround'X(topleft).13FIG.S9.SettingonlyψX=ψX(=0),athighforce(12pN)wefindacrestofsub-optimalfitsthatgoofftheoriginalparameters(aX,'X)(blackcross).14FIG.S10.Effectofthecoarse-grainingonthezero-writheapproximation.TheresultsofFig.5ofthemaintextareunchangedwhenusingadifferentlevelofdiscretization(heren=4).15FIG.S11.Snapshotofasimulationofa1kb-longmolecule(100cylinders,f=1pN,σ=−0.06)wherethecylinderatthecenter(number50)hasnodenaturationpenalty(γD,i=50=0).Thedenaturationbubble(red)appearsandstaysatthecenterduringthewholesimulation.16
1811.05805	2	1811	2019-06-06T16:48:18	Trail-Mediated Self-Interaction	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	A number of microorganisms leave persistent trails while moving along surfaces. For single-cell organisms, the trail-mediated self-interaction will influence its dynamics. It has been discussed recently [Kranz \textit{et al.} Phys. Rev. Lett. \textbf{117}, 8101 (2016)] that the self-interaction may localize the organism above a critical coupling $\chi_c$ to the trail. Here we will derive a generalized active particle model capturing the key features of the self-interaction and analyze its behavior for smaller couplings $\chi < \chi_c$. We find that fluctuations in propulsion speed shift the localization transition to stronger couplings.	physics.bio-ph	physics	Trail-Mediated Self-Interaction W. Till Kranz1, 2, a) and Ramin Golestanian3, 2 1)Institute for Theoretical Physics, Universitat zu Koln, Zulpicher Strasse 77, 50937 Koln, Germany 2)Rudolf Peierls Centre for Theoretical Physics, University of Oxford, Oxford OX1 3NP, United Kingdom 3)Max Planck Institute for Dynamics and Self-Organization (MPIDS), Am Fassberg 17, 37077 Gottingen, Germany (Dated: 7 June 2019) A number of microorganisms leave persistent trails while moving along surfaces. For single-cell organisms, the trail-mediated self-interaction will influence the dynamics. It has been discussed recently [Kranz et al. Phys. Rev. Lett. 117, 038101 (2016)] that the self-interaction may localize the organism above a critical coupling χc to the trail. Here we will derive a generalized active particle model capturing the key features of the self-interaction and analyze its behavior for smaller couplings χ < χc. We find that fluctuations in propulsion speed shift the localization transition to stronger couplings. I. INTRODUCTION Motility or active motion of an organism is most useful if it can occur in response to external stimuli.1 -- 3 It has been recognized from the early days4 that directed motion is already realized in both prokaryotic and eukaryotic single- celled microorganisms.5 -- 7 On the microbial scale, chemical signals and the corresponding response -- chemotaxis -- are the most widespread but by no means the only stimuli that are used as information. Surface dwelling microorganisms like the bacteria from the species Pseudomonas, Neisseria and amoeboid slime molds are known to leave trails8,9 of high molecular weight biopolymers like polysaccharides.10 -- 12 It is believed that these trails are used as a means of cell-cell communi- cation in particular in the process of colony and spore formation.6,9,13 -- 19 The precise mechanisms, however, are still under active investigation. Living organisms have intricate signal processing path- ways, even on the microbial scale.20 Therefore the re- sponse to a stimulus may be the result of a complicated control algorithm. On the other hand there is evolution- ary value in robustness.21 Simplicity often lends itself to robustness. In recent years there has been a grow- ing recognition that mechanistic models that forgo the intracellular chemical signal processing may be able to explain surprisingly complicated behavior.22 -- 26 Here we follow this active particle approach and do not address the question of how much internal signal processing is involved in trail interaction. Rapidly diffusing, small molecules such as cyclic adeno- sine monophosphate (camp) are also employed in cell-cell communication in the form of auto-chemotaxis.27,28 This form of inter-microbial communication is relatively well understood both on the level of the intra-cellular signal- ing path-ways and on the level of collective effects. Most artificial active particles are auto-chemotactic in that they a)Electronic mail: [email protected] create and are propelled by local chemical gradients.26,29 All these systems are characterized by particles or organ- isms whose typical speed v0 or typical effective diffusivity D is much smaller than the diffusivity of the the signaling molecules Dm (cid:29) D, v0R not least because their size R is much bigger than molecular length scales. In effect, an organism's self-generated concentration field is to a lowest order approximation isotropic with respect to the organism itself even if the organism is in motion. Self- interaction of the organism with its own auto-chemotactic field due to direct coupling with the translational degrees of freedom has been studied in the past30 and shown not to be strong enough to lead to trapping.31 The trail material, on the other hand, consists of en- tangled macromolecules32 with a very low diffusivity Dm (cid:28) D, v0R. A moving microorganism will therefore encounter an anisotropic distribution of its own trail. It obviously leaves a trail behind and not all around. As a consequence, the only observable dynamics is the ef- fective dynamics that results from the interplay of the organism's propulsion mechanism and the trail-mediated self-interaction. We have recently shown that this self- interaction may profoundly alter the dynamics.33,34 Here we will discuss this mechanism in more detail and on a more general basis. We will start by specifying the model in Sec. II and de- rive the effective dynamics in Sec. III. Using this effective description we will analyze the orientational dynamics in Sec. IV and the translational diffusivity in Sec. V. We will briefly discuss the influence of speed fluctuations in Sec. VI before closing in Sec. VII. II. BARE DYNAMICS In the following we will be exclusively concerned with the dynamics of a single microorganism on a pristine, essentially flat surface. The state of a microorganism at time t is fully described by its position r(t), its orientation n(t) = (cos ϕ(t), sin ϕ(t)) and the trail it has left so far ψ(x, t). We assume active propulsion along the current 9 1 0 2 n u J 6 ] h p - o i b . s c i s y h p [ 2 v 5 0 8 5 0 . 1 1 8 1 : v i X r a (cid:112) orientation n(t) with a mean speed v0 and small fluctua- tions on top, characterized by a translational diffusivity Dv, dr(t) = n(t)(v0 dt + 2Dv dWt). (1) Here Wt denotes a Wiener process. Orientational persis- tence is observed to be limited in microorganisms such that the rotational diffusivity D0 r should be substantial. It has been found that a torque is naturally generated by gradients of the trail field, ∇ ψ(x, t), perpendicular to the microorganism's instantaneous orientation. To be precise, the organism may not react to the actual trail field but only to the trail field it senses via some transfer function Ξ(x, t) that may perform some spatial averaging and potentially some time integration. The orientation dynamics is therefore of the following form, dϕ(t) = χ∇⊥(t)(Ξ ∗ ψ)(r(t), t) dt + 2D0 r dZt, (2) (cid:112) where Z is a second, independent Wiener processes,35 χ characterizes the sensitivity of the microorganism to the trail and ∇⊥(t) := eϕ(t) · ∇ where eϕ(t) ⊥ n(t). The as- terisk denotes a convolution in space and time and Ξ(x, t) trajectory r(t) := {r(t(cid:48))}t(cid:48)<t, the microorganism deposits trail material with a constant rate k and distributed ac- cording to a compact profile Ψ(x2) normalized such that is normalized such that(cid:82) d2x(cid:82) dt Ξ(x, t) = 1. Along its (cid:82) Ψ(x2) d2x = 1, ψ(x, tr(t)) = k (cid:90) t −∞ Ψ(cid:0)[x − r(t(cid:48) )]2(cid:1) dt(cid:48). Equations. (1 -- 3) constitute a set of stochastic integro- differential equations for the time evolution of a single crawling microorganism. We note that in order to un- derstand the dynamics of the microorganism, we do not need the full trail field ψ(x, t) but only what the organism senses ∇⊥ψ(t) := ∇⊥(t)(Ξ ∗ ψ)(r(t), t). We imagine a primarily mechanic response to the trail such that the transfer function is instantaneous on the time scales of the noise, Ξ(x, t) ∝ δ(t). Then it is sufficient to consider an effective profile Ψ := Ξ ∗ Ψ, i.e., dt(cid:48) −∞ (cid:90) t )]2(cid:1) = ∇⊥ψ(t) = k∇⊥(t) ∇⊥(t)Ψ(cid:0)[r(t) − r(t(cid:48) = 2eϕ(t) · [r(t) − r(t(cid:48) )]Ψ We may then expand the gradient as Ψ(cid:0)[r(t) − r(t(cid:48) )]2(cid:1). (cid:48)(cid:0)[r(t) − r(t(cid:48) )]2(cid:1), (3) (4) (5) where the prime on Ψ denotes the derivative with respect to the argument. 2 FIG. 1. Sketch of the trail profile (as sensed by the organism) ψ(r2) and its (negative) derivative −ψ(cid:48)(r2) as a function of the squared distance from the trail's center. We assume well defined trail edges, ∆R/R (cid:28) 1, throughout. defined (see Fig. 1) with a characteristic size R and a trail boundary of width ∆R (cid:28) R. Then we may approximate Ψ(x2) ≈ Θ(R2 − x2)/πR2 where Θ(x) is the Heaviside step-function. Likewise, we need the active propulsion speed to be sufficiently well defined, i.e., Dv (cid:28) v0R. Then the characteristic time to cross a trail τ is narrowly distributed around R/v0. We now assume a priori that trails are sufficiently straight that self crossings can be neglected. We will find below that the self-interaction renormalizes the rotational diffusivity D0 r to an effective value Dr, i.e., we assume Dr (cid:104)τ(cid:105) (cid:28) 1 where (cid:104)τ(cid:105) is the mean crossing time. If the trajectories are straight enough so that the organism only rarely crosses its own trail, we can safely ignore these self-crossings. In the following we will adopt units such that R = v0 = 1 and k = π. With the above assumptions, Eq. (1 -- 3) reduce to dr(t) = n(t)(dt + dϕ(t) = χ∇⊥ψ(t) dt + 2Dv dWt) (cid:112) (cid:112) (cid:90) t −∞ 2D0 r dZt dt(cid:48)rtt(cid:48)δ(1 − rtt(cid:48)) ∇⊥ψ(t) = eϕ(t) · where rtt(cid:48) ≡ r(t) − r(t(cid:48)). (6) (7) (8) Due to the Dirac delta in Eq. (8) and our assumption that self-crossings are negligible, we only need to integrate over a time interval of order unity. To this end, we iterate the equations of motion (6,7) to lowest order and find r(t) − r(t − t(cid:48) ) = n(t)t(cid:48) 2Dv n(t)W−t(cid:48) + (cid:112) (cid:90) t−t(cid:48) (cid:112)DvD0 t ×(cid:104) + eϕ(t) du × (cid:90) u (cid:112) (cid:90) t−t(cid:48) t χ∇⊥ψ(w) dw + 2D0 r dZw r eϕ(t) Zu dWu, t (cid:105) (9) III. EFFECTIVE DYNAMICS + 2 To make progress we have to make a number of as- sumptions. We need the trail profile to be sufficiently well where we have used that eϕ(w) ≈ eϕ(t) to lowest order. In the following we are going to drop the last term in Eq. (9) because it contains the product of the small parameters Dv, D0 r (cid:28) 1. The first term in Eq. (9) turns the Dirac delta in Eq. (8) into a first passage time problem of a one-dimensional Brownian motion with unit drift, t + 2DvWt, on the positive half line with a reflecting boundary at 0, i.e., √ dt(cid:48)rtt(cid:48)δ(1 − rtt(cid:48)) = r(t) − r(t − τ ), (10) (cid:90) t −∞ (cid:104) (cid:90) τ 0 where τ is a random variable, the first passage time. The characterization of the first passage time distribution p(τ ) is a non-trivial task that we are not going to pursue here but note that for Dv → 0, p(τ ) → δ(τ − 1). closed equation for ∇⊥ψ(t), Making use of Eqs. (9) and (10) in Eq. (8) we find a ∇⊥ψ(t) = χ∇⊥ψ(w) dw + 2D0 r dZw . (11) Once we have a solution of Eq. (11) we can use it in Eqs. (6, 7) to analyze the effective dynamics of a microorganism under trail-mediated self-interaction. Let us start by considering a different representation (cid:90) t−τ (cid:90) u (cid:104) du t t (cid:105) of Eq. (11), (cid:90) τ 0 ∇⊥ψ(t) = (τ−w) χ∇⊥ψ(t − w) dw + (cid:112) (cid:112) (cid:105) 2D0 r dZt−w (12) to investigate the mean gradient (cid:104)∇⊥ψ(t)(cid:105)Z = χ dw(τ − w)(cid:104)∇⊥ψ(t − w)(cid:105)Z . (13) This is solved by the ansatz (cid:104)∇⊥ψ(t)(cid:105)Z ∼ eαt given the rate α solves the equation λτ (α) = 0 where [cf. Eq. (A5)] (cid:20) (cid:0)e−ατ − 1(cid:1)(cid:21) λτ (α) = 1 − χ α τ + 1 α For vanishing speed fluctuations, Dv → 0, τ ≡ 1, and we recover the behavior of Ref. 33, i.e., the average gradient relaxes to zero, α < 0, for weak coupling to the trail, χ < 2, whereas it grows exponentially above the critical value χc = 2. For a discussion of the localization transition that occurs for χ > χc we refer to Ref. 33. the ensemble average, λ(α) :=(cid:82) ∞ For significant fluctuations, Dv > 0, we need to analyze 0 dτ p(τ )λτ (α), over the unknown first passage time distribution p(τ ). For the time being we assume that we may approximate λ(α) ≈ λ(cid:104)τ(cid:105)(α), i.e., by replacing the random variable τ by its mean which is known exactly,36 (cid:16) (cid:17) (cid:39) 1 − Dv. (cid:104)τ(cid:105) = 1 + Dv e−1/Dv − 1 (15) From λ(cid:104)τ(cid:105)(α) we find a critical coupling strength χc = 2/(1− Dv)2 which is shifted to larger values for increasing speed fluctuations. (cid:112) 3 For χ < χc, ∇⊥ψ(t) represents a stochastic process with zero mean and Eq. (11) can be solved in the Fourier domain37 (cf. Sec. A) (cid:93)∇⊥ψ(ω) = 2D0 r χ × [λ−1 τ (iω) − 1]iω Zω. (16) Note that by the Wiener representation theorem, iω Zω ∼ N (0, 1) are iid normal random variables. In other words ∇⊥ψ(t) is essentially filtered white noise. In particular, ϕ(ω) = r λ−1 τ (iω) Zω 2D0 (17) (cid:112) is a stationary Gaussian process for χ < χc. This implies that the joint probability, P (ϕ1 − ϕ2, t) = P (ϕ1, ϕ2, t(cid:48), t(cid:48) + t) = 1(cid:112) δϕ2(t) =(cid:10)[ϕ(t(cid:48) 2πδϕ2(t) )]2(cid:11) . + t) − ϕ(t(cid:48) e−(ϕ1−ϕ2)2/2δϕ2(t), (18) is fully specified by the angular mean square displacement (19) , IV. THE ANGULAR MEAN SQUARE DISPLACEMENT r Using Eq. (17) we find dτ p(τ )λτ (s)−2, (cid:90) ∞ s2(cid:100)δϕ2(s) = 2D0 where f (s) = LT[f ](s) = (cid:82) ∞ s2(cid:100)δϕ2(s) = 2D0 0 dtf (t)e−st is the Laplace transform of f (t), s = σ + iω ∈ C. To make progress and derive the results discussed in Ref. 33, we will again use the mean first passage time approximation, rλ(cid:104)τ(cid:105)(s)−2 (20) (21) 0 . (14) and will analyze λ−1(cid:104)τ(cid:105)(s), cf. Eq. (14). In order to understand the angular mean square dis- placement in the time domain, we need to know the analytical structure of λ−1(cid:104)τ(cid:105)(s). Due to the oscillating factor e−iω there are infinitely many poles in the complex plain and an analytical inverse Laplace transform is not tractable. We may, however, consider the asymptotic limits t → ∞ and t → 0. Note that, apart from the trivial scale factor D0 r , the only control parameters that affect δϕ2(t) is the coupling strength χ and the speed fluctuations Dv. A. Short-Time Asymptotics Expanding λ(cid:104)τ(cid:105)(σ) in powers of σ−1 we find λ(cid:104)τ(cid:105)(σ) = 1 − χ(1 − Dv)σ−1 + χσ−2 + O(cid:0)σ−2e−1/σ−1(cid:1). (22) 4 FIG. 2. Influence of the speed fluctuations Dv on the effective dynamics for strong coupling to the trail χ = 1.8. (a,b) Angular mean square displacement δϕ2(t) and translational mean square displacement δr2(t)/R2 as a function of time t normalized by the mean crossing time (cid:104)τ(cid:105) for three different magnitudes of the speed fluctuations Dv/v0R = 0 (blue), 0.05 (yellow), 0.1 (green) r (cid:104)τ(cid:105) = 0.1. (c) Sample trajectories r(t) for the three different Dv/v0R = 0 and a common value of the orientational diffusivity D0 (i), 0.05 (ii), and 0.1 (iii) color coded as a function of time for a total duration 50(cid:104)τ(cid:105) and D0 r (cid:104)τ(cid:105) = 0.01. See Sec. C for details on the numerics. The first three terms are dominant as long as e−1/σ−1 (cid:28) 1, i.e., as long as σ−1 (cid:28) 1, or upon reinstating units, for times t (cid:28) (cid:104)τ(cid:105). Then we have λ(cid:104)τ(cid:105)(σ → ∞)−2 (cid:39) 1 + 2(1 − Dv)χσ−1 − 2χ(1 + χ/χc)σ−2, and in the time domain δϕ2(t → 0)/2D0 r (cid:39) t+χ(1−Dv)t2− 1 3 χ(1+χ/χc)t3. (24) In other words, the angular mean square displace- ment starts with the bare diffusivity D0 r before the self-interaction becomes visible on times of the order 1/χ(1 − Dv). B. Long-Time Asymptotics Expanding λ(cid:104)τ(cid:105)(σ) in powers of σ we find λ(cid:104)τ(cid:105)(σ) = 1 − χ/χc + (1 − Dv)σ + O(σ2). (25) This shows that δϕ2(t → ∞) = 2Drt will asymptotically always be diffusive with a renormalized diffusivity38 χ 3χc Dr/D0 r = λ(cid:104)τ(cid:105)(0)−2 = 1 + which diverges for χ → χc. χ χc × 2 − χ/χc (1 − χ/χc)2 (26) The validity of the long time asymptotics is bounded by the radius of convergence of the Taylor expansion, Eq. (25). The latter is determined by the location of the pole of λ−1(cid:104)τ(cid:105)(s) closest to the origin of the complex plane. In Sec. B we show that Eq. (25) holds for s < 3(χc/χ − 1)/(cid:104)τ(cid:105), i.e., for times The onset of the asymptotic regime, t∗, diverges with the rotational diffusivity as χ → χc. For χ → χc we write χ = χc(1 − δχ). Assuming the smallest pole σ = O(δχ) to be confirmed below we expand to lowest order (23) λ(cid:104)τ(cid:105)(σ) = δχ + 1 3 (1 − Dv)σ + O(δχ2, σ2, σδχ). (28) Close to the critical coupling strength we therefore find λ(cid:104)τ(cid:105)(σ)−2 (cid:39) 9 (1 − Dv)2σ2 × 1 [1 + 3δχ/(1 − Dv)σ]2 , i.e., a superballistic behavior δϕ2(t) = 3 2 χcD0 r t3 in a diverging time window (cid:104)τ(cid:105) (cid:28) t (cid:28) 1/3δχ. (29) (30) V. THE TRANSLATIONAL MEAN SQUARE DISPLACEMENT The velocity autocorrelation function C(t) := (cid:104) r(t + t(cid:48)) · r(t(cid:48))(cid:105) = 2Dvδ(t) + Cnn(t) where Cnn(t) := (cid:104)n(t + t(cid:48)) · n(t(cid:48))(cid:105) can be determined explicitly with the help of Eq. (18),39 Cnn(t) = dϕP (ϕ, t) cos ϕ = e−δϕ2(t)/2. (31) (cid:90) ∞ −∞ (cid:90) ∞ 0 The asymptotic translational diffusivity is then given by a Green-Kubo integral 2Dvδ(t) + e−δϕ2(t)/2(cid:105) (cid:104) , (32) D/D0 = D0 r dt t/(cid:104)τ(cid:105) (cid:29) t∗ := χ/χc 1 − χ/χc . (27) where D0 = 1/D0 r is the diffusivity for χ = Dv = 0.39 101100101t/1011001011021032(t)101100101t/102101100101102r2(t)/R2-10 0 10-10 0 10 20(i)(ii)(iii)(a)(b)(c)The (translational) mean square displacement δr2(t) := (cid:10)[r(t + t(cid:48)) − r(t(cid:48))]2(cid:11), (cid:90) t (cid:90) u (cid:90) t 0 δr2(t) = 2 du dwC(w) 0 = 2Dvt + 2 dw(t − w)e−δϕ2(w)/2 D/D0 = Dv/D0 + D0 r (33) (34) 5 patching together the short time, bare diffusion 2D0 r t, and the intermediate time, superballistic behavior, ∝ t3. The asymptotic diffusion is irrelevant here because δϕ2(t) (cid:29) 1 before it sets in. Using this ansatz in Eq. (32), we find (cid:90) (cid:104)τ(cid:105) 0 + D0 r r t dte−D0 (cid:90) ∞ (cid:104)τ(cid:105) dte−3D0 r χct3/4. (40) The second integral can be expressed in terms of the generalized exponential integral En(x), D/D0 = 1 + Dv/D0 − e−D0 r(cid:104)τ(cid:105) r (cid:104)τ(cid:105) E2/3(3D0 D0 + 1 3 Consistent with our assumption D0 expand this to yield D/D0 (cid:39) Γ(4/3)(D0 r (cid:104)τ(cid:105))2/3 − ( 3(cid:112) r (cid:104)τ(cid:105) /2). (41) r (cid:104)τ(cid:105) (cid:28) 1 we need to 3/2(cid:104)τ(cid:105) − 1)D0 r , (42) where Γ(x) is the Euler Gamma-function. Note that for a r → 0, the translational perfectly persistent organism, D0 diffusivity will obviously diverge D ∼ (D0 r (cid:104)τ(cid:105))−1/3. VI. THE EFFECT OF SPEED FLUCTUATIONS Nonzero speed fluctuations Dv > 0 have multiple effects as can bee seen by the examples in Fig. 2. For the angular mean square displacement, the influence lies mostly in the distance to the critical point χc. At vanishing fluc- tuations, Dv = 0, the chosen coupling strength χ = 1.8 is close to the critical value χc = 2 and the trajectories already begin to violate the assumptions of the derivation by turning quickly. For increasing fluctuations Dv, the critical value χc = 2/(1 − Dv)2 shifts to higher values. In effect, both the intermediate regime of the angular mean square displacement as well as the asymptotic diffusivity Dr decrease and the trajectories become straighter. For the translational mean square displacement δr2(t), the effects are less drastic but can be seen in both the short and the long time limit. For short times, the ballis- tic regime, δr2(t) ∝ t2 is replaced by diffusive behavior, δr2(t) = Dvt. At intermediate times, the directed motion prevails if the short time diffusivity is small enough as it has been discussed by Peruani and Morelli 40. For long times the straighter trails enhance translational diffusiv- ity for increasing fluctuations Dv but rather mildly so because the differences in δϕ2(t) mostly occur at large displacements δϕ2(t) (cid:29) 1. 0 cannot be given in closed form. However, the form of Eq. (33) indicates, that it will be dominated by the small angle behavior, δϕ2(t) < 1, of the angular mean square displacement. In the following we will derive analytic expressions for the diffusivity D in certain limiting cases. A. Short Persistence Regime For parameters such that δϕ2((cid:104)τ(cid:105)) (cid:29) 1, we may use the short time expansion, Eq. (24). With this, the condition r (cid:29) 1/[1 + χ(1 − Dv)] which shows that this reads 2D0 regime applies for small coupling strength χ and large intrinsic rotational diffusivity D0 r . Given this is fulfilled, we have (cid:90) ∞ (cid:112)π/κ erfc(cid:0)1/ κ(cid:1)e1/κ + Dv/D0 dte−D0 √ 0 r [t+χ(1−Dv)t2] = = 1 + Dv/D0 − κ/2 + 3κ2/4 + O(κ3), (37) r (cid:28) 1 is a kind of Peclet number where κ := 4χ(1−Dv)/D0 relating the "convective" rate χ to the diffusive rate D0 r . D/D0 = Dv/D0 + D0 r (35) (36) B. Long Persistence Regime The opposite limit is given by an angular mean square displacement which reaches the value one well into the asymptotic regime, δϕ2(t∗) (cid:28) 1. A condition which may be estimated as Dr/(1 − χ/χc) (cid:28) 1. Then we may approximate δϕ(t) = 2Drt for all relevant times and directly find D/D0 = D0 r /Dr, or, equivalently, D = 1/Dr. (38) In other words for very persistent trails, i.e., small intrin- sic directional noise D0 r and/or small coupling strength χ (cid:28) χc, the asymptotic translational diffusivity is in- versely proportional to the asymptotic rotational diffusiv- ity. Higher order terms are given in Ref. 33. C. Critical Regime Close to the critical coupling strength, χ → χc, we use VII. CONCLUSION Eq. (30), to make the ansatz δϕ2(t)/2D0 r = tΘ((cid:104)τ(cid:105) − t) + 3χct3Θ(t − (cid:104)τ(cid:105))/4, (39) We have started by motivating a model of a self pro- pelled particle (the microorganism) on a two-dimensional plane that interacts with its own trail, cf. Eqs (1 -- 3). A simplified version of this model has been introduced by us before.33 Here we argued that in reality the microorgan- ism will not interact with the trail, ψ, itself but with its observation of the trail, ψ. Given that ψ has well defined edges, ∆R/R (cid:28) 1, that the trails are reasonably straight, Dr (cid:104)τ(cid:105) (cid:28) 1 and the propulsion speed v0 is well defined, Dv/Rv0 (cid:28) 1, we showed how to decouple the equation for the trail's gradient, Eq. (11), from the equation of motion of the particle, Eqs. (6,7). Analyzing the effective trail gradient, Eq. (11), we showed that it fails to regress to a zero mean beyond a critical coupling strength kχc πv2 0R = 2 (1 − Dv/v0R)2 ≥ 2. (43) However for χ < χc, both the effective trail gradient, Eq. (16), as well as the orientation, Eq. (17), turn out to be filtered white noise. The filter function λ−1 τ (iω), Eq. (14), therefore, is crucial for the dynamics. In essence we derived generalized expressions for the effective angular [Eq. (26)] and translational diffusivity [Eqs. (37, 38, 42)], special cases of which have been presented in Ref. 33. The effect of the self-interaction becomes apparent on a timescale t/(cid:104)τ(cid:105) ∼ 1/χ which indicates the start of an intermediate regime displaying angular superdiffusion that extends to times t/(cid:104)τ(cid:105) ∼ χ/χc(1 − χ/χc) beyond which the dynamics is effectively diffusive again. The onset of the asymptotic regime diverges for χ → χc. The translational dynamics is unaffected by the asymptotic behavior of the angular motion due to the exponential suppression in Eq. (31). Consequently, the translational diffusivity remains finite at the critical coupling χ → χc, Eq. (42). A detailed analysis of the localized phase χ > χc requires a new approach that includes the effect of frequent self-crossings neglected here and is left to future work. We note that trail-mediated self-interactions have been recently studied in the context of aggregation and collec- tive motion of myxobacteria, where experiments revealed that the collective motion of the wild type of Myxococcus Xanthus is organized in a network structure made out of trails, in contrast with signaling-deficient mutants of the species.41,42 These results have been rationalized theoret- ically using a phenomenological agent-based model with implemented trail-following rules.43 It will be interesting to extend our study to mechanistically study the case of myxobacteria by incorporating both positional and orientational coupling to the trail and making predictions about the collective behaviour as as been performed for the case of Pseudomonas aeruginosa.33,34 We also note that the self-trapping reported here is somehow remi- niscent of the milling patterns that appear in a similar cognitive flocking model of animals.44 6 ACKNOWLEDGMENTS We acknowledge helpful discussions with Anatolij Ge- limson. This work was supported by the Human Frontier Science Program RGP0061/2013. W. T. K. thanks the dfg for partial funding through KR 4867/2-1. Appendix A: Solving Eq. (11) To this end we start from yet another representation of Eq. (11), ∇⊥ψ(t) = (cid:112) (cid:90) τ 2D0 r du(Zt − Zt−u) (cid:90) τ 0 0 + χ (cid:90) t du t−u dw∇⊥ψ(w). (A1) Employing the Fourier representation, we find for the first du(Zt − Zt−u) = 0 = (cid:90) ∞ −∞ dω Zωeiωt (cid:90) τ 0 du(1 − e−iωu) (A2) and for the second term (cid:90) t du t−u dw∇⊥ψ(w) = (cid:90) ∞ (cid:90) τ (cid:90) t = dω −∞ 0 du t−u Upon performing the w-integral this yields dw(cid:93)∇⊥ψ(ω)eiωw. (A3) term(cid:90) τ (cid:90) τ 0 (cid:90) τ 0 (cid:90) t (cid:90) ∞ t−u dw∇⊥ψ(w) = (cid:93)∇⊥ψ(ω) dω −∞ iω du (cid:90) τ 0 Together with du(1 − e−iωu) = τ + eiωt du(1 − e−iωu). 0 (cid:90) τ (cid:0)e−iωτ − 1(cid:1) (cid:112) 1 iω (A4) (A5) (cid:17) (cid:16) this implies that Eq. (11) in the Fourier domain reads (cid:93)∇⊥ψ(ω) = 1 − λτ (iω) χ χ(cid:93)∇⊥ψ(ω) + 2D0 r iω Zω (A6) which can easily be solved for (cid:93)∇⊥ψ(ω) to yield Eq. (16). Appendix B: Poles near the Origin For χ < χc we can rule out a pole at the origin. Then we may rewrite the condition λ(cid:104)τ(cid:105)(s) = 0 as s2 − χ(cid:104)τ(cid:105) s − χ(cid:0)e−s(cid:104)τ(cid:105) − 1(cid:1) = 0, (B1) and to third order in s χ(cid:104)τ(cid:105)3 s3 + (1 − χ(cid:104)τ(cid:105)2 /2)s2 = 0. (B2) 1 6 This is solved by s∗ = −3(χc/χ − 1)/(cid:104)τ(cid:105). Appendix C: Details of the Numerics We determined δϕ2(t) using a numerical inverse Laplace transformation of Eq. (21) by the method of Abate and Valk´o 45 implemented in Python with the help of the multi-precision library mpmath.46 The translational mean square displacement δr2(t), we determined by numerical integration of Eq. (33) using SciPy's47 quad method. For the trajectories we used SciPy's inverse fast Fourier transform ifft to determine ϕ(t) from Eq. (17) and then Euler integration of Eq. (6). 1T. Fenchel, Science 296, 1068 (2002). 2J. Adler, Science 153, 708 (1966). 3I. Chet and R. Mitchell, Annu. Rev. Microbiol. 30, 221 (1976). 4T. W. Engelmann, Arch. gesamte Physiol. 26, 537 (1881). 5L. L. Burrows, Annu. Rev. Microbiol. 66, 493 (2012). 6B. Rodiek and M. J. B. Hauser, EPJ ST 224, 1199 (2015). 7B. Maier and G. C. L. Wong, Trends Microbiol. 23, 775 (2015). 8A. J. Merz and K. T. Forest, Curr. Biol. 12, R297 (2002). 9C. R. Reid, T. Latty, A. Dussutour, and M. Beekman, Proc. Natl. Acad. Sci. 109, 17490 (2012). 10R. P. Burchard, J. Bacteriol. 152, 495 (1982). 11B. E. Christensen and W. G. Characklis, in Biofilms, Ecological and Applied Microbiology, edited by W. G. Characklis and K. C. Marshall (John Wiley & Sons, New York, 1990) pp. 93 -- 130. 12K. Zhao, B. S. Tseng, B. Beckerman, F. Jin, M. L. Gibiansky, J. J. Harrison, E. Luijten, M. R. Parsek, and G. C. Wong, Nature 497, 388 (2013). 13J. T. Bonner and L. J. Savage, J. Exp. Zool. 106, 1 (1947). 14R. M. Harshey, Mol. Microbiol. 13, 389 (1994). 15T. Nakagaki, Res. Microbiol. 152, 767 (2001). 16D. L. Higashi, S. W. Lee, A. Snyder, N. J. Weyand, A. Bakke, and M. So, Infect. Immun. 75, 4743 (2007). 17D. Kaiser, Curr. Biol. 17, R561 (2007). 18E. Bernitt, C. Oettmeier, and H.-G. Dobereiner, in 6th World Congress of Biomechanics (Springer, 2010) pp. 1133 -- 1136. 19G. Amselem, M. Theves, A. Bae, E. Bodenschatz, and C. Beta, PloS ONE 7, e37213 (2012). 20S. L. Porter, G. H. Wadhams, and J. P. Armitage, Nature Rev. Microbiol. 9, 153 (2011). 21J. A. G. M. de Visser, J. Hermisson, G. P. Wagner, L. A. Meyers, H. Bagheri-Chaichian, J. L. Blanchard, L. Chao, J. M. Cheverud, S. F. Elena, W. Fontana, et al., Evolution 57, 1959 (2003). 7 22S. Saha, R. Golestanian, and S. Ramaswamy, Phys. Rev. E 89, 062316 (2014). 23J. Palacci, S. Sacanna, A. Abramian, J. Barral, K. Hanson, A. Y. Grosberg, D. J. Pine, and P. M. Chaikin, Sci. Adv. 1, e1400214 (2015). 24J. R. Gomez-Solano, A. Blokhuis, and C. Bechinger, Phys. Rev. Lett. 116, 138301 (2016). 25A. Zottl and H. Stark, J. Phys. Condensed Matt. 28, 253001 (2016). 26P. Illien, R. Golestanian, and A. Sen, Chem. Soc. Rev. 46, 5508 (2017). 27A. Sengupta, S. van Teeffelen, and H. Lowen, Phys. Rev. E 80, 031122 (2009). 28A. Gelimson and R. Golestanian, Phys. Rev. Lett. 114, 028101 (2015). 29B. Liebchen and H. Lowen, Acc. Chem. Res. 51, 2982 (2018). 30Y. Tsori and P.-G. de Gennes, Europhysics Letters (EPL) 66, 599 (2004). 31R. Grima, Phys. Rev. Lett. 95, 128103 (2005). 32I. W. Sutherland, Microbiology 147, 3 (2001). 33W. T. Kranz, A. Gelimson, K. Zhao, G. C. L. Wong, and R. Golestanian, Phys. Rev. Lett. 117, 038101 (2016). 34A. Gelimson, K. Zhao, C. K. Lee, W. T. Kranz, G. C. L. Wong, and R. Golestanian, Phys. Rev. Lett. 117, 178102 (2016). 35While there may be processes that generate fluctuations in trans- lation and orientation at the same time34, additional sources of noise and the coarse-graining of time to the diffusional time scale largely decorrelate the fluctuations. 36R. Mahnke, J. Kaupusz, and I. Lubashevsky, Physics of Stochastic Processes: How Randomness Acts in Time, 1st ed. (Wiley-VCH, Weinheim, 2009). 37f (t) = FT−1[ f ](t) ≡(cid:82) ∞ −∞ dω f (ω)eiωt. 38This corrects a misprint in Eq. (4) of Ref. 33. 39M. Doi and S. F. Edwards, The Theory of Polymer Dynamics, International Series of Monographs on Physics, Vol. 73 (Clarendon Press, Oxford, 1988). 40F. Peruani and L. G. Morelli, Phys. Rev. Lett. 99, 010602 (2007). 41J. Starruss, F. Peruani, V. Jakovljevic, L. Sogaard-Andersen, A. Deutsch, and M. Bar, Interface Focus 2, 774 (2012). 42S. Thutupalli, M. Sun, F. Bunyak, K. Palaniappan, and J. W. Shaevitz, Journal of The Royal Society Interface 12, 20150049 (2015). 43R. Balagam and O. A. Igoshin, PLOS Computational Biology 11, e1004474 (2015). 44L. Barberis and F. Peruani, Phys. Rev. Lett. 117, 248001 (2016). 45J. Abate and P. P. Valk´o, Int. J. Numer. Meth. Engng. 60, 979 (2004). 46F. Johansson et al., mpmath: a Python library for arbitrary- (2013), precision floating-point arithmetic (version 0.18) http://mpmath.org/. 47E. Jones, T. Oliphant, P. Peterson, et al., "SciPy: Open source scientific tools for Python," (2001 -- ), http://www.scipy.org.
1012.4153	2	1012	2011-03-28T21:20:51	Modelling DNA Response to THz Radiation	[ "physics.bio-ph" ]	Collective response of DNA to THz electric fields is studied in a simple pair bond model. We confirm, with some caveats, a previous observation of destabilising DNA breather modes and explore the parameter-dependence of these modes. It is shown that breather modes are eliminated under reasonable physical conditions and that thermal effects are significant.	physics.bio-ph	physics	Modelling DNA Response to THz Radiation Eric S. Swanson∗ Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh, PA 15260, USA. (Dated: October 22, 2018) Collective response of DNA to THz electric fields is studied in a simple pair bond model. We confirm, with some caveats, a previous observation of destabilising DNA breather modes and explore the parameter-dependence of these modes. It is shown that breather modes are eliminated under reasonable physical conditions and that thermal effects are significant. 1 1 0 2 r a M 8 2 ] h p - o i b . s c i s y h p [ 2 v 3 5 1 4 . 2 1 0 1 : v i X r a PACS numbers: 87.14.gk, 87.50.U-, 87.16.A- I. INTRODUCTION There is longstanding speculation that nonionising ra- diation can damage biological function at the cellular level[1]. More specifically, it has also been suggested that nonionising radiation of varying frequency causes cancer[2]. Since nonionising radiation cannot directly disrupt DNA structure, such genotoxic effects must de- rive from resonance phenomena driven by external elec- tromagnetic radiation1 We shall shortly see that the natural frequency of os- cillation of DNA base pair separations is approximately 1 THz, thus THz radiation is of special interest. Further- more, it is very likely that this is the unique frequency range of relevance to bio-resonance effects in DNA. Inter- est in this issue has recently been heightened due to the deployment of full body scanners in airports that employ millimeter wave (typically 30-300 GHz) technology. Motivated by these observations, Alexandrov et al. have examined the effects of coupling an electric field driving force to a model of dsDNA bond dynamics[4]. The resulting model of damped, driven, coupled non- linear oscillators can naturally exhibit exotic collective behaviour (for similar earlier conclusions see Ref. [5]). For example, the familiar period-doubling approach to chaotic dynamics is present. Of more immediate inter- est is the discovery of a nonlinear discrete breather mode that arises in response to a specific perturbation of the system. This mode stores energy for very long times and can lead to unbinding effects in dsDNA, with obvious implications for the genotoxicity of THz radiation. While the results of Alexandrov et al. are compelling, it is unclear if the model is sufficiently robust to per- mit application to physically realisable situations (such as body scanners). In particular, this paper critically ex- amines the choice of parameter values, investigates the effect of including thermal fluctuations, and examines the stability of breather modes in a variety of scenar- ios. It will be shown that parameter variation can elim- inate breather modes entirely, or make them unrealis- tically strong, that thermal noise completely dominates the external influences of the system, and that it is ex- tremely unlikely that dsDNA denaturing can be induced by experimentally accessible THz radiation. II. MODEL DEFINITION AND THE AGBUR BREATHER The model of Alexandrov et al. is based on a model of dsDNA pairing dynamics due to Dauxois et al. (the PBD model)[6]. The PBD model employs a Morse potential to model hydrogen bonding between base pairs (and other effects) and an inter-pair stacking potential. Since the nucleotide bonding interactions are much weaker than those of the phosphate-sugar backbone, the degrees of freedom associated with the backbone are neglected. The model also ignores degrees of freedom associated with the helicoidal structure of dsDNA. The resulting model is described by mi yi = −U(cid:48) i (yi) − W (cid:48)(yi+1, yi) − W (cid:48)(yi, yi−1). √ where yi/ of the i'th base pair. The Morse potential is given by 2 is the deviation from equilibrium distance (1) Ui(y) = Di[exp(−aiy) − 1]2 (2) and the stacking potential between consecutive base pairs is modelled as2 1 2 W (yi, yi−1) = k(yi − yi−1)2 (1 + ρ exp(−β(yi + yi+1)) . (3) In general the parameters can depend on the linked base pairs and hence can be labelled ki,i−1, etc. Properties of this model, including the melting transition, were studied in Refs. [6, 9]. Alexandrov et al. chose to supplement the PBD model with periodic driving and frictional terms to model the ∗[email protected] 1 It is worth remembering that biological electric noise generates 2 There is an obvious error in the definition of W in Ref. [4]. This internal fields with strengths up to 0.1 V/m[3]. has been repeated in Ref. [7]. interactions of dsDNA with an electric field. They state that the interactions of the base pairs with an external electric field are difficult to model and therefore they as- sume a simple harmonic driving force. The additional terms are then − miγ yi + A cos Ωt. (4) Evidently the drag term is not required to model the interactions with an electric field, however such a term is required to produce collective nonlinear phenomena. Fi- nally, Alexandrov et al. assumed a homogeneous poly(A) DNA molecule with 64 base pairs. Parameters employed were m = 300 amu (this was not specified in Ref. [4] -- I assume the value given in Ref. [6]), D = 0.05 eV, a = 4.2 1/A, k = 0.025 eV/A2, β = 0.35 1/A, and ρ = 2.0. A relaxation time typical to water of γ = 1.0/ps was used. As expected, this system displays complicated nonlin- ear dynamics. Of particular interest is a breather mode found by Alexandrov et al. (which we call the AGBUR breather) under a perturbation specified by δyi(t) = δ(t−t0) 0.42 cos[ π 4 (i−i0)] θ(−4 ≤ i ≤ 4)(A) (5) at frequency Ω = 2.0 THz and with a drive force of A = 144 pN. The breather was localised to be approxi- mately four base pairs wide and had a maximum ampli- tude of approximately 0.3 A. The authors note that fluc- tuations like that of Eq. 5 can occur thermally and hence transcription and genotoxic effects can be expected. A. Breather Characteristics In preparation for a detailed examination of these claims, we first seek to reproduce the AGBUR breather. Solutions were obtained via a microcanonical molecu- lar dynamics simulation employing the coupled Runge- Kutta (RK4) algorithm. This proved extremely accurate (with relative deviations in total energy of order 5 · 10−6 over 10 ps) and fast. The Verlet method was also im- plemented, yielding results in agreement with RK4, al- though less accurate. We follow Ref. [4] and employ 64 base pairs with periodic boundary conditions. A breather mode was found at Ω = 1.0 THz, some- what smaller than the 2.0 THz employed in Ref. [4]. Al- though it was similar in shape, the maximum amplitude of this breather was found to be about 4 A. Note that a compression of 0.42 A, such as generated by the pertur- bation of Eq. 5, raises the energy of a single bond pair by approximately 1.2 eV, which represents an enormous insertion of energy. In fact the bond length can then be expected to recoil to very large distances, with damping supplied by the stacking potential. Thus one anticipates that large amplitude breathers, such as found here, are to be expected. Another breather with a double-lobe structure was found at a frequency of Ω = 1.5 THz. This novel mode is shown in Fig. 1. 2 FIG. 1: (Colour online) A breather mode at Ω = 1.5 THz. The compressive perturbation was applied at 2.0 ps. The colour scale represents y (A). One expects that the particular form of the perturba- tion is not important for the formation of breathers. This has been confirmed by using a compression of the form δyi(t) = δ(t − t0) Yp θ(1 ≤ i ≤ n). (6) For Yp = −0.42 A, it was found that all perturbations with n > 4 generated breathers (with the curious ex- ception of n = 8). The double-lobe breather was also obtained at Ω = 1.5 THz with this perturbation. At n = 5 (and A = 144 pN, Ω = 1 THz) one requires a compression of greater than 0.3 A to achieve a breather mode. The parameters of Alexandrov et al. are not the same as those of Dauxois et al.; in particular the value of ρ was changed from 0.5 to 2.0[8]. A run with n = 5, Yp = −0.42 A, and the PBD parameters3 reveals that the breather spreads with time, until the entire DNA molecule melts after approximately 140 ps (the same happens with the perturbation of Eq. 5). This is our first indication that breather dynamics are subtle and that model results can depend crucially on parameters. It should be noted that the parameters of Refs. [4, 6] are not universally employed. For example, Barbi et al. have developed a similar model that couples base pair bond extension to helical twist[10]. They take D = 0.15 eV, a = 6.3 1/A, β = 0.5 1/A, and kρ = 0.65 eV/A2. 3 m = 300 amu, D = 0.04 eV, a = 4.4.5 1/A, k = 0.02 eV/A, β = 0.35 1/A, and ρ = 0.5. We implement this by assuming ρ = 2.0 and setting k = 0.325 eV/A2. Notice that these parameters lead to a considerably stiffer collection of nonlinear oscillators. In- deed, running with the previous drive parameters (A, Ω, Yp, n) reveals that the breather damps out within tens of ps. This remains true for all drive parameters that were tested. The results make it clear that the relatively large stacking interaction disperses the putative breather (see Fig. 2). FIG. 2: (Colour online) System response to a compressive perturbation at 2.0 ps with Barbi parameters. Ω = 3.1 THz. The colour scale represents y (A). Finally, the effects of allowing two base pair types are examined. We model this by setting D = 0.05 eV for AT pairs and D = 0.075 eV for GC pairs[11]; all other parameters are left at their AGBUR values. We find that alternating base pairs or a random configuration of base pairs destabilises the putative breather after approx- imately 20 ps, again illustrating the fragility the breather mode with respect to parameter variation. (cid:90) ∞ ∆U p(∆U ) = ρM B(E)dE ∼ 2√ π √ xe−x 3 (7) where ρM B is the Maxwell-Boltzmann energy distribu- tion and x = ∆U/(kBT ). The asymptotic form of the error function has been used to obtain this result. A 0.42 A compression at room temperature yields x = 46.8 and a probability p ≈ 10−20. We now assume that n pair bonds must be compressed, 108 base pairs per dsDNA, 102 dsDNA per cell, 1011 skin cells4, and a solute collision rate of 1.0/ps, to obtain the estimate P = 10−20(n−2) (8) where P is the probability of obtaining one breather fluc- tuation in n base pairs per person per year. We previ- ously established that n > 4 and hence conclude that such an occurrence is essentially impossible. Motivated by this result and the observation that the fluctuation-dissipation theorem links friction with ther- mal noise, we have explored the properties of the model with the addition of Langevin thermal forcing. Thus the term ηi(t) (9) has been added to the right hand side of Eq. 1. We assume memory-free noise. The fluctuation-dissipation theorem then implies (cid:104)ηi(t)ηj(t(cid:48))(cid:105) = 2miδijγkBT δ(t − t(cid:48)). (10) The molecular dynamics algorithm steps in temporal units ∆t and therefore we employ the average noise over a time interval (tn, tn + ∆t): ηi(t)dt. (11) (cid:90) tn+∆t ¯ηi = 1 ∆t tn III. THERMAL EFFECTS Hence To this point, the genesis of breather modes has re- lied on the imposition of a perturbative shock (Eqs. 5, 6) that insert substantial energy into the system. How reasonable are these shocks? Presumably they must be generated by noise within the cell nucleus. This can be due to a variety of biological processes such as cell mem- brane activity or by simple thermal fluctuations. Here we focus on thermal noise and ask the question: how likely is it to perturb a system by δyi? Restricting attention to a single base pair, a compression of 0.4 A corresponds to an insertion of ∆U = 1.2 eV of potential energy to the system. The probability of such a fluctua- tion is (cid:104)¯η2 i (cid:105) = 2miγkBT ∆t . (12) Thus average noise forces are chosen from a Gaussian distribution 1(cid:112)2π(cid:104)¯η2(cid:105) · e−¯η2/(2(cid:104)¯η2(cid:105)). ρth(¯η) = (13) 4 THz radiation is heavily attenuated and only penetrates 1-2 mm into the body[13]. To test the effect of thermal fluctuations we revert to the homogeneous system with AGBUR parameters and driving forces and eliminate the perturbative shocks of the previous section. System response was computed at a variety of temperatures. At low temperature one sees a nonlinear mode with period of approximately 8 ps that is created by the coupled damped driven oscilla- tors. Increasing temperature leads to rapidly increasing bond length excursions. This can be quantified by plot- ting the distribution of y versus temperature, as in Fig. 3. One observes that the distribution is largely invariant for kBT <∼ 0.003 eV, kBT = 0.004 is a transition tem- perature, and temperatures greater than 0.005 eV seem to yield an invariant distribution for y >∼ 1/2 A. Thus it appears the dsDNA with AGBUR driving melts at ap- proximately 0.004 eV (46K). If one were to take this result seriously it would im- ply complete chromosomal denaturation in all skin cells in the presence of THz radiation. But the previous sec- tion warns of large parameter sensitivity in this system and one must not arrive at conclusions too hastily. In fact using a larger solute relaxation time γ = 2.0/ps or increasing the Morse potential strength to D = 0.12 eV stabilises the system at room temperature. It should be noted that there is a subtlety concern- ing the thermal properties of the PBD model. The form of the interaction implies that the equilibrium configu- ration is two widely separated strands. This is reflected in the partition function, which necessarily diverges. Of course this situation is never realised experimentally be- cause the system is embedded in a complex environment of approximately µm extent. In our case, the issue does not arise because the molecular dynamics simulation is microcanonical and because it only need be run for hun- dreds of picoseconds, too short a time scale to probe the asymptotic dynamics of the system. More details con- cerning the thermal properties of the PBD model can be found in Refs. [8, 12]. I simply add the observation that Barbi parameters yields tiny fluctuations; in fact the mean bond extension from equilibrium is (cid:104)y(cid:105) = 0.03 A at 350K, which is inconsistent with the experimentally observed melting transition. IV. REALISTIC DRIVING The AGBUR model of driven dsDNA leads to unrea- sonable results in the presence of thermal fluctuations. We have already seen how parameter variation can al- leviate this problem. But another explanation is pos- sible, namely the physical assumptions underlying the model could be inaccurate. We follow this by idea by fo- cussing on the drive term of Eq. 4. Indeed, this term is of immediate concern, since the authors of Ref. [4] state, "One complication is that the specific physical nature of the interactions between DNA and the THz electromag- netic field is not known in detail. ... We will here sim- ply augment the PBD [model] to include a drive in the 4 FIG. 3: Distribution curves for y for various kBT with AG- BUR driving. THz frequency range." In fact, I find that the AGBUR breather (at zero temperature, with the AGBUR pertur- bative shock) requires A >∼ 140 pN to occur. What is a reasonable physical value for the driving force? Legal limits on THz radiation power densities range from 5-10 mW/cm2[13]. The mean magnitude of the Poynting vector (cid:104)S(cid:105) = E2 m/(2µ0c) relates5 this to the maximum electric field strength, Em. Employing the up- per limit gives Em <∼ 30V/m. (14) Assuming that a nucleotide is singly charged then yields a maximum force A <∼ 4 · 10−18N, (15) far smaller than that assumed by AGBUR. In fact the driving force is coupled to the base pair displacement and therefore depends on the nonuniformity of the field over the range (cid:104)y(cid:105). Assuming an incident plane wave (with wavevector (cid:126)k) reduces the strength of the drive force by a factor of (cid:104)(cid:126)k · (cid:126)y(cid:105) ∼ 10−6 hence A <∼ 4 · 10−24N. (16) But this assumes that the pair bonds all lie in an op- timal direction (the force must be along y but is propor- tional to (cid:126)k · (cid:126)y, thus the pair bond must lie in a plane defined by the wave propagation direction and the direc- tion of electric field oscillation). In reality, DNA is em- bedded in a heavily hierarchical structure, ranging from 5 The permeabilities of air and water are essentially equal to that of the vacuum. 0 20000 40000 60000 80000 100000 120000 140000(cid:233)1 0 1 2 3 4 5 6arb. units<y> (Å)kT=0.001kT=0.002kT=0.003kT=0.004kT=0.005kT=0.006kT=0.007the DNA molecule itself to the chromosome. This effec- tively randomises the pair bond direction with respect to any external field. Thus one should compute A after av- eraging over bond directions. The result is proportional kj . . . k(cid:96) which is zero to contractions of tensors like Ei to all orders. Thus the driving force relies on remnant order in chromosomal structure and A must be much smaller than 10−24 N. Finally, approximately one half of the incident radiation is reflected[14], cell membranes and cytoplasm are extremely efficient at screening elec- tric fields, even in the THz regime, and the electric charge may be mobile[15]. All of these effects reduce the cou- pling further. One must conclude that the electric field driving force is many orders of magnitude smaller than that required to generate breather modes. V. DISCUSSION AND CONCLUSIONS The PBD model of base pair dynamics is sufficiently rich that interesting collective behaviour can exhibited. Under assumptions concerning drag and drive forcing, breather modes can be generated at certain resonant fre- quencies. Thus, although this work disagrees on the de- tails, it agrees with the main conclusions of Ref. [4]. The stability of breathers under parameter variation has been addressed here. We have seen that changing ρ from 2.0 to 0.5 or including thermal noise are sufficient to dissociate dsDNA under AGBUR driving. Alternatively, employing the stiffer Barbi parameters or allowing for a mixture of AT and GC base pairs seems to disallow breather forma- 5 tion. All of these conclusions are based on drive frequencies near the resonant frequency of the system, a drag term, and a driving term with a magnitude of approximately 100 pN. However, it has been argued that the magni- tude of the driving term is much smaller than this. The physical reason is that the source power is rather weak, and DNA is heavily screened from external influences by the cell membrane, the cytoplasm, and the nucleoplasm. The coupling to electric fields is further reduced by the ef- fectively random orientation of a base pair displacement vector. The field strength necessary (estimated gener- ously) to generate breather modes is approximately 109 V/m, which is much greater than the dielectric break- down threshold of air (∼ 106 V/m). Thus it appears that the analysis of Refs. [4, 7] is not relevant to physically realisable situations. Although strong THz radiation is artificial, DNA has evolved in a noisy electrical and thermal environment, and it might be expected that the molecule and the pro- cesses in which it takes part will be stable with respect to external nonionising radiation. Similarly, one would expect that all molecular level biological processes are immune to low intensity nonionising radiation; although, of course, this speculation needs to be confirmed with rigorous experiment. Acknowledgments The author is grateful for discussions with D. Boy- anovsky, L. Chong, R. Coalson, and D. Jasnow. [1] For an early example of concerns over mmw radiation see mussen, arXiv:1012.2565. H. Frohlich, Int. J. Quantum Chem. 2, 64 (1968). [8] A. Campa and A. Giansanti, Phys. Rev. E 58, 3585 [2] For a reviews see A.H. Aparola and A. Merkle, pg 177, in "Biological Effects of Electric and Magnetic Fields", eds. D.O. Carpenter and S.Ayrapetyan (Academic Press, New York, 1994); S.M. Michaelson, pg. 339, in "CRC Hand- book of Biological Effects of Electromagnetic Fields", eds, C. Polk and E. Postow (CRC Press, Boca Raton, 1986). [3] R.K. Adair, Phys. Rev. A 43, 1039 (1991). [4] B.S. Alexandrov, V. Gelev, A.R. Bishop, A. Usheva, and K.O. Rasmussen, Phys. Lett. 374A, 1214 (2010). [5] E.W. Prohofsky and J.M. Eyster, Phys. Lett. 50A, 329 (1974). [6] T. Dauxois, M. Peyrard, and A.R. Bishop, Phys. Rev. E 47, R44 (1993). [7] P. Maniadis, B.S. Alexandrov, A.R. Bishop, K.O. Ras- (1998). [9] M. Peyrard and A.R. Bishop, Phys. Rev. Lett. 62, 2755 (1989). [10] M. Barbi, S. Lepri, M. Peyrard, and N. Theodorakopou- los, Phys. Rev. E 68, 061909 (2003). [11] N.L. Komarova and A. Soffer, Bull. Math. Bio. 67, 701 (2005). [12] T.S. van Erp, S. Cuesta-L´opez, and M. Peyrard, Eur. Phys. J. E 20, 421 (2006). [13] K.L. Ryan, J. A. D'Andrea, J.R. Jauchem, and P.A. Ma- son, Health Physics 78, 170 (2000). [14] See the second of Ref. [2], pg. 275. [15] D. Hall et al., Nature 382, 73 (1996); X. Guo et al., Nature Nano. 3, 163 (2008).
1507.00716	1	1507	2015-07-02T19:52:34	Superdiffusion dominates intracellular particle motion in the supercrowded space of pathogenic Acanthamoeba castellanii	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	Acanthamoebae are free-living protists and human pathogens, whose cellular functions and pathogenicity strongly depend on the transport of intracellular vesicles and granules through the cytosol. Using high-speed live cell imaging in combination with single-particle tracking analysis, we show here that the motion of endogenous intracellular particles in the size range from a few hundred nanometers to several micrometers in Acanthamoeba castellanii is strongly superdiffusive and influenced by cell locomotion, cytoskeletal elements, and myosin II. We demonstrate that cell locomotion significantly contributes to intracellular particle motion, but is clearly not the only origin of superdiffusivity. By analyzing the contribution of microtubules, actin, and myosin II motors we show that myosin II is a major driving force of intracellular motion in A. castellanii. The cytoplasm of A. castellanii is supercrowded with intracellular vesicles and granules, such that significant intracellular motion can only be achieved by actively driven motion, while purely thermally driven diffusion is negligible.	physics.bio-ph	physics	Superdiffusion dominates intracellular particle motion in the supercrowded cytoplasm of pathogenic Acanthamoeba castellanii Julia F. Reverey1, Jae-Hyung Jeon2, Han Bao1, Matthias Leippe3, Ralf Metzler4,5, Christine Selhuber-Unkel1* 1 Institute for Materials Science, Biocompatible Nanomaterials, Christian-Albrechts- Universität zu Kiel, Kaiserstr. 2, D-24143 Kiel, Germany. 2 School of Physics, Korea Institute for Advanced Study, Seoul 130-722, Republic of Korea. 3 Zoological Institute, Comparative Immunobiology, Christian-Albrechts-Universität zu Kiel, Olshausenstr.40, D-24098 Kiel, Germany. 4 Institute of Physics & Astronomy, University of Potsdam, D-14776 Potsdam-Golm, Germany. 5 Department of Physics, Tampere University of Technology, FI-30101 Tampere, Finland. * Corresponding author; email address: [email protected] Abstract Acanthamoebae are free-living protists and human pathogens, whose cellular functions and pathogenicity strongly depend on the transport of intracellular vesicles and granules through the cytosol. Using high-speed live cell imaging in combination with single-particle tracking analysis, we show here that the motion of endogenous intracellular particles in the size range from a few hundred nanometers to several micrometers in Acanthamoeba castellanii is strongly superdiffusive and influenced by cell locomotion, cytoskeletal elements, and myosin II. We demonstrate that cell locomotion significantly contributes to intracellular particle motion, but is clearly not the only origin of superdiffusivity. By analyzing the contribution of microtubules, actin, and myosin II motors we show that myosin II is a major driving force of intracellular motion in A. castellanii. The cytoplasm of A. castellanii is supercrowded with intracellular vesicles and granules, such that significant intracellular motion can only be achieved by actively driven motion, while purely thermally driven diffusion is negligible. 1 Introduction Intracellular motion is an essential process for a multitude of vital functions, such as cell motility, cell division, and phagocytosis. For the active transport of cargo inside a cell, the interplay of cytoskeletal filaments and molecular motors plays a key role. Examples include myosin motors on actin filaments and kinesin motors on microtubules1. An additional role has been attributed to Brownian motion and subdiffusion2,3. Both are frequently observed in intracellular motion of single molecules4,5 and of endogenous and endocytosed particles6-8, but can also be found for particles moving in in vitro biopolymer networks9-11, which are viscoelastic at intermediate timescales. The type of diffusion present in the intracellular space is strongly determined by the intracellular architecture of the cell, mostly by the arrangement and density of the cellular cytoskeleton and by the action of molecular motors. Theoretically, the existence of subdiffusion has been explained by the presence of obstacles that hinder particles from carrying out normal diffusion12-16. With increasing obstacle density, the diffusion exponent decreases significantly17. This effect has also been observed in living cells: during the cell cycle, cytoskeletal elements are rearranged, polymerized, and depolymerized. Such structural changes have enormous effects on the diffusive behavior of macromolecules18 and endogenous granules19. Due to its relatively simple, well-understood architecture and its non-motile behavior, fission yeast S. pombe is an excellent model system to study different theoretical aspects and concepts of subdiffusion20 in living cells, e.g. weak ergodicity breaking7 and the mean maximal excursion method3. In contrast to intracellular motion in non-motile S. pombe cells, the situation is very different in motile cells. There, intracellular particle motion is always superimposed by the locomotion of the cell body. In spite of such large-scale movements, subdiffusion has been reported to still be present inside motile cells, e.g. for endothelial cells21, for the social amoeba Dictyostelium discoideum22, and for keratinocytes23. Here, we focus on intracellular motion in Acanthamoeba castellanii, a free-living amoeba that is often found in soil and water reservoirs. Acanthamoebae are of considerable medical relevance24,25, as some Acanthamoeba species are highly pathogenic. These amoebae are the causative agents of granulomatous amebic encephalitis and amebic keratitis26, which are difficult to cure. Pathogenic amoebae destroy host tissues and kill host cells in a contact- dependent reaction, in which the release of cytolytic factors, such as pore-forming toxins or metalloproteinases, is involved27-30. For the highly pathogenic Acanthamoeba culbertsoni, a unique pore-forming protein, termed acanthaporin, has recently been comprehensively characterized31. Prior to host cell destruction, acanthamoebae form close contact with host 2 cells. Subsequently, endogenous granules move to the contact site where they presumably release cytolytic factors into the extracellular space. Intracellular motion is therefore a crucial prerequisite for acanthamoeba pathogenicity. Understanding the basic mechanisms involved in intracellular motion and transport inside acanthamoebae is hence essential to get a complete picture of the pathogenicity of these protozoan parasites. This investigation is particularly interesting, as the intracellular space in acanthamoebae appears densely packed with different types of endogenous particles, such as digestive and contractile vacuoles, smaller vesicles, and granules (Fig. 1). The largest vacuole in acanthamoebae is the contractile vacuole, which is responsible for osmotic regulation. Digestive vacuoles contain precipitates and amorphous substrates32. Lysosomes store various hydrolytic enzymes and are responsible for the digestion of endocytosed food and other foreign particles inside the phagolysosome, e.g. bacteria, fungi, and viruses33. Fig. 1: Schematic picture of the intracellular architecture of an acanthamoeba, showing that many different types of vacuoles and granules densely fill the cytoplasm. Actin, microtubules, and myosin motors are also ubiquitous in acanthamoebae. Microtubules originate from the so-called centrosome near the nucleus and spread out in a dense 3D network throughout the cell body34. Baumann and Murphy showed that mitochondria co- localize with the microtubules in A. castellanii and that the movement of mitochondria and small particles (< 1µm) is inhibited by colchicine, a microtubule depolymerizing substance35. Actin plays a key role in the contraction of the contractile vacuole, as myosin IC is located around the vacuole prior to contraction36-39. Myosin I is mainly present near the rear edge of the acanthamoebae and in the filopodia, which are here typically called acanthopodia39-42, whereas myosin II is present within the whole cytoplasm and is concentrated in the cell cortex37. The state of macromolecular crowding in biological cells with proteins and 3 cytoskeletal structures was previously referred to as `superdense6. Here we use the notion of `supercrowded´ volume to point out that in the A. castellani cells considered here we additionally find a large amount of vacuoles of several microns in size. Here, we focus on investigating the intracellular motion of endogenous particles, such as vesicles and granules, in A. castellanii under different experimental conditions. We emphasize the relation between intracellular motion, acanthamoeba locomotion and the contribution of cytoskeletal elements. Intracellular motion and acanthamoeba locomotion are investigated by high-speed live cell imaging and theoretical analysis concepts, including mean-square displacement (MSD) and velocity autocorrelation function analyses. Our results show a predominance of superdiffusion that cannot be explained by the a priori statement that intracellular particles are swept along with the locomoting acanthamoeba. We observe a striking involvement of myosin II motors, which turns out to be essential for maintaining motion in the supercrowded intracellular volume of A. castellanii. Results Following the tracks of individual particles in the intracellular space reveals valuable information about the physical properties of their intracellular motion. Fig. 2A shows a representative phase contrast image of an A. castellanii trophozoite. The white halo surrounding the cell body arises from the ellipsoidal shape of the acanthamoeba, which only slightly flattens during attachment. 2D tracks of individual intracellular particles from the acanthamoeba in Fig. 2A are plotted in Fig. 2B. The size of the tracked particles ranged from a few hundred nanometers to several micrometers. For almost all tracked particles we observe a consistent directionality in motion. This directionality appears to be defined by the direction of movement of the acanthamoeba, so that intracellular particles are swept along with the cell body. Still, we find a few exceptions from this strongly directional motion, showing that not only cell drift causes intracellular particles to move, but that additional mechanisms for individual particle motion must exist, e.g. by the contribution of molecular motors. Analyzing the TA MSDs (see Methods section) of the particle tracks revealed that the diffusion exponent 𝛼 is close to 2 for almost all particle tracks (Fig. 3A, Table 1). Some single particles show normal diffusive behavior, subdiffusive motion is negligible. 4 In order to investigate the influence of acanthamoeba locomotion on the trajectories of individual particles, we assume that particle motion in a moving acanthamoeba is the sum of the drift due to acanthamoeba locomotion and its additionally induced intracellular motion 𝑟!𝑡 =(𝑥!𝑡,𝑦!𝑡), such that 𝑟𝑡 =𝑟!(𝑡)+𝑟!𝑡 . Here, acanthamoeba locomotion is characterized by the position of its centroid 𝑟!𝑡 , which is determined from the outline of the acanthamoeba after image segmentation (Suppl. Fig. S7). 𝑟𝑡 are the raw particle trajectories and 𝑟!(𝑡) represents the particle motion relative to the centroid of the acanthamoeba. The trajectories 𝑟!(𝑡) have a very different appearance (Fig. 2C) with a much less pronounced directionality than the raw trajectories 𝑟𝑡 shown in Fig. 2B. Still, our TA MSD analysis reveals that intracellular particle motion relative to the centroid of the acanthamoeba is superdiffusive (Fig. 3B). This effect can also be demonstrated by plotting <r2(Δ)>t/Δ, <r2(Δ)>t/Δ2 (Suppl. Fig. S8) and stacked TA MSDs (Suppl. Fig. S9). 5 Fig. 2: Intracellular particle motion in an acanthamoeba and its relation to locomotion. (A) Phase contrast image of an A. castellanii trophozoite crawling on a surface. Scalebar: 10 µm. The arrow indicates the direction of acanthamoeba locomotion. (B) Particle trajectories tracked inside the acanthamoeba shown in A. Most tracks follow the direction of movement of the acanthamoeba centroid (bold red line). However, not all particles obey this trend. (C) Particle trajectories xI(t), yI(t) of intracellular particles relative to the position of the acanthamoeba centroid. No significant directionality of particle motion can be observed. 6 Colors shown here are consistent with the colors used for the tracks in Fig. 3A. Fig. 3: Representative TA MSD plots of intracellular particle trajectories in A. castellani. (A) TA MSD of intracellular particles. Most of the particles show superdiffusive motion with a diffusion exponent close to α ~ 2. (B) TA MSD of particle trajectories relative to the centroid of the acanthamoeba. individual particle trajectories 𝑟𝑡 we evaluated particle motion is superdiffusive, the velocity autocorrelation is always positive at all times and its relaxation occurs above the saturation value. Note that the velocity autocorrelation From their corresponding velocity autocorrelation functions, as described in the Methods section and in the Supplementary Information. Fig. 4A depicts the theoretical curves of the drift-FBM model defined below for three different cases: (i) if intracellular motion is subdiffusive, the velocity autocorrelation has a minimum after the time interval 𝛿𝑡 and then relaxes towards its saturation value; (ii) if intracellular motion is normally diffusive, essentially there is no relaxation after 𝛿𝑡; (iii) if eventually saturates to a constant 𝑉!!/𝑉!!+4𝐾!𝛿𝑡!!! so that the saturation value is always positive irrespective of the drift direction, and for 𝛿𝑡≠1 it has a different value if α is varied FBM model, the long-time relaxation towards its saturation decays as ~𝑡!!!. This means albeit the drift velocity is the same. Such a constant drift velocity is in good agreement with the experimentally determined motion of the acanthamoeba centroid (Suppl. Fig. S7C). In the that the relaxation of superdiffusive motion is always slower than that of subdiffusion43,44. 7 Fig. 4: Velocity autocorrelation analysis of particle trajectories. (A) Theoretical velocity autocorrelation functions (VACF) for FBM-like intracellular motion assuming a constant drift of the amoeba for α=0.5, 1, 1.5, and 1.8. Parameters: 𝛿𝑡=1,𝐾!=1,𝑉!=1. (B) 2D plot (x(t),y(t)) of arbitrarily chosen, long intracellular particle trajectories. (C) VACFs of the particle trajectories shown in B. These profiles are representative VACFs of intact acanthamoebae. Color codes in B and C are the same. Fig. 4B shows raw particle trajectories used in the VACF analysis. Fig. 4C depicts the corresponding VACF curves. Here, we observe the following: (i) The obtained pattern of VACF curves is similar to the case of superdiffusive FBM considered in Fig 4A. The curves continuously decay from unity with time, without a dip, towards a positive constant. Consequently, the intracellular motion of particles in the acanthamoeba studied is superdiffusive, presumably due to the activity of molecular motors and strong intracellular fluid flow (see supplementary movie). This directional persistency in intracellular diffusion has a characteristic time of about a second. (ii) The fact that the VACF saturates towards a constant value tells us that the acanthamoeba indeed moves with an almost constant drift as in the assumption of our model. Such a constant movement appears to be in the time window of a few seconds. (iii) There is a trajectory-to-trajectory fluctuation in the saturation value of the VACF. Although a constant drift motion was observed, such a fluctuation is expected because the saturation value of VACF depends on the diffusion exponent α and the diffusivity of the particles, as well as on the drift speed. In order to investigate the role of myosin II for intracellular particle motion, we treated acanthamoebae with different concentrations of blebbistatin, a myosin II specific inhibitor 45. At concentrations smaller than 50 µM, blebbistatin did not inhibit intracellular motion completely: at concentrations above 50 µM, most acanthamoebae detached from the surface (data not shown). Accordingly, we chose 50 µM for intracellular particle tracking experiments. Fig. 5A shows a phase contrast image of a blebbistatin-treated A. castellanii. Intracellular structures, such as vacuoles, granules, and intracellular organelles are not as clearly visible as in an intact acanthamoeba. Right after the addition of blebbistatin, acanthamoeba locomotion disappeared almost completely, and also intracellular dynamics 8 were drastically reduced (Fig. 5B). Even for these comparably long tracks of more than 20 sec duration, intracellular particle motion is almost negligible. Specifically, particles exhibit subdiffusion where 𝑟!(Δ)!~∆! with 𝛼~0.2 at ∆≲1 sec, and almost normal diffusion at ∆≳1 sec (Fig. 5C). The normally diffusive motion for these confined particles reflects the fact that the intracellular components in acanthamoebae move altogether in a diffusive way at long times. Active intracellular motion was not observed for any of the tracks, revealing a dramatic change of intracellular dynamics due to myosin II inhibition, from superdiffusion in intact acanthamoebae to almost completely suppressed intracellular motion in myosin II inhibited acanthamoebae. Notably, some time after addition of blebbistatin we observed a change in cell morphology (Fig. 5D) and intracellular movement restarted (Fig. 5E). However, particle trajectories look at first glance different from trajectories of intact A. castellanii (Fig. 2B and Fig. 2C), as they are not yet very directional. The TA MSDs show that intracellular motion of many particles is superdiffusive in this situation (Fig. 5F). Fig. 5: Analysis of intracellular particle motion in blebbistatin-treated acanthamoebae. (A) Phase contrast image of A. castellanii treated with blebbistatin for inhibiting myosin II motors. Scalebar: 10 µm (B) Particle trajectories indicate that intracellular motion is almost eliminated. (C) Strongly confined motion is observed in the TA MSDs. (D) Intracellular particle tracks for the same amoeba 1 h after the addition of blebbistatin. Intracellular motion (E) and superdiffusion (F) can be observed. 9 In Fig. 6A we plot representative VACF curves from single particle trajectories of a blebbistatin-treated acanthamoeba. It shows that there is almost no drift in the motion (as the relaxation decays to almost zero) and there is a dip (negative correlation) after Δ=0.1 sec. Such a VACF curve is typical for subdiffusive particle motion without drift. The experimental VACF curve is very similar to the theoretical curve of FBM with α=0.2 (black solid line). As demonstrated in theoretical and experimental studies10,13,46-48 viscoelastic fluids induce strong anti-correlation of particle displacements, leading to FBM-like anomalous diffusion15,49. Thus, the negative value in the VACF curve shown here suggests that the highly restricted passive motion of particles is connected to the viscoelastic nature of the supercrowded intracellular fluid. Fig. 6: Velocity autocorrelation function for intracellular particles in a blebbistatin-treated acanthamoeba (A) and after 1 h of recovery from inhibitor treatment (B). (A) Blue and red lines correspond to VACFs of particle trajectories. The black line denotes the theoretical VACF curve for FBM with α=0.2. (B) Red, blue and grey lines show different types of VACFs received from the trajectory analysis. The black line denotes the average of 37 analyzed trajectories. A clear difference is observed when investigating the velocity autocorrelation function for particle trajectories in acanthamoeba after 1 h of recovery. In Fig. 6B we plot three such motion. This result is consistent with the trajectories (Fig. 5E) and the TA MSDs (Fig. 5F). representative VACFs that clearly show that the particles carry out various types of diffusive The average VACF shows superdiffusive motion plus drift, in agreement with our observation that the acanthamoeba is motile again. Due to the strong effect of myosin II on intracellular motion, actin was our next target of investigation. We used latrunculin A to inhibit the polymerization of the actin filaments and a concentration of 5 µM latrunculin A turned out to be optimal, as higher concentrations often induced an immediate detachment of acanthamoebae. At lower concentrations acanthamoebae 10 still moved on the surface and no pronounced effect of latrunculin A was observed. At concentrations higher than 5 µM, most acanthamoebae detached from the surfaces. A few minutes after latrunculin A addition we typically observe that cell locomotion was slowed down. The shape of acanthamoebae became relatively smooth with much fewer protrusions compared to intact cells (Fig. 7A). Within acanthamoebae, particles still moved and many small particles (size < 500 nm) could be observed close to the membrane, where normally large particles, presumably vacuoles, are crowded together and make the optical identification of tiny particles impossible. However, many acanthamoebae detached during the experiments, presumably because actin is a key player for their adhesion. As shown in Fig. 7B, latrunculin A does not stop intracellular motion and intracellular flow, even though locomotion of the acanthamoeba itself is stopped, and there is no distinct directionality in the particle trajectories. In particular, the TA MSD analysis proves that the particles mainly carry out superdiffusive motion with a diffusion exponent of about 𝛼~1.6 (Fig. 7C). Velocity autocorrelations for selected trajectories (Fig. 7D) also show positive correlations, in agreement with the superdiffusive TA MSD and the observation of intracellular flow. 11 Fig. 7: Intracellular particle motion in an acanthamoeba treated with 5 µM latrunculin A. (A) Phase contrast image. Scalebar: 10 µm. (B) 2D intracellular particle trajectories of the acanthamoeba shown in A. The tracks do not show strong directionality. (C) TA MSDs reveal superdiffusion. (D) Velocity autocorrelation function for selected representative trajectories. The bold black line denotes the average of 13 analyzed trajectories. 40 µM nocodazole treated with Microtubules are suggested to be key players for intracellular particle motion, also in amoebae50. Acanthamoebae for microtubule depolymerization have a comparably rough shape (Fig. 8A) and their locomotion is drastically reduced compared to intact acanthamoebae. At lower concentrations, such a pronounced effect of nocodazole was not observed. Within acanthamoebae there were many vacuoles, but only a few tiny particles were visible. In contrast to latrunculin A treatment, the adhesion of nocodazole-treated acanthamoebae to the surface was not impaired. The motion of acanthamoebae was drastically affected by nocodazole, but not completely inhibited. Intracellular motion was observed (Fig. 8B), but no clear directionality was found for particle motion. Comparable to latrunculin A, the TA MSDs here reveal diffusion exponents of about α ~ 1.7 (Fig. 8C). The difference between latrunculin A and nocodazole treatments becomes particularly visible in the velocity autocorrelation functions. Under the influence of nocodazole, the velocity autocorrelation shows a constant drift motion. This can be a result of the remaining slow drift motion of the acanthamoeba during the experiment (Fig. 8D). Fig. 8: Influence of nocodazole on intracellular motion in acanthamoebae. (A) Particle tracks in an acanthamoeba treated with 40 µM nocodazole. Scalebar: 10 µm (B) 2D intracellular particle trajectories of the acanthamoeba shown in A. In contrast to the intact acanthamoeba, the tracks do not show strong directionality. (C) TA MSD analysis reveals superdiffusion. (D) VACF of the red, blue, and grey tracks shown in B and C. 12 In order to clearly visualize the effect of different drugs, we summarized the distributions obtained for the diffusion exponent α in the different drug treatment situations (Fig. 9). Whereas a striking effect is obtained between blebbistatin and all other situations, the difference between latrunculin A and nocodazole treatment is negligible and the values of α are for both treatments only slightly smaller than in the case of intact acanthamoebae. In table 1 averaged values of α are shown and information on the size of the datasets is given. Fig. 9: Influence of different treatments on the diffusion exponent α. The influence of the different treatments is clearly visible, particularly the extremely strong impact of blebbistatin, giving rise to the pronounced subdiffusion with 0<α<0.5. In all other cases shown here we observe superdiffusion with α >1. Discussion We have used time-lapse imaging and particle tracking algorithms to investigate the role of locomotion, cytoskeletal elements, and myosin II motors on intracellular particle movement in the protozoon and human pathogen A. castellani by analyzing mean-square displacements 13 and velocity autocorrelations. The experiments were complex because acanthamoebae move extremely fast and exhibit quick shape changes25. For intact acanthamoebae we found that superdiffusion dominates the motion of endogenous intracellular particles. This is very different from previous results on non-motile and shape- conserved S. pombe yeast cells, where subdiffusion dominates intracellular particle motion3,7,8,19. Surprisingly, in A. castellanii even the motion of intracellular particles relative to the acanthamoeba centroid is superdiffusive (Fig. 3B). At first glance this result contradicts recently published work by Kinneret et al., who showed that the motion of intracellular quantum dots relative to the centroid of a keratinocyte is subdiffusive23. However, it has to be considered that acanthamoebae are, in contrast to keratinocytes, densely packed with vacuoles and granules. As already noted, acanthamoebae also frequently change their shape during locomotion, whereas keratinocytes hardly do. Accordingly, one may assume that intracellular motion in acanthamoebae relies on different principles to ensure the transport of intracellular cargo, in spite of their dense packing. Hence, we believe that active superdiffusive motion is a strategy of acanthamoebae to allow particle movement in its supercrowded intracellular volume. This becomes particularly evident in the experiments in which myosin II motors were blocked, as intracellular motion and acanthamoeba locomotion vanished consistently in the TA MSD and in the VACF analysis, leading to a diffusion exponent α~0.2. These results show that the intracellular volume is considerably more supercrowded than the bacterial cytoplasm6. A well-known result of inhibition of actin polymerization is a decreased diffusion exponent α both for non-motile8 and motile cells22,51. We also observed a slight decrease of α in our experiments, with a decrease from in average about 1.8 to about 1.5 and 1.6 in response to latrunculin A and nocodazole treatments, respectively. This also means that intracellular motion stays superdiffusive in spite of inhibiting the polymerization of cytoskeletal elements. Hence, cytoskeletal elements themselves play a minor role for intracellular motion in acanthamoebae, in agreement with the results on the cytoplasm of E. coli bacteria6. Instead, myosin II is an essential driving force to maintain intracellular particle motion. In contrast to intracellular particle tracking in breast cancer cells, where only a minor influence of latrunculin A and blebbistatin treatments on the diffusion exponent was detected52, we here clearly observe an impact of both inhibiting myosin II and actin polymerization. Interestingly, our diffusion exponent in the case of untreated acanthamoebae is larger than the one detected in previous publications on superdiffusive intracellular particle motion52,53. Therefore it is even more striking that motion appears to be almost completely blocked by myosin II 14 inhibition in acanthamoeba. As we studied the motion of particles endogenous to acanthamoeba, our results are intrinsically different from other results on superdiffusive intracellular motion, where the motion of engulfed artificial particles52,53, also under the influence of mechanical disturbances54, was investigated. Our data show that the TA MSD analysis and the VACF analysis are consistent and support our assumption that intracellular particle motion in A. castellanii is dominated by superdiffusion. In particular, the VACFs reveal that intracellular motion follows an FBM-type Gaussian process and is therefore consistent with the drift-FBM model (see Methods). We note that superdiffusion also occurs in vacancy-dominated, confined crowded systems55-57. In conclusion, intracellular particle motion in A. castellanii is a prime example for the importance of superdiffusion in a supercrowded, motile, and spatially dynamic volume. We have shown that superdiffusivity not only arises from the fast migration of acanthamoebae, but also from superdiffusive motion relative to the centroid of an acanthamoeba. Myosin II turned out to play a crucial role for the functionality of intracellular and cellular transport systems: if myosin II was inhibited, acanthamoebae neither migrated nor exhibited intracellular particle motion. Although inhibiting actin polymerization and depolymerizing microtubules changed the overall cell shape and could almost stop acanthamoeba migration, intracellular particle motion was not completely eliminated. Inhibiting myosin II caused strongly confined intracellular motion, so that we consider myosin II motors to be key players for maintaining such motion. Brownian motion and subdiffusion appear to play negligible roles for intracellular transport in acanthamoebae because the extreme crowding would limit all transport-related biological functions. With regard to amoebic pathogenicity, the dominant role of active particle motion agrees well with the notion that it relies – at least in part – on the fast and targeted intracellular transport of toxic factors stored in lysosome-like cytolytic granules through a supercrowded intracellular volume. Our study may serve as a starting point for further investigations on the interplay between crowding, diffusion, and active transport in cells in general and their role for the pathogenicity of amoebic parasites. 15 Methods Culture of acanthamoebae A. castellanii (ATTC 30234) were cultured in PYG medium at room temperature in 75 ml tissue culture bottles (Sarstedt, Germany) as described previously58,59. Prior to experiments, about 105 freshly harvested acanthamoebae (A. castellanii) per ml medium were seeded on a flat surface (tissue culture bottle or six-well plate, Sarstedt AG & Co., Nümbrecht, Germany) and incubated at room temperature for 30 min to 1 h to ensure their attachment to the tissue culture surface. Time-lapse imaging Acanthamoeba locomotion and intracellular motion were observed with an inverted phase contrast microscope (Olympus CKX41 and IX81, Olympus Deutschland GmbH, Hamburg, Germany) with either a LCACHN 40X PHP, a LUCPLANFLN 40X PH2 or a UPLFLN 60X (long distance) objective (all from Olympus). Sequences having a length of at least 30 sec were recorded with a high-speed camera (Hamamatsu C9300, Hamamatsu, Japan) at frame rates between 80 and 99 frames per second (fps). Exposure times chosen for image recording are not expected to cause errors in the analysis of particle trajectories60. Inhibition of actin polymerization Latrunculin A inhibits the polymerization of actin filaments as it binds to the subunits of actin filaments and avoids further reaction of the subunits to longer filaments61. For our experiments, Latrunculin A (Sigma-Aldrich Chemie GmbH, Munich, Germany) was dissolved in DMSO (Carl Roth GmbH & Co. KG, Karlsruhe, Germany) at a 1 mM stock solution concentration. Acanthamoebae were cultured as described above and latrunculin A was added to reach final concentrations of 2 µM, 5 µM, 7.5 µM or 10 µM. High-speed sequences were recorded a few minutes after latrunculin A addition and again after 1-2 h to ensure acanthamoeba survival. Inhibition of microtubule polymerization Nocodazole was used to inhibit microtubule polymerization. It binds to the subunits of the microtubules and thus prevents polymerization62. Stock solutions of nocodazole (Sigma- 16 Aldrich Chemie GmbH, Munich, Germany) in DMSO (Carl Roth GmbH & Co. KG, Karlsruhe, Germany) were prepared at concentrations of 1 mM. Acanthamoebae were cultured as described above and nocodazole was added to reach final concentrations of 10 µM, 20 µM and 40 µM. Live-cell imaging was carried out as described before and was started a few minutes after addition of nocodazole. Acanthamoebae viability was verified 1- 2 h after the experiment. Inhibition of myosin II motors For inhibiting myosin II motors, (-)-blebbistatin was used. The reagent binds to ATPase, which is a catalyzer for ATP hydrolysis and hence blocks the myosin heads in their unbound state as it slows down ATP hydrolysis, especially the release of phosphate45. (-)-blebbistatin (Sigma-Aldrich, Germany) was dissolved in DMSO (Roth, Germany) at 1 mM concentration. (-)-blebbistatin was added to the medium of the cultured acanthamoebae to reach different concentrations (50 µM, 60 µM, and 70 µM). Right after the addition of (-)-blebbistatin, time- lapse imaging was started. Here, we employed for the experiments a phase contrast microscope (Olympus IX81, Olympus Deutschland GmbH, Hamburg, Germany) with a 40× (LCACHN 40X PHP) objective. After a period of 15 min to 1 h the effect of (-)-blebbistatin decreased and time-lapse sequences were recorded again to prove the viability of the cells by monitoring intracellular motion and locomotion. Acanthamoebae had completely normal appearance after 1-2 h of relaxation from the drug treatment. Evaluation of time-lapse image sequences For tracking the position of intracellular particles, an automated image processing algorithm (Polyparticletracker) specially designed for tracking endogenous cellular particles was used in Matlab (The MathWorks, Natick, MA)63,64. Image segmentation and acanthamoeba centroid detection were carried out with a Matlab program of our own based on built-in algorithms for segmenting cells in phase contrast images. Centroid position was fitted by a polynomial function (see Supplementary Information) in order to remove strong fluctuations arising from short-term acanthopodia extensions and retractions. 17 Mean-square displacement analysis !!!! 𝑟!(Δ)!= !!!! 𝑑𝑡𝑟𝑡+Δ −𝑟(𝑡)! The time-averaged mean-square displacement (TA MSD) of the tracked particles was calculated from with 𝑟(𝑡)=(𝑥𝑡,𝑦𝑡). Δ denotes the lag time in seconds and 𝑇 is the overall measurement time. For typical timescales, the TA MSD is characterized by 𝑟!(Δ)!~4𝐾!∆! where α is the diffusion exponent with 0<𝛼≤2 and Kα is the generalized diffusion coefficient of dimension [cm2/secα]. For 𝛼<1 particles carry out subdiffusion, for 𝛼=1 normal diffusion (Brownian motion), and for 𝛼>1 superdiffusion. In order to determine the distribution of α (1) values, we fitted the linear region of the TA MSDs in the double-logarithmic plot between Δ=0.1 s and Δ=1 s using MATLAB. 𝐶!(∆)= !!!!!!" !!!!!" ! Analysis of the velocity autocorrelation function 𝒱!"(𝑡+Δ) ∙𝒱!"(𝑡)𝑑𝑡 (2) The time-averaged velocity autocorrelation function (VACF) was calculated using for a given trajectory 𝑟𝑡 and the time interval 𝛿𝑡. 𝒱!"𝑡 refers to the average velocity at time t over the time interval 𝛿𝑡, which is defined as 𝒱!"𝑡 =(𝑟𝑡+𝛿𝑡 −𝑟𝑡)/𝛿𝑡. All quantity. In our analysis, the time interval 𝛿𝑡 was fixed to 𝛿𝑡=10Δ𝑡!"#$%≈0.12 𝑠. By results for the velocity autocorrelation function presented below are this time-averaged studying the relaxation profile of the velocity autocorrelation we can obtain information on the drift of an acanthamoeba as well as on the stochastic properties of intracellular motion. In particular, we investigated the case where intracellular motion follows a Fractional Brownian motion (FBM)-type Gaussian process that can be obtained for subdiffusive, normal diffusive, or superdiffusive motion43,44,65. As demonstrated theoretical and experimental studies10,13,46,47, FBM-like anomalous diffusion typically occurs in viscoelastic environments, which give rise to long-time correlation in spatial displacements. It has been shown that intracellular diffusion is in many cases governed by FBM, at least on the relevant intermediate time scales, due to the viscoelastic nature of cytoplasmic solutions5,7,20,66-68. Based on this knowledge, we consider here the case that intracellular particles follow FBM and the acanthamoeba hosting the particles moves with a constant drift speed 𝑉!. More details are given in the Supplementary Information. 18 in References 1 Vale, R. D. & Milligan, R. A. The Way Things Move: Looking Under the Hood of Molecular Motor Proteins. Science 288, 88-95, (2000). Metzler, R. & Klafter, J. The Random Walk's Guide to Anomalous Diffusion: A Fractional Dynamics Approach. Phys. Rep. 339, 1-77, (2000). Tejedor, V. et al. Quantitative Analysis of Single Particle Trajectories: Mean Maximal Excursion Method. Biophys. J. 98, 1364-1372, (2010). Barkai, E., Garini, Y. & Metzler, R. Strange kinetics of single molecules in living cells. Phys. Today 65, 29-35, (2012). Burnecki, K. et al. Universal Algorithm for Identification of Fractional Brownian Motion. A Case of Telomere Subdiffusion. Biophys. J. 103, 1839-1847, (2012). Golding, I. & Cox, E. C. Physical Nature of Bacterial Cytoplasm. Phys. Rev. Lett. 96, 098102-098104, (2006). Jeon, J. H. et al. In Vivo Anomalous Diffusion and Weak Ergodicity Breaking of Lipid Granules. Phys. Rev. Lett. 106, 048103, (2011). Tolic-Nørrelykke, I., Munteanu, E.-L., Thon, G., Oddershede, L. & Berg-Sørensen, K. Anomalous Diffusion in Living Yeast Cells. Phys. Rev. Lett. 93, 078102, (2004). Wong, I. et al. Anomalous Diffusion Probes Microstructure Dynamics of Entangled F- Actin Networks. Phys. Rev. Lett. 92, 178101, (2004). Jeon, J. H., Leijnse, N., Oddershede, L. B. & Metzler, R. Anomalous diffusion and power-law relaxation of the time averaged mean squared displacement in worm-like micellar solutions. New J. Phys. 15, 045011, (2013). Pan, W. et al. Viscoelasticity in Homogeneous Protein Solutions. Phys. Rev. Lett. 102, 058101, (2009). Höfling, F., Franosch, T. & Frey, E. Localization Transition of the Three-Dimensional Lorentz Model and Continuum Percolation. Phys. Rev. Lett. 96, 165901, (2006). Godec, A., Bauer, M. & Metzler, R. Collective dynamics effect transient subdiffusion of inert tracers in flexible gel networks. New J. Phys. 16, 092002, (2014). 2 3 4 5 6 7 8 9 10 11 12 13 15 16 17 18 14 Weiss, M., Elsner, M., Kartberg, F. & Nilsson, T. Anomalous Subdiffusion Is a Measure for Cytoplasmic Crowding in Living Cells. Biophys. J. 87, 3518-3524, (2004). Ernst, D., Hellmann, M., Kohler, J. & Weiss, M. Fractional Brownian motion in crowded fluids. Soft Matter 8, 4886-4889, (2012). Höfling, F. & Franosch, T. Anomalous transport in the crowded world of biological cells. Rep. Prog. Phys. 76, 046602, (2013). Banks, D. S. & Fradin, C. Anomalous Diffusion of Proteins Due to Molecular Crowding. Biophys. J. 89, 2960-2971, (2005). Pawar, N., Donth, C. & Weiss, M. Anisotropic Diffusion of Macromolecules in the Contiguous Nucleocytoplasmic Fluid during Eukaryotic Cell Division. Curr. Biol. 24, 1905-1908. Selhuber-Unkel, C., Yde, P., Berg-Sørensen, K. & Oddershede, L. B. Variety in intracellular diffusion during the cell cycle. Phys. Biol. 6, 025015 (2009). 20 Metzler, R., Jeon, J.-H., Cherstvy, A. G. & Barkai, E. Anomalous diffusion models and their properties: non-stationarity, non-ergodicity, and ageing at the centenary of single particle tracking. Phys. Chem. Chem. Phys. 16, 24128-24164, (2014). Leijnse, N., Jeon, J. H., Loft, S., Metzler, R. & Oddershede, L. B. Diffusion inside living human cells. Eur. Phys. J. Spec. Top. 204, 75-84, (2012). Otten, M. et al. Local Motion Analysis Reveals Impact of the Dynamic Cytoskeleton on Intracellular Subdiffusion. Biophys. J. 102, 758-767, (2012). Keren, K., Yam, P. T., Kinkhabwala, A., Mogilner, A. & Theriot, J. A. Intracellular fluid flow in rapidly moving cells. Nat. Cell Biol. 11, 1219-1224, (2009). 19 21 22 23 19 24 Marciano-Cabral, F. & Cabral, G. Acanthamoeba spp. as Agents of Disease in Humans. Clin. Microbiol. Rev. 16, 273-307, (2003). Siddiqui, R. & Khan, N. Biology and pathogenesis of Acanthamoeba. Parasit. Vectors 5, 6, (2012). 26 Marciano-Cabral, F., Puffenbarger, R. & Cabral, G. A. The Increasing Importance of 42 Miyata, H., Bowers, B. & Korn, E. D. Plasma membrane association of Acanthamoeba myosin I. J. Cell Biol. 109, 1519-1528, (1989). Burov, S., Jeon, J.-H., Metzler, R. & Barkai, E. Single particle tracking in systems showing anomalous diffusion: the role of weak ergodicity breaking. Phys. Chem. Chem. Phys. 13, 1800-1812, (2011). Sokolov, I. M. Models of anomalous diffusion in crowded environments. Soft Matter 8, 9043-9052, (2012). 20 31 Michalek, M. et al. Structure and function of a unique pore-forming protein from a Acanthamoeba Infections. J. Eukaryot. Microbiol. 47, 29-36, (2000). Leippe, M. & Herbst, R. Ancient Weapons for Attack and Defense: the Pore-forming Polypeptides of Pathogenic Enteric and Free-living Amoeboid Protozoa. J. Eukaryot. Microbiol. 51, 516-521, (2004). Leher, H., Silvany, R., Alizadeh, H., Huang, J. & Niederkorn, J. Y. Mannose Induces the Release of Cytopathic Factors from Acanthamoeba castellanii. Infect. Immun. 66, 5-10, (1998). Alsam, S. et al. Acanthamoeba interactions with human brain microvascular endothelial cells. Microb. Pathog. 35, 235-241, (2003). Hurt, M., Neelam, S., Niederkorn, J. & Alizadeh, H. Pathogenic Acanthamoeba spp. Secrete a Mannose-Induced Cytolytic Protein That Correlates with the Ability To Cause Disease. Infect. Immun. 71, 6243-6255, (2003). pathogenic acanthamoeba. Nat. Chem. Biol. 9, 37-42, (2013). Bowers, B. & Korn, E. D. The fine structure of Acanthamoeba castellanii: I. The Trophozoite. J. Cell Biol. 39, 95-111, (1968). Khan, N. Acanthamoeba: Biology and Pathogenesis. (Calster Academic Press, 2009). Preston, T. M. A prominent microtubule cytoskeleton in Acanthamoeba. Cell Biol. Int. Rep. 9, 307-314, (1985). Baumann, O. & Murphy, D. B. Microtubule-associated movement of mitochondria and small particles in Acanthamoeba castellanii. Cell Motil. Cytoskeleton 32, 305-317, (1995). Baines, I. C., Corigliano-Murphy, A. & Korn, E. D. Quantification and localization of phosphorylated myosin I isoforms in Acanthamoeba castellanii. J. Cell Biol. 130, 591- 603, (1995). Baines, I. C. & Korn, E. D. Localization of myosin IC and myosin II in Acanthamoeba castellanii by indirect immunofluorescence and immunogold electron microscopy. J. Cell Biol. 111, 1895-1904, (1990). Doberstein, S. K., Baines, I. C., Wiegand, G., Korn, E. D. & Pollard, T. D. Inhibition of contractile vacuole function in vivo by antibodies against myosin-I. Nature 365, 841-843, (1993). Yonemura, S. & Pollard, T. D. The localization of myosin I and myosin II in Acanthamoeba by fluorescence microscopy. J. Cell Sci. 102, 629-642, (1992). Kong, H.-H. & Pollard, T. D. Intracellular localization and dynamics of myosin-II and myosin-IC in live Acanthamoeba by transient transfection of EGFP fusion proteins. J. Cell Sci. 115, 4993-5002, (2002). González-Robles, A. et al. Acanthamoeba castellanii: Identification and distribution of actin cytoskeleton. Exp. Parasitol. 119, 411-417, (2008). 25 27 28 29 30 32 33 34 35 36 37 38 39 40 41 43 44 Kovács, M., Tóth, J., Hetényi, C., Málnási-Csizmadia, A. & Sellers, J. R. Mechanism of Blebbistatin Inhibition of Myosin II. J. Biol. Chem. 279, 35557-35563, (2004). Goychuk, I. Viscoelastic subdiffusion: From anomalous to normal. Phys. Rev. E 80, 046125, (2009). Szymanski, J. & Weiss, M. Elucidating the Origin of Anomalous Diffusion in Crowded Fluids. Phys. Rev. Lett. 103, 038102, (2009). Bronstein, I. et al. Transient Anomalous Diffusion of Telomeres in the Nucleus of Mammalian Cells. Phys. Rev. Lett. 103, 018102, (2009). 49 Weiss, M. Single-particle tracking data reveal anticorrelated fractional Brownian motion in crowded fluids. Phys. Rev. E 88, 010101, (2013). Clarke, M., Köhler, J., Heuser, J. & Gerisch, G. Endosome Fusion and Microtubule- Based Dynamics in the Early Endocytic Pathway of Dictyostelium. Traffic 3, 791-800, (2002). Bruno, L., Levi, V., Brunstein, M. & Despósito, M. A. Transition to superdiffusive behavior in intracellular actin-based transport mediated by molecular motors. Phys. Rev. E 80, 011912, (2009). Goldstein, D., Elhanan, T., Aronovitch, M. & Weihs, D. Origin of active transport in breast-cancer cells. Soft Matter 9, 7167-7173, (2013). Caspi, A., Granek, R. & Elbaum, M. Enhanced Diffusion in Active Intracellular Transport. Phys. Rev. Lett. 85, 5655-5658, (2000). 54 Wilhelm, C. Out-of-Equilibrium Microrheology inside Living Cells. Phys. Rev. Lett. 45 46 47 48 50 51 52 53 55 56 57 58 59 60 101, 028101, (2008). Bénichou, O., Mejía-Monasterio, C. & Oshanin, G. Anomalous field-induced growth of fluctuations in dynamics of a biased intruder moving in a quiescent medium. Phys. Rev. E 87, 020103, (2013). Bénichou, O., Illien, P., Mejía-Monasterio, C. & Oshanin, G. A biased intruder in a dense quiescent medium: looking beyond the force-velocity relation. J. Stat. Mech. 2013, (2013). Bénichou, O. et al. Geometry-Induced Superdiffusion in Driven Crowded Systems. Phys. Rev. Lett. 111, 260601, (2013). Gutekunst, S. B., Grabosch, C., Kovalev, A., Gorb, S. N. & Selhuber-Unkel, C. Influence of the PDMS substrate stiffness on the adhesion of Acanthamoeba castellanii. Beilstein J. Nanotechnol. 5, 1393-1398, (2014). Reverey, J. F., Fromme, R., Leippe, M. & Selhuber-Unkel, C. In vitro adhesion of Acanthamoeba castellanii to soft contact lenses depends on water content and disinfection procedure. Cont. Lens Anterior Eye 37, 262-266, (2014). Savin, T. & Doyle, P. S. Static and Dynamic Errors in Particle Tracking Microrheology. Biophys. J. 88, 623-638, (2005). 62 63 61 Morton, W. M., Ayscough, K. R. & McLaughlin, P. J. Latrunculin alters the actin- monomer subunit interface to prevent polymerization. Nat. Cell Biol. 2, 376-378, (2000). Samson, F., Donoso, J. A., Heller-Bettinger, I., Watson, D. & Himes, R. H. Nocodazole action on tubulin assembly, axonal ultrastructure and fast axoplasmic transport. J. Pharmacol. Exp. Ther. 208, 411-417, (1979). Rogers, S. S., Waigh, T. A., Zhao, X. & Lu, J. R. Precise particle tracking against a complicated background: polynomial fitting with Gaussian weight. Phys. Biol. 4, 220, (2007). http://personalpages.manchester.ac.uk/staff/salman.rogers/polyparticletracker/. 64 65 Mandelbrot, B. B. & Van Ness, J. W. Fractional Brownian Motions, Fractional Noises and Applications. SIAM Review 10, 422-437, (1968). 21 Guigas, G., Kalla, C. & Weiss, M. Probing the Nanoscale Viscoelasticity of Intracellular Fluids in Living Cells. Biophys. J. 93, 316-323, (2007). Taylor, M. A. et al. Biological measurement beyond the quantum limit. Nat. Photon. 7, 229-233, (2013). 68 Weber, S. C., Spakowitz, A. J. & Theriot, J. A. Bacterial Chromosomal Loci Move Subdiffusively through a Viscoelastic Cytoplasm. Phys. Rev. Lett. 104, 238102, (2010). Acknowledgements We thank Heidrun Liessegang for her advice in culturing acanthamoebae, Ali Raza for support with image segmentation, Brook Shurtleff and Steven Huth for proofreading the manuscript. R. M. acknowledges support by the Academy of Finland (Suomen Akatemia) within the Finland Distinguished Professorship program. M.L. was supported by the “Cluster of Excellence Inflammation at Interfaces”, Cluster laboratory X, of the German Research Foundation (DFG). C.S. and J.R. acknowledge support from the German National Academy of Sciences Leopoldina by grant LPD 9901/8-164, as well as from the DFG through the Emmy Noether program (grant SE-1801/2-1), and the SFB 677 (project B11). Author contributions J. F. R. and H. B. conducted experiments and analyzed data. J.-H. J. and R. M. analyzed data and designed theory, M. L. provided acanthamoebae and contributed to the design of the study, C. S. designed the experiments and analyzed data. All authors contributed to writing the manuscript. Additional information Supplementary information and supplementary videos accompany this paper. There are no competing financial interests. 66 67 22 Table 1: Average diffusive exponents for different drug treatments and numbers of analyzed acanthamoebae and intracellular particles. # (amoebae) # (particles) 8 8 5 11 167 174 157 112 inhibitor none latrunculin A nocodazole blebbistatin α 1.79 ± 0.12 1.53 ± 0.21 1.60 ± 0.16 0.15 ± 0.13 23
1106.6150	1	1106	2011-06-30T09:06:46	Quantum effects in energy and charge transfer in an artificial photosynthetic complex	[ "physics.bio-ph", "quant-ph" ]	We investigate the quantum dynamics of energy and charge transfer in a wheel-shaped artificial photosynthetic antenna-reaction center complex.This complex consists of six light-harvesting chromophores and an electron-acceptor fullerene. To describe quantum effects on a femtosecond time scale, we derive the set of exact non-Markovian equations for the Heisenberg operators of this photosynthetic complex in contact with a Gaussian heat bath. With these equations we can analyze the regime of strong system-bath interactions, where reorganization energies are of the order of the intersite exciton couplings. We show that the energy of the initially-excited antenna chromophores is efficiently funneled to the porphyrin-fullerene reaction center, where a charge-separated state is set up in a few picoseconds, with a quantum yield of the order of 95%. In the single-exciton regime, with one antenna chromophore being initially excited, we observe quantum beatings of energy between two resonant antenna chromophores with a decoherence time of $\sim$ 100 fs. We also analyze the double-exciton regime, when two porphyrin molecules involved in the reaction center are initially excited. In this regime we obtain pronounced quantum oscillations of the charge on the fullerene molecule with a decoherence time of about 20 fs (at liquid nitrogen temperatures). These results show a way to directly detect quantum effects in artificial photosynthetic systems.	physics.bio-ph	physics	Quantum effects in energy and charge transfer in an artificial photosynthetic complex Pulak Kumar Ghosh1∗, Anatoly Yu. Smirnov1,2, and Franco Nori1,2 1 Advanced Science Institute, RIKEN, Wako, Saitama, 351-0198, Japan 2 Physics Department, The University of Michigan, Ann Arbor, MI 48109-1040, USA (Dated: August 5, 2021) Abstract We investigate the quantum dynamics of energy and charge transfer in a wheel-shaped artificial photosynthetic antenna-reaction center complex. This complex consists of six light-harvesting chromophores and an electron-acceptor fullerene. To describe quantum effects on a femtosecond time scale, we derive the set of exact non-Markovian equations for the Heisenberg operators of this photosynthetic complex in contact with a Gaussian heat bath. With these equations we can analyze the regime of strong system-bath interactions, where reorganization energies are of the order of the intersite exciton couplings. We show that the energy of the initially-excited antenna chromophores is efficiently funneled to the porphyrin-fullerene reaction center, where a charge-separated state is set up in a few picoseconds, with a quantum yield of the order of 95%. In the single-exciton regime, with one antenna chromophore being initially excited, we observe quantum beatings of energy between two resonant antenna chromophores with a decoherence time of ∼ 100 fs. We also analyze the double-exciton regime, when two porphyrin molecules involved in the reaction center are initially excited. In this regime we obtain pronounced quantum oscillations of the charge on the fullerene molecule with a decoherence time of about 20 fs (at liquid nitrogen temperatures). These results show a way to directly detect quantum effects in artificial photosynthetic systems. 1 1 0 2 n u J 0 3 ] h p - o i b . s c i s y h p [ 1 v 0 5 1 6 . 6 0 1 1 : v i X r a ∗ Author to whom correspondence should be addressed. Electronic mail: [email protected] 1 I. INTRODUCTION The multistep energy-transduction process in natural photosystems begins with capturing sunlight photons by light-absorbing antenna chromophores surrounding a reaction center [1, 2]. The antenna chromophores transfer radiation energy to the reaction center directly or through a series of accessory chromophores. The reaction center harnesses the excitation energy to create a stable charge-separated state. Energy transfer in natural and artificial photosynthetic structures has been an intriguing issue in quantum biophysics due to the conspicuous presence of long-lived quantum coher- ence observed with two-dimensional Fourier transform electronic spectroscopy [3, 4]. These experimental achievements have motivated researchers to investigate the role of quantum coherence in very efficient energy transmission, which takes place in natural photosystems [5 -- 9]. Quantum coherent effects surviving up to room temperatures have also been ob- served in artificial polymers [10]. Artificial photosynthetic elements, mimicking natural photosystems, might serve as building blocks for efficient and powerful sources of energy [11, 12]. Some of these elements have been created and studied experimentally in Refs. [13 -- 18]. The theoretical modelling of artificial reaction centers has been recently performed in Refs. [19, 20]. Here we study energy transfer and charge separation in a wheel-shaped molecular complex (BPF complex, see Fig. 1) mimicking a natural photosynthetic system. This complex has been synthesized and experimentally investigated in Ref. [17]. It has four antennas - two bis(phenylethynyl)anthracene (BPEA) molecules and two borondipyrromethene (BDPY) chromophores, as well as two zinc porphyrins (ZnPya and ZnPyb). These six light-absorbing chromophores are attached to a central hexaphenylbenzene core. Electrons can tunnel from the zinc porphyrin molecules to a fullerene F (electron acceptor). Thus, two porphyrins and the fullerene molecule form an artificial reaction center (ZnPya − F − ZnPyb). The BPEA chromophores strongly absorb around 450 nm (the blue region), while the BDPY moieties have good absorptions around 513 nm (green region). Porphyrins have absorption peaks at both red and orange wavelengths. Therefore, the BPF complex can utilize most of the rainbow of sunlight -- from blue to red photons. It is shown in [17] that the absorption of photons results in the formation of a porphyrin-fullerene charge-separated state with a life- time of 230 ps; in doing so, excitations from the BPEA and BDPY antenna chromophores 2 FIG. 1: (Color online) Schematic diagram of the wheel-shaped artificial antenna-reaction center complex reported in ref [17]. We use the short notation, BPF Complex, to denote this photosyn- thetic device. The antenna-reaction center complex contains six light-harvesting pigmets: (i) two bis (phenylethynyl)anthracene chromophores, BPEAa and BPEAb, (ii) two borondipyrromethene chromophores, BDPYa and BDPYb, and (iii) two zinc tetraarylporphyrin chromophores, ZnPya and ZnPyb. All the chromophores are attached to a rigid hexaphenyl benzene core. In addition to the antenna components, the photosystem contains a fullerene derivative (F) containing two pyridyl groups, acting as an electron acceptor. The fullerene derivative F is attached to the both ZnPy chromophores via the coordination of the pyridyl nitrogens with the zinc atoms. For structural details of the BPF Complex we refer [17, 28]. are transferred to the porphyrins with a subsequent donation of an electron from the ex- cited states of the porphyrins to the fullerene moiety. This process takes a few picoseconds, suggesting that the excitonic coupling between chromophores is sufficiently strong. The elec- tronic coupling between the porphyrins and the fullerene controlling tunneling of electrons in the artificial reaction center also should be quite strong. It should be noted, however, that spectroscopic data [15 -- 17] show that the absorption spectrum of the BPF complex is approximately represented as a superposition of contributions from the individual chro- mophores with almost no perturbations due to the links between the chromophores. This means that the chromophores comprising the light-harvesting complex can be considered 3 ZnPyaFZnPybBDPYbBDPYaBPEAaBPEAbas individual interacting units, but not as an extended single chromophore. We can expect that, at these conditions, quantum coherence is able to play an important role in energy and charge transfer dynamics, manifesting itself in quantum beatings of chromophore popula- tions as well as in quantum oscillations of the charge accumulated on the fullerene molecule. In principle, these oscillations could be measured by a sensitive single-electron transistor, thus providing a direct proof of quantum behavior in the artificial photosynthetic complex. Since these phenomena occurs at very short time scales (a few femtoseconds), these could be within the reach of femtosecond spectroscopy in the near future. The main goal of this study is to explore quantum features of the energy and charge transfer in a wheel-shaped antenna-reaction center complex at subpicosecond timescales. II. MODEL AND METHODS A. Hamiltonian Each chromophore has one ground and one excited state, whereas the electron acceptor fullerene F has just one energy level with energy EF . We introduce creation (annihilation) operators, a † k (ak), of an electron on the kth site. The electron population operators are † kak. We assume that each electron state can be occupied by a single electron, as spin degrees of freedom are neglected. The basic Hamiltonian of the system has defined as nk = a the form: H0 = (cid:88) (Eknk + Ek∗nk∗) + EF nF + HC + k (cid:88) k(cid:54)=l † k∗ak a Vkla l al∗ −(cid:88) † σσ(cid:48) ∆σσ(cid:48)a† σaσ(cid:48), (1) where the first part incorporates the energies of the electron states (hereafter k, l = BPEAa, BPEAb, BDPYa, BDPYb, ZnPya, ZnPyb), and the second term is related to a fullerene † F aF . The pair (k, k∗) denotes a ground energy level EF with a population operator nF = a (k) and an excited (k∗) state of an electron located on the site k with the corresponding energy Ek (Ek∗). The term HC represents the contribution of Coulomb interactions between electron-binding sites. This term is given in Appendix A. The fourth term of Eq. (1) describes excitonic couplings between the chromophores k and l. The matrix element Vkl is a measure of an interchromophoric coupling strength. The last term in Eq. (1) describes the electron tunneling from excited states of the porphyrin molecules ZnPya, ZnPyb to the electron 4 acceptor F characterized by the tunneling amplitudes ∆σσ(cid:48), where σ, σ(cid:48) = ZnPy∗ F. a, ZnPy∗ b, The interaction of the system with the environment (heat bath), represented here by a sum of independent oscillators with Hamiltonian (cid:88) j Henv = (cid:32) p2 j 2mj (cid:33) , + j x2 mjω2 j 2 is given by the term He−ph = −(cid:88) jk mjω2 j xjkxjnk, (2) (3) where xj and pj are the position and momentum of the jth oscillator having an effective mass mj and a frequency ωj. The coefficients xjk define the strength of the coupling between the electron subsystem and the environment. The contribution of the energy-quenching mechanisms responsible for the recombination processes in the system is given by the Hamiltonian Hquen = −(cid:88) † l a † l∗al + qla † l al∗). (4) (q l For the sake of simplicity, we include the radiation damping of the excited states into the energy-quenching operator ql. The first term in the Hermitian Hamiltonian Hquen is related to the excitation of the l−chromophore by the quenching bath, whereas the second term corresponds to the reverse process, namely, to the absorption of chromophore energy by the bath. Both processes are necessary to provide correct conditions for the thermodynamic equilibrium between the system and the bath. The total Hamiltonian of the system is H = H0 + He−ph + Henv + Hquen. (5) We omit here the Hamiltonian of the quenching (radiation) heat bath. B. Diagonalization of H0 We choose 160 basis states M(cid:105) of the complex including a vacuum state, where all chromophores are in the ground state and the F site is empty. We diagonalize the Hamil- tonian H0 (1) to consider the case where the excitonic coupling between chromophores, 5 described by coefficients Vlm, and the porphyrin-fullerene tunneling, which is determined µ(cid:105) = (cid:80) by amplitudes ∆σσ(cid:48), cannot be analyzed within perturbation theory. In the new basis, M M(cid:105)(cid:104)Mµ(cid:105), the Hamiltonian H0 is diagonal with the energy spectrum {Eµ}, so that the total Hamiltonian of the system H has the form H = Here (cid:88) Eµµ(cid:105)(cid:104)µ −(cid:88) µ µν Aµν = Qµν + qµν Aµνµ(cid:105)(cid:104)ν + Henv. (6) (7) (8) (9) (cid:88) is the combined operator for both heat baths with fluctuating in time variables (xjk(cid:104)µnkν(cid:105) + xjk∗(cid:104)µnk∗ν(cid:105))], j xj[xjF(cid:104)µnFν(cid:105) + (cid:88) Qµν = mjω2 j and (cid:88) l qµν = k (cid:104)µa l al∗ν(cid:105) ql + H.c. † To distinguish the processes of energy transfer, where the number of electrons on each chro- mophore remains constant, from the processes of charge transfer, where the total population of the site changes, we introduce the following operators Sl = nl + nl∗, Ml = nl − nl∗, together with coefficients ¯xjl = xjl + xjl∗ 2 , xjl = xjl − xjl∗ 2 . Thus, the environment operator Qµν can be rewritten as (cid:88) Qµν = mjω2 j xjΛµν j j with (cid:88) l Λµν j = {¯xjl(cid:104)µSlν(cid:105) + xjl(cid:104)µMlν(cid:105)} + xjF(cid:104)µnFν(cid:105) C. Non-Markovian equations for the system operators (10) (11) (12) (13) ρµν = µ(cid:105)(cid:104)ν; with W =(cid:80) An arbitrary electron operator W can be expressed in terms of the basic operators µν Wµν ρµν, and Wµν = (cid:104)µWν(cid:105). The operator ρµν denotes a ma- trix with zero elements, with the exception of the single element at the crossing of the µ−row 6 and the ν−column. The matrix elements Wµν of any electron operator can be easily calcu- lated (see, e.g., Eqs. (S10) and (S11) in the Supporting Information for Ref. [20]). For exam- ple, an electron localized in a two-well potential [21], with the right and left states 1(cid:105) and 2(cid:105), is described by the Pauli matrices {σx, σy, σz} : σz = 1(cid:105)(cid:104)1−2(cid:105)(cid:104)2, σx = 1(cid:105)(cid:104)2 +2(cid:105)(cid:104)1, and σy = i(2(cid:105)(cid:104)1 − 1(cid:105)(cid:104)2), which are expressed in terms of the basic operators µ(cid:105)(cid:104)ν with µ, ν = 1, 2. In the Heisenberg picture, the operator W evolves in time according to the equation: i (∂W/∂t) = [W, H]− . This evolution can be described with the time-evolving operators, ρµν(t) = (µ(cid:105)(cid:104)ν)(t), which satisfy the Heisenberg equation: (Aναρµα − Aαµραν), (14) = [ρµν, H]− = − ωµνρµν −(cid:88) α i ∂ρµν ∂t where ωµν = Eµ− Eν, and the heat bath operator Aµν is defined in Eq. (7). Here, we use the fact that the Hamiltonian H Eq. (6) is also expressed in terms of the operators ρµν taken at the same moment of time t. For two of these operators, ρµν(t) and ραβ(t), we have simple multiplication rules: ρµνραβ = δναρµβ. These rules allow to calculate commutators of basic operators taken at the same moment of time. We note that at the initial moment of time the operator, ρµν(0) ≡ µ(cid:105)(cid:104)ν, is represented by the above-mentioned zero matrix with a single unit at the µ-ν intersection. The matrix elements of the electron operators in Eqs. (9,13) are taken over the time-independent eigenstates of the Hamiltonian H0. The bath operators Aµν fluctuate in time since they depend on the environmental variables, {xj(t)}, and on the variables {ql(t)} of the quenching bath. It is known that the dissipative evolution of the two-state system can be described by the Heisenberg equations for the Pauli matrices {σx, σy, σz} with the spin-boson Hamiltonian [see Eq. (1.4) in Ref. [21]], which includes environmental degrees of freedom. The artificial photosynthetic complex analyzed in the present paper has 160 states. A dissipative evolution of this complex is described by the Hamiltonian H in Eq. (6), written in terms of the Heisenberg operators ρµν(t) = (µ(cid:105)(cid:104)ν)(t) taken at the moment of time t. Instead of the time-dependent Pauli matrices, the time evolution of the two-state dissipative system can be described by the basic operators 1(cid:105)(cid:104)1, 1(cid:105)(cid:104)2, 2(cid:105)(cid:104)1, 2(cid:105)(cid:104)2, evolving in time. In a similar manner, the evolution of the multi-state photosynthetic complex is described by the set of the time-dependent Heisenberg operators ρµν(t), which obey the equation (14). As its spin-boson counterpart, the Hamiltonian H in Eq. (6) contains the Hamiltonian, Henv, of 7 the heat bath as well as the system-bath interaction terms. Here, we generalize the spin- boson model from the case of two states to the case of 160 states. With a knowledge of the operators ρµν(t), it is possible to find the time evolution of any Heisenberg operator of the system. Only at the initial moment of time, t = 0, the operators ρµν(0) form the basis of the Liouville space. Note that we work in the Heisenberg representation, without using the description based on the von Neumann equations for the density matrix. To obtain functions that can be measured in experiments, we have to average the operator ρµν(t) and the equation (14) over the initial state Ψ0(cid:105) of the electron subsystem as well as over the Gaussian distribution, ρT = exp(−H (0) bath/T ), of the equilibrium bath, (cid:104). . .(cid:105)T , with temperature T and with a free Hamiltonian H (0) bath, which is comprised of the free environment Hamiltonian and the free Hamiltonian of the quenching bath. The notation (cid:104). . .(cid:105) means double averaging: (cid:104). . .(cid:105) = (cid:104)(cid:104)Ψ0 . . .Ψ0(cid:105)(cid:105)T . (15) The quantum-mechanical average value of the initial basic matrix, (cid:104)Ψ0ρµν(0)Ψ0(cid:105) = (cid:104)Ψ0µ(cid:105)(cid:104)νΨ0(cid:105), is determined by the product of amplitudes to find the electron subsystem at the initial moment of time in the eigenstates µ(cid:105) and ν(cid:105) of the Hamiltonian H0. A standard density matrix, ¯ρ = {¯ρµν}, of the electron subsystem is a deterministic function which allows to calculate the average value of an arbitrary operator W with the formula: (cid:104)W (t)(cid:105) = T r[¯ρ(t) W ] = (cid:88) (16) The same average value can be written as (cid:104)W (t)(cid:105) = (cid:80) µν Wµν (cid:104)ρµν(t)(cid:105), which means that the average matrix, (cid:104)ρµν(t)(cid:105) = ¯ρνµ(t), has matrix elements related to the transposed density matrix ¯ρ(t). Wµν ¯ρνµ(t). µν It should be emphasized that the time evolution of the heat-bath operators {xj, pj} and {ql}, as well as their linear combinations Qµν, qµν, and Aµν, are determined by the total Hamiltonian H in Eq. (6). In the absence of an interaction with the dynamical system (the other baths, q(0) electron-binding sites), the free-phonon operators Q(0) µν , as well as the free operators of the µν , are described by Gaussian statistics [23], as in the case of an environ- ment comprised of independent linear oscillators with the Hamiltonian Henv (2). Using the Gaussian property, Efremov and coauthors [24] derived non-Markovian Heisenberg-Langevin equations, without using perturbation theory, that assumes a weak system-bath interaction. 8 Recently, a similar non-perturbative approach has been developed by Ishizaki and Fleming in Ref. [25]. Due to Gaussian properties of the free bath, the total operator Aµν of the combined dissipative environment is a linear functional of the operators ρµν, ¯µ¯ν (t1)]−(cid:105)θ(t − t1)ρ¯µ¯ν(t1), Aµν(t) = A(0) µν (t),A(0) (cid:104)i[A(0) (cid:88) µν (t) + (17) (cid:90) ¯µ¯ν It is shown in Ref. [24] that the average value of the free operator A(0) where θ(τ ) is the Heaviside step function. We note that this expansion directly follows from the solution of the Heisenberg equations for the positions {xj} and {ql} of the bath µν (t) oscillators. multiplied by an arbitrary operator B(t) is proportional to the functional derivative of the operator B over the variable A(0) (cid:88) µν (t)B(t)(cid:105) = (cid:42) δB(t) µν (t)A(0) ¯µ¯ν (t1)(cid:105) × dt1 (cid:104)A(0) (cid:104)A(0) µν (t): (cid:43) (18) (cid:90) , δA(0) ¯µ¯ν (t1) with ¯µ¯ν δB(t) δA(0) ¯µ¯ν (t1) = i [B(t), ρ¯µ¯ν(t1)]− θ(t − t1). (19) Substituting Eqs. (17,18,19) into Eq. (14) we derive the exact non-Markovian equation for the Heisenberg operators ρµν of the dynamical system (chromomorphic sites + fullerene) dt1 (cid:90) t (cid:88) − (cid:104)A(0) − (cid:104)A(0) (cid:110)(cid:104)A(0) να (t)A(0) ¯µ¯ν (t1)A(0) α¯µ¯ν 0 interacting with a Gaussian heat bath, (cid:104) ρµν(cid:105) − i ωµν(cid:104)ρµν(cid:105) = ¯µ¯ν (t1)A(0) να (t)(cid:105)(cid:104)ρ¯µ¯ν(t1)ρµα(t)(cid:105) ¯µ¯ν (t1)(cid:105)(cid:104)ρµα(t)ρ¯µ¯ν(t1)(cid:105) + (cid:104)A(0) αµ(t)(cid:105)(cid:104)ρ¯µ¯ν(t1)ραν(t)(cid:105)}. αµ(t)A(0) ¯µ¯ν (t1)(cid:105)(cid:104)ραν(t)ρ¯µ¯ν(t1)(cid:105) (20) The time evolution of the average operator (cid:104)ρµν(cid:105) is determined by the second-order corre- lation functions of the system operators as well as by the correlation functions of the free dissipative environment. Here we do not impose any restrictions on the spectrum of the environment. It should be emphasized that the exact non-Markovian equation (20) goes far beyond the von Neumann equation, i ¯ρ = [¯ρ, H]−, for the density matrix ¯ρ of the electron subsystem. D. Beyond the system-bath perturbation theory. We assume that the coupling of the system to the quenching heat bath determined by the Hamiltonian Hquen (4) is weak enough to be analyzed perturbatively. However, an 9 interaction of the chromophores with the protein environment cannot be treated entirely within perturbation theory since the reorganization energies are of the order of the intersite couplings. As in the theory of modified Redfield equations [26, 27], the phonon operator Qµν in Eq. (12) can be represented as a sum of diagonal Qµ = Qµµ and off-diagonal Qµν parts: Qµν = Qµδµν + (1 − δµν) Qµν. (21) We derive equations for diagonal and off-diagonal elements of the matrix (cid:104)ρµν(t)(cid:105) (see Ap- pendix B for details about the derivation), where the interaction with the off-diagonal ele- ments of the environment operators Qµν are considered within perturbation theory, and the effects of the diagonal elements Qµ are treated exactly. The time dependence of the electron distribution (cid:104)ρµ(cid:105) (diagonal elements) over eigenstates of the Hamiltonian H0 is governed by the equation (cid:88) (cid:104) ρµ(cid:105) + γµ(cid:104)ρµ(cid:105) = γµα(cid:104)ρα(cid:105), (22) where the relaxation matrix γµα contains a contribution, γµα, from the non-diagonal envi- ronment operators [see Eq. (B22)] as well as a contribution from the quenching processes, α γquen µα [see Eq. (B30)], γµα = γµα + γquen µα , with the total relaxation rate γµ =(cid:80) arbitrary operator W of the system: (cid:104)W (t)(cid:105) =(cid:80) are given by Eq. (B31) in Appendix B. µν(cid:104)µWν(cid:105)(cid:104)ρµν(t)(cid:105). Equations (22,B31) allow us to determine the time evolution of an average value for an α γαµ. The time evolution of the off-diagonal elements (23) III. ENERGIES AND OTHER PARAMETERS A. Energy levels and electrochemical potentials The energies of the excited states of chromophores BPEA, BDPY, and ZnPy, in the BPF complex are estimated from an average between the longest wavelength absorption band and the shortest wavelength emission band of the chromophores. The average excited state energies of the chromophores BPEA, BDPY and ZnPy are 2610 meV, 2370 meV, and 2030 meV, respectively, if we count from the corresponding ground energy levels [16, 17]. 10 TABLE I: This table presents the chosen values of the excitonic couplings (V ) and reorganization energies for energy transfer (Λ) of the six antenna chromophores. We choose two sets of parameters, one set (denoted by I) corresponds to V > Λ and the other set (II) to the opposite limit V < Λ. The calculated values of the time constants using both sets of parameters agree with the experimental values. Chromophores Set I Coupling (V) BPEAa ↔ BPEAb, BPEAb ↔ BPEAa BPEAa ↔ BDPYa, BPEAb ↔ BDPYb BDPYa ↔ ZnPya, BDPYb ↔ ZnPyb BPEAa ↔ ZnPya, BPEAb ↔ ZnPyb BPEAb ↔ ZnPya, BPEAa ↔ ZnPyb 50 meV 30 meV 60 meV 50 meV 60 meV Set I Reorganization energy (Λ) ΛBPEAa = 20 meV ΛBPEAb = 20 meV ΛBDPYa = 15 meV ΛBDPYb = 15 meV ΛZnPya = 20 meV ΛZnPyb = 20 meV - - Set II Coupling (V) 30 meV 17 meV 25 meV 40 meV 40 meV Set II Reorganization energy (Λ) ΛBPEAa = 40 meV ΛBPEAb = 40 meV ΛBDPYa = 30 meV ΛBDPYb = 30 meV ΛZnPya = 40 meV ΛZnPyb = 40 meV - - Cyclic voltammetric studies [17] of reduction potentials with respect to the standard calomel electrode show that the first reduction potential of the fullerene derivative, F, is about -- 0.62 V and the first oxidation potential of ZnPy is about 0.75 V. From these data we calculate that the energy of the charge separated state ZnPy+−F− is about 1370 meV. This energy is a sum of the energy of an electron on site F and a Coulomb interaction energy between a positive charge on ZnPy and a negative charge on F. The Coulomb energy can be calculated with the formula u = e2/4π0r, where 0 is the vacuum dielectric constant. The dielectric constant of 1,2 diflurobenzene (a solvent used in all experimental measurements of Ref. [17]) is about 13.8. If the distance r between porphyrin ZnPy and fullerene F is about 1 nm, the Coulomb interaction energy is about 105 meV. Thus, the estimated energy of the electron on F can be of the order of 1475 meV. 11 B. Reorganization energies and coupling strengths The reorganization energies for exciton and electron transfer processes and electronic coupling strengths between the chromophores depend on the mutual distances and orienta- tions of the components, strengths of chemical bonds, solvent polarity and other structural details of the system. Precise values of these parameters are not available. However, time constants for energy transfer between different chromophores in the BPF complex, as well as rates for transitions of electrons between the fullerene F and porphyrin chromophores ZnPy, have been reported in Ref. [17]. We fit the experimental values of these time constants with the rates following from our equations with the goal of extracting reasonable values for the reorganization energies and the electronic and excitonic couplings. In principle, many com- binations of reorganization energies and coupling constants could be possible. For the sake of simplicity, we consider two sets of parameters, for two limiting situations. One parameter set (denoted by I in Table I) corresponds to a larger excitonic couplings, V , compared to the reorganization energies, Λ, whereas another set of parameters (denoted by II in Table I) considers the opposite case: where the reorganization energies are larger than the excitonic couplings. These two sets of parameters are presented in Table I. In addition to the parame- ters listed in Table I, we consider the following values for the charge-transfer reorganization energies (set I): λF = 200 meV, λlM = 100 meV, and λF = 230 meV, λlM = 120 meV (set II), where l = ZnPya, ZnPyb. The values of the reorganization energies for energy-transfer processes are much smaller than those for charge transfer. References [17, 28] reported a very fast electron transfer (with a time constant τ ∼ 3 ps) between excited states of zincporphyrins (ZnPya,ZnPyb) and the fullerene derivative F. This fact indicates a good porphyrin-fullerene electronic coupling, which is due to the short covalent linkage and close spatial arrangement of the components [28]. Hereafter, we assume that the ZnPy-F tunneling amplitudes ∆ are about 100 meV (parameter set I) and 80 meV (parameter set II). These parameters provide a quite fast electron transfer, despite of a significant energy gap between the ZnPy excited states and the fullerene energy level. To describe recombination processes, we introduce a coupling of the l-th chromophore to a quenching heat-bath characterized for simplicity by the Ohmic spectral density: χ(cid:48)(cid:48) l (ω) = αl ω with a dimensionless constant αl. We assume that the shifts of the en- ergy levels caused by the quenching bath are included into the renormalized parameters 12 TABLE II: This table presents a comparison between the calculated values of the time constants (using the parameters sets I and II) to the experimental values reported in Ref. [17]. Process τ (Set I) τ (Set II) τ (Experimental) BPEAa → BPEAb, BPEAb → BPEAa BPEAa → BDPYa, BPEAb → BDPYb BDPYa → ZnPya, BDPYb → ZnPyb BPEAa → ZnPya, BPEAb → ZnPyb BPEAb → ZnPya, BPEAa → ZnPyb ZnPyb → F, ZnPya → F ∼ 0.4 ps ∼ 0.2 ps 0.4 ps ∼ 5 ps ∼ 5.4 ps 5-13 ps ∼ 5 ps ∼ 3.9 ps 2-15 ps ∼ 12 ps ∼ 12 ps ∼ 10 ps ∼ 12 ps ∼ 3 ps ∼ 3 ps 7 ps 6 ps 3 ps of the electron subsystem. The experimental values [17, 28] of the lifetimes τ e l for excited BPEA = 2.82 ns, τ e states of chromophores BPEA, BDPY and ZnPy: τ e BDPY = 0.26 ns, ZnPY = 0.45 ns, can be achieved with the following set of coupling constants: and τ e αBPEA ∼ 10−7, αBDPY ∼ 10−6, and αZnPy ∼ 7 × 10−7. IV. RESULTS AND DISCUSSIONS Using Eqs. (B31,22) and two sets of parameters discussed in Sec. III, here we study electron and energy transfer kinetics in the BPF complex with special emphasis on the femtosecond time range, where the effects of quantum coherence can play an important role. We consider both single- and double-exciton regimes. A. Evolution of a single exciton in the BPF complex In Fig. 2 we show the time evolution of the excited states populations provided that only the BPEAa chromophore is excited at t = 0 (single-exciton regime). We use here the param- 13 FIG. 2: (Color online) Site populations as a function of time for the parameter set I. The inset plots depict the features of site populations for short times, at two different temperatures: T = 300 K and 77 K. The site populations of the BPEA moieties oscillate with a considerably large amplitude, while the oscillations of the other site populations are hardly observable. eter set I, where excitonic couplings are larger than reorganization energies (see Sec. III). The process starts with quantum beatings between the resonant BPEAa and BPEAb chro- mophores, with a decoherence time of the order of 100 fs (at T = 300 K). In a few picoseconds, the excitation energy is subsequently transferred to the adjacent BDPY moieties and to the ZnPy chromophores. Later on, an electron moves from the excited energy level of the por- 14 02600.20.40.60.81 Time (ps)Site populations 00.40.8BPEAaBPEAbBDPYaBDPYbZnPyaZnPybF00.200.40.8 Time (ps)Site populationsT = 300 KT = 77 KT = 300 KParameterset−1Fig2FIG. 3: (Color online) This figure presents site populations as a function of time for the parameter set II. The inset plots show the site populations for short times, at two different temperatures: T = 300 K and 77 K. The amplitudes of the site-population oscillations are much smaller and die out earlier, compared to Fig. 2. This figure indicates that even for Λ > V , the energy transfer between BPEA chromophores is dominated by wave-like coherent motion. phyrins to the fullerene moiety; thus, producing a charge-separated state, ZnPy+−F−, with a quantum yield 95%, which is in agreement with experimental results [16]. It is evident from Fig. 2 that excited state populations of the BDPY chromophores oscillate with much lower amplitudes and die out within a very short time, t < 10 fs, at both temperatures: T 15 02600.20.40.60.81 Time (ps)Site populations BPEAaBPEAbBDPYaBDPYbZnPyaZnPybF00.40.8Site populations 00.100.40.8 Time (ps)T = 300 KT = 300 KT = 77 KFig3FIG. 4: (Color online) Site populations as a function of time for the parameter set I, when the ZnPya chromophore is in the excited state and all the other chromophores are in the ground state at t = 0. The inset plots depict the site populations at short times for two temperatures: T = 300 and 77. Lowering the temperature enhances the oscillations of the charge density on the fullerene moiety. Despite the huge energy difference between ZnPy∗−F and ZnPy+−F−, the charge of the fullerene site exhibits oscillatory behavior for short times, specially at lower temperatures. = 300 K and 77 K. The populations of the other sites of the BPF complex do not exhibit any oscillatory behavior. This can be ascribed to incoherent hopping becoming dominant because of significant energy mismatch between these chromophores. Figure 3 shows the time-dependence of the excited state populations of chromophores for 16 02600.20.40.60.81 Time (ps)Site populations ZnPyaZnPybF00.40.80204000.040.08Time (fs)Site populations 00.40.80204000.040.08Time (fs)Site populations T = 300 KT = 300 KT = 77Kthe parameter set II, where the reorganization energies are larger than the excitonic couplings between chromophores. At t = 0 the BPEAa chromophore is excited (single-exciton regime). Then, after a few picoseconds, the charge-separated state is formed with a quantum yield of the order of 97%. However, owing to a stronger system-environment coupling, quantum beats between the BPEAa and BPEAb chromophores have a lower amplitude and shorter decoherence time (∼50 fs) than in the previous case when we use the parameter set I. We note that no quantum oscillations of the fullerene population (site F) are visible in Figs. 2 and 3. No significant oscillations of the site populations were observed (not shown here) when the BDPY chromophores were initially (at t = 0) excited. In this case, due to the consid- erable energy gaps between the BDPY and the adjacent BPEA and ZnPy chromophores, incoherent hopping dominates over the coherent transfer of excitons. Furthermore, the struc- ture of the BPF complex [15, 28] does not allow direct energy transfer between two BDPY chromophores. Figures 4 and 5 demonstrate charge- and energy-transfer dynamics for two parameter sets, I and II, for the case when one of the porphyrin chromophores (ZnPya) is excited. Here we do not show the time evolution of the BPEA and BDPY chromophores since these moieties have higher excitation energies than the ZnPy chromophore and they are not excited in the process. As evident from Figs. 4 and 5, the excited porphyrin molecule rapidly transfers an electron to fullerene, thus, producing a charge-separated state ZnPy+−F− with a quantum yield of about 98%. The most important feature here is that the population and charge of the fullerene molecule oscillates in time due to a quantum superposition of the porphyrin excited state and the state of an electron on the fullerene. The amplitude of these quantum beats is very small and the decoherence time is quite short (∼10 fs at T = 77 K). This fact can be explained by the significant energy mismatch between the ZnPy∗−F and ZnPy+−F− states as well as by the strong influence of the environment on the electron dynamics. B. Evolution of double excitons in the BPF complex In the previous subsection, we consider a single exciton case with just one chromophore initially being in the upper energy state. Here we analyze a situation where two porphyrin molecules (ZnPya and ZnPyb) are excited at t = 0. Figures 6a and 6b show the coherent 17 FIG. 5: (Color online) Time evolution of the site populations for the parameter set II, starting with an exciton on the chromophore ZnPya at t = 0. The inset plots depict the features of the site populations for a shorter time regime and at two temperatures: T = 300 K and 77 K. Lowering the temperature enhances oscillations of the charge density on the fullerene derivative. These results indicate that the population of the site F oscillates for short times, even for Λ > V . These oscillations are more pronounced at lower temperatures. dynamics of the fullerene population (and the fullerene charge) for the parameter sets I (Fig. 6a) and II (Fig. 6b) at two different temperatures, T = 77 K and T = 300 K. We also compare the double-exciton case with the previously analyzed single-exciton case. It is apparent from Fig. 6, that the double excitation significantly enhances the amplitude of quantum oscillations of the fullerene charge for both sets of parameters. As one might expect, the frequency of the quantum beatings and the decoherence time are not affected by the number of excitons. 18 00.40.80204000.040.08Time (fs)Site populations 02600.40.8 Time (ps)Site populations ZnPyaZnPybF0204000.040.08Time (fs)Site populations00.40.800.40.8Site populations0204000.040.08Time (fs)T = 77KFig5T = 300 KT = 300 KFIG. 6: (Color online) Time evolution of the populations on the site F, for both sets of parameters, I and II, comparing the double-exciton case (the two ZnPy chromophores are excited) with the single-exciton case. (a) Time evolution of the populations on the site F for the parameter set I. (b) Time evolution of the populations on the site F for the parameter set II. Note that the double- excitation significantly enhances the amplitude of the charge oscillations at the fullerene site for both sets of parameters, either at low or high temperatures. C. Amplification of charge oscillations In the previous discussion we observed that lowering the temperature and the simultane- ous excitation of both porphyrins significantly enhances quantum oscillations of the fullerene charge. In this subsection we show that these oscillations can also be controlled by tuning 19 00.060.120.17Population on the site, F 0204000.060.12 Time (fs)Double−exciton, 300 KDouble−exciton, 77 KSingle−exciton, 300 KSingle−exciton, 77 K(a)(b)FIG. 7: (Color online) Time evolution of the population on the site F for the parameters set II when both ZnPy chromophores are excited at t = 0. (a) Effects of the coupling ∆ on the time evolution of the populations on the site F. (b) Effects of the energy gap between an excited state of a ZnPy chromophore and the charge-separated state, Ech, on the time evolution of populations on the site F. (c) Effects of the reorganization energy λ on the time evolution of populations on the site F. As can be seen from these plots, the contribution of wave-like coherent motion to electron- transfer dynamics is significantly enhanced when strengthening the coupling between fullerene and porphyrin, lowering the energy gap between the fullerene and porphyrin sites, and decreasing the reorganization energy. 20 00.10.2 00.20.4Population on the site, F Ech = 1370 meVEch = 1500 meVEch = 1650 meV∆ = 20 meV∆ = 50 meV∆ = 80 meV∆ = 120 meV0204000.070.14 Time (fs)Population on the site, F λF = 320 meVλF = 200 meVλF = 80 mev(b)(a)(c)T = 77 KT = 77 KT = 77 KParameterthe following parameters: 1. Electron tunneling amplitude ∆ . The electronic coupling between the fullerene electron acceptor and zinc porphyrins has a strong effect on the quantum oscillations of the fullerene charge. To explore this effect, in Fig. 7a we plot the electron population of the fullerene as a function of time, for different values of the coupling ∆. Figure 7a clearly shows that, with increasing ∆, the amplitude of the charge oscillations is significantly enhanced. This coupling can be increased by attaching the fullerene to porphyrins with better ligands which form much stronger covalent bonds. 2. Energy of the charge-separated state Ech . The energy Ech ∼ 1370 meV, of the charge separated state, ZnPy+−F− is much lower than the energy of the zinc porphyrin excited state, EZnPy∗ ∼ 2030 meV. It is evident from Fig. 7b that increasing the energy Ech, which leads to a decrease of the porpyrin-fullerene energy mismatch, results in a pronounced amplification of the quantum oscillations of the fullerene charge. The energy of the fullerene can be changed by placing nearby a charge residue, electrostatically coupled to the fullerene. 3. Reorganization energy λF . In Fig. 7c we present the time evolution of the fullerene population for different values of charge transfer reorganization energy λF . This parameter can be decreased by replacing the polar solvent with another one which has a much lower polarity. As can be seen from Fig. 7c, the quantum oscillations of the fullerene charge survive much longer times for smaller values of the reorganization energy, which correspond to weaker system-environment couplings. A similar effect is expected when the porphyrin reorganization energy is changed. V. CONCLUSIONS We theoretically studied the energy and electron-transfer dynamics in a wheel-shaped artificial antenna-reaction center complex. This complex [17], mimicking a natural photo- 21 system, contains six chromophores (BPEAa, BPEAb, BDPYa, BDPYb, ZnPya, ZnPyb) and an electron acceptor (fullerene, F). Using methods of dissipative quantum mechanics we derive and solve a set of equations for both the diagonal and off-diagonal elements of the density matrix, which describe quantum coherent effects in energy and charge transfer. We consider two sets of parameters, one corresponding to the case where the energy-transfer reorganization energy Λ is less than the resonant coupling V between the chromophores, Λ < V , and another regime where Λ > V . For these two sets of parameters we examine the electron and exciton dynamics, with special emphasis on the short-time regime (∼ femtosec- [17], the onds). We demonstrate that, in agreement with experiments performed in Ref. excitation energy of the BPEA antenna chromophores is efficiently funneled to porphyrins (ZnPy). The excited ZnPy molecules rapidly donate an electron to the fullerene electron acceptor, thus creating a charge-separated state, ZnPy+−F−, with a quantum yield of the order of 95%. There is no observable difference in energy transduction efficiency for these two sets of parameters. In the limit of strong interchromophoric coupling, coherent dynam- ics dominates over incoherent-hopping motion. In the single-exciton regime, when one of the BPEA chromophores is initially excited, quantum beatings between two resonant BPEA chromophores occur with decoherence times of the order of 100 fs. However, here the elec- tron transfer process is dominated by incoherent hopping. For the case where one porphyrin molecule is excited at the beginning, we obtain small quantum oscillations of the fullerene charge characterized by a short decay time scale (∼ 10 fs). More pronounced quantum os- cillations of the fullerene charge (with an amplitude ∼ 0.1 electron charge and decoherence time of about 20 fs at T = 77 K) are predicted for the double-exciton regime, when both porphyrin molecules are initially excited. We also show that the contribution of wave-like coherent motion to electron-transfer dynamics could be enhanced by lowering the temper- ature, strengthening the fullerene-porphyrin bonds, shrinking the energy gap between the zinc porphyrin and fullerene moieties (e.g., by attaching a charged residue to the fullerene), as well as by decreasing the reorganization energy (by tuning the solvent polarity). Acknowledgements. FN acknowledges partial support from the Laboratory of Physical Sciences, National Security Agency, Army Research Office, DARPA, Air Force Office of Scientific Research, National Science Foundation grant No. 0726909, JSPS-RFBR contract No. 09-02-92114, Grant-in-Aid for Scientific Research (S), MEXT Kakenhi on Quantum Cybernetics, and Funding Program for Innovative Research and Development on Science 22 and Technology (FIRST). Appendix A: Coulomb interaction energies The Coulomb interactions between the electron states are, HC = −uF [(1 − ¯nZnPya)nF + (1 − ¯nZnPyb)nF] + uPy(1 − ¯nZnPya)(1 − ¯nZnPyb) + uZnPyanZnPyanZnPy∗ a + uZnPybnZnPybnZnPy∗ b , (A1) where, ¯nZnPya = nZnPya + nZnPy∗ a and ¯nZnPyb = nZnPyb + nZnPy∗ b . The first term of (A1) represents the electrostatic attraction (so the minus sign) between the positively charged ZnPy chromophores and the negatively-charged fullerene. The sec- ond term is due to the Coulomb repulsion (so the plus sign) between two ZnPy chro- mophores. The last two terms are the repulsive interaction energies when both the excited and ground states of the ZnPy chromophores are occupied by electrons. The coefficients uF, uPy, uZnPya, and uZnPya represent the magnitude of the electrostatic interactions and these are calculated using the Coulomb formula. We have assumed that the empty ZnPy chromophores (nZnPy + nZnPy∗ = 0) have positive charges and the acceptor state F becomes negatively-charged when it is occupied by an electron. Appendix B: Derivation of equations for the matrix (cid:104)ρµν(cid:105) Our derivation of the equations for the matrix (cid:104)ρµν(cid:105) is based on the exact solution for the operator ρµν = (µ(cid:105)(cid:104)ν)(t) of the system influenced only by diagonal fluctuations of the bath. In this case the "system + bath" Hamiltonian has the form mjω2 j Λµ j xjµ(cid:105)(cid:104)µ, (B1) (cid:88) Hdiag = Eµµ(cid:105)(cid:104)µ + µ j (cid:88) (cid:32) p2 j 2mj (cid:33) −(cid:88) (cid:88) µ j + mjω2 j x2 j 2 where Λµ j = Λµµ j [see Eq. (13)]. The time evolution of the exciton operators ρµν is governed by the Heisenberg equation i ρµν = − ωµνρµν + (cid:88) j 23 mjω2 j (Λµ j − Λν j )xjρµν. (B2) It is possible to verify that the solution of Eq. (B2) is given by the equation ρµν(t) = exp[iΩµν(t − t0)] × exp (cid:88) −i Ωµν = ωµν −(cid:88) exp mjω2 j j 2 j where (cid:104) i (cid:88) j pj(t)(Λµ pj(t0)(Λµ j − Λν j ) j − Λν j )  ×  ρµν(t0), j )2(cid:105) , j )2 − (Λν (Λµ and pj is the Heisenberg operator of the dissipative environment. The evolution begins at time t = t0. The diagonal operators ρµ = ρµµ are constant, ρµ(t) = ρµ(t0), in the presence of a strong interaction with the diagonal operators of the protein environment. uncorrelated when µν (t(cid:48))(cid:105) = 0, the contribution of the environment to the non-Markovian equation environment off-diagonal operators, diagonal and For (cid:104)Q(0) α (t) Q(0) (20) consists of two parts: (cid:104)−i[ρµν, He−ph]−(cid:105) = (cid:104)−i[ρµν, H diag e−ph]−(cid:105) + (cid:104)−i[ρµν, H n−diag e−ph ]−(cid:105). The diagonal elements, Qµ, of the environment contribute to the first part, (cid:104)−i[ρµν, H diag e−ph]−(cid:105) = (cid:90) t dt1(cid:104)(Q(0) (cid:90) t dt1(cid:104)Q(0) 0 0 µ − Q(0) ν )(t)Q(0) ¯ν (t1)(cid:105)(cid:104)ρµν(t)ρ¯ν(t1)(cid:105) − ν )(t)(cid:105)(cid:104)ρ¯ν(t1)ρµν(t)(cid:105), µ − Q(0) ¯ν (t1)(Q(0) (B3) (B4) (B5) (B6) (B7) whereas the second part is due to a contribution of the non-diagonal (abbreviated as n-diag in the super-index) operators, Qµν, e−ph ]−(cid:105) = − (cid:104)−i[ρµν, H n−diag να (t) Q(0) 0 (cid:90) t dt1(cid:104) Q(0) (cid:90) t dt1(cid:104) Q(0) (cid:90) t dt1(cid:104) Q(0) (cid:90) t 0 0 ¯µ¯ν (t1) Q(0) αµ(t) Q(0) ¯µ¯ν (t1)(cid:105)(cid:104)ρµα(t)ρ¯µ¯ν(t1)(cid:105) + να (t)(cid:105)(cid:104)ρ¯µ¯ν(t1)ρµα(t)(cid:105) + ¯µ¯ν (t1)(cid:105)(cid:104)ραν(t)ρ¯µ¯ν(t1)(cid:105) − αµ(t)(cid:105)(cid:104)ρ¯µ¯ν(t1)ραν(t)(cid:105). dt1(cid:104) Q(0) ¯µ¯ν (t1) Q(0) 0 We note that the time evolution of the diagonal elements of the system operator, ρµ = ρµµ, is determined by the non-diagonal operators Qµν as well as by quenching terms. Strong diagonal fluctuations of the environment have no effect on the evolution of the diagonal 24 elements of the matrix. Thus, in Eq. (B6) we assume that ρ¯ν(t1) = ρ¯ν(t), so that Eq. (B6) can be rewritten as (cid:104)−i[ρµν, H diag e−ph]−(cid:105) = −(Γdiag µν + iδΩdiag µν )(t)(cid:104)ρµν(t)(cid:105), (B8) where the time-dependent rate, Γdiag µν (t), and the frequency shift, δΩdiag µν , can be found from the following expression µν (t) + iδΩdiag Γdiag µν (t) = (cid:90) t 0 dt1 (cid:110)(cid:68) µ − Q(0) (Q(0) ν )(t)Q(0) ν (t1) (cid:69) −(cid:68) Q(0) µ (t1)(Q(0) µ − Q(0) ν )(t) (cid:69)(cid:111) .(B9) The rate Γdiag µν (t) determines the fast decay of quantum coherence in our system. For an environment composed of independent oscillators we obtain (cid:104)(Q(0) µ − Q(0) −(cid:88) ν )(t)Q(0) mjω3 j j 2 i ν (t1)(cid:105) − (cid:104)Q(0) j − Λν (Λµ j )2 coth (cid:88) mjω3 j (cid:104) (Λµ 2 j µ (t1)(Q(0) (cid:18) ωj 2T j )2 − (Λν (cid:19) µ − Q(0) ν )(t)(cid:105) = cos ωj(t − t1) − j )2(cid:105) sin ωj(t − t1). (B10) The fluctuations of the diagonal operators of the environment can be described by the set of spectral functions, together with the corresponding reorganization energies, We also introduce a spectral function, Jµν(ω), which characterizes the non-diagonal (µ (cid:54)= ν) environment fluctuations, Jµν(ω) = mjω3 j 2 Λµν j 2δ(ω − ωj), (B13) where Λµν (13) taken at µ (cid:54)= ν. With Eq. (B10) we calculate the contributions of the diagonal environment fluctuations into the decoherence rate and the frequency shift of the j = Λµν j 25 Jµ(ω) = mjω3 j j )2δ(ω − ωj), (Λµ 2 j − Λν (Λµ j )2δ(ω − ωj), ¯Jµν(ω) = λµ = (cid:90) ∞ ¯λµν = dω ω ¯Jµν(ω) = 0 (cid:88) Jµ(ω) = (cid:88) j j mjω2 j 2 mjω2 j (Λµ j )2, 2 j − Λν (Λµ j )2. (cid:88) j mjω3 j 2 (cid:88) j (cid:90) ∞ 0 dω ω (cid:88) j (B11) (B12) off-diagonal elements of the system matrix (cid:104)ρµν(cid:105) in (B8), (cid:18) ω (cid:19) (cid:90) ∞ (cid:90) ∞ 0 dω ω 0 Γdiag µν (t) = δΩdiag µν (t) = ¯Jµν(ω) coth dω sin ωt, ω [Jµ(ω) − Jν(ω)](1 − cos ωt). 2T (B14) The contribution of the non-diagonal fluctuations of the environment to the evolution of the electron operators (cid:104)ρµν(cid:105) is defined by Eq. (B7). To calculate the products of exciton variables taken at different moments of time, for example, ρµα(t)ρ¯µ¯ν(t1), we use Eq. (B3), which describes the evolution of exciton operators in the presence of strong coupling to the diagonal operators, Qµ, of the environment. We assume that the interaction with the non- diagonal environment operators, Qµν, is weak. With Eq. (B3) we express the operators at time t1 in terms of operators taken at time t: ρ¯µ¯ν(t1) = exp [−iΩ¯µ¯ντ ] exp [iu¯µ¯ν(τ )] exp [−iv¯µ¯ν(t, t1)] ρµν(t), ρ¯µ¯ν(t1) = ρµν(t) exp [−iΩ¯µ¯ντ ] exp [−iu¯µ¯ν(τ )] exp [−iv¯µ¯ν(t, t1)] , where τ = t − t1, and (cid:90) ∞ dω ¯Jµν(ω) sin ωτ, ω j − Λν j )[pj(t) − pj(t1)]. 0 uµν(τ ) = vµν(t, t1) = (Λµ (B15) (B16) (cid:88) j (cid:29) (cid:28)1 2 (cid:32)¯hωj (cid:33) 2T Here we assume that pj(t), pj(t1) are free-evolving momentum operators of the environment, which are described by Gaussian statistics with a correlation function [ pj(t), pj(t1)]+ = ¯hmjωj 2 coth cos ωj(t − t1). (B17) The operator function vµν(t, t1) does not commute with the exciton matrix ρµν(t), and, therefore, we need two expressions for the operator ρµν(t1), which are distinguished by the order of the operators ρµν(t) and exp [−ivµν(t, t1)] . For the average value of the operator exp [−ivµν(t, t1)] we obtain (cid:40) (cid:90) ∞ 0 (cid:32) ¯hω (cid:33) 2T (cid:41) [1 − cos ω(t − t1)] . (B18) (cid:104)exp [−ivµν(t, t1)](cid:105) = exp − dω ω2 ¯Jµν(ω) coth Substituting Eqs. (B15) to Eq. (B7) and using the secular approximation we obtain a contribution of the non-diagonal environment operators, Qµν, to the evolution of diagonal exciton operators (cid:104)ρµ(cid:105), e−ph ]−(cid:105) = −(cid:88) (cid:104)−i[ρµ, H n−diag (cid:88) α γαµ(t)(cid:104)ρµ(cid:105) + γµα(t)(cid:104)ρα(cid:105), (B19) α 26 characterized by the following relaxation matrix, (cid:90) t 0 γµα(t) = 0 where dt1(cid:104) Q(0) (cid:90) t αµ(t) Q(0) µα)(t1)(cid:105)e−iΩµα(t−t1)e−iuµα(t−t1)(cid:104)e−ivµα(t,t1)(cid:105) + µα)(t)(cid:105)e−iΩαµ(t−t1)eiuαµ(t−t1)(cid:104)e−ivαµ(t,t1)(cid:105), αµ(t1) Q(0) dt1(cid:104) Q(0) (cid:104) Q(0) (cid:21) αµ(t) Q(0) − 1 eiω(t−t1) + µα)(t1)(cid:105) = (1/2) (cid:20) (cid:19) (cid:18) ω coth + 1 2T (cid:90) ∞ Jαµ(ω) × (cid:27) (cid:21) e−iω(t−t1) 0 . (cid:26)(cid:20) coth (cid:18) ω (cid:19) 2T (B20) (B21) When the environment is at high temperatures (2T (cid:29) ω) and at low frequencies of the diagonal fluctuations (ωτ (cid:28) 1) we have: uµν(τ ) (cid:39) ¯λµντ, and (cid:104)exp[−ivµν(t, t1)](cid:105) (cid:39) exp[−¯λµνT (t − t1)2]. With these assumptions the relaxation matrix has a simple form (cid:40) exp (cid:34) −(ω + Ωαµ − ¯λαµ)2 4¯λαµT (cid:35) γµα = (cid:18) ω (cid:19) T + exp (cid:115) π (cid:90) ∞ (cid:34) ¯λαµ −(ω − Ωαµ + ¯λαµ)2 dω Jαµ(ω) n(ω) × (cid:35)(cid:41) exp 0 4¯λαµT , (B22) where n(ω) = [exp(ω/T ) − 1]−1 is the Bose distribution function at the temperature T . The moment of time t in the expression (B20) for the relaxation matrix is usually higher than the effective retardation time, τc ∼ (¯λαµT )−1/2, of the integrand in Eq. (B20): t (cid:29) τc. Therefore, we assume that t (cid:39) ∞, so that γµα(t) (cid:39) γµα(∞) = γµα. It follows from Eq. (B7) that a contribution of the non-diagonal environment operators Qµν to the evolution of the off-diagonal elements ρµν is given by the formula (cid:104)−i[ρµν, H n−diag e−ph ]−(cid:105) = −(Γµν + iδ Ωµν)(t)(cid:104)ρµν(t)(cid:105), (B23) where Γµν(t) + iδ Ωµν(t) = (cid:90) t 0 να (t) Q(0) dt1(cid:104) Q(0) (cid:90) t αν (t1)(cid:105)e−iΩαν (t−t1)e−iuαν (t−t1)(cid:104)e−ivαν (t,t1)(cid:105) + αµ(t1)(cid:105)e−iΩµα(t−t1)eiuµα(t−t1)(cid:104)e−ivµα(t,t1)(cid:105). dt1(cid:104) Q(0) µα(t) Q(0) 0 (B24) 27 A small frequency shift, δ Ωµν, can be hereafter ignored. The dephasing rate, Γµν, has two parts, Γµν = Γµ + Γν, where (cid:115) π (cid:90) ∞ (cid:88) (cid:34) (cid:18) ω (cid:19) −(ω − Ωµα + ¯λµα)2 dω Jµα(ω)n(ω) × (cid:35)(cid:41) ¯λµαT exp α 0 1 2 4¯λµαT T . (B25) (cid:40) (cid:34) −(ω + Ωµα − ¯λµα)2 Γµ = (cid:35) + exp exp We note that Γµ = (1/2)(cid:80) 4¯λµαT α γαµ, and Ωµν = ωµν − λµ + λν from Eq. (B4),(B12). Assuming that the environment fluctuations acting on each electron-binding site are independent and using Eq. (13) for the coefficients Λµν j (cid:88) (cid:104) JlS(ω)(cid:104)µSlν(cid:105)2 + JlM (ω)(cid:104)µMlν(cid:105)2(cid:105) Jµν(ω) = , we obtain + JF (ω)(cid:104)µnFν(cid:105)2, (B26) l where JlS(ω) = JlM (ω) = JF (ω) = j (cid:88) (cid:88) (cid:88) j j mjω3 j 2 mjω3 j 2 jlδ(ω − ωj), ¯x2 jlδ(ω − ωj), x2 mjω3 j 2 jF δ(ω − ωj). x2 (B27) The results obtained above are valid for an arbitrary frequency dependence of the spectral densities JlS(ω), JlM (ω), JF (ω). Hereafter we assume that these functions are described by the Lorentz-Drude formula characterized by a common inverse correlation time, γc = τ−1 and by a corresponding reorganization energy λlS, λlM , or λF , e.g. , c (B28) Quenching processes also contribute to the decay of the off-diagonal elements, (cid:104)ρµν(cid:105), with JlS(ω) = 2 ω2 + γ2 c . ωγc λlS π the following decoherence rates: Γquen µ + Γquen ν , where µν = Γquen l al∗α(cid:105)2χ(cid:48)(cid:48) † (cid:104)µa (cid:88) lα Γquen µ = l (ωµα) coth (cid:20) (cid:18) ωµα 2T (cid:19) (cid:21) . + 1 (B29) Here we consider an Ohmic quenching heat-bath with the spectral density χ(cid:48)(cid:48) l (ω) = αlω, which is determined by a set of site-dependent dimensionless coupling constants αl (cid:28) 1. The contribution of quenching to the relaxation of the diagonal elements of the electron matrix, (cid:104)ρµ(cid:105), is determined by the standard Redfield term γquen µν = ((cid:104)µa l al∗ν(cid:105)2 + (cid:104)νa † l al∗µ(cid:105)2)χ(cid:48)(cid:48) † l (ωµν) coth . (B30) (cid:20) (cid:18) ωµν (cid:19) (cid:21) − 1 2T (cid:88) l 28 As a result, we find that the time evolution of the off-diagonal elements of the electron matrix is determined by the expression (cid:104)ρµν(cid:105)(t) = exp ( i ωµν t − ¯λµν T t2 ) × exp (−Γµν t ) ρµν(0), (B31) with the decoherence rates Γµν = Γµ + Γν, where the coefficient Γµ contains contributions of . The evolution starts at the moment t = 0 with the initial matrix the off-diagonal fluctuations of the environment (B25) as well as quenching processes Γquen (B29): Γµ = Γµ + Γquen ρµν(0). An effect of diagonal environment fluctuations is determined by the rate where ¯λµν is the reorganization energy defined by Eq. (B12) and T is the temperature of the environment. µ µ (cid:113)¯λµν T , [1] R.E. Blankenship Molecular mechanisms of photosynthesis (Blackwell Science, Oxford, UK, 2002). [2] H. V. Amerongen, L. Valkunas, R. Van Grondelle Photosynthetic Excitons (World Scientific, Singapore, 2000). [3] G. S. Engel, T. R. Calhoun, E. L. Read, T. K. Ahn, T. Mancal, Y. C. Cheng, R. E. Blankenship and G. R. Fleming, Nature 446, 782 (2007). [4] G. Panitchayangkoon, D. Hayes, K.A. Fransted, J.R. Caram, E. Harel, J. Wen, R.E. Blanken- ship, and G.S. Engel, Proc. Natl. Acad. Sci. USA 107, 12766 (2010). [5] P. Rebentrost, M. Mohseni, I. Kassal, S. Lloyd and A. Aspuru-Guzik, New J. Phys. 11, 033003 (2009); P. Rebentrost, R. Chakraborty, and A. Aspuru-Guzik, J. Chem. Phys. 131, 184102 (2009). [6] M.B. Plenio and S.F. Huelga, New J. Phys. 10, 113019 (2008). [7] A. Ishizaki and G. R. Fleming, Proc. Natl Acad. Sci. USA 106, 17255 (2009). [8] Y-C. Cheng and G. R. Fleming, Annu. Rev. Phys. Chem. 60, 241 (2009). [9] M. Sarovar, A. Ishizaki, G.R. Fleming, and K.B. Whaley, Nature Physics 6, 462 (2010). [10] E. Collini and G.D. Scholes, Science 323, 369 (2009). [11] J. Barber, Chem. Soc. Rev. 38, 185 (2009). [12] A. W. D. Larkum, Current Opinion in Biotechnology 21, 271 (2010). 29 [13] G. Steinberg-Yfrach, P. A. Liddell, S. C. Hung, A. L. Moore, D. Gust, and T. A. Moore, Nature 385, 239 (1997). [14] G. Steinberg-Yfrach, J. L. Rigaud, E. N. Durantini, A. L. Moore, D. Gust, T. A. Moore, Nature 392, 479 (1998). [15] G. Kodis, Y. Terazono, P. A. Liddell, J. Andr´easson, V. Garg, H. Hambourger, T. A. Moore, A. L. Moore, and D. Gust, J. Am. Chem. Soc. 128, 1818 (2006). [16] D. Gust, T. A. Moore and A. L. Moore, Acc. Chem. Res. 42, 1890 (2009). [17] Y. Terazono, G. Kodis, P. A. Liddell, V. Garg, Andr´easson J, Garg V, T. A. Moore, A. L. Moore, and D. Gust, J. Phys. Chem. B 113, 7147 (2009). [18] H. Imahori, H. Yamada, Y. Nishimura, I. Yamazaki, Y. Sakata, J. Phys. Chem. B 104, 2099 (2000). [19] P. K. Ghosh, A. Yu. Smirnov, and F. Nori, J. Chem. Phys. 131, 035102 (2009). [20] A. Yu. Smirnov, L. G. Mourokh, P. K. Ghosh, and F. Nori, J. Phys. Chem. C 113, 21218 (2009). [21] A.J. Leggett, S. Chakravarty, A.T. Dorsey, M.P.A. Fisher, A. Garg, W. Zwerger, Rev. Mod. Phys. 59, 1 (1987). [22] A. Yu. Smirnov, L. G. Mourokh, and F. Nori, J. Chem. Phys. 130, 235105 (2009). [23] Y. Tanimura, J. Phys. Soc. Jpn. 75, 082001 (2006). [24] G.F. Efremov and A.Yu. Smirnov, Sov. Phys. JETP 53, 547 (1981); G.F. Efremov, L.G. Mourokh, and A.Yu. Smirnov, Phys. Letters A 175, 89 (1993); A.Yu. Smirnov, Phys. Rev. B 68, 134514 (2003). [25] A. Ishizaki and G. R. Fleming, J. Chem. Phys. 130, 234111 (2009). [26] W. M. Zhang, T. Meier, V. Chernyak, and S. Mukamel, J. Chem. Phys. 108, 7763 (1998). [27] M. Yang and G.R. Fleming, Chem. Phys. 282, 163 (2002). [28] Y. Terazono, G. Kodis, P. A. Liddell, V. Garg, M. Gervaldo, T. A. Moore, A. L. Moore, G. Gust, Photochem. Photobiol. 83, 464469 (2007). 30
1511.03790	2	1511	2016-07-13T06:06:46	Velocity condensation for magnetotactic bacteria	[ "physics.bio-ph", "cond-mat.soft", "cond-mat.stat-mech", "physics.data-an" ]	Magnetotactic swimmers tend to align along magnetic field lines against stochastic reorientations. We show that the swimming strategy, e.g. active Brownian motion versus run-and-tumble dynamics, strongly affects the orientation statistics. The latter can exhibit a velocity condensation whereby the alignment probability density diverges. As a consequence, we find that the swimming strategy affects the nature of the phase transition to collective motion, indicating that L\'evy run-and-tumble walks can outperform active Brownian processes as strategies to trigger collective behavior.	physics.bio-ph	physics	Velocity condensation for magnetotactic bacteria Jean-Fran¸cois Rupprecht,1, 2 Nicolas Waisbord,3 Christophe Ybert,3 C´ecile Cottin-Bizonne,3 and Lyd´eric Bocquet1, ∗ 1Ecole Normale Sup´erieure, Laboratoire de Physique Statistique, UMR CNRS 8550, 24 rue Lhomond, Paris (France). 2Mechanobiology Institute, National University of Singapore, Singapore. 3Institut Lumi`ere Mati`ere, UMR CNRS 5306, Universit´e Lyon 1, Lyon (France). (Dated: October 7, 2018) Magnetotactic swimmers tend to align along magnetic field lines against stochastic reorientations. We show that the swimming strategy, e.g. active Brownian motion versus run-and-tumble dynamics, strongly affects the orientation statistics. The latter can exhibit a velocity condensation whereby the alignment probability density diverges. As a consequence, we find that the swimming strategy affects the nature of the phase transition to collective motion, indicating that L´evy run-and-tumble walks can outperform active Brownian processes as strategies to trigger collective behavior. Bacteria, spermatozoa or algae have in common the ability to propel themselves in low-Reynolds fluids in or- der to explore space [1, 2]. The directed motion of these swimmers is always affected by stochastic impulses due to noise in the propulsion mechanism. Swimmers under- going white noise perturbations, which lead to persistent small-amplitude fluctuations of the orientation [3, 4], are usually called active Brownian particles (ABPs). In con- trast, Bacteria like E. Coli exhibit sudden reorientations of their velocity vector (called tumbles) which are due to stochastic switches in the direction of rotation of pro- pelling flagella [5]. Such dynamics are usually coined as Run-and-Tumble (RT). Though ABPs and RTs corre- spond to two different swimming strategies, in the ab- sence of external torques, their long-time dynamics are similar and lead in both cases to an effective diffusion process [2]. Biological microswimmers can also orient themselves in response to external stimuli, either of chemical or me- chanical nature. In particular, the RT walk is essentially thought to provide a mean to move along chemical gra- dients, called chemotaxis, in which the run duration is modulated with respect to the direction of the stimulus. Other micro-organisms have also developed the ability to orient their propelling direction under external mechani- cal fields, for example under gravity (gravitaxis) or shear or flow gradients (gyro- and rheo- taxis) [6 -- 9]. Similar behavior have been recently reproduced with artificial catalytic swimmers [10 -- 13]. Here we consider the dynamics of swimmers driven un- der magnetic torques, keeping in mind that results gen- eralize to a larger class of mechanical torques. A repre- sentative example are magnetotactic bacteria (MB) that behave as self-propelled compasses, due to iron-based or- ganelles orienting the propelling flagella along the mag- netic field lines. Since their discovery in 1975, theoret- ical studies of MB focused on the case of white noise perturbations on the orientation [14, 15]. Recent work also demonstrated how superparamagnetic beads could be attached to E-Coli bacteria, making them reactive to magnetic fields [16]. Here we show that in the presence of an external align- ing field, the orientation distribution strongly differs for the two swimming strategies, ABPs and RTs. For RTs, we report a velocity condensation phenomenon which is associated with a divergence of the orientation distribu- tion function in the direction of the field and which occurs above a critical magnetic field. We point that the result- ing behavior is significantly different from a chemotactic response. In the final paragraph, we consider the onset of the collective phase of a swarm with nematic interac- tions. We show that the nature of the alignment diver- gence shapes the phase diagram of the isotrope-nematic transition. √ The ABPs dynamic is a diffusion process on the di- rection of the velocity vector V , with a fixed speed V = V0 [3, 4, 14]; hence the dynamics of the align- ment angle θ is described by an Ito equation: dθ = 2dtDr ζ, where ζ is a Gaussian white noise f (θ)dt/τB + with (cid:104)ξ(t)ξ(t(cid:48))(cid:105) = δt,t(cid:48) and Dr is a rotational diffusion coefficient, and f (θ) = − sin(θ) is the magnetic torque. The magnetic relaxation time τB can be expressed as τB = ξ0/(mBa), where m is the magnetic moment and ξ0 is a rotational drag coefficient. The stationary prob- ability distribution for θ corresponds to the Boltzmann statistics P∞(θ) = µ(θ) exp [1/(τBDr) cos θ] /Zd, (1) where Zd a dimension dependent normalization factor and µ(θ) is the uniform probability measure (µ(θ) = 1/π in 2D or sin(θ)/2 in 3D). In 3D, the mean velocity Vz = (cid:104)cos θ(cid:105) reduces to Vz = V0f (1/(τBDr)), where f (x) = coth x − x−1 is the classical Langevin function [14]. Indeed, Eq. (1) corresponds to the distribution of a passive magnet in a thermal noise [15], with an effective temperature defined as kBTeff = Drξ0. From an experimental point of view, MB bacteria usu- ally behave as ABP particles in standard chemical envi- ronment (see [18], [19] and Fig. 1a. c. below). However, it has been recently reported that some specific environ- ments can trigger non-Brownian reorientations of MB, as demonstrated on the strains MO-1 [17] and MC-1 [19]. 6 1 0 2 l u J 3 1 ] h p - o i b . s c i s y h p [ 2 v 0 9 7 3 0 . 1 1 5 1 : v i X r a In particular, it has been shown in [19] that the trajec- tory of the bacteria presents sudden changes of direction when the concentration in growing medium is reduced (see Fig. 1b). These kinks are then detected by a track- ing algorithm, which show that the run durations are exponentially distributed (see [19]). Using the experi- mental data reported in [19], we build the experimental histogram presented in Fig. 1d from 500 trajectories. The histogram is peaked in the direction of the magnetic field (θ = 0) while maintaining a substantial statistical weight for the anti-parallel orientation (θ = π). These two features can not be consistently accounted by an ABP model, which fails to match the statistical weights both in the parallel and in the anti-parallel directions of the magnetic field, as highlighted see in Fig. 1d). This inconsistency of the ABP model to reproduce experimen- tal results calls for a shift from the classical Langevin paradigm [14], which can only describe accurately the behavior of magnetotactic bacteria in a medium favor- able to growth. As indicated in [19], this change in the behavior of magnetotactic bacteria in a lesser favorable environment could be related to an evolutionary advan- tage. 2 Figure 1: (color online) Trajectories of MB in (a) rich growing medium and (b) poor growing medium environments. (c -- d) Alignment angle distribution at Ba = 7 · 10−5 T in (c) rich growing medium and (d) poor growing medium environments: (black crosses) experimental histogram and (solid blue line) best ABP fit with (a) B/D⊥ = 11 and (b) B/D⊥ = 4.5. In (d), there is a 10% discrepancy between the experimental and fitted cumulative distributions (i.e. Kolmogorov-Smirnov norm [20]). (Insets) The anti-parallel response is enhanced in a poor environment (see [19]). RT walk -- The RT dynamics is composed of runs at a fixed speed V0 interrupted by instantaneous reorienta- tions. During runs under a magnetic field Ba, the evolu- tion of the alignment angle θ (between V and the applied magnetic field Ba) is deterministic, with θ = f (θ)/τB. Furthermore, we assume that the duration of each run, x, (i) is drawn according to a given probability density ρ(x/τr), (ii) is independent of the previous runs (iii) is independent of the alignment direction θ (in contrast to chemotaxis). After a tumble, the alignment angle θ0 is drawn according to the uniform probability measure µ(θ0). We finally define the magnetotactic dimensionless parameter B as B = τr , (2) mBa ξ0 where τr is the mean run time, so that B = τr/τB. Remarkably, the estimated values for the magnetotac- tic number B appear to be of the order unity for MB in typical geomagnetic fields (see [19] and [15]). θ1 Angular distribution and velocity condensation -- . We seek to obtain the expression for the stationary probabil- ity P∞(θ) density for the RT walk, defined so that the probability that the angle θ belongs to the interval [θ1, θ2] P∞(θ)dθ. Partitioning on successive events, we 0 dθ0 µ(θ0) δ(θ − θt(θ0)), reads(cid:82) θ2 find that P∞(θ) = (cid:82) ∞ 0 dt π(t)(cid:82) π where π(t) is the distribution of 'run' time since the last tumble, and which is to be determined from the dis- tribution of run duration ρ(x) by renewal process the- ory [1]; θ0 is the outgoing angle after the tumble; and θt(θ0) is the time-dependent evolution operator. For a torque with angular dependence f (θ), the latter is for- mally defined as θt(θ0) = F (−1)[F [θ0] + Bt], with F a primitive of 1/f and F (−1) the reciprocal function of F . Using that the function θ0 → θ − θt[θ0] is canceled for θ0 = θ∗ 0(θ, t) = F (−1)[F [θ] − Bt], one gets dt π(t)(µ.f )[θ∗ P∞(θ)f (θ) = (cid:90) ∞ 0(θ, t)]. (3) 0 Equation (3) holds for an arbitrary torque f (θ) and spent time distribution π(t). We first consider a magnetic torque f (θ) = − sin(θ) and ρ(t) = exp(−t) (exponential RT). The spent time distribution is then π(t) = ρ(t) [1]. Following the previ- ous definition, we obtain θ∗ 0 = 2 arctan[tan(θ/2) exp(Bt)]. In 2D, we apply Eq. (3) and we obtain (cid:90) π (tan θ/2)1/B sin θ θ P∞(θ) = 1 B dφ µ(φ) (tan φ/2)1/B (4) A key feature which emerges from Eq. (4) is that the low- θ behavior drastically differs above and below the critical value Bc = 1 (see Fig. 2a). For B < Bc, P∞(θ) takes a finite value at θ = 0. However, for B > 1 we find that P∞(θ) ∼ θ→0 γ−1 d θ−(1−1/B), (5) where γ2 = 21/BB cos (π/(2B)) in 2D and γ3 = 21/B+1B2 sin(π/(2B))/π in 3D. At B = Bc a dynami- cal transition occurs above which the probability density diverges -- a property that we call the velocity condensa- tion. Extension and robustness -- We first remark that the condensation phenomenon is maintained for alternative aligning torques, provided the torque is strong enough abcdBBaround θ = 0. Consider that f (θ) ∼ −θn for θ (cid:28) 0, then: (i) if n > 1 and for exponential RT, the torque term is too weak for the velocity condensation to occur, and (ii) if n < 1, the condensation always occurs, as the θ = 0 state can be attained after a finite run time. Sec- ond, the strict mathematical divergence disappears in the presence of a rotary Brownian noise on the velocity orien- tation during runs, characterized by a diffusion coefficient r. As θ → 0, the diffusive noise eventually dominates D(cid:48) over the vanishing torque and the orientation probabil- ity scales as P∞(θ) ∝ exp(−θ2Bτr/(2D(cid:48) r)) in the region θ ∈ [0, D(cid:48) r/(τrB)]). However, provided that the rotary diffusion coefficient noise is relatively small (D(cid:48) r/τr < 1), the probability density of RTs is sharply peaked when B > 1. RT fit of experiments -- We now compare the predic- tion of the RT model to the experimental histogram pre- sented in 2b. In contrast to the ABP model fit, which can not account for the sharp peak in the orientation dis- tribution without underestimating it in the anti-parallel directions, the RT walk provides the appropriate statis- tical weight to both the parallel and the anti-parallel di- rections. The increase in the quality of the fit can be measured through the Kolmogorov-Smirnov norm [20], which quantifies the discrepancy between the cumulative distributions. The quality of the fit is increased by con- sidering a RT walk in which runs are perturbed by a mild rotary noise (D(cid:48) r/τr = 0.15). We conclude that, in spite of this small rotary diffusion that affects the orientation of bacteria, the distribution in Fig. 2 is still very sharply peaked in the direction of the magnetic field. We conclude that the RT walk is more efficient than the ABP process to sample both the parallel and anti- parallel directions to the magnetic field. Our intuitive explanation is that, in contrast to a diffusion process, all orientations are sampled after a tumble, and in particular the anti-parallel directions to the magnetic field. Mean velocity and diffusion -- From the orientation dis- tribution, we can calculate the mean velocity of the RTs. The averaged velocity in the direction of the magnetic field is defined as Vz = V0 (cid:104)cos θ(cid:105). Using the previous expressions for the distribution function, one gets Vz = V0 × 1 B dw w1/B−1gd(w). (cid:90) 1 (cid:1)(cid:9) /B − 1, with ψ0(z) = (6) 0 2B 2B (cid:1) − ψ(0)(cid:0) 1 (cid:8)ψ(0)(cid:0) B+1 g3(w) = (cid:0)1 − w4 + 4w2 log(w)(cid:1) /(cid:0)w2 − 1(cid:1)2 In 2D, g2(w) = (1 − w)/(1 + w) and Vz/V0 = Γ(cid:48)(z)/Γ(z) and Γ(z) is the Gamma function [22]. In 3D, and Eq. (6) reads Vz/V0 = ψ(1) (1/(2B)) /(2B2)−1/B−1, where ψ1(z) stands for the derivative of ψ0(z) [22]. This result is plotted in Fig. 2 against the results obtained for ABP in terms of the Langevin equation. Interestingly, there is no strong signature of the onset of the velocity condensation on the mean velocity. Furthermore, while the expression 3 Figure 2: (color online) (a) Probability density P∞(θ) for RTs (2D) for (red solid line) B = 4 and (blue circle line) B = 0.25. (Inset) The distribution in the opposite direction to the magnetic field remains comparable for both B = 0.25 and B = 4. (b) Fit of the experiments by the RT model: (black crosses) experimental histogram; (positive side, red solid line) RT fit with B = 2.3; (negative side, magenta solid line) RT fit with a mild rotary diffusion with D(cid:48) r/τr = 0.15 and B = 3.2; (both sides, blue solid line) ABP fit with B/D⊥ = 4.5. The Kolmogorov-Smirnov error is of 7% for the pure RT model and of 3% for the perturbed RT model. Figure 3: (color online) Forward mean velocity Vz in 3D, nor- malized by the velocity norm V0: (a) (red line) exact expres- sion for the velocity of a RT with an average step duration τr; (blue solid line) ABP with Dr = 1/τr; (blue dashed line) ABP with Dr = 1/(2τr). (b) RT L´evy walks case (ρ(x) ∼ x−β, β ≤ 3): (red crosses) β = 2.30 and (blue circles) β = 3.30. Inset: log-scale behavior at θ (cid:28) 1. Inset: in the limit B (cid:28) 1, Vz scales as Bξ (black line), with ξ = β − 2 = 0.30 for β = 2.30 < 3 and ξ = 1 for β = 3.30 ≥ 3. for Vz differs from the Langevin prediction for ABPs, we observe that for a general B, the curve of the RT velocity lies in between those for ABPs with Dr = 1/(2τr) and Dr = 1/(τr) (see Fig. 7a). We obtain similar conclusions concerning the diffusion coefficient D⊥ in the transverse direction (xOy) (see [23]), suggesting that, in spite of the velocity condensation, RTs are as efficient as ABPs in exploring their environment. Chemotaxis -- Chemotaxis refers to an angular depen- dence in the mean duration of a run [5]. We first consider the parallel chemotaxis case τr(θ) = 1 + χ cos(θ), which favors runs in the direction of the magnetic field when χ > 0. In units of τr, the Fokker-Planck equation reads [27] B∂θ(sin θP ) − (1 + χ sin θ)P (θ) = − (1 + χ(cid:104)P sin θ(cid:105)) , abab−πdu P (φ) sin(φ). We show that the (cid:90) π θ dφ where(cid:10)P(cid:12)(cid:12) sin θ(cid:11) =(cid:82) π solution reads sin(φ)−χ/B tan(φ/2)1/B , tan(θ/2)1/B sin(θ)1−(χ/B) P (θ) = γ for θ > 0, where γ = 1 + χ(cid:10)p(cid:12)(cid:12) sin θ(cid:11). The constant (cid:10)p(cid:12)(cid:12) sin θ(cid:11) is found as a solution of the self-consistency equation: (cid:10)p(cid:12)(cid:12) sin θ(cid:11) = (cid:10)fp (cid:12)(cid:12)c(cid:11)). From Eq. (cid:12)(cid:12)c(cid:11)/(λ0 − (cid:10)fp (7) (7), we find that a positive parallel chemotaxis lowers the value of the critical magnetotatic constant above which the velocity condensation occurs, as Bc = 1 − χ. In contrast, a transverse chemotactic field, as defined by τr(θ) = 1 + χ sin(θ), will not change the value Bc = 1 (see SI). L´evy walks -- We show that the velocity condensation phenomenon is further amplified for systems exhibiting a L´evy statistics of the run period. L´evy walks are char- acterized by heavy -- tailed distribution of run duration: ρ(x) = 1x>1(β − 1)/xβ with 2 < β < 3. The value β = 2 corresponds to a predicted optimal search strategy [28]. To apply Eq. (3), we notice that π(t) = (β − 2)/(β − 1)t1−β, when t > 1 [1], and that the function t → sin(θ(cid:63) 0) is sharply peaked around t∗ = − log(tan(θ/2))/B. We find that the velocity condensation occurs for any β > 2 as P∞(θ) ∼ θ→0 γ Bβ−1(β − 2) (β − 1) 1 θ(log(1/θ))β−1 , (8) where γ = 0.46 . . . both in 2D and 3D. This expres- sion corresponds to an enhanced condensation compared to exponentially-distributed runs. The mean velocity is found in terms of an expression analogous to Eq. (6). We truncate the functions gd(w) to its first order expansion at w = 1 to obtain: Vz/V0 ∼ B→0 γd Γ(3 − β) β − 1 Bβ−2, (9) where γ2 = 1/2 and γ3 = 2/3. The non-analytical scal- ing Bβ−2 in Eq. (9) corresponds to a highly sensitive directional response at the onset of the stimulus detec- tion (from B = 0 to B > 0). In comparison, the velocity Vz is proportional to B when B (cid:28) 1 for ABPs as well as for RTs with a finite second moment for the run duration (e.g. β ≥ 3 in Fig. 7b). Collective behavior -- We finally consider the conse- quence of the velocity condensation on the collective be- havior of a swarm [29 -- 36]. We adapt the Maier-Saupe mean-field treatment for a highly concentrated swarm of interacting self-propelled rods [29] (see also [30 -- 32]). In contrast to previous results which assumed a Boltzmann distribution for the orientation distribution Eq. (1), we use here the precise statistic of the orientation from Eq. (3). Following [29, 32], we consider that interactions be- tween bacteria result in an effective torque acting uni- formly on each bacterium: f (θ) = −U0A[P∞] sin(2θ), 4 Figure 4: (color online) Phase diagrams of the order parame- ter S∗ in 3D in terms of the interaction strength U0, and for several values of B (see Eq. (2)): (a) RTs with exponentially- distributed runs, (b) L´evy walks with β = 2.6: S∗ > 0 for U0 > 0 even at B = 0 (blue circle curve), hence the isotropic phase is intrinsically unstable. (cid:82) π 0 du cos(2u)P∞(u) in 2D and A[P∞] =(cid:82) π where U0 is the interaction strength and A measures of the local nematic order and is defined as A[P∞] = 0 du (3 cos2 u− 1)P∞(u) in 3D. Using Eq. (3), we compute the prob- ability distribution P∞(S) that corresponds to an im- posed value S = A[P∞]. The order parameter, denoted S(cid:63), is then found as the solution of the following self- consistency equation: S(cid:63) = A[P∞(S(cid:63))]. For RTs with exponential runs, the isotropic phase (S(cid:63) = 0) is desta- bilized above a critical value of the interaction strength 0 > 1.87.τ−1 U (c) in favor of the nematic phase. The phase diagram is alike for ABP swimmers [32]. However, the probability distribution diverges in both directions θ = 0 and θ = π for RTs within the nematic phase (3D). Within Onsager's theory, the quantity 1/U (c) can be interpreted as an excluded volume induced by steric interactions [37]. For RT L´evy walks, the phase diagram is drastically changed due to destabilization of the isotrope phase (see Fig. 4). Indeed, the order parameter should satisfy by the following self-consistency equation: 0 r S(cid:63) = γd Γ(3 − β) β − 1 (U0S(cid:63))β−2, S(cid:63) (cid:28) 1, (10) (10), Eq. where γ3 = (2β − 4)/5 in 3D. Due to the behavior of the (9) displays probability distribution in Eq. an non-analytical behavior with S(cid:63) (cid:28) 1. This Sβ−2 behavior implies that a solution S(cid:63) > 0 necessarily exists for any value of the interaction strength U0 > 0. Hence, we find that the isotropic phase is intrinsically unstable, as U (c) 0 = 0, which corresponds to a diverging excluded volume. Similar conclusions can be drawn in 2D (γ2 = 1/2 in Eq. (10)). Experimentally, it appears that bacteria [38, 39] and immune cells [40] can perform L´evy walks with an ex- ponent β < 3, which is within our predicted regime of a high sensitivity at the onset of the stimulus detection and to collective motion (see Eqs. 9 and 10). Conclusion -- In this paper, we exhibit a divergence in the orientation response of the RT walk under torque. abisotropenematicnematicExp. RTLévy RTnematicThis divergence is required to account the high direc- tional response of tumbling magnetotactic swimmer, even when perturbed by a Brownian rotary noise (see 2b). Experiments on MC -- 1 bacteria confirm the observation that tumbling bacteria exhibit a stronger parallel or anti- parallel response to the magnetic field, which cannot be described by the standard ABP model. Based on our an- alytical expressions for the orientation distribution, we find that the noise statistic has a crucial impact on the onset of collective motion. The fact that for L´evy runs, the transition occurs in the limit of an infinite excluded volume hints at a possible extension of Onsager's the- ory [37] in terms of a dynamical excluded volume that depends on the noise statistic. We thank J. Prost for suggesting the idea of a dynam- ical excluded volume. We also thank Fran¸cois Detchev- erry for fruitful discussions. NW was supported by the AXA fond. ∗ Electronic address: [email protected] [1] E. M. Purcell, American Journal of Physics 45, 3 (1977). [2] M. E. Cates, Reports on Progress in Physics 75, 042601 (2012), arXiv:1208.3957 . [3] B. ten Hagen, S. van Teeffelen, and H. Lowen, Journal of physics. Condensed matter : an Institute of Physics journal 23, 194119 (2011), arXiv:1005.1343 . [4] P. Romanczuk, M. Bar, W. Ebeling, B. Lindner, and L. Schimansky-Geier, The European Physical Journal Special Topics 202, 1 (2012). [5] H. C. Berg, ed., PloS one, Biological and Medical Physics, Biomedical Engineering, Vol. 7 (Springer New York, New York, NY, 2004) p. 133. [6] W. M. Durham, E. Climent, and R. Stocker, Physical Review Letters 106, 1 (2011). [7] D.-P. Hader, R. Hemmersbach, and M. Lebert, Gravity and the Behavior of Unicellular Organisms (Cambridge University Press, 2005) p. 258. [8] K. Fukui and H. Asai, Biophysical journal 47, 479 (1985). [9] X. Garcia, S. Rafaı, and P. Peyla, Physical Review Letters 5 J. Ruan, D. Murat, C. L. Santini, T. Song, T. Kato, P. No- tareschi, Y. Li, K. Namba, A. M. Gu´e, and L. F. Wu, Environmental Microbiology Reports 6, 14 (2014). [18] R. B. Frankel, R. P. Blakemore, and R. S. Wolfe, Science (New York, N.Y.) 203, 1355 (1979). [19] N. Waisbord, C. Lef`evre, L. Bocquet, C. Ybert, and C. Cottin-Bizonne, arxiv 1603.00490 (2016). [20] F. J. Massey, Journal of the American Statistical Asso- ciation 46, 68 (1951). [1] W. Feller, Wiley Series, Vol. 2 (1968) p. 509. [22] I. M. Ryzhik and I. S. Gradstein, Tables of Series, Prod- ucts and Integrals (1957) p. 438. [23] See Supplemental Material below, which includes Refs [[2 -- 4]] . [2] M. Schienbein and H. Gruler, Bulletin of Mathematical Biology 55, 585 (1993). [3] C. W. Gardiner, Handbook of Stochastic Methods for Physics, Chemistry and Natural Sciences (Springer, 2004). [4] R. Balagam and O. A. Igoshin, PLoS Computational Bi- ology 11, 1 (2015), . [27] R. N. Bearon and T. J. Pedley, Bulletin of mathematical biology 62, 775 (2000). [28] G. M. Viswanathan, S. V. Buldyrev, S. Havlin, M. G. E. da Luz, E. P. Raposo, and H. E. Stanley, Nature 401, 911 (1999). [29] M. Doi and S. F. Edwards, The theory of polymer dy- namics (Clarendon Press, 1988). [30] F. Bolley, J. a. Canizo, and J. a. Carrillo, Applied Math- ematics Letters 25, 339 (2012), . [31] P. Degond, A. Frouvelle, 10.1007/s00332-012-9157-y, . and J.-G. Liu, (2011), [32] B. Ezhilan, M. J. Shelley, and D. Saintillan, Physics of Fluids 25, 1 (2013). [33] E. Bertin, A. Baskaran, H. Chat´e, and M. C. Marchetti, Phys. Rev. E 92, 042141 (2015), . [34] F. Peruani, J. Starruss, V. Jakovljevic, L. Søgaard- Andersen, A. Deutsch, and M. Bar, Physical Review Let- ters 108, 1 (2012), . [35] A. Baskaran and M. C. Marchetti, Physical Review E - Statistical, Nonlinear, and Soft Matter Physics 77, 1 (2008), . [36] A. Baskaran and M. C. Marchetti, Physical Review Let- ters 101, 1 (2008), . [37] L. Onsager, Annals of the New York Academy of Sciences 51, 627 (1949). 110, 1 (2013), arXiv:arXiv:1301.2431v1 . [38] E. Korobkova, T. Emonet, J. M. G. Vilar, T. S. Shimizu, and P. Cluzel, Nature 428, 574 (2004). [39] G. Ariel, A. Rabani, S. Benisty, J. D. Partridge, R. M. Harshey, and A. Be'er, Nature Communications 6, 8396 (2015). [40] T. H. Harris, E. J. Banigan, D. A. Christian, C. Kon- radt, E. D. Tait Wojno, K. Norose, E. H. Wilson, B. John, W. Weninger, A. D. Luster, A. J. Liu, and C. A. Hunter, Nature 486, 545 (2012). [10] J. Palacci, S. Sacanna, A. Abramian, J. Barral, K. Han- and P. M. Chaikin, son, A. Y. Grosberg, D. J. Pine, Science Advances 1, e1400214 (2015). [11] B. Hagen, F. Ku, R. Wittkowski, D. Takagi, H. Lo, and C. Bechinger, (2014), 10.1038/ncomms5829. [12] A. Zottl and H. Stark, Physical Review Letters 108, 1 (2012), arXiv:arXiv:1201.0629v1 . [13] W. Gao, D. Kagan, O. S. Pak, C. Clawson, S. Cam- puzano, E. Chuluun-Erdene, E. Shipton, E. E. Fullerton, L. Zhang, E. Lauga, and J. Wang, Small 8, 460 (2012). [14] R. P. Blakemore, Annual review of microbiology 36, 217 (1982). [15] R. Nadkarni, S. Barkley, and C. Fradin, PLoS ONE 8 (2013). [16] M. M. van Oene, L. E. Dickinson, F. Pedaci, M. Kober, D. Dulin, J. Lipfert, and N. H. Dekker, Physical Review Letters 114, 1 (2015). [17] S. D. Zhang, N. Petersen, W. J. Zhang, S. Cargou, We recall that P∞(θ) =(cid:82) ∞ 0 dt π(t)(cid:82) π argument of the delta function is canceled for θ0 = θ∗ the reciprocal function of F . We finally obtain: Supplemental Material 6 0 dθ0 µ(θ0) δ(θ − θt(θ0)), which corresponds to Eq. (2) in the main text. The 0(θ, t) = F (−1)[F [θ] − Bt], with F a primitive of 1/f and F (−1) (cid:90) ∞ P∞(θ) = 1 sin θ 0 dt π(t) sin(θ∗ 0(θ, t)). We recall the identity sin(θ∗ 0) = 2eBt tan (θ/2) 1 + e2Bt tan2 (θ/2) . The change of variable Bt = F (θ) leads to P∞(θ) = − 1 Bf (θ) (cid:90) π θ µ(φ)dφ π ({F [θ] − F [φ]} /B) . The expression Eq. (13) lead to the identities of Eqs. (5) and (6) in the main text. Exponentially distributed runs Let us define the function φ(v) = e−ve2Bv/(cid:0)1 + u2 e2Bv(cid:1)2 Approximate expressions for the probability distribution , with u = tan(θ/2), which corresponds to the integrand (11). We first notice that the function φ exhibits a maximum for a positive v = v(cid:63) for B > 1/2 and in Eq. u2 < (2B + 1)/(2B − 1). Otherwise the function φ decays smoothly to zero. Hence we distinguish two cases that will leads to two different approximation schemes. Large B: B > 1/2 and u2 < (2B + 1)/(2B − 1) For B → ∞, the function is peaked around this maximum at v(cid:63). One calculates exp(2Bv(cid:63)) = (2B−1)/(2B+1)×1/u2 and v(cid:63) = 1/(2B) × log[(2B − 1)/(2B + 1)1/u2]. We write φ(v) = exp[S[v]] and we expand S(v) around its maximum at v(cid:63): (11) (12) (13) (14) (15) (16) with 1/σ2 = − d2 dv2 [log[φ(v)]]v=v(cid:63) = (4B2 − 1)/2. The integral over v yields S(v) = log[φ(v(cid:63))] − 1 (cid:90) ∞ dv φ(v) (cid:39) φ(v(cid:63)) where 0 φ(v(cid:63)) = (cid:18) 4B2 − 1 B2 2σ2 (v − v(cid:63))2 + O((v − v(cid:63))3) (cid:90) +∞ (cid:19)(cid:18) 2B + 1 √ = φ(v(cid:63)) − 1 2σ2 (v−v(cid:63))2 −∞ dv e 1 2B − 1 16(tan θ)2−1/B , 2πσ2, (cid:19)1/(2B) × (cid:17)− 1 2B (cid:16) (cid:1) tan1/B(cid:0) θ 2 sin(θ) From Eq. (15), we obtain the following approximate expression for the distribution P∞(θ), which is expected to be exact in the limit B (cid:29) 1 and u < 1, √ √ π 4B2 − 1 1 − 2 2B+1 P∞(θ) (cid:39) (17) with u = tan(θ/2. As visible on Fig. 6, Eq. (19) is efficient for B (cid:29) 1 and u < 1. For B (cid:29) 1, Eq. (17) has the same behavior as the expression from Eq. (5) in the main text: P∞(θ) ∼ Lθ−1 with a prefactor L = π/2 = 1.13 which is comparable to the exact value L = 1 presented in the main text (see inset of Fig 6). 4B2 √ 7 Figure 5: (color online) Stationary probability for the alignment angle θ, for B = 4 (red solid line) and B = 0.25 (blue solid line): (left) 3D case, with (black crosses) the high -- B approximate expression from Eq. (17) and (magenta crosses for B = 4, blue circles for B = 0.25) the low -- B approximate expression from Eq. (19) (right) 2D case, with (black crosses) the high -- B approximate expression from Eq. (20) and (magenta crosses for B = 4, blue circles for B = 0.25) the low -- B approximate expression from Eq. (21) Small B: B < 1/2 or {B > 1/2 and u2 > (2B + 1)/(2B − 1)} In the case B < 1/2 or {B > 1/2 and u2 > (2B + 1)/(2B − 1)}, the function φ(v) decays smoothly to zero -- exponentially for a large v. We make the simplifying substitution that φ(v) ≈ φ(0) exp[−γt], 0 dv φ(v) (cid:39) φ(0)2/φ(cid:48)(0), and we obtain that: with γ = −φ(cid:48)(0)/φ(0). Under this assumption,(cid:82) ∞ As visible on Fig. 6, the approximate expression Eq. (19) works best for B (cid:28) 1/2 and θ → π. P∞(θ) (cid:39) sin θ 2 1 1 − 2B cos θ . Equations in 2D In 2D, the analogous equation to Eq. (17) is for low B (cid:29) 1 (cid:115) (cid:18) tan2 (cid:18) θ 2 (cid:19) (cid:19)(cid:32) + 1 (cid:1)(cid:33) B−1 2B , (B + 1) tan2(cid:0) θ B − 1 2 1 + B B2 P∞(θ) (cid:39) 1 π 2(B2 − 1) and for B (cid:28) 1, the analogous equation to Eq. (19) is P∞(θ) (cid:39) B 1 π (1 − B cos θ) . As visible on Fig. 6, the approximate expression Eq. (21) works best for B (cid:29) 2 and and θ → π. Projected angle (18) (19) (20) (21) (22) (23) In tracking experiments, the accessible information is often limited to a projection of trajectories within the 2D focal plane. The observed alignment angle ψ in the focal plane is related to the alignment angle θ by the relation tan(ψ) = tan(θ) sin(φ), where φ is the azimutal direction (see Fig. 6). The probability distribution P∞(ψ) for ψ is (cid:90) 2π (cid:90) π P∞(ψ) = dφ dθ P∞(θ)δ(ψ − arctan(tan(θ) sin(φ))). Simplification of the δ function leads to 0 0 (cid:90) π/2 (cid:112)cos2(ψ) − cos2(θ) cos(θ)P∞(θ) dθ P∞(ψ) = 1 π cos(ψ) 0 θ0123P∞(θ)00.511.522.5RTB=4B=0.25log10(θ)-5-4-3-2-1log10(P∞(θ))24App.Exactθ0123P∞(θ)00.511.522.5RTB=4B=0.25log10(θ)-5-4-3-2-1log10(P∞(θ))24App.ExactCombining the asymptotic behavior from Eq. (5) in the main text and Eq. (23), we obtain the following asymptotic behavior, P∞(ψ) ∼ ψ 1 B −1 (B − 1)B sin(cid:0) π 2− B+1 B 2B (cid:1) , ψ (cid:28) 1, 8 (24) which is valid for B > 1. From the latter Eq. (24), we conclude that the velocity condensation is observable on the projected angle ψ for a sufficiently strong magnetotactic constant. Figure 6: Trajectories of the velocity vector V , with a zenith angle θ and an azimuthal angle φ, for (a) a RT walk and (b) an ABP; the magnetotactic constant is (blue line) B = 4 and (red line) B = 1.8. The angle ψ is the projected alignment angle in the plane y = 0. Approximate expressions for the velocity The averaged velocity in the direction of the magnetic field (Vz = V0 (cid:104)cos θ(cid:105)) is expressed in terms of the function gd(w), defined as: (cid:90) π (cid:34) (cid:35)d−1 , (25) gd(w) = dθ cos(θ) γd sin(θ) 2w tan (θ/2) with γ2 = π and γ3 = 2. We find that g2(w) = (w− 1)/(w + 1) in 2D and g3(w) = 2(cid:0)w4 − 4w2 log(w) − 1(cid:1) /(cid:0)w2 − 1(cid:1)2 (1 + w2 tan (θ/2)2) 0 in 3D. We obtain an approximate expression for Eq. (32) by expanding the function gd(w) around w = 1. In 2D, we expand the function g2(w) at the 5-th order around w = 1, and we obtain: Vz/V0 ≈ 465B5 + 435B4 + 169B3 + 30B2 + 2B 4(B + 1)(2B + 1)(3B + 1)(4B + 1)(5B + 1) . The expression of Eq. (26) is exact for B (cid:28) 1 and reaches at B (cid:29) 1 its maximal error, which is less than 4%. In 3D, we expand g3(w) around w = 1 at the 2-nd order, and we obtain: Vz/V0 ≈ 2(B + 3B2) 3(1 + B)(1 + 2B) . (26) (27) This expression Eq. (27) is exact for both B (cid:28) 1 and B (cid:29) 1 and it reaches at B ≈ 1 its maximal error that is less than 5%. (maximal relative error is less than 1%). ab9 Figure 7: (color online) Simulations of tumbling swimmers in the presence of a Brownian noise on the orientation (D(cid:48) r/τr = 0.10, B = 3). (a) Trajectories. Length scale is V0/τr. (b) Distribution of alignment angle with B = 3: (blue square) result of simulations at t = 20τr after an initial tumble (red solid line), RT model with no rotary diffusion, and (black circle-solid line) best ABP fit, with Dr/τr = 0.60. The maximal discrepancy with the cumulative distribution from simulation (i.e. Kolmogorov-Smirnov norm [1]) is 3% for the RT model, compared to 10% for the ABP model. Diffusion coefficient We now determine the diffusion coefficient D⊥ in the transverse direction (xOy), defined through the long-time limit of the transverse displacement(cid:10)∆y2(cid:11) ∼ 2D⊥t. We first notice that the variance of the transverse displacement after a single jump, denoted Var[Y1], can be obtained after averaging the quantity y(θ0, t) =(cid:82) t 0 sin(θt[θ0]) over all θ0 and t. As successive runs are independent, this quantity is equal to the transverse diffusion coefficient: D⊥/D0 = Var[Y1]/τr. In the large magnetic field limit B (cid:29) 1, the transverse diffusion reads: D⊥/D0 ∼ γd/B2, where γd = 3.3 . . . in 2D B(cid:29)1 and γd = 0.46 . . . in 3D. For ABPs, D⊥/D0 ∼ 2/B2 for B (cid:29) 1, as in this limit ABPs follow an Ornstein-Uhlenbeck dynamic with a 1/B spring constant (see [2, 3]). The fact that these transverse diffusion coefficients have the same scaling behavior suggest that, in spite of the velocity condensation, RTs are as efficient as ABPs in exploring their environment. Regular runs We now consider the case of runs of constant duration equal to τr = 1 (ρ(x) = δ(x − 1)). This distribution is in particular used to model the motion of myxo-bacteria [4]. The distribution of spent time since the last tumble reads: π(t) = 10<t<1. From Eq. (2), we find that the probability distribution reads 2 arctan eB tan , (2D), (28) Both expressions converge to a finite value at θ = 0, equal to(cid:8)eB − 1(cid:9) /(πB) in 2D, and 0 in 3D. The mean velocity reads Vz/V0 = −2 log(2)/B + 2 log(cid:0)eB + 1(cid:1) /B − 1 in 2D, and V /V0 = coth (B) − 1/B in 3D. 1/ tan(θ) − coth (B) / sin(θ) (3D). (29) , Hence the expressions for the forward velocity in 3D is the same for the ABPs or RTs with regular steps, provided that identification Dr and 1/τr are identified. This observation highlights the fact a Langevin fit of the velocity profile is not sufficient to discriminate between two microscopic models, namely the ABP model and the RT with regular steps. For a Pareto distribution ρ(x) = 1x>1(β − 1)/xβ with a finite mean run duration (β > 2): Pareto distribution π(t) = 1{0<t<1} β − 2 β − 1 + 1{1<t} β − 2 β − 1 t1−β. (30) (cid:26) (cid:18) 1 (cid:18) θ (cid:19)(cid:19) 2 (cid:27) − θ P∞(θ) = 1 πB sin θ = − 1 2B abzBxThe counter-intuitive effect known as the inspection paradox is that π(t) has a broader distribution and fatter tails than ρ(x) [1]. 10 Probability density In order to use Eq. (2) from the main text, we notice that the function t → sin(θ(cid:63) t = − log(tan(θ/2))/B. By substitution of the expression sin(θ(cid:63) obtain the following approximate expression: 0) is sharply peaked around 0) by a delta condition δ(t + log(tan(θ/2))/B), we P∞(θ) ∼ θ→0 γd Bβ−1(β − 2) (β − 1) 1 θ(log(1/θ))β−1 , (31) with γd = 4/(5π) = 0.254 (2D) and γd = 8/25 = 0.32 (3D), which are in close agreement to the value γ = 0.46 obtained by fits to simulations, see Eq. (7). Averaged velocity In terms of the function gd(w) defined in Eq. (25), the averaged velocity reads (cid:90) 1 0 (cid:90) ∞ 1 Vz/V0 = β − 2 β − 1 dt gd(e−Bt) + β − 2 β − 1 dt t1−βgd(e−Bt). (32) We now determine the asymptotic behavior of Vz/V0 at B (cid:28) 1. In the limit B (cid:28) 1, we approximate g2(w) ≈ (w−1)/2 and g3(w) ≈ 2(w − 1)/3, hence when β ≤ 3: Vz/V0 = γdBβ−2 Γ(3 − β) (β − 1) , (33) where γ2 = 1/2 and γ3 = 2/3, see Eq. (8). The physical significance of the non-analytical Vz/V0 ∝ Bβ−2 behavior is discussed in the main text. Non-magnetic torque functions In this section, we consider torques that take the general expression f (θ) = −θn for all θ ∈ [0, π]. We define the 0 = F (−1)[F [θ] − Bt] -- where F a primitive of 1/f and F (−1) is function θ0 → θ − θt[θ0], that is canceled for θ0 = θ∗ 0(θ, t) exceed the value π for t > tc = (F (π)− F (θ0)/B, i.e. for a run the reciprocal function of F . The function t → θ∗ (cid:90) tc(θ) duration larger than tc(π), the point θ cannot be reached from an angle θ0 that is below π. With a modified torque P∞(θ) = 1 f (θ) 0 dt π(t)(µ.f )(θ∗ 0(θ, t)). (34) If F (π) is infinite, tc(θ) = ∞ and we retrieve the expression of Eq. (2). If n < 1, the θ = 0 state can be attained in a finite time. Therefore, there is a finite probability that the particle reaches θ = 0, the probability density has a an atom at θ = 0. Perpendicular chemotaxis We now consider that the modulation of the run duration is given by τr(θ) = 1 + χ sin(θ). In this case, the Fokker-Planck equation reads: B∂θ(sin θP ) − (1 + χ sin θ)P (θ) = − (1 + χ(cid:104)P sin θ(cid:105)) . (35) The solution to Eq. (35) reads: P (θ) = γ exp(χθ) tan(θ/2)λ0 sin(θ) (cid:90) π θ exp(−χφ) tan(φ/2)λ0 . dφ 11 (36) As exp(χθ) → 1 when θ → 0, we conclude that a finite value of χ > 0 will not change of the critical magnetic field Bc = 1 obtained for χ = 0. ∗ Electronic address: [email protected] [1] W. Feller, Wiley Series, Vol. 2 (1968) p. 509. [2] M. Schienbein and H. Gruler, Bulletin of Mathematical Biology 55, 585 (1993). [3] C. W. Gardiner, Handbook of Stochastic Methods for Physics, Chemistry and Natural Sciences (Springer, 2004). [4] R. Balagam and O. A. Igoshin, PLoS Computational Biology 11, 1 (2015), arXiv:1506.00681 .
1701.05662	2	1701	2017-01-26T17:44:29	Detection and tracking of chemical trails by local sensory systems	[ "physics.bio-ph", "physics.flu-dyn", "q-bio.CB" ]	Many aquatic organisms exhibit remarkable abilities to detect and track chemical signals when foraging, mating and escaping. For example, the male copepod { \em T. longicornis} identifies the female in the open ocean by following its chemically-flavored trail. Here, we develop a mathematical framework in which a local sensory system is able to detect the local concentration field and adjust its orientation accordingly. We show that this system is able to detect and track chemical trails without knowing the trail's global or relative position.	physics.bio-ph	physics	Detection and tracking of chemical trails by local sensory systems Yangyang Huang1, Jeannette Yen2 and Eva Kanso1 ∗ 1 Aerospace and Mechanical Engineering Department, University of Southern California, Los Angeles, CA 90089 and 2 School of Biology, Center for Biologically-Inspired Design, Georgia Institute of Technology, Atlanta, GA 30332 (Dated: April 4, 2018) Abstract Many aquatic organisms exhibit remarkable abilities to detect and track chemical signals when foraging, mating and escaping. For example, the male copepod T. longicornis identifies the female in the open ocean by following its chemically-flavored trail. Here, we develop a mathematical framework in which a local sensory system is able to detect the local concentration field and adjust its orientation accordingly. We show that this system is able to detect and track chemical trails without knowing the trail's global or relative position. 7 1 0 2 n a J 6 2 ] h p - o i b . s c i s y h p [ 2 v 2 6 6 5 0 . 1 0 7 1 : v i X r a ∗ [email protected] 1 I. INTRODUCTION The response to olfactory signals and pheromones plays an important role in a variety of biological behaviors [1 -- 3] such as homing by the Pacific salmon [4], foraging by seabirds [5], lobsters [6, 7] and blue crabs [8], and mate-seeking and foraging by zooplanktons and insects [9, 10]. These dissimilar organisms and behaviors share similar mechanisms of sensing and responding to chemical signals [2]. The underlying mechanisms could be applied or adapted to design artificial devices for purposes such as source detecting in various environments, see examples in [11 -- 14]. Evidence suggests that many organisms respond to concentration difference (= signal strength) and orient themselves to the desired direction, either locating towards or escaping from a source [2, 15, 16]. Biological and physical gradients also act as signals for tracking processes in smaller organisms; see, e.g., [17] and references within. Copepods, a type of zooplankton about 0.1 cm in length, are known to aggregate at the boundaries of different water bodies in the ocean [18]. This aggregation is thought to be a result of the response to oceanic structures involving spatial gradients of flow velocities and densities [17]. Copepods adjust their swimming speed or turning frequencies with respect to these physical gradients in the water environment. Also, copepods sense biological gradients in mate-seeking [19]. In careful laboratory experiments by Jeannette Yen that focus on the mating behavior of the copepod Temora longicornis, a chemically-scented trail that mimics the pheromone-laden trail of the female is introduced into a quiescent water tank. Male copepods are able to detect and successfully track the trail mimic to its source as shown in FIG 1. In this work, we are loosely inspired by the copepod tracking ability of the female chemical trail. We develop an idealized, simple model where a moving chemical source generates a trail in an infinite two dimensional space and a tracker is able to locally sense the chemical field and adjust its orientation accordingly to locate and track the trail. The organization of this work is as follows. We first illustrate the chemical trail in section II by reformulating the problem in the moving frame attached to the source. In section III, we study the conditions for successful tracking using a gradient-based tracking scheme. In the situation where the tracker is far away from the chemical trail such that the gradient information is not reliable, a random-walk phase is introduced to first detect the chemical signal before switching to the gradient-tracking method. The detection algorithm and results are described in Section IV. 2 (a) b2 (b) b1 FIG. 1. (a) Copepod Temora longicornis (∼ 1 mm in length). Sensing of chemical signals is mediated by a distribution of small mechanoreceptive organs on its antennae. (b) Trail tracking by copepod: sequences showing the progression of the copepod T. longicornis while navigating a chemically-flavored laminar trail mimic. The copepod first follows the trail in the direction away from the source then corrects its heading direction and traces the trail in the direction of increasing chemical signal. The trail mimic is created by releasing fluid via syringe pump (0.01 mL/min) and small bore tubing (1mm). Dextran, a large molecular weight, was added to increase refractive index of the trail, enabling us to see both the deformation of the signal and movement of the tracking copepod. We conclude by summarizing our findings and discussing their potential implications to understanding the behavior of copepods in section V. II. PROBLEM DESCRIPTION Consider a chemical source moving at a constant velocity U from right to left in a fixed frame (X, Y ), shown in FIG. 2(a). The concentration field is governed by the diffusion equation ∂C ∂t = K ∂2C ∂Y 2 + Qδ(X + Ut)δ(Y ), (1) where Q is the rate of generation of the chemicals, K is the mass diffusivity of the chemicals, and δ is the Dirac-delta function. In (1), we neglected diffusion in the X-direction. This assumption can be readily justified by calculating the P´eclet number Pe, defined as the ratio of advective to diffusive transport rate. Large Pe (Pe ≫ 1) implies that advection is dominant while for small Pe (Pe ≪ 1) diffusion is dominant. In the X-direction, P´eclet number is given by Pe= LU/K where L and U are the characteristic length and speed, which for a swimming copepod take the values L = 0.1cm and U = 1cm/s [20]. The diffusivity 3 (a) (b) high (c) Fixed frame (X,Y) Moving frame (x,y) chemical source U Y X y x V θ tracker Sensory feedback s1, s2 of the tracker b2 b1 θ ω tracker V θ U low FIG. 2. Chemical trail left behind a source moving at constant speed U to the left: (a) in fixed inertial frame (X, Y ) and (b) in a frame (x, y) moving with the source. A sensory system (black ellipse) moving at a constant velocity V senses the chemical gradient in body-fixed frame b1, b2 and adjusts its orientation θ accordingly (c). coefficient involved in small biological organisms is of the order K = 10−5cm2/s [21]. Thus, Pe ∼ 104 and diffusion is negligible in the X-direction. It is convenient to rewrite equation (1) in a reference frame (x, y) moving with the chem- ical source at a speed U, shown in FIG. 2(b). The moving frame (x, y) is related to the fixed inertial frame (X, Y ) via the transformation x = X + Ut, y = Y (2) Therefore, ∂C ∂t = ∂C ∂t + U ∂C ∂x and equation (1) becomes an advection-diffusion equation as ∂C ∂t + U ∂C ∂x = K ∂2C ∂y2 + Qδ(x)δ(y). The steady-state solution of (3) is given by (3) (4) C = exp(cid:18)− y2 4(K/U)x(cid:19) . Q/U p4π(K/U)x A color map of this concentration is shown in FIG. 2 with pink indicating higher concentra- tion values. Next we study the tracking behavior of a sensory system, or a chemical tracker, in response to these chemical signals. One natural example of chemical tracking is in the mating behavior of copepods, where the female swims at a roughly constant speed along a straight path, while the male swims at faster speeds along a sinuous route until it detects the chemical trail left by the female and follows it [20]. The female copepod has body length L = 0.1cm and speed 4 around U = 1cm/s, leaving a trail of chemicals where Q/U = 0.253µg/cm3, or equivalently, the source rate is Q = 0.253µg/(cm2 · s). We inherit these parameter values for our current study. In our simulation, we choose mass scale m∗ = 0.1µg, velocity scale U ∗ = U = 1cm/s and length scale L∗ = 10L = 1cm to non-dimensionalize the problem. This choice of length scale makes it more feasible to treat the tracker as a point particle. III. TRACKING OF CHEMICAL TRAILS Consider a sensory system moving at a swimming speed V and let (b1, b2) be an or- thonormal frame attached to the sensory system such that b1 is aligned along the swimming direction; see FIG. 2. Let θ denote the orientation of the b1-axis measured from the e1 direc- tion. The sensory system is able to sense the directional concentration gradients s1 = ∇C ·b1 and s2 = ∇C · b2 and adjust its orientation, but not speed, based on the gradients it senses. In the moving frame (x, y), we have x = U + V cos θ, y = V sin θ, θ = F (s1, s2). Here, we postulate a simple form of the function F (s1, s2), namely, F (s1, s2) = ω sgn(s2)H(γ − s1) (5) (6) where ω is a constant rotation rate, sgn(·) is the sign function and H(·) is the heaviside function. According to (6), if the concentration gradient s1 = ∇C · b1 in the b1-direction is larger than a threshold value γ, then θ = 0 and the sensor continues to move in the same direction. If s1 is less than γ, one has H(γ − s1) = 1 and θ = ω sgn(s2). In this case, the sensor turns with angular velocity ω into the direction of increasing concentration, indicated by the sign of s2 = ∇C · b2. Note that the tracking scheme depends on the sign of the signals instead of their exact values. Therefore, the results are not sensitive to the distance between the tracker and the chemical source, especially in the x-direction. We simulate the trajectory of our sensory system by integrating equation (5) in time using the adapted-time-step function 'ode45' in MATLAB. Basic parameter values are chosen as follows: source rate Q = 0.1, diffusivity K = 10−5 and and threshold γ = 0.1. The initial location of the sensory system is x(0) = 6, y(0) = −3. FIG. 3 shows the trajectories for the same initial orientation θ(0) = π/3 and swimming speed V = 2 but two sets of control 5 (a) y 3 2 1 0 -1 -2 -3 U = 1 Successful tracking ω=2π (b) y 3 2 1 0 -1 -2 -3 U = 1 Unsuccessful tracking ω=π 9 8 7 6 5 4 3 2 1 0 2 4 6 8 10 0 2 4 6 8 10 x x FIG. 3. Successful and unsuccessful tracking for parameter values (a) ω = 2π, V = 2 and (b) ω = π, V = 2. Other parameters are set as Q = 0.1, K = 10−5, and γ = 0.01. The initial location of the sensory system is x(0) = 6, y(0) = −3 and its initial orientation is θ(0) = π/3. Colors represent the steady-state spatial distribution of the chemicals. (a) 3π l ω y t i c o e v r a u g n a l 5π/2 2π 3π/2 π π/2 0 0.5 successful tracking V=U (b) (c) unsuccessful tracking U = 1 U = 1 (b) y (c) y 1 0 −1 −2 −3 1 0 −1 −2 −3 V=0.8 backward tracking V=1.2 forward tracking 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0 4 8 12 16 20 swimming speed V x FIG. 4. Tracking behavior in the parameter space of swimming speed V and angular velocity ω. Other parameter values and initial conditions are the same as in FIG. 3. (a) Parameter space (V, ω) of successful versus unsuccessful tracking. (b) Successful backward tracking with swimming speed V = 0.8 less than the source speed U = 1. (c) Successful forward tracking with V = 1.2 > U . parameters: (a) ω = 2π and (b) ω = π. In (a), the tracker successfully follows the chemical trail while in (b) the tracker encounters the trail but fails to track it. This is because its angular velocity ω = π is not large enough for the sensory system to make a quick turn into the chemical trail. It is worth noting that the oscillatory trajectory in successful tracking is also found in the copepod experiments [20]. We now examine the tracking behavior of the sensory system with respect to the two 6 control parameters: angular velocity ω and swimming speed V . Other parameter values and initial conditions remain the same as those in FIG. 3. We map the unsuccessful and successful tracking on the two dimensional space (V, ω) in FIG. 4(a). It shows that as the tracker swims faster, the required angular velocity ω for successful tracking also increases. The transition from unsuccessful to successful tracking displays a linear relationship between the angular velocity ω and the swimming speed V . Note that the chemical source has a speed U = 1. When the tracker's speed V < U, the tracking is in the opposite direction of the source location, which we denote as backward tracking. See the example in FIG. 4(b) for V = 0.8 and ω = π. Backward tracking of a chemical trail has already been observed in copepod experiments [19]. When the tracker's speed is larger than that of the source, V > U, it tracks the chemical trail in the direction towards the location of the source, termed as forward tracking, shown in FIG. 4(c) for V = 1.2 and ω = π. The boundary separating these two types of successful tracking is illustrated as a dashed line at V = U. This boundary can be easily inferred from equation (5) by setting θ = π where the tracker is heading into the direction of the source. To achieve forward tracking, the horizontal velocity x must be in the negative x-direction; namely x = U − V < 0. Both backward and forward tracking are successful in tracking the chemical trail but differ in their ability to locate the source. The parameter space displayed in FIG. 4(a) is specified for one initial orientation θ(0) = π/3. We now explore the two dimensional space (V, ω) in FIG. 5(a)-(f) with respect to six different initial orientations: θ(0) = 0, π/3, π/2, 2π/3, 5π/6 and π. When θ(0) = 0 or π, the tracker moves in the positive or negative x-direction parallel to the chemical trail without ever turning or intercepting the trail. The sensory system fails to approach the chemical trail and therefore the tracking is unsuccessful irrespective of the values of V and ω. For initial conditions that intercept the trail, both unsuccessful and successful tracking can be achieved, as shown in FIG. 5(b)-(e). Note that plot (b) is the same as the one in FIG. 4. The boundary marking the transition from unsuccessful to successful tracking is given by a linear relationship between V and ω. The slope of the linear boundary gets steeper as θ(0) increases. In other words, as the angle between the tracker θ(0) and the trail becomes more obtuse, the tracker requires faster rotational motion for successful tracking. As θ(0) → π, the slope of the transition between unsuccessful and successful tracking tends to infinity. In FIG. 5(b)-(e), the transition from backward to forward tracking is independent of θ(0) and 7 l ω y t i c o e v r a u g n a l l ω y t i c o e v r a u g n a l (a) 3π 5π/2 2π 3π/2 π π/2 (d) 3π 5π/2 2π 3π/2 π π/2 (b) (c) successful tracking successful tracking unsuccessful tracking unsuccessful tracking unsuccessful tracking (e) (f) successful tracking successful tracking unsuccessful tracking unsuccessful tracking unsuccessful tracking 0.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 swimming speed V swimming speed V swimming speed V FIG. 5. Parameter space of swimming speed V and angular velocity ω as a function of six different initial orientations θ(0) = 0, π/3, π/2, 2π/3, 5π/6 and π, shown in (a)-(f) respectively. Dashed line is the interface between backward and forward tracking. bifurcates at V = U. IV. DETECTION OF CHEMICAL TRAILS The gradient-based model for trail tracking is not feasible for the initial detection of the trail because at distances far away from the trail the gradient is too shallow to be accurately sensed. However, the local concentration itself can be sensed [22]. Therefore, we introduce a detection step to first find the strong chemical trail by comparing the local chemical concentration to a threshold value Co. If the former is larger, then the chemical trail is detected and the tracker enters the tracking step using gradient information in (6). During the detection, the tracker executes a random walk that resembles the run-and-tumble behavior of bacteria [23, 24]. That is to say, the tracker runs in the same orientation if the detected concentration is increasing otherwise it tumbles by randomly choosing a direction. An illustration of the detection algorithm is shown in FIG. 6. According to FIG. 6, a tracker initially at (xm, ym) detects the local concentration as Cm = C(xm, ym) and picks a random direction θm to start moving. After a given time 8 Detection step:td = mΔt Initialize Next state time counter m = 0 θm = rand(1)2π xm, ym Cm = C(xm,ym) m = m+1 θm = θm-1 xm = xm-1 + (U+Vcosθm-1)Δt ym = ym-1 + VsinθΔt Cm = C(xm,ym) If m>mmax, stop detection YES Cm > Co? Cm > Cm-1? NO Random Walk if mod(m,N) = 0, then θm = rand(1)2π NO YES Gradient tracker x = U+Vcosθ y = Vsinθ θ = ω sgn(s2)H(γ-s1) FIG. 6. The run-and-tumble detection step of the chemical tracker. Co denotes the concentration threshold for the chemical trail. The tracker can detect the concentration Cm at its current location (xm, ym) and also remembers the previous concentration Cm−1. If Cm < Co, the tracker runs and tumbles in the detection step. If Cm > Cm−1, it remains the current orientation θm. Otherwise, it executes a random walk with frequency N such that θm is randomly chosen in between 0 and 2π every N iterations of time step △t. When Cm > Co, the tracker stops the detection step and enters the tracking behavior. If m > mmax, then the detection is unsuccessful. tm = tm−1 + △t, where ∆t is the time step and m is a positive integer, the tracker's position is given by xm = xm−1 + V cos(θm−1)∆t and ym = ym−1 + V sin(θm−1)∆t. It senses a new concentration Cm at the new location (xm, ym). If Cm > Co, then the chemical trail is detected. Otherwise, the tracker executes the run-and-tumble behavior by comparing the current concentration Cm to the previous one Cm−1. If Cm > Cm−1, the tracker runs without changing its orientation θm = θm−1. If Cm < Cm−1, then the tracker picks a random direction θm and follows that direction for N time steps. That is, 1/(N∆t) can be interpreted as the "frequency" of random walk. At the time when the local concentration Cm achieves the threshold value Co, the detection step ends and transitions to the gradient-tracking step. The total detection time is calculated as td = m△t. Yet, if during a simulation, the time count m is greater than a given maximum value mmax, then the detection is considered unsuccessful within the given amount of time tmax = mmax△t. Note that if time is long enough, the tracker is guaranteed to detect the chemical trail in a two dimensional plane [25, 26]. The detection step is governed by four control parameters: the concentration threshold Co, time step △t, frequency N, and maximum detection time tmax. Here, we choose Co = 9 U = 1 (a) y 1 0 -1 -2 -3 0 V=2 5 backward tracking td = 3.89 10 x 15 20 U = 1 (b) y 1 0 -1 -2 -3 0 V=2 5 forward tracking td = 4.81 15 20 10 x FIG. 7. Two sample cases of random walk in the detection of chemical trail. Control parameters are set as follows: concentration threshold Co = 10−5, time step △t = 0.01, random-walk frequency N = 100 and maximum detection time tmax = 20. The two trackers have the same initial conditions x(0) = 6, y(0) = −3, θ(0) = π/3 and parameter values ω = 2π, V = 2 but end up with different tracking behaviors: (a) backward tracking and (b) forward tracking. 10−5, △t = 0.01, N = 100 and tmax = 20 and study the effect of swimming speed V on the detection time td. In FIG. 8, we show two sample trajectories starting at the same initial position x(0) = 6, y(0) = −3 and random initial orientation θ(0) = π/3 for the same parameter values ω = 2π and V = 2. The two trajectories are distinct owing to the random nature of the search motion such that backward tracking occurs in (a) and forward tracking in (b). The detection time td, which we define as the total time it takes the sensory system to first detect the trail, is also not the same in the two simulations: td = 3.89 in (a) and td = 4.81 in (b). Note that the detection time td is independent of the angular velocity ω, which only participates in the tracking behavior, but depends on the swimming speed V . In the copepod experiments [19], the dimensional detection time is up to 10 seconds. FIG. 8 depicts the histogram or distribution of detection time td of successful detections obtained from 1000 distinct simulations for the same initial locations shown in FIG. 7 and three parameter values of V = 1, 2, 3. We fit the probability distributions to smooth expo- nential functions P (t) = λe−λt, shown as red curves, such that the average detection time is htdi = 1/λ. Note that the exponential fit is not perfect, but it is a closer analytical fit to the resulting distribution compared to a Poisson and normal distribution. The discrep- ancy between the analytical fit and the numerical data has minimal implications on the following results. As velocity increases, the decrease of the probability density function is steeper (larger λ) and the averaged detection time htdi decreases from 9.83 to 6.39 and 4.39. Therefore, larger swimming speed results in faster detection. In addition, out of the 1000 simulations we run, we keep track of the number of simulations which resulted in unsuccess- 10 ful detection in tmax = 20 . We find that the ratio of unsuccessful detection to total number of simulations is 0.57, 0.3 and 0.26 for V = 1, 2 and 3, respectively. That is to say, faster swimming is also beneficial to more successful detections in a given amount of time. f d p .1 .08 .06 .04 .02 0 V=1 td = 9.83 8 16 4 detection time td 12 .2 .16 .12 .08 .04 0 20 4 V=2 td = 6.39 8 12 16 detection time td 3 .25 .2 .15 .1 .05 0 20 4 V=3 td = 4.39 8 12 16 detection time td 20 FIG. 8. Probability distribution of detection time td with respect to different swimming speed V . The red curves are the fitted exponential probability density functions P (t) = λe−λt with λ = 1/ htdi, where htdi is the averaged detection time. For each value of V , 1000 distinct simulations are conducted to obtain the distribution of successful detections. The values of control parameters and also the initial locations of the tracker are the same as those in FIG. 7. We finally evaluate the average detection time htdi as a function of the tracker's initial location x(0), y(0). The region of interest is chosen to be [0.1, 10] × [−5, −1] as shown in FIG. 9. The colors indicate the values of the detection time at the corresponding initial locations, with red specifying longer detection time. We can see that htdi varies little in the x-direction especially in the range of x(0) > 2 meanwhile the average detection time decrease significantly when the horizontal distance between the tracker and the source is small x(0) < 1. The average detection time grows with increasing distances in the y- direction. These finding are consistent with the intuition that closer distance between the tracker and the source results in faster detection. V. CONCLUSIONS Odor tracking plays an important role in the behavior of organisms at different scales and in different environments and could have significant implications on engineering and robotic applications. Inspired by the odor-tracking abilities of male copepods in their mating behav- ior, we simulated a sensory system that tracks and detects a two-dimensional chemical trail 11 y(0) 0 -1 -2 -3 -4 -5 averaged detection time td 9 8 7 6 5 4 3 1 2 3 4 5 6 7 8 9 x(0) FIG. 9. The averaged detection time htdi as a function of the tracker's initial location x(0), y(0) in the region of [0.1, 10] × [−5, −1]. Red colors represent longer detection time. generated by a moving chemical source. The tracker can sense the local chemical gradients in its own body-fixed frame. The sensed gradients are used to control the orientation of the tracker such that it turns into the direction of increasing concentration. We identify the tracking behavior as successful if the trajectory of the tracker ends up oscillating around or directly moving inside the chemical trail. Otherwise, the tracking is unsuccessful. Successful tracking consists of either backward or forward tracking. If the sensory system successfully tracks the trail but moves away from the source, then it is backward tracking. Backward tracking occurs when the speed of the tracker is less than that of the chemical source. We then mapped the tracking behavior onto the parameter space consisting of the speed V of the tracker normalized by the speed of the chemical source and the angular velocity ω of the tracker. The results show that higher V requires larger ω for successful tracking. The boundary marking the transition from unsuccessful to successful tracking follows a linear growth of ω as a function of increasing V . The parameter space (V, ω) changes with respect to the initial orientation θ(0) such that when the angle between the tracker and the trail becomes more obtuse (i.e., the angle between the velocity of the tracker and the velocity of the source is more shallow) the tracker requires both larger speed and angular velocity to succeed in tracking. That is to say, the tracker should speed up to successfully track the trail when the orientation of its velocity is close to that of the source. A detection step is introduced when the chemical gradient is too weak to be accurately sensed by the tracker such as when the tracker is located far from the chemical trail. In this situation, the sensory system detects the chemical concentration first until the sensed local concentration is larger than a threshold value. The detection step is adapted from the run- 12 and-tumble behavior of bacteria such as E. coli, which runs when sensing a chemical signal or tumbles otherwise. In our implementation, the tracker continues in the same orientation if it senses an increasing concentration in that direction. If not, the tracker randomly picks an orientation θ from 0 to 2π. If the detection takes longer than a given amount of time (the total simulation point), then the detection is unsuccessful. We illustrated the distribution of the detection time td obtained from 1000 distinct simulations and calculated the average detection time of successful ones. We found that the average detection time decreases with increasing speed V and the ratio of successful detection is higher when V is larger. Therefore, for a more successful detection, a fast speed V is preferred. We also showed that closer initial location to the source results in smaller detection time. The two main results obtained from this study -- the fact that both successful detection and successful forward tracking require the tracker to have larger swimming speed than the source and that the tracker's speed should be even larger when it swims nearly parallel to the source -- are consistent with experimental observation of the copepod mating behavior [19]. Male copepods are known to swim faster than female copepods. While the reasons may be biological, this difference in speed between the male and female seems to have significant implications on successful detection and tracking of the female. Further, the detection time scale obtained here is consistent with experimental measurements of copepods [19]. Male copepods are reported to detect the chemical trail in time intervals up to 10 s, which is similar to the average detection time reported in this study. Note that here one dimensionless unit time scales to 1s. A few remarks on the limitations of the model and future directions are in order. We considered a simple gradient-tracking model where the speed of the tracker and its turning rate are not affected by the intensity of the chemical signal. While this model was able to track the chemical trail, it would be interesting in future studies to compare this model to more complex models where the speed and turning rate to change with the chemical signal. This study was restricted to two-dimensional tracking and detection but in many aquatic organisms, this behavior is inherently three-dimensional. Also, we considered the chemical signal to diffuse in a quiescent environment. In many real-world applications, the environment is often characterized by unsteady and at time turbulent flows. Future work will extend the framework presented here to account for three-dimensional effects and the effect of flows and patchiness in the chemical signal [8, 27 -- 30]. It is also interesting to couple 13 the sensory and control framework presented here to more accurate models of the swimming mechanics; see [31] for an analysis of the details of such drag-based swimming and [32] for an experimental study of the flow field generated by the swimming motion. Acknowledgment. This work is partially supported by the Office of Naval Research through the grant ONR 14-001. [1] J. Partridg, Trends in Ecology & Evolution 8, 262 (1993). [2] N. J. Vickers, The Biological Bulletin 198, 203 (2000). [3] R. K. Zimmer and C. A. Butman, The Biological Bulletin 198, 168 (2000). [4] A. D. Hasler and A. T. Scholz, Olfactory imprinting and homing in salmon: Investigations into the mechanism of the imprinting process, Vol. 14 (Springer Science & Business Media, 2012). [5] G. A. Nevitt, The Biological Bulletin 198, 245 (2000). [6] J. Basil and J. Atema, Biol. Bull 187, 272 (1994). [7] D. V. Devine and J. Atema, The Biological Bulletin 163, 144 (1982). [8] M. J. Weissburg and R. K. Zimmer-Faust, Journal of Experimental Biology 197, 349 (1994). [9] R. T. Card´e, Olfaction in mosquito-host interactions 200, 54 (1996). [10] R. T. Card´e and A. Mafra-Neto, in Insect pheromone research (Springer, 1997) pp. 275 -- 290. [11] F. W. Grasso, The Biological Bulletin 200, 160 (2001). [12] P. Pyk, S. B. i Badia, U. Bernardet, P. Knusel, M. Carlsson, J. Gu, E. Chanie, B. S. Hansson, T. C. Pearce, and P. F. Verschure, Autonomous Robots 20, 197 (2006). [13] T. Nakatsuka, Y. Kagawa, H. Ishida, and S. Toyama, in 2006 5th IEEE Conference on Sensors (IEEE, 2006) pp. 416 -- 419. [14] A. Dhariwal, G. S. Sukhatme, and A. A. Requicha, in Robotics and Automation, 2004. Proceedings. ICRA'04. 2004 IEEE International Conference on, Vol. 2 (IEEE, 2004) pp. 1436 -- 1443. [15] L. B. Buck, Cell 100, 611 (2000). [16] N. S. Johnson, A. Muhammad, H. Thompson, J. Choi, and W. Li, Behavioral Ecology and Sociobiology 66, 1557 (2012). [17] C. Woodson, D. Webster, M. Weissburg, and J. Yen, Limnology and Oceanography 50, 1552 14 (2005). [18] D. Holliday, R. Pieper, C. Greenlaw, and J. Dawson, MicroCAT sets the NEW standard in accurate moored CT instruments 11, 18 (1998). [19] M. H. Doall, S. P. Colin, J. R. Strickler, and J. Yen, Philosophical Transactions of the Royal Society of London B: Biological Sciences 353, 681 (1998). [20] C. B. Woodson, D. R. Webster, M. J. Weissburg, and J. Yen, Integrative and Comparative biology 47, 831 (2007). [21] F. Lombard, M. Koski, and T. Kiørboe, Limnology and Oceanography 58, 185 (2013). [22] W. Li, J. A. Farrell, S. Pang, and R. M. Arrieta, IEEE Transactions on Robotics 22, 292 (2006). [23] J. Adler, Science 153, 708 (1966). [24] H. C. Berg, D. A. Brown, et al., Nature 239, 500 (1972). [25] J. M. Borwein, D. H. Bailey, and D. Bailey, Mathematics by experiment: Plausible reasoning in the 21st century (AK Peters Natick, MA, 2004). [26] F. Spitzer, Principles of random walk, Vol. 34 (Springer Science & Business Media, 2013). [27] J. S. Kennedy and D. Marsh, Science 184, 999 (1974). [28] H. Ishida, K. Hayashi, M. Takakusaki, T. Nakamoto, T. Moriizumi, and R. Kanzaki, Sensors and Actuators A: Physical 51, 225 (1996). [29] R. Kanzaki, Robotics and Autonomous Systems 18, 33 (1996). [30] J. H. Belanger and M. A. Willis, in Intelligent Control (ISIC), 1998. Held jointly with IEEE International Symposium on Computational Intelligence in Robotics and Automation (CIRA), Intelligent Systems and Semiotics (ISAS), Proceedings (IEEE, 1998) pp. 265 -- 270. [31] S. Alben, K. Spears, S. Garth, D. Murphy, and J. Yen, Journal of The Royal Society Interface 7, 1545 (2010). [32] K. B. Catton, D. R. Webster, J. Brown, and J. Yen, Journal of Experimental Biology 210, 299 (2007). 15
1307.0644	1	1307	2013-07-02T09:38:50	Force Spectroscopy with Dual-Trap Optical Tweezers: Molecular Stiffness Measurements and Coupled Fluctuations Analysis	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.BM" ]	Dual trap optical tweezers are often used in high-resolution measurements in single-molecule biophysics. Such measurements can be hindered by the presence of extraneous noise sources, the most prominent of which is the coupling of fluctuations along different spatial directions, which may affect any optical tweezers setup. In this paper we analyze, both from the theoretical and the experimental points of view, the most common source for these couplings in Dual Trap Optical Tweezers setups: the misalignment of traps and tether. We give criteria to distinguish different kinds of misalignment, to estimate their quantitative relevance and to include them in the data analysis. The experimental data is obtained in a novel dual trap optical tweezers setup which directly measures forces. In the case in which misalignment is negligible we provide a method to measure the stiffness of traps and tether based on variance analysis. This method can be seen as a calibration technique valid beyond the linear trap region. Our analysis is then employed to measure the persistence length of ds-DNA tethers of three different lengths spanning two orders of magnitude. The effective persistence length of such tethers is shown to decrease with the contour length, in accordance with previous studies.	physics.bio-ph	physics	Force Spectroscopy with Dual-Trap Optical Tweezers: Molecular Stiffness Measurements and Coupled Fluctuations Analysis M. Ribezzi -- Crivellari Departament de Fisica Fonamental, Universitat de Barcelona F. Ritort Departament de Fisica Fonamental, Universitat de Barcelona, & Ciber-BBN de Bioingeneria, Biomateriales y Nanomedicina June 16, 2021 3 1 0 2 l u J 2 ] h p - o i b . s c i s y h p [ 1 v 4 4 6 0 . 7 0 3 1 : v i X r a Abstract Dual trap optical tweezers are often used in high-resolution measurements in single-molecule biophysics. Such measurements can be hindered by the pres- ence of extraneous noise sources, the most prominent of which is the coupling of fluctuations along different spatial directions, which may affect any optical tweezers setup. In this paper we analyze, both from the theoretical and the experimental points of view, the most common source for these couplings in Dual Trap Optical Tweezers setups: the misalignment of traps and tether. We give criteria to distinguish different kinds of misalignment, to estimate their quantitative relevance and to include them in the data analysis. The experimental data is obtained in a novel dual trap optical tweezers setup which directly measures forces. In the case in which misalignment is negligi- ble we provide a method to measure the stiffness of traps and tether based on variance analysis. This method can be seen as a calibration technique valid beyond the linear trap region. Our analysis is then employed to measure the persistence length of ds-DNA tethers of three different lengths spanning two orders of magnitude. The effective persistence length of such tethers is shown to decrease with the contour length, in accordance with previous studies. Key words; keyword 1; keyword 2; keyword 3; keyword 4 Introduction Optical Tweezers (OT) have been often employed to measure the elastic properties of polymers tethered between dielectric beads. A direct measure- ment of the tether stiffness is possible through fluctuation analysis (1). This kind of measurements are appealing from the experimental point of view because they require the measurement of a single quantity, either force or extension, so that they do not need a prior determination of the trap stiff- ness. One major source of error in these measurements is the coupling of fluctuations along different spatial directions, mainly in and out of the focal plane. This is a nontrivial effect which is expected to affect, although to different extents, any OT setup. The relevance of such effect is not limited to fluctuation measurements: it can also affect the correct measurement of force-extension curves, especially for short tethers. A clear understanding of the physical basis of such couplings is useful to determine under which con- ditions they lead to systematic errors in the measurements. In this paper we study fluctuation coupling in Double-Trap Optical Tweezers (DTOT) setups which are currently used in high resolution force spectroscopy. The most important coupling sources are misalignmnt effects affecting both traps and tether. Through theoretical modeling we establish a criterion to identify the kind of misalignment which causes the coupling and give explicit formulas to quantify its effect on the basis of the parameters characterizing the experi- mental setup. These considerations are also relevant for Single Trap Optical Tweezers (STOT). Moreover we show that, if coupling effects are negligible, the analysis of the variance of the measured signals, either force or posi- tion, may be used to simultaneously measure the tether and trap stiffnesses, including possible asymmetries between the two traps. This kind of mea- surements can be used as a calibration technique, suitable for any DTOT, which works beyond the linear region of the traps. When instead couplings are non-negligible we show how to include them in the data analysis. To illustrate this methodology we carried out measurements in a novel DTOT setup which uses counterpropagating beams and measures forces directly us- ing linear momentum conservation. This system is especially well suited for stiffness measurements as the force measurement calibration is independent of the shape and size of the trapped object. This is not true when using other techniques, e.g. back focal plane interferometry, which require a spe- cific calibration of the position measurement for each bead. We performed direct measurements of the stiffness of ds-DNA tethers whose contour length spans two decades. 2 Materials and Methods Optical Tweezers Setup The DTOT setup is shown in Figure 1 and is very similar to the one designed by Smith et al. (2) and described in (3), which operates with a single trap and a pipette. Force measurement is based on the conservation of linear momentum (2), making force calibration very robust. Calibration factors are determined by the optical setup and the detector response but they are independent of other details of the experimental setup, such as the index of refraction of the trapped object, its size or shape, the refractive index of the buffer medium and laser power. Unless the optics or the detector are changed, there is no need for continued calibration (2). Fluctuation measurements were performed using an acquisition board (Agilent Technologies) with a 50 kHz bandwidth, which is higher than corner frequencies of the measured signals. In a typical experiment fluctuations were measured for 10 seconds, by increasing the force in 2 pN steps between subsequent measurements. Molecular synthesis For our experiments we used ds-DNA tethers of four different lengths. The 24 kb tether was obtained by digesting the phage-λ plasmid with the XbAI restric- tion enzyme. The tether was then ligated to a biotin-labeled oligo on one side and with a dig labeled oligo on the other side. The 3 kb tether was obtained by PCR amplification of a section of the phage-λ plasmid using biotin-modified primers. The amplified segment was then restricted with Xba I and ligated to a dig-modified oligo. The 1.2 kb and 58 b tethers were synthesized according to the protocol described in (4). Experiments were performed in a microfluidic chamber formed by two coverslips interspaced with parafilm. Anti-dig coated beads were first incubated with the molecule of interest and then introduced one at a time in the microfluidics chamber trough a dispenser tube. Once the anti-dig coated bead was trapped a streptavidin coated bead was introduced through a second dispenser tube and trapped in the second trap. The connection was then formed directly inside the microfluidics chamber. All experiments on DNA tethers were performed in PBS buffer solution at pH 7.4, NaCl 1M, at 25oC. This buffer solution was found to greatly reduce the nonspecific adsorption of DNA on silica. We dis- solved 1mg/µl Bovine Serum Albumin in the buffer in order to reduce non-specific silica-silica interactions. 3 Results Coupled fluctuations in a plane. In order to introduce the main subject of this paper it is useful to consider a pedagogical example that shows how a small coupling between two fluc- tuating degrees of freedom can have a large effect on their variances. Let us consider a particle moving on a plane while constrained by two harmonic springs (Fig. 2 A). The position of the particle is given by a vector p = (y, z) and the springs are oriented along the coordinate axes (y, z) (Fig.2A). The stiffness matrix ¯k is given by: = ky , α = kz/ky, (1) where ky,kz are the stiffnesses along y,z respectively. The equilibrium Boltz- mann distribution for p can be written in compact form as: (cid:18) ky 0 (cid:19) 0 kz ¯k = (cid:19) (cid:18) 1 0 0 α (cid:18) −p · ¯kp (cid:18) 2kBT (cid:19) −p · ¯kp (cid:18) 1 0 2kBT 0 1 α (cid:19) exp 1 Z (cid:90) ρ(p) = Z = dzdy exp (cid:19) , (2) (3) where kB is the Boltzmann constant, T the absolute temperature and Z the partition function. The variance of p is thus: (cid:104)p2(cid:105) = kBT ¯k−1 (cid:39) kBT ky . (4) In the case ky (cid:29) kz (α (cid:28) 1) the variance of spatial fluctuations along z is much bigger than the variance along y and the level curves of the probability distribution in Eq. (2) are highly eccentric ellipses (Fig. 2 C, red solid curves). If the springs are misaligned by an angle θ with respect to the reference frame (Fig. 2 B), the stiffness tensor changes as: ¯k(cid:48) = ¯R(θ)¯k¯RT (θ) = (cid:18) ky cos2 θ + kz sin2 θ (ky − kz) cos θ sin θ (ky − kz) cos θ sin θ kz cos2 θ + ky sin2 θ = (cid:19) . (5) where ¯k(cid:48) is the new stiffness and ¯R(θ) is a rotation of angle θ. In this situation, since ky (cid:54)= kz, the off -- diagonal terms in Eq. (5) couple the motion of the 4 particle along y and z. We define = sin θ as the coupling parameter. To lowest order in we get: For a small coupling the variance of y is increased to: (cid:104)p2(cid:105) = kBT ¯k(cid:48)−1 (cid:39) (cid:39) kBT ky (cid:18) 1 +(cid:0) 1 α − 1(cid:1) 2 (cid:0)1 − 1 (cid:1) (cid:18) 1 (cid:18) (cid:19) (cid:18) 1 + α α 1 + , 2 α (cid:104)y2(cid:105) = kBT ky (cid:39) kBT ky (cid:19) (cid:1) α (1 − 2) + 2 (cid:19) (cid:0)1 − 1 (cid:19) α 1 − 1 2 (cid:39) (α (cid:28) 1). . (6) (7) This shows that the effect of a small coupling (small misalignment of the springs) on the variance of y can be large if 2 small α value acts as a lever arm, amplifying the effect of a small rotation (Fig. 2 C). Only when α ∼ 1. In other words, the 2 (cid:28) α , (8) the coupling of fluctuations can be ignored. So far we gave the general picture of the effect of a coupling between two fluctuating degrees of freedom. In the next section we will show how this simple model can be extended to study fluctuations in a DTOT. Generally speaking, when a molecule is pulled at high forces the rigidity of the tether is much higher along the pulling direction (y) than in the perpendicular direction (z) and Eq.(8) can be violated even if is very small ((cid:39) 0.1). At low forces and for long tethers Eq.(8) is instead fulfilled and the coupling can be ignored. A quantitative treatment of these effects and the corresponding data analysis are described next. Coupled fluctuations in a DTOT setup Correlation functions of fluctuation measurements in a DTOT do often show a double exponential nature. This is usually interpreted in terms of a cou- pling of the fluctuations in the optical plane and along the optical axis (5, 6). Couplings affect the variance of the measured force (or position) signal and the spurious contribution must be removed from the measurements. Yet, coupling effects reflect some defect in the design of the experimental setup and their study can provide valuable diagnostic tools for the fine tuning of a DTOT setup, crucial for high resolution measurements. Most prominent coupling sources are misalignment effects, due to at least two causes. A first cause is tether misalignment, i.e. a configuration in which the centers of the 5 traps lie in different planes (Fig. 3C). A second case is trap misalignment, when the principal axes of the traps are tilted with respect to the direc- tion along which the tether is stretched (Fig. 3D). Remarkably these two scenarios lead to different coupling structures making possible the identifi- cation of misalignment effects. To discuss coupling effects it is necessary to consider a 4-dimensional configuration space, describing the position of the beads both along the optical axis and the pulling direction. The laser beams (black arrows in Fig. 3A) define the optical axis, which we identify with the z direction in our reference frame. In the optical plane (perpendicular to the optical axis), we shall use the coordinate y to denote the direction along which the tether is oriented. The coordinate x, perpendicular to both y and z, will play no role in our analysis. The effect of both traps and tether on the dynamics of the beads can be modeled by a potential energy function, which depends on the positions of the beads in the y − z plane. When con- sidering equilibrium fluctuations at constant trap -- to -- trap distance, a linear approximation around the equilibrium positions can be used: U (y1, y2, z1, z2) = (9) with p = (y1, y2, z1, z2), and ¯K(cid:48) the stiffness tensor (tensors are primed when they are represented in the y − z coordinate system). In the ideal case (Fig. 3B), the stiffness tensor, which is now 4 × 4, reads: pT ¯K(cid:48)p, 1 2  ¯K(cid:48) = y1 y2 z1 z2 y1 ky + km −km 0 0 y2 −km ky + km 0 0 z1 0 0 kz + f r0 − f r0 z2 0 0 − f r0 kz + f r0 . (10) (11) (12) The stiffness tensor is two-block diagonal, with the first block describing the effect of traps and tether along the y axis, and the second block describing the effect of traps and tether in the z direction, sharing the same structure. Here ky is the stiffness of the trap in the y direction (the two traps are assumed identical for simplicity), kz is the trap stiffness in the z direction, km is the stiffness of the tether connecting the two beads, f is the mean force and r0 the mean distance between the centers of the beads. In this case the stiffness tensor can be diagonalized switching to the coordinate system defined by: y+ = z+ = y1 + y2√ 2 z1 + z2√ 2 , , y− = z− = 6 y1 − y2√ z1 − z2√ 2 , 2 where y+, z+ represent the center of mass position and y−, z− represent the differential coordinate. In this second coordinate system, the stiffness tensor reads:  y+ ky 0 0 0 ¯K = y+ y− z+ z− y− 0 ky + 2km 0 0 z+ 0 0 kz 0 z− 0 0 0 kz + 2 f r0 , which shows that the four different fluctuation modes are uncoupled (we dropped the prime because we switched to a different reference frame). The tensor ¯K can be written as the sum of two contributions: with: ¯KT = which accounts for the trap contribution to the stiffness, and: ¯K = ¯KT + 2¯Km, 0 ky 0 0 0 0 kz 0 0 0 0 kz 0 0 0  ky  0 ¯Km = 0 0 0 km 0 0 0 0 0 0 0 0 0 0 f r0  ,  , (13) (14) (15) (16) (17) which accounts for the tether contribution to the stiffness. Tether misalignment effects In the presence of tether misalignment, i.e. if the two traps are focused at different depths along the optical axis (Fig. 3C), the tether forms an angle θ with respect to the y axis. This can be incorporated in the stiffness tensor through a rotation of ¯Km, Eq. (16), of the same angle:  0 ¯Km() = 0 0 u() 0 0 w() 0 0 0 0 w() 0 v() 0 0  , 7 with and = sin(θ). u() = km(1 − 2) + (cid:18) v() = w() = f 2 r0 (1 − 2) + km2 f r0 km − f r0 (cid:19)√ 1 − 2, The total stiffness tensor is now:  ¯K() = y+ y− z+ z− y+ ky 0 0 0 y− 0 ky + 2u() 0 w() z+ 0 0 kz 0 z− 0 w() 0 kz + 2v() (18) (19) (20) (21) , which must be compared to Eq. (13). Looking at the non diagonal terms of this tensor it is evident that the coupling will only affect the two differential modes, y−, z− leaving the motion of the center of mass, y+, z+, unchanged. Trap misalignment effect In the second configuration the trap stiffness tensor is rotated (Fig. 3D), and the stiffness tensor becomes:  y+ p() 0 q() 0 y− 0 p() + 2km 0 q() z+ q() 0 m() 0 z− 0 q() 0 m() + 2 f r0 . ¯K() = y+ y− z+ z− with (22) (23) (24) (25) p() = ky(1 − 2) + kz2 m() = kz(1 − 2) + ky2 1 − 2, q() = (ky − kz) √ In contrast to the previous case, the non-diagonal terms and = sin(θ). show that the coupling will affect both the differential and center of mass coordinates. This important difference provides a simple tool to distinguish between tether and trap misalignments. 8 On the relevance of coupling effects In the case of tether misalignment the off diagonal terms in Eq. (21) cou- ple the fluctuations along the y− and z− directions. A reduced description, addressing these two coordinates is possible using a subtensor of ¯K(), (21): (cid:18) y− (cid:19) z− w() kz + 2v() ¯K−() = y− ky + 2u() z− w() The variance of y− is given by: (cid:104)y2−(cid:105) = kBT(cid:0)¯K(cid:48) −()−1(cid:1) y−y− , which, to leading order in , is: (cid:104)y2−(cid:105) (cid:104)y2−(cid:105)=0 = 1 + 2 α with α = (ky + 2km)(kz + 2f /r0) (2km + kz)(2km − 2f /r0) . . (26) (27) (28) (29) The value of α can be computed if the stiffnesses are known. In Fig. 3 E we show the behavior of α as a function of the force for two different ds-DNA tethers (3kbp and 24 kbp) and for two different trap stiffnesses. The continuous curves show the value of α computed in a low stiffness setup (ky = 0.02 pN/nm, kz = 0.001 pN/nm, the condition in which our DTOT operates) while the dotted lines show the value of α for high trap stiffness (ky = 0.2 pN/nm, kz (cid:39) 0.01 pN/nm, as reported for the DTOT in (7, 8)). Even in the case of high trap stiffness the attained value of α is small ((cid:39) 10−1) at high forces for the shorter tether. In Fig. 3 E the shaded area shows the values of α for which a misalignment (cid:39) 0.1 causes a 10% error in the measurements of y− fluctuations Eq. (28). In our DTOT setup such coupling in negligible when using long molecules (8 µm) at low and moderate forces (up to (cid:39)10 pN). In the case of trap misalignment both the variances of y− and y+ are affected and similar formulas hold: (cid:104)y2−(cid:105) (cid:104)y2−(cid:105)=0 = 1 + 2 γ (cid:104)y2 +(cid:105) (cid:104)y2 +(cid:105)=0 = 1 + 2 δ γ = (ky + 2km)(kz + 2f /r0) (ky − kz)(ky + 2f /r0) 9 (30) (31) δ = kz ky − kz . (32) Equations (28),(31),(32) estimate the error due to misalignment as a func- tion of the stiffnesses contributing to the experimental setup (Fig. 3E). As such they are useful to understand in which force regimes misalignment is going to be important. Neglecting coupling effects: variance analysis In experimental setups where force is directly measured, the trap stiffness is used to measure the extension of the fiber. Conversely, when the bead displacement is measured (e.g. by video microscopy or by back focal plane interferometry (9)), the stiffness is used to measure force. In both cases the trap stiffness depends on the details of the experimental setup i.e. laser power, bead size and shape or buffer medium (via its refraction index). Opti- cal traps are assumed to be linear close to the trap center, so that a stiffness measurement at zero force can be used to characterize the trap shape in this region. This kind of calibration has been shown to achieve 1% accuracy (10). Nevertheless when one needs to do measurements at high forces, nonlinear effects may become relevant, especially at low trap power and the full force field of the trap should be measured (11). When the coupling is negligible, motions along the y and z directions are uncoupled and the stiffnesses of traps and tethers can be obtained from a straightforward analysis of the ex- perimental force (or position) variances. We will model traps and thether by the dumbbell shown in Fig. 4. It consists of three serially connected harmonic springs. In this setting we allow for a different stiffness in the two traps: k1, k2 (Fig. 4 A). Together with the tether stiffness km we get the stiffness tensor: (cid:18) k1 + km −km (cid:19) , (33) k2 + km ¯K(cid:48) = −km which is just the first block of Eq. system is written as: (10), and the potential energy of the U (y1, y2) = (y1, y2) · ¯K(cid:48)(y1, y2) 1 2 (34) where (y1, y2) have the same meaning as in the previous sections. The equi- librium distribution for (y1, y2) is related to ¯K(cid:48) by: (cid:18) (cid:19) P (y1, y2) = 1 Z exp 10 −U (y1, y2) kBT . (35) The variances and covariance of (y1, y2) are linked to the inverse of the stiff- ness tensor. In terms of the experimentally measured variances and covari- ances: k1 = κ k2 = κ km = κ , , (cid:104)y2 2(cid:105) − (cid:104)y1y2(cid:105) kBT 1(cid:105) − (cid:104)y1y2(cid:105) (cid:104)y2 kBT (cid:104)y1y2(cid:105) kBT , . (36) (37) (38) with κ−1 = (cid:104)y2 1(cid:105)(cid:104)y2 2(cid:105)−(cid:104)y1y2(cid:105)2 (kBT )2 In experimental set-ups which directly measure forces different formulas ij = (cid:104)fifj(cid:105)−(cid:104)fi(cid:105)(cid:104)fj(cid:105), i = 1, 2) apply: the experimental values (σ2 and the model parameters (k1, k2, km) are related by: 12, σ2 11, σ2 22, σ2 k1 = k2 = km = 11 + σ2 σ2 12 kBT 22 + σ2 σ2 12 kBT 12 (σ2 σ2 1 kBT 11 + σ2 σ2 11σ2 12) (σ2 22 − σ4 12 22 + σ2 12) (39) (40) (41) . The method we have introduced in this section is similar to the one used by Meiners and Quake (1), the difference being that here we use equal -- time force covariance whereas (1) uses time -- dependent correlation functions. Both methods can be affected by the presence of extraneous noise sources. The effect of low frequency noises (such as line noise, drift or air flows) can be minimized by computing the variance of the force (or distance) signal on traces which are much longer than the corner frequency of the dumbbell, but short enough so that the low frequency noise sources never dominate the power spectrum (Fig. 4B). The method just presented can be used in any DTOT setup to calibrate the two traps (by measuring k1, k2) at differ- ent forces. Equations (36),(37) and (38) are useful in experimental set-ups which directly measure bead positions, while Equations (39),(40),(41) refer to set-ups which directly measure forces. Fig. 4C shows the results of direct stiffness measurements, based on equations (39),(40),(41) on a 24kbp ds- DNA molecule, in the shaded region misalignment effects gain importance and the measured stiffness departs from the WLC fit obtained at low forces. Note that the effect of tether misalignment on trap stiffness measurements is negligible (Fig. 4D). This happens because the measurement of trap stiffness, 11 in the absence of large asymmetries between the traps, does only depend on the variance of the center-of-mass coordinate. In our setup a force dependent trap stiffness calibration appears to be crucial if we want to measure the force vs molecular extension curve as the trap response is strongly non-linear (Fig. 4). This is also the case for many other DTOT and STOT set-ups, especially at high enoughforces. Measurement of the molecular stiffness The method discussed in the previous section is only useful if coupling effects are negligible, i.e. for long molecules (104 bp) at small forces (< 10 pN). To extend the applicability of direct stiffness measurements to shorter molecules or wider force ranges coupling effects must be taken into account. Since re- moving couplings while performing experiments is generally unpractical, it may be convenient to include them data analysis. To illustrate how this is done we will use fluctuation measured on ds -- DNA tethers of four different lengths: two sections of the phage -- λ genome of 24 kb and 3 kb respectively, and two tethers (1.2 kb and 58 b) which are used as handles in single molecule experiments (4). The differential, y− and center-of-mass coordinate y+ cor- relation functions, measured in our counter propagating DTOT setup are shown in Fig. 5. While the correlation function for the center of mass shows a single expo- nential behavior (upper panels), the differential coordinate displays a double exponential behavior which derives from the coupling of fluctuations (lower panels). This is the expected phenomenology for tether misalignment, which is the dominant effect in our setup. On the contrary, in the case of trap misalignment, both the differential and center of mass coordinates would be affected. According to Eq (27),(28) and neglecting the trap stiffness with respect to the molecular stiffness (km (cid:29) ky, kz; f /r0 (cid:29) kz), we have that, in presence of tether misalignment, the variance of the differential coordinate y− changes to: . (42) (cid:18) (cid:18) r0km f (cid:19) (cid:19) − 1 2 (cid:104)y2−(cid:105) (cid:39) kBT 2km 1 + As the value of is not known the derivation of the molecular stiffness cannot be performed on the basis of variance analysis. Fortunately, as already noted in (5), the decay rate of fluctuations in the optical plane ω+ is much bigger than that of fluctuations along the optical axis ω−. The correlation function for y− (derived in the Supplementary Material) reads: (cid:104)y−(0)y−(t)(cid:105) =(cid:0)1 − 2(cid:1) e−2ω+t + 2 r0e−2ω−t 2f , (43) 2km 12 and the molecular stiffness can still be recovered by selecting the amplitude of the fast decaying component of the correlation function (Fig. 6). Using this data analysis technique we measured the force dependent stiffness of the different tethers (Fig. 7). The nonlinear elasticity of ds-DNA is usually modeled with the Worm Like Chain (WLC) model. When a contribution for enthalpic stretching and overwinding (12) is added, the so -- called Extensible WLC (EWLC) (13) model is obtained. The Marko-Siggia approximation formula for the force dependent stiffness of a EWLC is: (cid:32) (cid:114) km(f ) = 1 (cid:96)0 1 4 kbT P (cid:33)−1 (cid:18) 1 (cid:19)3/2 f + 1 S , (44) where P is the persistence length of the polymer, (cid:96)0 its contour length and S is the stretch modulus. This approximation is valid for molecules whose countour length is larger than the persistence length, (l0 > P ) (13). The EWLC model was fitted to the data letting the persistence length P and the stretch modulus S vary. Fit results are shown in Table 1. In all cases, the re- sults are compatible with the existing literature, although they were derived in a different force range. In particular the 3kbp molecule shows a decrease in persistence length which is compatible with theoretical predictions based on finite-size effects as recently shown in (14). Data for the 1.2 kbp and 58 bp tethers are instead consistent with measurements performed in a STOT (4). In principle, 58 bp data should not be fitted with the Marko-Siggia ap- proximation (l0 < P ) and a semiflexible rod model is required. Nevertheless we decided to include the results for such short tether in Figure 7 to stress the agreement between our dual trap measurements and those reported in Ref. (4) for a STOT. Several recent measurements on ds-DNA suggest a strong reduction of the persistence length as the contour length of the molecule decreases. This could be the result of scale dependent DNA elasticity, as already found in microtubules (15), or a finite-size effect due to the boundary conditions (14). Here we carry out persistence length measurements in the highly stretched regime (2 − 15 pN), i.e. when the molecule is almost fully stretched and the enthalpic contribution is important (13, 16). In this regime and in contrast to low force measurements (14, 17, 18), excluded volume effects between the trapped beads are negligible. Moreover the effect of fluctuating boundary conditions on the force-extension curve should become less and less relevant as f Rb grows, with f the mean tension along the tether and Rb the bead radius ((cid:39)2 µm in our experiments). Seol et al. propose a phenomenological 13 scaling equation: P ((cid:96)0) = P∞ 1 + a P∞ (cid:96)0 . (45) We fitted such empirical formula to our persistence length measurements, leaving out the shortest thether whose persistence length cannot be correctly measured using the Marko-Siggia appriximation. The parameters obtained from the fit are P∞ = 49± 2 nm, in accordance to what measured in (14) and a = 4 ± 1 which is bigger than the value (2.78) measured in (14) and later confirmed in (17). It must be stressed that our experiments are performed in a dumbbell configuration, while both (14) and (17) have one of the ends of the molecule attached to a surface, and the boundary conditions imposed on the molecule are different in the two cases. The finite size WLC theory (14) does indeed predict a faster decrease of the effective presistence length with the contour length in the dumbbell configuration. Conclusions Misalignment effects may affect any optical tweezers setup and the effect of a small misalignment can be enhanced by the large difference in stiffness between fluctuations along the pulling direction and along the optical axis. Analyzing these effects we have provided tools to estimate their relevance, we have shown that trap and tether misalignments lead to different coupling structures and explained how to include couplings in the data analysis. When misalignment is negligible, it is possible to measure the stiffnesses of traps and tether from a straightforward analysis of the measured variance of force (or position) signals. This technique may be used as a calibration technique, valid beyond the linear trap region, in any DTOT. Otherwise it is possible to include misalignment effects in data analysis. Such techniques have been used to measure the stiffness of ds-DNA tethers of three different lengths, from 24 kbp to 1200 bp. The stiffness was interpreted in the framework of EWLC model and we confirmed the decrease in apparent persistence length previously reported in (14, 17), although our measurements were performed in a wider force range and with a different set-up. Financial Support FR is supported by MICINN FIS2007-3454, HFSP Grant No. RGP55-2008, and ICREA Academia grants. MR is supported by HFSP Grant No. RGP55- 2008. 14 References 1. Meiners, J., and S. Quake, 2000. Femtonewton force spectroscopy of single extended DNA molecules. Physical Review Letters 84:5014 -- 5017. 2. Smith, S. B., Y. Cui, and C. Bustamante, 2003. Optical-trap force trans- ducer that operates by direct measurement of light momentum. Methods in Enzymology 361:134 -- 162. 3. Huguet, J. M., C. V. Bizarro, Forns, S. B. Smith, C. Bustamante, and F. Ritort, 2010. Single-molecule derivation of salt dependent base-pair free energies in DNA. Proceedings of the National Academy of Sciences 107:15431. 4. Forns, N., S. de Lorenzo, M. Manosas, K. Hayashi, J. M. Huguet, and F. Ritort, 2011. Improving Signal/Noise Resolution in Single-Molecule Experiments Using Molecular Constructs with Short Handles. Biophys- ical Journal 100:1765 -- 1774. 5. Meiners, J. C., and S. R. Quake, 1999. Direct measurement of hydrody- namic cross correlations between two particles in an external potential. Physical review letters 82:2211 -- 2214. 6. Bustamante, C., Y. Chemla, and J. Moffitt, 2009. High-Resolution Dual- Trap Optical Tweezers with Differential Detection: Alignment of Instru- ment Components. Cold Spring Harbor Protocols 2009:pdb -- ip76. 7. Gebhardt, J., T. Bornschlogl, and M. Rief, 2010. Full distance-resolved folding energy landscape of one single protein molecule. Proceedings of the National Academy of Sciences 107:2013. 8. Comstock, M., T. Ha, and Y. Chemla, 2011. Ultrahigh-resolution optical trap with single-fluorophore sensitivity. Nature Methods 8:335 -- 340. 9. Neuman, K., and S. Block, 2004. Optical trapping. Review of Scientific Instruments 75:2787. 10. Toli´c-Nørrelykke, S., E. Schaffer, J. Howard, F. Pavone, F. Julicher, and H. Flyvbjerg, 2006. Calibration of optical tweezers with positional detec- tion in the back focal plane. Review of scientific instruments 77:103101. 11. Jahnel, M., M. Behrndt, A. Jannasch, E. Schaffer, and S. Grill, 2011. Measuring the complete force field of an optical trap. Optics letters 36:1260 -- 1262. 15 12. Gore, J., Z. Bryant, M. Nollmann, M. Le, N. Cozzarelli, and C. Busta- mante, 2006. DNA overwinds when stretched. Nature 442:836 -- 839. 13. Marko, J., and E. Siggia, 1995. Stretching DNA. Macromolecules 28:8759 -- 8770. 14. Seol, Y., J. Li, P. Nelson, T. Perkins, and M. Betterton, 2007. Elasticity of Short DNA Molecules: Theory and Experiment for Contour Lengths of 0.6-7 µm. Biophysical Journal 93:4360 -- 4373. 15. Pampaloni, F., G. Lattanzi, A. Jon´as, T. Surrey, E. Frey, and E. Florin, 2006. Thermal fluctuations of grafted microtubules provide evidence of a length-dependent persistence length. Proceedings of the National Academy of Sciences 103:10248 -- 10253. 16. Marko, J., 1997. Stretching must twist DNA. EPL (Europhysics Letters) 38:183. 17. Chen, Y., D. Wilson, K. Raghunathan, and J. Meiners, 2009. Entropic boundary effects on the elasticity of short DNA molecules. Physical Review E 80:020903. 18. Chen, Y., G. Blab, and J. Meiners, 2009. Stretching submicron biomolecules with constant-force axial optical tweezers. Biophysical jour- nal 96:4701 -- 4708. 19. Smith, S., Y. Cui, and C. Bustamante, 1996. Overstretching B-DNA: the elastic response of individual double-stranded and single-stranded DNA molecules. Science 271:795. 16 Molecule Fit Results 24 kbp 3 kbp 1.2 kbp 58 bp P (nm) 48 ± 5 39 ± 4 34 ± 5 1.4 ± 0.8 S (pN) 1400 ± 300 1800 ± 400 850 ± 100 20 ± 2 Table 1: Persistence length and stretch modulus for different ds-DNA tethers. Errors are standard deviations over different molecules. In all cases the measurements were obtained on at least three different molecules. The Marko- Siggia approximation is not valid in the case of the shortest tether and Eq. (44) cannot consistently be used to estimate the persistence length of such tether. Nev- ertheless we included this result to compare it with that obtained in Ref. (4) using a STOT. The data for the persistence length of two shortest molecules agree with those obtained using a STOT (4), which are P = 31± 3 nm, for the 1.2kb molecule and P = 1.6 ± 0.3 nm for the 58 bp molecule 17 Figure Captions Figure 1 Experimental Setup. The scheme of the optical setup, with the optical paths of the lasers and the led. Fiber-coupled diode lasers are focused inside a fluidics chamber to form optical traps using underfilling beams in high NA objectives. All the light leaving from the trap is collected by a second objective and sent to a Position Sensitive Detector which integrates the light momentum flux, measuring changes in light momentum (2). The laser beams share part of their optical paths and are separated by polarization. Part of the laser light ((cid:39) 5%) is deviated by a pellicle before focusing and used to monitor the trap position (Light Lever). Each trap is moved by pushing the tip of the optical fiber by piezos coupled to a brass tube (wigglers). Figure 2 Coupled fluctuations in a plane. A) A particle (P) is constrained by two springs which are oriented along the y, z coordinate axes (shown by the two arrows). ky, kz denotes the spring stiffnesses along y, z. B) A particle (P) constrained by two springs that are misaligned by an angle θ with respect to the coordinate axes. C) Level curves and marginal distributions for the joint equilibrium probability distribution of the particle position in the two systems shown in panels A (solid curves) and B (dashed curves). If the two springs have very different stiffnesses, the level curves form highly eccentric ellipses. In this situation a rotation by a small angle θ in the joint distri- bution can lead to a large change in the marginal distribution for y. The large difference in stiffness acts as a lever arm amplifying the strength of the coupling. Figure 3 Misaligned experimental configurations. A) The coordinate system used throughout the text. The direction of light propagation (black horizon- tal arrows) defines the optical (z) axis. The stretching direction, perpendic- ular to z, defines the y axis. The positions of the beads ((y1, z1) and (y2, z2)) are measured with respect to the equilibrium positions and r0 denotes the mean separation between the centers of the beads. B) Aligned configuration, the tether is perfectly oriented along the y axis. C) Misaligned tether, the two traps are focused at different positions along the optical axis and the tether forms an angle θ with the y axis. D) Misaligned traps, the principal 18 axes of the traps form an angle θ with the y − z reference frame. E) The value of α, Eq. (28), as a function of the mean force, for different tethers (3kbp and 24kbp ds-DNA) and trap stiffnesses. The continuous lines (low trap stiffness) describes a setup similar to that used for the measurements discussed in the present paper, with ky (cid:39) 0.02 pN/nm, kz (cid:39) 0.001 pN/nm. The dotted line (high trap stiffness) describes a setup ten times stiffer (such as that described in Ref. (7, 8)). The shaded area denotes the values of α for which a coupling (cid:39) 0.1 causes a 10% error (2/α (cid:39) 0.1). Figure 4 Trap and molecular stiffness measurements. A) A linear dumbbell model, where three elastic elements with different stiffnesses are arranged in series: Trap 1 (k1), Trap 2 (k2) and the tether (km). B) Measured force vari- ance as a function of trace length. The two data sets refer to the variance in each trap, measured on a 24kbp tether and pulled at 10 pN. Force fluctuations in each trap are the superposition of two different linear modes. The solid curves are fits to the expected behavior in the case of a superposition of two modes (Supp. Mat.). The good agreement between theory and experiment shows that the effect of low frequency noise is not relevant on our experi- mental timescales (less than 1%). C) Molecular stiffness (km) measured and averaged over different molecules. The continuous line shows a fit to the ex- tensible WLC model Eq. (44), giving a persistence length P = 52±4 nm and a stretch modulus S = 1000 ± 200 pN, consistent with what it is reported in the literature (19). The shaded area denotes the region where misalignment is expected to be relevant. The fair points are not included in the fit and show the effect of misalignment. D) Comparison of the stiffness values of the two traps, k1 and k2, measured through Eqs. (39),(40),(41) (solid symbols) with those measured by immobilizing the bead on the micropipette (open symbols), see Materials and Methods Section. Measurements agree within experimental errors. Note that stiffness is measured correctly even when misalignment is relevant (shaded region). This happens beacause the mea- surement of trap stiffness is mostly based on the center of mass coordinate which is not affected in the case of tether misalignment. Figure 5 Time correlation functions in ds-DNA tethers of varying contour length at different forces. Time correlation functions were measured along the y-axis ((cid:104)y+(0)y+(t)(cid:105), (cid:104)y−(0)y−(t)(cid:105)) and normalized by the vari- ances ((cid:104)y2 +(cid:105)). Upper panels: correlation function for fluctuations of the +(cid:105), (cid:104)y2 19 center of mass. The correlation function shows a simple exponential decay as expected for a single component noise. The correlation function changes with force due to trap nonlinearity (change in trap stiffness) but does not show dependence on the length of the tether. Lower panels: correlation function for the distance between the centers of the beads. These correlation functions show a double -- exponential behavior (Fig. 6) which denotes the presence of two relaxational processes. Data for 58 bp and 1.2 kbp DNA tethers are not shown at 16 pN as these experiments were carried out on tethers with an inserted hairpin which unfolds around 14 pN (4), and the released ss-DNA would affect the stiffness measurement. Figure 6 Fast and slow components of the correlation function of the differ- ential coordinate. Double exponential fits to (cid:104)y−(0)y−(t)(cid:105) in semi log plot. Dots show the experimental data, the continuous fair curve superimposed on the data shows a double exponential fit to the measured data. The dark solid curves show the fast and slow components of the double exponential fit. Every plot reports the value of , the coupling strength as obtained from Eq. (42). As the molecules get shorter, the relative weight of slow fluctuations increases indicating a stronger coupling. Figure 7 Measurement of the molecular stiffness. Main Figure: measured molec- ular stiffness for four tethers 58bp (triangles), 1.2 kbp (diamonds), 3 kbp (squared), 24 kbp (circles), as a function of the mean force along the tether. Symbols are measured quantities, data from at least three different molecules have been averaged in the four cases. Solid lines are an EWLC fit to the data (main text). The Marko-Siggia approximation is not valid in the case of the shortest (58 bp) molecule. In this case the fit is only meant to compare the results obtained in the DTOT with those reported in Ref. (4) obtained in a STOT. The fit results are shown in Table 1. Upper right panel, compar- ison of the measured persistence length (P) of the three longer tethers to the empirical scaling law proposed by Seol et al. Eq. (45) (solid line). Fit parameters are discussed in the Main Text. Lower right panel: the measured stretch modulus for the three longer tethers. Errors in P and S values are standard deviation over at least 3 different molecules. 20 Figure 1 21 (cid:76)(cid:97)(cid:115)(cid:101)(cid:114)(cid:32)(cid:100)(cid:105)(cid:111)(cid:100)(cid:101)(cid:32)(cid:65)(cid:76)(cid:97)(cid:115)(cid:101)(cid:114)(cid:32)(cid:100)(cid:105)(cid:111)(cid:100)(cid:101)(cid:32)(cid:66)(cid:87)(cid:105)(cid:103)(cid:103)(cid:108)(cid:101)(cid:114)(cid:87)(cid:105)(cid:103)(cid:103)(cid:108)(cid:101)(cid:114)(cid:80)(cid:101)(cid:108)(cid:108)(cid:105)(cid:99)(cid:108)(cid:101)(cid:80)(cid:101)(cid:108)(cid:108)(cid:105)(cid:99)(cid:108)(cid:101)(cid:80)(cid:83)(cid:68)(cid:80)(cid:83)(cid:68)(cid:80)(cid:66)(cid:83)(cid:80)(cid:66)(cid:83)(cid:108)(cid:47)(cid:52)(cid:79)(cid:98)(cid:106)(cid:101)(cid:99)(cid:116)(cid:105)(cid:118)(cid:101)(cid:79)(cid:98)(cid:106)(cid:101)(cid:99)(cid:116)(cid:105)(cid:118)(cid:101)(cid:108)(cid:47)(cid:52)(cid:80)(cid:66)(cid:83)(cid:80)(cid:66)(cid:83)(cid:80)(cid:83)(cid:68)(cid:66)(cid:117)(cid:108)(cid:108)(cid:101)(cid:121)(cid:101)(cid:80)(cid:104)(cid:111)(cid:116)(cid:111)(cid:100)(cid:101)(cid:116)(cid:101)(cid:99)(cid:116)(cid:111)(cid:114)(cid:66)(cid:117)(cid:108)(cid:108)(cid:101)(cid:121)(cid:101)(cid:80)(cid:104)(cid:111)(cid:116)(cid:111)(cid:100)(cid:101)(cid:116)(cid:101)(cid:99)(cid:116)(cid:111)(cid:114)(cid:80)(cid:83)(cid:68)(cid:67)(cid:111)(cid:108)(cid:108)(cid:105)(cid:109)(cid:97)(cid:116)(cid:105)(cid:110)(cid:103)(cid:32)(cid:108)(cid:101)(cid:110)(cid:115)(cid:67)(cid:111)(cid:108)(cid:108)(cid:105)(cid:109)(cid:97)(cid:116)(cid:105)(cid:110)(cid:103)(cid:32)(cid:108)(cid:101)(cid:110)(cid:115)(cid:67)(cid:67)(cid:68)(cid:32)(cid:99)(cid:97)(cid:109)(cid:101)(cid:114)(cid:97)(cid:82)(cid:101)(cid:108)(cid:97)(cid:121)(cid:32)(cid:108)(cid:101)(cid:110)(cid:115)(cid:82)(cid:101)(cid:108)(cid:97)(cid:121)(cid:32)(cid:108)(cid:101)(cid:110)(cid:115)(cid:76)(cid:69)(cid:68)(cid:70)(cid:108)(cid:117)(cid:105)(cid:100)(cid:105)(cid:99)(cid:115)(cid:32)(cid:99)(cid:104)(cid:97)(cid:109)(cid:98)(cid:101)(cid:114)(cid:77)(cid:111)(cid:116)(cid:111)(cid:114)(cid:105)(cid:122)(cid:101)(cid:100)(cid:32)(cid:115)(cid:116)(cid:97)(cid:103)(cid:101)(cid:65)(cid:115)(cid:112)(cid:104)(cid:101)(cid:114)(cid:105)(cid:99)(cid:97)(cid:108)(cid:108)(cid:101)(cid:110)(cid:115)(cid:65)(cid:115)(cid:112)(cid:104)(cid:101)(cid:114)(cid:105)(cid:99)(cid:97)(cid:108)(cid:108)(cid:101)(cid:110)(cid:115)(cid:76)(cid:105)(cid:103)(cid:104)(cid:116)(cid:45)(cid:108)(cid:101)(cid:118)(cid:101)(cid:114)(cid:76)(cid:105)(cid:103)(cid:104)(cid:116)(cid:45)(cid:108)(cid:101)(cid:118)(cid:101)(cid:114)Figure 2 22 Figure 3 23 Figure 4 24 Figure 5 25 Figure 6 26 Figure 7 27 Supplementary Information 1 Derivation of equations (36-41) in the Main text If misalignment is negligible the experimental setup is described by two co- ordinates y1, y2, as discussed in the Main Text. The equilibrium probability distribution for y1, y2 is a Gaussian distribution: (cid:18)(y1, y2) · ¯K(cid:48)(y1, y2) (cid:19) kBT P (y1, y2) = 1 Z exp , (46) so that the covariance matrix is easily obtained as: (cid:19) . (47) (48) (49) (50) (51) (52) (53) (cid:18) (cid:104)y2 1(cid:105) (cid:104)y1y2(cid:105) (cid:19) (cid:104)y1y2(cid:105) 2(cid:105) (cid:104)y2 = If we set: then, from (47) we get: = kBT ¯K(cid:48)−1 kBT (cid:18) k2 + km km k1k2 + kmk1 + k2km km k1 + km κ = k1k2 + kmk1 + k2km k1 κ k2 κ km κ = = = , , 2(cid:105) − (cid:104)y1y2(cid:105) (cid:104)y2 kBT (cid:104)y2 1(cid:105) − (cid:104)y1y2(cid:105) kBT (cid:104)y1y2(cid:105) kBT . Moreover, using the identity: we get: κ−1 = 1 α = α α2 = kmk2 α2 k1k2 α2 + (cid:104)y2 1(cid:105)(cid:104)y2 k1km α2 + 2(cid:105) − (cid:104)y1y2(cid:105)2 (kBT )2 . In experimental set-ups where forces are measured directly, it is convenient to extract the trap stiffness on the basis of force fluctuation measurements. Force and bead positions have an affine relation: fi = kiyi + f 0 i , 28 (54) where f 0 i put in a matrix form as: is the mean tension measured in trap i. This affine relation can be with f = (f1, f2), y = (y1, y2) and ¯kD = f = ¯kDy + f0 (cid:18) k1 0 0 k2 (cid:19) . (55) (56) Given the affine relations, Eq. (54),(55), the force covariance matrix for the force is now given by: (cid:19) (cid:18) σ2 11 σ2 12 12 σ2 σ2 22 (cid:0)¯K(cid:48)(cid:1)−1 ¯kD = (k2+km)k2 1 k1k2+k1km+k2km k1k2km = kBT ¯kD (cid:32) = kBT (cid:33) (57) k1k2km k1k2+k1km+k2km (k1+km)k2 2 , with σ2 ij = (cid:104)fifj(cid:105) − (cid:104)fi(cid:105)(cid:104)fj(cid:105), i = 1, 2. Using Eq. (57) it is easy to show that: k1k2+k1km+k2km k1k2+k1km+k2km k1 = k2 = 11 + σ2 σ2 12 kBT 22 + σ2 σ2 12 kBT (58) (59) These formulae can be used to invert any element of the covariance matrix to get km: km = 1 kBT σ2 12 (σ2 11 + σ2 σ2 11σ2 12) (σ2 22 − σ4 12 22 + σ2 12) (60) 2 Experimental Variance as a function of mea- surement length The power spectrum of a fluctuating linear mode, x (Ornstein-Uhlembeck process) is: 2ωc S(ω) = σ (61) where σ is the variance (cid:104)δx2(cid:105) of the process and ωc its corner frequency. Inte- grating the power spectrum in the range between the inverse of the acquisition time T and the acquisition bandwidth B yields the expected variance: π(ω2 + ω2 c ) , (cid:90) 2πB 2π/T (cid:104)δx2(cid:105)B,T = dωS(ω) = 2σ π (cid:18)2πB (cid:19) ωc (cid:18) arctan 29 (cid:18) 2π (cid:19)(cid:19) T ωc − arctan (62) If the acquisition bandwidth is much larger than the corner frequency (B (cid:29) ωc) this can be approximated as: (cid:18) (cid:104)δx2(cid:105)T = σ 1 − 2 π (cid:18) 2π (cid:19)(cid:19) T ωc arctan . (63) If a signal y is the superposition of two linear modes with variances σ1, σ2 and corner frequencies ω1, ω2 the expected behavior for the variance (cid:104)δx2(cid:105)T as a function of the acquired trace is: (cid:18) 2π (cid:19)(cid:19) T ω1 (cid:18) 1 − 2 π (cid:18) 2π (cid:19)(cid:19) T ω2 + σ2 arctan , (64) (cid:104)δy2(cid:105)T = σ1 1 − 2 π arctan (cid:18) which is the form of the fit used in Figure 5B. 3 Dumbbell dynamics The discussion in the main text shows that, in absence of misalignment, using the differential and center of mass coordinates, the stiffness tensor is diagonalized and the four fluctuation modes are uncoupled. In non-ideal cases the decoupling is not complete, but the four dimensional problem is reduced into independent lower dimensional problems. Both in the case of trap and of tether misalignment the dynamics of the center of mass is decoupled from that of the differential coordinate: the off diagonal terms couple either y− with z− or y+ with z+ but never y− with z+ or y+ with z−. This fact does greatly simplify the description of the dynamics of the dumbbell in Fig. 3A of the main text: instead of considering a four dimensional problem we can consider two independent two dimensional problems: (cid:0)−¯K+R + ηR (cid:1) , (cid:0)−¯K−r + ηr (cid:1) . R = ¯µR r = ¯µr (65) (66) Here we arranged the coordinates in two vectors: R = (y+, z+), r = (y−, z−), ¯K+ is the subtensor of ¯K which affects y+ and z+, and ¯K− is the subtensor which affects y− and z−. For example, in the case of tether misalignment: (cid:18) y− ¯K− = y− ky + 2u() z− w() (cid:19) z− w() kz + 2v() . (67) 30 with, and v() = w() = u() = km(1 − 2) + (cid:18) f 2 r0 (1 − 2) + km2 f r0 km − f r0 (cid:19)√ (cid:19) (cid:18) y+ z+ 1 − 2, ¯K+ = y+ z+ ky 0 0 kz . Moreover ¯µR, ¯µr are tensors describing both viscous friction on each particle and hydrodynamic interactions while ηR, ηr are Gaussian noises with zero mean and correlations: (cid:104)ηR(t)ηR(s)(cid:105) = 2kBT ¯µRδ(t − s) (cid:104)ηr(t)ηr(s)(cid:105) = 2kBT ¯µrδ(t − s). After Bachelor (1) we set: (68) (69) (70) (71) (72) (73) and + I − r0 ⊗ r0 r2 0 r2 0 ¯µR =(cid:0)γ−1 + Γ−1(cid:1) r0 ⊗ r0 +(cid:0)λ−1 + Λ−1(cid:1)(cid:18) ¯µr =(cid:0)γ−1 − Γ−1(cid:1) r0 ⊗ r0 +(cid:0)λ−1 − Λ−1(cid:1)(cid:18) r2 0 + I − r0 ⊗ r0 r2 0 (cid:19) (cid:19) , where λ, γ, Λ, Γ are scalar parameters depending on r0. In brief, γ (λ) is the hydrodynamic friction coefficient for motions along (perpendicular to) r0, while Γ (Λ) is the intensity of hydrodynamic interactions along (perpen- dicular to) r0 (the vector connecting the centers of the beads in Fig.3A of the main text). It is important to bear in mind that ¯µR, ¯µr, ky, kz, km, r0 are functions of the trap -- to -- trap distance RT or, equivalently, of the mean ten- sion along the tether. The equilibrium probabilities generated by (65),(66) 31 are given by the Boltzmann distribution i.e.: Qeq(R) = Peq(r) = 1 ZR 1 Zr with exp exp −U (R) kBT −V (r) kBT (cid:18) (cid:18) (cid:19) (cid:19) , U (R) = V (r) = R · ¯K+R, r · 2¯K−r 1 2 1 2 , (74) (75) (76) (77) and ZR, Zr partition functions. The variance of equilibrium fluctuations in R and r is connected to the elastic properties of traps and tether by: (cid:104)r ⊗ r(cid:105) =(cid:0)2¯K−(cid:1)−1 kBT. (78) (cid:104)R ⊗ R(cid:105) = ¯K−1 + kBT, Information about hydrodynamic interactions can be obtained from the time- dependent correlation functions (tensors) of R and r: ¯CR(t) = (cid:104)R(t) ⊗ R(0)(cid:105) ¯Cr(t) = (cid:104)r(t) ⊗ r(0)(cid:105), (79) (80) which characterizes the decay of fluctuations and allows to distinguish the presence of different contributions to the total variance. The computation of the correlation functions yields: ¯CR(t) kBT ¯Cr(t) kBT = e−¯µR + ¯K+t ¯K−1 −¯µr(2 ¯K−)t(cid:0)2¯K−(cid:1)−1 . = e (81) (82) 4 Analysis of fluctuations: the uncoupled case = 0 The simplest and most desirable experimental condition is that in which the tether is perfectly aligned to the y axis ( = 0, Fig.3B). In this case fluctuations along the two axis are uncoupled. In the model this corresponds 32 to the vanishing of all off-diagonal elements in the hydrodynamic and elastic tensors. Indeed, when = 0, 0 0 0 kz (cid:18) γ−1 + Γ−1 (cid:18) γ−1 − Γ−1 (cid:18) ky (cid:19) (cid:18) ky + 2km (cid:32) e−ν+t (cid:32) e−ω+t ky 0 = 0 0 = ky+2km 0 ¯µR = ¯µr = ¯K+ = ¯K− = ¯CR(t) kBT ¯Cr(t) kBT 0 λ−1 + Λ−1 0 λ−1 − Λ−1 (cid:19) (cid:19) (cid:19) . 0 kz + 2 f r0 (cid:33) 0 e−ν−t kz 0 e−ω−t kz+2f /r0 (cid:33) . , , (83) (84) (85) (86) (87) (88) The correlation functions are also diagonal in this case: The above expressions shows the presence of 4 different frequencies in the fluctuation spectrum: γ (cid:18) 1 (cid:18) 1 (cid:18) 1 (cid:18) 1 γ λ λ ky (cid:19) (cid:19) (cid:19) (cid:19)(cid:18) kz + + 1 Γ 1 Λ − 1 Γ − 1 Λ ν+ = ν− = ω+ = ω− = (ky + 2km) kz + 2 f r0 (89) (90) (91) (92) . (cid:19) From the measurement of ¯CR(t) and ¯Cr(t) it is possible to obtain the stiffness of both traps and molecule: ky = km = = kBT (cid:0)¯CR(0)(cid:1) (cid:32) (cid:0)¯Cr(0)(cid:1) kBT 1 2 yy yy 33 kBT (¯σ2 R)yy (cid:33) . (93) (94) The time correlation function for fluctuations of R and r, Eq. (87),(88) carries further information regarding hydrodynamic interactions, which can be retrieved once the stiffnesses ky, km are known: log((cid:0)¯CR (cid:1) log((cid:0)¯Cr (cid:12)(cid:12)(cid:12)(cid:12)t=0 (cid:12)(cid:12)(cid:12)(cid:12)t=0 yy) (cid:1) yy) 1 γ 1 γ + 1 Γ − 1 Γ d dt = − 1 ky = − 1 2km d dt (95) (96) . 5 Analysis of fluctuations with tether mis- In presence of tether misalignment ( (cid:54)= 0) we have: alignment (cid:54)= 0 (cid:18) ky (cid:18) ¯K+ = 0 (cid:19) y− 0 kz (cid:19) . z− w() kz + 2v() y− ky + 2u() z− w() ¯K− = (cid:1) (cid:104)y2 +(cid:105) = kBT(cid:0)¯K−1 (cid:104)y2−(cid:105) = kBT(cid:0)2¯K−1 (cid:1) (cid:18) (cid:18) r0km m + yy = = kBT 2km 1 + f = y+y+ kBT ky (cid:19) (cid:19) 2 + O(3). − 1 (97) (98) (99) (100) Since we will be interested in tether misalignment for short tethers (≤ 3 kbp), in the last expression we have neglected the trap stiffness with respect to the tether stiffness: (km (cid:29) ky, kz; f /r0 (cid:29) kz). Note that whereas the variance of R is not affected by the coupling , the variance of r does. The increased (cid:104)y2−(cid:105) is due to the superposition of two contributions, one due to fluctuations in the optical plane and the other due to fluctuations along the optical axis. In order to separate these two types of fluctuations we need to characterize their correlation function. The ¯µr appearing in the correlation function Eq.(82) is left invariant under a rotation of r0. This is also approximately true for ¯K− if km (cid:29) ky, kz; f /r0 (cid:29) kz. As a consequence the correlation function in presence of coupling can be computed as a rotation of ¯Cr(t) obtained in the previous section (Eq. (88)). If we denote by ¯Cr(t, ) the correlation function 34 in presence of a coupling of strength and by ¯R() a rotation of an angle θ ( = sin θ) we get: (cid:19) (cid:18) √ ¯R() = 1 − 2 − √ 1 − 2 (101) (102) (103) and (cid:0)¯Cr(t, )(cid:1) kBT ¯Cr(t, ) = ¯R()T ¯Cr(t)¯R(), =(cid:0)1 − 2(cid:1) e−2ω+t yy + 2 r0e−2ω−t 2f . 2km Similar although more cumbersome formulas can be obtained in more gen- eral cases, i.e. when the trap stiffness ky, kz are comparable or larger than km, f /r0 respectively. Summarizing, in presence of misalignment along the z axis we expect the correlation function of the relative distance to be a dou- ble exponential exhibiting two widely separated timescales: a fast timescale + ) due to fluctuations in the optical plane and a slow timescale (ω−1− ) due (ω−1 to fluctuations along the optical axis. Once the two components have been separated through fitting, as shown in the Main Text, the same analysis as in the uncoupled case can be performed on the fast component of the cor- relation function, to measure the molecular stiffness and the hydrodynamic parameters. From the slow contribution of the correlation function (second term in the r.h.s. of Eq. (103)) it is also possible to extract the coupling pa- rameter, since the ratio f /r0 is independently known. In all the experiments presented in this paper, the coupling parameter was not higher than 0.25, which corresponds to an angle θ (cid:39) 15o. In our setup, especially for short tethers, we can have 2 (cid:39) α, making the slow contribution to the variance (42) comparable or even bigger than the one due to fast fluctuations. References 1. Batchelor, G. K., 1976. Brownian diffusion of particles with hydrodynamic interaction. Journal of Fluid Mechanics 74:1 -- 29. 35
1608.07469	1	1608	2016-08-26T14:22:20	Reply to Comment on "Enhanced diffusion of enzymes that catalyze exothermic reactions"	[ "physics.bio-ph", "q-bio.BM" ]	Catalytically active enzymes have recently been observed to exhibit enhanced diffusion. In a recent work [C. Riedel et al., Nature 517, 227 (2015)], it has been suggested that this phenomenon is correlated with the degree of exothermicity of the reaction, and a mechanism was proposed to explain the phenomenon based on channeling the released heat into the center of mass kinetic energy of the enzyme. I addressed this question by comparing four different mechanisms, and concluded that collective heating is the strongest candidate out of those four to explain the phenomenon, and in particular, several orders of magnitude stronger than the mechanism proposed by Riedel et al. In a recent preprint (arXiv:1608.05433), K. Tsekouras, C. Riedel, R. Gabizon, S. Marqusee, S. Presse, and C. Bustamante present a comment on my paper [R. Golestanian, Phys. Rev. Lett. 115, 108102 (2015); arXiv:1508.03219], which I address here in this reply.	physics.bio-ph	physics	Reply to Comment on "Enhanced diffusion of enzymes that catalyze exothermic reactions" Ramin Golestanian Rudolf Peierls Centre for Theoretical Physics, University of Oxford, 1 Keble Road, Oxford, OX1 3NP, UK (Dated: September 10, 2018) Catalytically active enzymes have recently been observed to exhibit enhanced diffusion. In a recent work [C. Riedel et al., Nature 517, 227 (2015)], it has been suggested that this phenomenon is correlated with the degree of exothermicity of the reaction, and a mechanism was proposed to explain the phenomenon based on channeling the released heat into the center of mass kinetic energy of the enzyme. I addressed this question by comparing four different mechanisms, and concluded that collective heating is the strongest candidate out of those four to explain the phenomenon, and in particular, several orders of magnitude stronger than the mechanism proposed by Riedel et al. In a recent preprint (arXiv:1608.05433), K. Tsekouras, C. Riedel, R. Gabizon, S. Marqusee, S. Press´e, and C. Bustamante present a comment on my paper [R. Golestanian, Phys. Rev. Lett. 115, 108102 (2015); arXiv:1508.03219], which I address here in this reply. PACS numbers: 87.14.ej, 65.80.-g, 87.10.Ca, 87.16.Uv In a recent paper by Riedel et al. [1], empirical ev- idence was presented to suggest that the recently dis- covered phenomenon of enhanced diffusion of catalyti- cally active enzymes correlates with the amount of heat of reaction released into the solution during the cataly- sis, and a theoretical explanation was proposed for the phenomenon. By examining this phenomenon from a number of different angles, I presented a critique of their theoretical proposal in Ref. [3] and argued that a sys- tematic derivation of their result yields values for the diffusion enhancement that are 6-7 orders of magnitude too small. Moreover, I suggested that two other mecha- nisms, namely stochastic swimming and collective heat- ing, are stronger than the proposed mechanism by Riedel et al. [1]. In a comment [2], Tsekouras et al state that they have eliminated alternative explanations of their re- sults through experimental controls, implying that they believe their acoustic wave mechanism is the strongest candidate. I disagree with this statement, which involves numerical evaluations of various contributions that have not been performed consistently for all mechanisms (as discussed in Ref. [3]), favoring one specific mechanism over the others without sufficient evidence. Moreover, this assertion is based on the assumption that there is only one mechanism that governs all the observed exper- iments, whereas the complexity of the system suggests that there are likely several contributions at work which could add up with varying degrees of significance depend- ing on the particular condition in each experiment. Tsekouras et al [2] bring up the point that in the case of collective heating, I have done my estimate for a station- ary state while their experiment is not yet in stationary state; it is in a transient state. In Ref. [3], I have briefly discussed what happens when one needs to consider the transient behavior of the resulting nonlinear heat diffu- sion equation, using an analogy to the phenomenology of flame propagation. The equation admits a propagating wave solution, reminiscent of the Fisher-PKK equation. This occurs as a quick nucleation of a stationary solu- tion that is separated by a sharp boundary from a null solution, with the front moving at a characteristic veloc- ity that is defined as the square root of the product of the effective rate and the heat diffusion coefficient. Using the numbers I use in Ref. [3], this speed comes out as ∼ 1 µm/s. Therefore, the prediction will be that when the effect is observed, it is presumably because the illu- minated spot in which diffusion is probed is inside the stationary part of the wave front, and thus for that part the stationary solution will suffice as an estimate. The speed is fast enough to justify rapid observation of the enhancement effect in the observation domain. The tran- sient nature will then also mean that this effect will decay with substrate depletion. This effect can simply be in- corporated in the estimation by taking into account the effective catalytic reaction rates over the period of obser- vation. The control experiment performed by Riedel et al [1] does not rule out collective heating. Unlike the narra- tive of the observation in Ref. [1], the experiments on unlabeled active catalase plus labeled inactive urease do seem to show an increase in the effective diffusion coeffi- cient consistent with a few percentage rise that is typical in these systems, albeit with large error bars. It is not clear why they have chosen to ignore this clear trend in Ref. Interestingly, the original paper by Ayus- man Sen's group [4] shows a similar control experiment that indeed has an exact same trend of increase by a few percent for "tracers"; see Fig. 8B in Ref. [4]. Consider- ing that the other mechanisms (including the proposed mechanism discussed in Ref. [1]) are several orders of magnitude off scale when one evaluates the quantities by realistic numbers, this suggestive trend that exhibits the right order of magnitude cannot be ignored. [1]. Tsekouras et al [2] mention the observation by Sen et al of separation of active enzymes [5] as proof that col- lective heating cannot explain enhanced diffusion. I have a number of comments on this statement. First, to my knowledge there is as yet no theoretical explanation of the intriguing observation of active enzyme separation. Second, it is not a priori clear that this phenomenon is governed by the same mechanism that leads to enhanced diffusion. Third, my understanding is that the acoustic wave mechanism proposed in Ref. [1] cannot explain the separation phenomenon, so I do not see how this point is relevant to the current debate. The existence of this un- explained phenomenon, however, does indicate that there are more things to discover about these fascinating sys- tems, and it is likely that as yet unknown mechanisms will close the gaps in all these discussions. [3] to convey. It is helpful to re-iterate the message that I had in- tended for Ref. In that paper, I have identified four possible mechanisms that can explain the observation (there could certainly be others), quantified their contributions using simple estimates that could cer- tainly be improved upon by adding more realistic details, and ordered them in terms of magnitude. The conclu- sion is that heating comes out as the strongest, and the mechanism proposed by Riedel et al. [1] based on acous- tic waves comes out as the third in line, and 6-7 orders 2 of magnitude too small if we take into account energy partitioning between deformation modes. While taking into account more realistic aspects such as the specific boundary condition and the transient nature of the ex- periment etc could change these estimates by an order of magnitude, I do not expect this ordering to change when one makes more or less conservative estimates, because of the separation of the orders of magnitude between them. In Ref. [1], a choice has been made to favor the acoustic wave mechanism over heating, while the former is five orders of magnitude smaller than the latter. [1] C. Riedel, R. Gabizon, C.A.M. Wilson, K. Hamadani, K. Tsekourasl, S. Marqusee, S. Presse, and C. Bustamante, Nature 517, 227 (2015). [2] K. Tsekouras, C. Riedel, R. Gabizon, S. Marqusee, S. Press´e, and C. Bustamante, arXiv:1608.05433. [3] R. Golestanian, Phys. Rev. Lett. 115, 108102 (2015). [4] S. Sengupta, K.K. Dey, H.S. Muddana, T. Tabouillot, M.E. Ibele, P.J. Butler, and A. Sen, J. Am. Chem. Soc. 135, 1406 (2013). [5] K. K. Dey, S. Das, M. F. Poyton, S. Sengupta, P. J. Butler, P. S. Cremer, and A. Sen, ACS Nano 8, 11941 (2014).
1707.09009	1	1707	2017-07-27T19:26:12	Determination and biological application of a time dependent thermal parameter and sensitivity analysis for a conduction problem with superficial evaporation	[ "physics.bio-ph" ]	A boundary value problem, which could represent a transcendent temperature conduction problem with evaporation in a part of the boundary, was studied to determine unknown thermophysical parameters, which can be constants or time dependent functions. The goal of this paper was elucidate which parameters may be determined using only the measured superficial temperature in part of the boundary of the domain. We formulated a nonlinear inverse problem to determine the unknown parameters and a sensitivity analysis was also performed. In particular, we introduced a new way of computing a sensitivity analysis of a parameter which is variable in time. We applied the proposed method to model tissue temperature changes under transient conditions in a biological problem: the hamster cheek pouch. In this case, the time dependent unknown parameter can be associated to the loss of heat due to water evaporation at the superficial layer of the pouch. Finally, we performed the sensitivity analysis to determine the most sensible parameters to variations of the superficial experimental data in the hamster cheek pouch.	physics.bio-ph	physics	Determination and biological application of a time dependent thermal parameter and sensitivity analysis for a conduction problem with superficial evaporation Natalia N. Salva ∗a,b,c, Mar´ıa S. Herrerad, Andrea Monti Hughese,f, Claudio Padraa,b,c, and Gustavo A. Santa Cruze aCONICET, San Carlos de Bariloche, 8400, Argentina bDepto. de Mec´anica Computacional, GIA, Comisi´on Nacional de Energ´ıa At´omica, Av. cCentro Regional Bariloche, Universidad Nacional del Comahue, Quintal 200, San Carlos de Bustillo 9500, San Carlos de Bariloche, 8400, Argentina dCETMIC, CIC-CONICET-CCT La Plata, Camino Centenario y 506, M. B. Gonnet, Bariloche, 8400, Argentina eDpto. de Radiobiolog´ıa, GAATEN, Comisi´on Nacional de Energ´ıa At´omica, Av. Gral. Paz 1499, Buenos Aires, B1650KNA, Argentina fCONICET, Godoy Cruz 2290, Buenos Aires, C1425FQB, Argentina B1897ZCA, Argentina Abstract A boundary value problem, which could represent a transcendent temperature conduction problem with evaporation in a part of the boundary, was studied to determine unknown thermophysical parameters, which can be constants or time dependent functions. The goal of this paper was elucidate which parameters may be determined using only the measured superficial temperature in part of the boundary of the domain. We formulated a nonlinear inverse problem to determine the unknown parameters and a sensitivity analysis was also performed. In particular, we introduced a new way of computing a sensitivity analysis of a parameter which is variable in time. We applied the proposed method to model tissue temperature changes under transient conditions in a biological problem: the hamster cheek pouch. In this case, the time dependent unknown parameter can be associated to the loss of heat due to water evaporation at the superficial layer of the pouch. Finally, we performed the sensitivity analysis to determine the most sensible parameters to variations of the superficial experimental data in the hamster cheek pouch. Keywords: Boundary value problem, Determination of parameters, Sensitivity analysis, Hamster cheek pouch. 1 Introduction Diffusion problems have been studied in many different areas, such as heat conduction [1], sediment transport in a river [2], the study of the weather [3], or in drying food technology [4, 5, 6]. The heat transport in a certain ∗Corresponding author: Tel.:+54 294 4445900 (5887). E-mail address: [email protected] Postal address: Laura 7749, (8400) San Carlos de Bariloche, R´ıo Negro, Argentina. 1 object may be modeled through a convection-diffusion differential equation. The boundary conditions usually considered are: a constant temperature (Dirichlet condition), a fixed heat flux (Neumann condition) or a mixed condition, where the heat flux depends on the temperature at the boundary. In [7] some thermal inverse problems were considered in a mathematical tumor model. They estimated simultaneously unknown thermophysical and geometrical parameters, through an evolutionary algorithm. All the parameters were constants and no evaporation was considered. In [8] a space-dependent convection param- eter was determined through a nonlinear least squares technique, using several temperature measurements at different times in different locations of the domain. In [9], we observed that exophytic tumors and irradiated tissues, principally those with ulcers, exhibit mass moisture transfer in the tissue-air interphase. In this work, we resolved a partial differential problem considering a time-dependent extra term in the mixed condition, which can be associated with a loss of heat due to superficial evaporation. Having measurements of temperature in a section of a particular domain during time, a nonlinear inverse problem was formulated to determine some thermophysical parameters such as the superficial heat loss or the heat conductivity. A sensitivity analysis was also performed. Finally the parameter determination and the sensitivity analysis were validated using our previous study on tissue temperature responses in the hamster cheek pouch [9]. 2 Description of the conduction problem Let us consider a one dimensional spatial domain Ω = [0, Xmax], and the following boundary value problem: c1 c5 ∂T ∂t ∂T ∂x (x, t) = c2 ∂2T ∂x2 (x, t) + c3T (x, t) + c4, ∀ (x, t) ∈ Ω × [0, tmax] (x, t) = c6T (x, t) + c7 + f (t), ∀ (x, t) ∈ Γu × [0, tmax] T (x, t) = g(t), T (x, 0) = h(x) ∀ (x, t) ∈ Γb × [0, tmax] ∀ x ∈ Ω (1a) (1b) (1c) (1d) where ci are real constants, f, g : [0, tmax] → R and h : Ω → R are real functions, and Γu = {Xmax}, Γb = {0}1. In general, the parameters ci, i = 1..7, f, g and h are fundamental in using this type of differential equation problem with boundary conditions given by Eqs. (1b), (1c) and (1d). Note that Eq. (1b) includes a time- dependent parameter, f (t), which is defined only in the boundary Γu. In particular, ci are often approximated, and sometimes their values are just guessed. If there is an experimental measure of the temperature T (x, t) on the boundary Γu, an inverse problem can be formulated defining the following cost function: J(T ) = Z tmax 0 ZΓu T (x, t) − T ∗(x, t)2dxdt , (2) where T ∗(x, t) is the measured superficial temperature. Therefore, some parameters can be determined mini- mizing the cost function J. The goal of this paper is to find which parameters ci may be determined using only the measured superficial temperature. First we proposed a method to identify the function f = f (t) such that the solution T of the problem (1) minimizes the cost function (Section 3). Then, we introduced a new way of computing a sensitivity analysis of a variable parameter (Section 4). As an example, in Section 5, we applied this problem to model tissue temperature changes, under transient conditions, in the hamster cheek pouch. In this biological problem, 1We chose one spatial dimension for simplicity. This procedure can be easily extended to more dimensions without loss of generality. 2 the function f (t) can be associated to the loss of heat due to water evaporation at the superficial layer of the cheek pouch. Finally, we performed the sensitivity analysis to determine the most sensible parameters to variations of the superficial experimental data (Section 5.3). 3 Determination of the function f The goal of this section is to determine the function f : [0, tmax] → R, assuming that the rest of the parameters are known constant. To solve the direct problem (1) we used the Finite Element Method (FEM), meshed the spatial-time2 domain Ω × [0, tmax] and obtained the triangulation T = {Ti, i = 1...M }. We define the following finite element spaces: ST = {v ∈ H 1(Ω × [0, tmax])/v ∈ P1(T ), ∀ T ∈ T}, where P1 is the space of polynomials of degree 1, (3) V = {v ∈ ST/v = g, in Γb × [0, tmax] ∧ v = h in Ω × {0}}, V0 = {v ∈ ST/v = 0, in Γb × [0, tmax] ∧ v = 0 in Ω × {0}}. The weak formulation of problem (1) is the following: where, (V P ) : Find T ∈ V / a(T, η) = l(η), ∀η ∈ V0 a(u, v) = Z c1 ∂u ∂t v + c2 ∂u ∂x ∂v ∂x − c3 uv dxdt − Z c2 c6 c5 uv dxdt Γu×[0,tmax] Ω×[0,tmax] l(v) = Z c4 v dxdt + Z c2 c5 (c7 + f (t))v dxdt . (4) (5) (6) (7) Ω×[0,tmax] Γu×[0,tmax] Let {(xi, ti), i = 1, .., N } be the nodes of the triangulation T, and I ⊂ {1, ..., M } be the subset of index whose nodes are in the boundary Γu × [0, tmax]. We define the function f as follows, f (t) = Xi∈I βiηi(Xmax, t), where {ηi}i=1..N is the base of the space V . The inverse problem of determining f consists in obtaining the values of {βi}i∈I such that the solution T of the weak formulation problem is close to the measured temperatures. Let T be a solution of (V P ). The cost function can be approximated by: J(T ) = Z Γu×[0,tmax] (T − T ∗)2 dxdt ≈ Nt Xj=1 (T (Xmax, tj) − T ∗(tj))2 . We defined the following optimization problem: 2The mesh used is a triangular uniform mesh, where the nodes are obtained by the Cartesian product of a discretization in time and a discretization in space. Using polynomials of degree one and this particular mesh makes the FEM equivalent to a finite difference scheme. 3 (V P A)  Z a(v, wβi) = −2 wβi = 0 (uβi − T ∗)v dxdt, ∀v ∈ V Γu×[0,tmax] in Γb × [0, tmax] Find B = {βi}i∈I such that minimizes the cost function j({βi}i∈I) = J(TB), where TB is a solution of (V P ) using the values of B = {βi}i∈I to define f (t). We used the Lagrange Method to determine the variation of j by each βi, and obtaining a direction of descent. Suppose that for r ∈ I − {i} the values of βr are fixed, and we vary only the parameter βi. We define the following function: L(u, w, βi) = J(u) + a(u, w) − lβi(w) (8) where a(·, ·) y l(·) is defined in (6) and (7), and the subscript βi means that we are using in f (t) the fixed values of {βr}r∈I−{i} and the value of βi (which may vary). Note that if uβi is a solution of (VP) then ∀w ∈ V0 : L(uβi, w, βi) = J(uβi) = j(βi), and therefore their derivatives are equal. Using the chain rule we obtained the derivative of the Lagrangian L respect to each βi: δL(uβi , wβi, βi) δβi = Z Γu×[0,tmax] ηiwβi dxdt (9) where wβi ∈ V (Ω) is the adjoint state, which is defined as the unique solution to the following variational problem: , · · · , ∂j(β 0) ∂β1 To find the optimal values of {βi} we used the gradient descent method. This method is based on the observation that if the function j(β) is defined and differentiable in a neighborhood of a point β 0 = {β0 N }, then j(β) decreases fastest if one goes from β 0 in the direction of the negative gradient of j at β 0 , e.g. 1, · · · , β0 −∇j(β 0) = n ∂j(β 0) ∂βN o. It follows that for γ small enough, the value of j in β 1 = β 0 − γ∇j(β 0) is smaller than j(β 0). We get a sequence β0, β1, β2, . . . such that βn+1 = βn − γn∇j(βn) and j(βn+1) ≤ j(βn), for all n ∈ N0. The convergence of this sequence to a local minimum of j depends on the properties of j, (for example, j convex and ∇j Lipschitz). The value of γn > 0 is different in every step. For each n, we searched for the greatest γn ≥ 0 such that it minimizes j(βn − γ∇j(βn)). This is achieved combining a dichotomy method and a method that approximates j by a parabola, for more information about this procedure, see page 51 of [10]. 4 Sensitivity analysis In [11] Blackwell and Dowding analyzed the usage of sensitivity parameters in connection with the estimation of thermal properties in the heat conduction equation. They stated that although parametric investigations may be done, sensitivity parameters are rarely computed (see also [12]). Sensitivity parameters help to understand the parametric dependence of an experiment and shape our experience and intuition for future cases . In inverse problems, the sensitivity parameter is the partial derivative of the output function (in our case the temperature) with respect to a parameter being determined, which is ∂T /∂p for a parameter p. Due to the general interest on the comparison of magnitudes for different parameters, a scaled (sometimes called "modified") sensitivity parameter is used: Sp := p ∂T ∂p . (10) Note that equation (10) has units of temperature for all parameters, therefore magnitudes for various pa- rameters can be directly compared. Small sensitivity parameters, or general insensitivity, are beneficial when the parameters are not well quantified, such as materials with no characterized thermal properties. Then the 4 parameter is not influential in the thermal response. To estimate a parameter, however, the measured response has to be sensible to that parameter. In this case, the scaled sensitivity parameters are desired to be larger in magnitude (compared to the representative temperature) and linearly independent (having different shapes). The more sensitive the temperature is, the more valuable the temperature measurements are. In a similar way, the estimation of multiple parameters requires that the sensitivity, or the effect on temperature of each parameter, is different or independent of one another for each parameter. If two parameters have similar effects on temperature, their individual influence is difficult to distinguish. In this work we analyzed the sensitivity of the superficial temperature, for which we defined the modified sensitivity parameter as follows: Sp(t) := lim ∆→0 p Tp+∆(Xmax, t) − Tp(Xmax, t) ∆ , (11) where Xmax indicates the superficial boundary, and Tα is the solution of the differential problem (1), using the value α in the parameter p. We used the method of finite differences [11] to determine the sensitivity parameters. First we solved the direct problem (1) using the values of the parameters p = (p1, p2, ..., pi, .., pn), and then we solved the direct problem again but with the following values: p△ = (p1, p2, ..., pi + △pi, .., pn). Finally we approximated the sensitivity parameters as follows: Spi (t) ≈ pi Tp△(Xmax, t) − Tp(Xmax, t) △pi . (12) We remark that as every partial derivative, it is dependent not only on time, but also depends strongly in the values assumed by the parameters pi, i = 1...n. This previous analysis works if the parameters are constant. In the case of f which is variable in time, we introduced a new way of defining the sensitivity parameter of a variable parameter: p = (p1, p2, ..., f, .., pn), p△ = (p1, p2, ..., (1 + △).f, .., pn) Sf (t) ≈ Tp△(Xmax, t) − Tp(Xmax, t) △ (13)   where (1 + △).f represents the product of a real number and a function. Observation 1 Note that in order to compute Spi we need to work in a concrete problem, and therefore we will perform the sensitivity analysis for the biological application, where f will be determined (Section 5.3). 5 Biological Application It was previously demonstrated that the hamster cheek pouch is useful for the study of tissue temperature affected by tissue superficial humidity [9, 13]. The hamster cheek pouch is widely used as a model of oral cancer and mucositis, an adverse side effect induced by several cancer therapies [14]. Our group is focused on the study of BNCT (Boron Neutron Capture Therapy), a binary treatment modality that can selectively target neoplastic tissue [15]. Particularly, we study BNCT therapeutic effect on tumors and BNCT induced mucositis in the hamster cheek pouch with a non-invasive complementary method called Dynamic Infrared Imaging (DIRI). This method is based on the observation of temperature changes under transient conditions associated with mass moisture transfer in the tissue-air interface of the pouch. In our previous studies, we described different temperature changes for normal and tumor tissue, and also for non-irradiated and irradiated pouches [9]. However, the study of the mass moisture transfer as a function of time was not quantified. 5 5.1 Dynamic Infrared imaging (DIRI) studies in the hamster cheek pouch Dynamic Infrared imaging (DIRI) is based on the acquisition of thermal images during transient processes, caused by sudden and sustained changes in surface temperature due to the application of a thermal stimulus (provocation test) that forces the neurovascular system to respond in order to maintain local and body tem- perature within normal parameters [16]. Other authors followed this concept, including our group [9, 13, 17], in different clinical research studies using thermography [18, 19, 20, 21]. DIRI provides a non-invasively supple- mentary in vivo information potentially useful to characterize normal and pathological tissues and their response to cancer therapy. The biological model, experimental setup and procedures of the DIRI studies can be found in [9]. Briefly, a total of 61 hamsters were examined under DIRI protocol. Following an acclimatization period in the room, the animals were anesthetized and the pouch was everted using a plastic pipette held by hand. Thermal responses were measured using a FLIR T420 infrared camera, before, during and after the provocation test, namely, Transient Equilibrium Phase (TEP), Provocation Test (PT) and Recovery Phase (RP), respectively. The PT consisted of a mild air current applied at ambient temperature for about 120 seconds. The purpose of an air stimulus is to eliminate the initial moisture condition of the tissue so that, in the RP, we can focus on its thermal behavior and evaporation process that occur in response of the PT. In TEP and RP no air was applied, leaving the pouch exposed to ambient conditions without perturbations during approximately 280 and 400 seconds, respectively. Figure 1(a) shows the normal hamster cheek pouch tissue. The measured temperature values during time were extracted from the thermal image (Figure 1(b)) and averaged in a user-defined region of interest (ROI) used to delineate the normal tissue. (a) (b) Figure 1: Example of a normal hamster cheek pouch (a) and a thermal image during the experiment. The region of interest (ROI), marked with a dashed line, is shown in the IR image (b). 5.2 Determination of the superficial temperature and superficial heat loss In this study, the differential equation model (1) was applied to model the hamster cheek pouch temperature, imposing a loss of heat in the superficial tissue, due to the water evaporation. For this simulation, we used the experimental temperature data as a function of time in a given ROI [9]. The heat transfer modeling in organs has proposed numerous equations, studied by Pennes since 1948 [22, 23]. He suggested that the heat transfer rate between blood and tissue was proportional to the product of the volumetric perfusion rate and the difference between the arterial blood temperature and the local tissue temperature. Therefore the temperature of a tissue depends on the rate of blood perfusion, the metabolic 6 activity and the heat conduction between the tissue and the environment. Taking into account the suggestion made by Pennes [23], Eq. (1a) of problem (1) takes the form: ρc ∂T (x, t) ∂t = k∆xT + ωbρbcb(Tb − T ) + qm, (x, t) ∈ Ω × [0, tmax] (14) where ρ (ρb) represents the tissue (blood) density, c (cb) is the tissue (blood) specific heat, k is the thermal conductivity, ωb is the blood perfusion coefficient, qm is the metabolic heat source and Tb is the constant blood temperature. The boundary condition Eqs. (1b), (1c) and (1d) are: = h(T − Tamb) + L(t), ∂T ∂x −k T = TD, T (x, 0) = F0(x) in Γu × [0, tmax] in Γb × [0, tmax] ∀t ∈ [0, tmax] (15)   Where, Γb = {0} represents the inner boundary and let Γu = {Xmax} represents the superficial boundary. Here, f (t) = L(t) (W/m2) represents the superficial heat loss due to water evaporation, h is the heat transfer coefficient between the tissue and the air and Tamb is the ambient temperature. We used the following initial temperature, that assures the continuity of the temperature values between the two boundaries Γb and Γu: F0(x) = (T0 − Tb) Xmax x + Tb , (16) where T0 is the initial superficial temperature measured. We used a linear function for simplicity. Since thermal responses in the hamster cheek pouch were assessed before, during and after the application of a thermal stimulus (TEP, PT and RP) [9], these three different phases were modeled considering that the heat transfer coefficient h assumed different values in each stage: h = h(t) =   h1 h2 h3 , if 0 < t ≤ 120 , if 120 < t ≤ 200 , if 200 < t ≤ 400 (17) The values used for h1, h2 and h3 and others parameters are shown in Table 1. T0 ◦C 28.715 Tamb ◦C 28.5 Tb ◦C 35 h1 TD ◦C W/m2 ◦C W/m2 ◦C W/m2 ◦C Tamb h3 h2 30 10 10 Table 1: Thermal coefficients depending on the experiment. To our knowledge, there are no published data related to the thermophysical properties (such as thermal conductivity, diusivity, etc.) of the hamster cheek pouch. In [24], Poppendiek et al. suggested that tissues may be considered accurately for thermal analysis as being composed of water, protein and fat. Thus, for the thermal properties needed to compute Eq. (10) in our biological application, we only considered water and protein to calculate the thermal properties, shown in Table 2. The other parameters not related to the composition of the tissue were obtained from [25]. k ωb ρ ρb c cb (W/m ◦C) (W/m3 ◦C) (Kg/m3) (Kg/m3) (J/Kg ◦C) (J/Kg ◦C) 0.445 0.0002 1200 1060 3300 3770 qm (W/m3) 368.1 Table 2: Thermal coefficients used in the calculation. 7 Figure 2 shows the normal hamster cheek pouch experimental and calculated thermal response as a function of time. In Figure 2(a) it can be seen the good approximation between the calculated superficial temperature and the experimental data. In particular, Figure 2(b) depicts the convective coefficient obtained using an initial constant coefficient L(t) = 1000 W/m2, and 15 steps of the gradient descent method described in Section 3. In Figure 2(b), we observed that at the beginning of each stage the superficial heat loss had oscillations. These oscillations can be seen also in the derivative of the functional (see Eq. (2)), and therefore it seems that they are intrinsic of this differential problem. Besides, during the provocation test, the superficial heat loss decreases, due to the air stimulus that helps to eliminate the tissue superficial moisture. (a) Superficial temperature. (b) Superficial heat loss. Figure 2: (a) Normal hamster cheek pouch experimental and calculated thermal response as a function of time. (b) Superficial heat loss obtained through experimental data of figure (a). Vertical dashed lines separate the transient equilibrium phase (TEP), Provocation Test (PT) and Recovery Phase (RP). 5.3 Sensitivity analysis Table 1 and 2 show those parameters used for the sensitivity analysis. We also used the parameter L(t) in Figure 2(b), obtained in the previous section. The initial temperature is in Eq. (16). Table 3 summarizes the physiological parameters considered, showing their units, and their position in the differential problem3, necessary for the sensitivity analysis. Note that some parameters, such as ωb and k, appeared in more than one place in the differential problem. Figure 3 shows different parameters sensitivities, grouped together depending on their order of the sensitivity. A summary of the order of the sensitivity parameters is shown in Table 4. The first important observation is that the most sensible parameter is the ambient temperature (Tamb) (Figure 3(a)). This could be explained by the thinness of the tissue and the Dirichlet condition in the inner surface. Therefore, this parameter should be determined with the smallest error possible. Secondly, we examined the linear dependence between parameters, studied in [26], which could be established by analyzing the shapes of the sensitivity parameters. If two parameters are linearly dependent, this means that there are infinite possible solutions, which implies that there will be infinite local minimums and therefore the minimization problem will not converge. Figure 3(b) shows that L and k are linearly dependent parameters, because their sensitivity are symmetric with respect to the horizontal axis. Therefore, although they have similar sensitivities, they should not be simultaneously determined. Figure 3(d) shows that ωb and Tb are linearly dependent parameters, because their sensitivities have the same behavior (increasing functions). Moreover, these parameters are linearly dependent 3The position in the differential problem, which was taken into account when the sensitivity was computed, changing the parameter p to p + ∆ whenever it appeared. 8 Symbol Units Represents Position in the differential problem∗ T (x, t) ωb k ρ ρb c cb α αb qm Q Tamb Tb TD T0 F0(t) h ◦C 1/s W/m ◦C Kg/m3 Kg/m3 J/Kg ◦C J/Kg ◦C m2/s 1/s W/m3 W/m3 ◦C ◦C temperature of tissue blood perfusion tissue conductivity tissue density blood density specific heat of tissue specific heat of blood diffusivity of tissue diffusivity of blood metabolic heat ωbρbcbTb + qm ambient temperature arterial temperature ◦C ◦C ◦C W/m2 ◦C tissue-air convective coeff. inferior temperature superficial temperature in t = 0 initial temperature L(t) W/m2 X m superficial heat loss tissue depth DE, BC. DE: indep. term, heat source DE: second order term BC: external normal derivative DE: temporal derivative DE: temporal derivative DE: temporal derivative DE: temporal derivative DE: second order term DE: independent term DE: heat source DE: heat source BC: convective condition DE: heat source BC: Dirichlet condition BC: Dirichlet condition BC: Dirichlet condition BC: Dirichlet condition BC: convective condition BC: convective condition EDP: Domain Table 3: Physiological parameters: Units and symbols. ∗DE: Differential equation, BC: Boundary conditions. Order of the sensitivity Parameters 10 1 10−1 10−2 10−4 Tamb L, k ρ, c h, ωb, Tb qm Table 4: Order of the sensitivity of the different parameters. with h1, h2 and h3. Therefore to determine a parameter, of this order of sensitivity, we should choose between ωb, Tb and {h1, h2, h3}. Finally the least sensible parameter is the metabolic heat qm, which may be justified by observing that the total heat source is Q = ωbρbcbTb + qm, and the predominant term is ωbρbcbTb. 9 (a) Sensitivity of Tamb. (b) Sensitivity of k and L. (c) Sensitivity of c (and ρ). (d) Sensitivity of h, ωb and Tb. Figure 3: Scaled sensitivity as a function of time of different parameters. 6 Conclusions We proposed a method for the determination of time dependent parameters using measured superficial tempera- tures in a conduction problem with evaporation, and a new way of computing a sensitivity analysis of a variable parameter. We applied this method successfully to a biological problem, modeling tissue temperature changes, under transient conditions, in the hamster cheek pouch. In this study we found a good approximation between the calculated superficial temperature and the experimental data. We performed a sensitivity analysis, which should be done whenever parameters are simultaneously determined. Based on temperature measurements and having calculated the loss of heat due to water evaporation at the superficial layer of the pouch, the sensitivity analysis determined which of the studied parameters were the most sensible to variations of the superficial experimental data. We found that ambient temperature should be measured with the smallest error, because it is the most sensible parameter in this problem. Moreover, we noted that the linear dependence between the conductivity and the superficial heat loss is not intuitive, in contrast with the dependence between the blood perfusion and the blood temperature (an increase in either of them would result in a raise in the superficial temperature). Therefore, we conclude that a choice has to be made between determining the conductivity or the superficial heat loss. In previous studies, we observed that tumors and particularly a precancerous tissue bearing ulcers after BNCT had high superficial humidity [9]. Thus, in future studies, the proposed mathematical model will be extended to explore the mass moisture transfer as a function of time in tumors and precancerous tissue in the hamster cheek pouch. 10 Acknowledgments This paper has been partially supported by the BNCT project of CNEA, and by CONICET. References [1] S. G. Bankoff, "Heat conduction or diffusion with change of phase," Advances in Chemical Engineering, vol. 5, pp. 75 -- 150, 1964. [2] J. J. Roering, J. W. Kirchner, and W. E. Dietrich, "Evidence for nonlinear, diffusive sediment transport on hillslopes and implications for landscape morphology," Water Resources Research, vol. 35 (3), pp. 853 -- 870, 1999. [3] L. S. Andrew, "Wildland surface fire spread modelling, 19902007. 1: Physical and quasi-physical models," International Journal of Wildland Fire, vol. 18, pp. 349 -- 368, 2009. [4] R. Baini and T. A. G. Langrish, "Choosing an appropriate drying model for intermittent and continuous drying of bananas," Journal of Food Engineering, vol. 79, pp. 330 -- 343, 2007. [5] M. P. Tolaba, R. J. Aguerre, and C. Su´arez, "Modeling cereal grain drying with variable diffusivity," Cereal Chem, vol. 74, pp. 842 -- 845, 2007. [6] C. Su´arez and P. E. Viollaz, "Shrinkage effect on drying behavior of potato slabs," Journal of Food Engi- neering, vol. 13, pp. 103 -- 114, 1991. [7] M. Paruch and E. Majchrzak, "Identification of tumor region parameters using evolutionary algorithm and multiple resiprocity boundary element method," Eng. Appl. Artificial Intelligence, vol. 20, pp. 647 -- 655, 2007. [8] F. S. Baz´an, L. Bedin, and L. S. Borges, "Space-dependent perfusion coefficient estimation in a 2d bioheat transfer problem," Computer Physics Communications, vol. 214, pp. 18 -- 30, 2017. [9] M. Herrera, A. Monti Hughes, N. Salva, C. Padra, A. Schwint, and G. A. Santa Cruz, "Non-invasive characterization of normal and pathological tissues through dynamic infrared imaging in the hamster cheek pouch oral cancer model," SPIE Proceedings: Thermosense: Thermal Infrared Applications XXXIX Paolo Bison; Douglas Burleigh, Editor(s), vol. 10214, 2017. [10] O. Pironneau, Optimal Shape Design for Elliptic Systems. Springer-Verlag, New York, 1984. [11] B. F. Blackwell and K. J. Dowding, Handbook of Numerical Heat Transfer, Chapter 14: Sensitivity analysis ans uncertainty propagation of computational models. Kluwer Academic, Netherlands, 1989. [12] K. Benke, K. Lowell, and A. Hamilton, "Parameter uncertainty, sensitivity analysis and prediction error in a water-balance hydrological model," Mathematical and Computer Modelling, vol. 47, pp. 1134 -- 1149, 2008. [13] G. A. Santa Cruz, S. J. Gonz´alez, A. Dagrosa, A. E. Schwint, M. Carpano, V. A. Trivillin, E. F. Boggio, J. Bertotti, J. Mar´ın, A. Monti Hughes , A. J. Molinari, and M. Albero, "Dynamic infrared imaging for biological and medical applications in boron neutron capture therapy," in Thermosense: Thermal Infrared Appl. XXXIII (M. Safai and J. R. Brown, eds.), vol. 8013, (Orlando, Florida, USA), pp. 7 -- 25, SPIE, 2011. [14] A. Monti Hughes, R. F. Aromando, M. A. P. andA. E. Schwint, and M. E. Itoiz, "The hamster cheek pouch model for field cancerization studies," Periodontol 2000, vol. 67, pp. 292 -- 311, 2015. [15] J. A. Coderre and G. M. Morris, "The radiation biology of boron neutron capture therapy," Radiat Res., vol. 151(1), pp. 1 -- 18, 1999. 11 [16] D. L. Kellogg, "In vivo mechanisms of cutaneous vasodilation and vasoconstriction in humans during thermoregulatory challenges," J. Appl. Physiol., vol. 100(5), pp. 1709 -- 1718, 2006. [17] G. A. Santa Cruz, S. J. Gonz´alez, J. Bertotti, and J. Mar´ın, "First application of dynamic infrared imaging in boron neutron capture therapy for cutaneous malignant melanoma," Med. Phys., vol. 36, pp. 4519 -- 4529, 2009. [18] N. Arora, D. M. andD. Ruggerio, E. Tousimis, A. J. Swistel, M. P. Osborne, and R. M. Simmons, "Effec- tiveness of a noninvasive digital infrared thermal imaging system in the detection of breast cancer," The American Journal of Surgery, vol. 196, pp. 523 -- 526, 2008. [19] C. Hildebrandt, C. Raschner, and K. Ammer, "An overview of recent application of medical infrared thermography in sports medicine in austria," Sensors, vol. 10, pp. 4700 -- 4715, 2010. [20] M. P. C¸ etingul and C. Herman, "Quantification of the thermal signature of a melanoma lesion," Clinics in Dermatology, vol. 13, pp. 329 -- 336, 2011. [21] G. Bhavani Bharathi, S. V. Francis, M. Sasikala, Sandeep, and D. Jaipurka, "Feature analysis for abnor- mality detection in breast thermogram sequences subject to cold stress," in Proceedings of The National Conference on Man Machine Interaction 2014 (M. H. Loew, ed.), NCMMI 2014, pp. 15 -- 21, 2014. [22] H. W. Huang, C. L. Chan, and R. B. Roemer, "Analytical solutions of pennes bio-heat transfer equation with blood vessel," J. of Biomechanical Engineering, vol. 116, pp. 208 -- 212, 1994. [23] H. Pennes, "Analysis of tissue ans arterial blood temperature in the resting human forearm," J. Appl. Physiol., vol. 1, pp. 93 -- 122, 1948. [24] H. F. Poppendiek, R. Randall, J. Breeden, J. E. Chambers, and J. R. Murphy, "Thermal conductivity measurements and predictions for biological fluids and tissues," Cryobiology, vol. 3, pp. 318 -- 327, 1966. [25] M. Pirtini Cetingul and C. Herman, "A heat transfer model of skin tissue for the detection of lesions: sensitivity analysis," Phys. Med. Biol., vol. 55, pp. 5933 -- 5951, 2010. [26] J. Beck and K. Arnold, Parameter estimation in engineering and sciences. John Wiley & Sons, 1977. 12
1601.00711	1	1601	2016-01-05T01:22:11	Open Markov processes: A compositional perspective on non-equilibrium steady states in biology	[ "physics.bio-ph", "cond-mat.soft", "math.CT" ]	In recent work, Baez, Fong and the author introduced a framework for describing Markov processes equipped with a detailed balanced equilibrium as open systems of a certain type. These `open Markov processes' serve as the building blocks for more complicated processes. In this paper, we describe the potential application of this framework in the modeling of biological systems as open systems maintained away from equilibrium. We show that non-equilibrium steady states emerge in open systems of this type, even when the rates of the underlying process are such that a detailed balanced equilibrium is permitted. It is shown that these non-equilibrium steady states minimize a quadratic form which we call `dissipation.' In some circumstances, the dissipation is approximately equal to the rate of change of relative entropy plus a correction term. On the other hand, Prigogine's principle of minimum entropy production generally fails for non-equilibrium steady states. We use a simple model of membrane transport to illustrate these concepts.	physics.bio-ph	physics	Open Markov processes: A compositional perspective on non-equilibrium steady states in biology Blake S. Pollard ∗ Department of Physics and Astronomy University of California Riverside, CA 92521 January 6, 2016 Abstract In recent work, Baez, Fong and the author introduced a framework for describing Markov pro- cesses equipped with a detailed balanced equilibrium as open systems of a certain type. These ‘open Markov processes’ serve as the building blocks for more complicated processes. In this paper, we describe the potential application of this framework in the modeling of biological sys- tems as open systems maintained away from equilibrium. We show that non-equilibrium steady states emerge in open systems of this type, even when the rates of the underlying process are such that a detailed balanced equilibrium is permitted. It is shown that these non-equilibrium steady states minimize a quadratic form which we call ‘dissipation.’ In some circumstances, the dissipation is approximately equal to the rate of change of relative entropy plus a correction term. On the other hand, Prigogine’s principle of minimum entropy production generally fails for non-equilibrium steady states. We use a simple model of membrane transport to illustrate these concepts. 1 Introduction Life exists away from equilibrium. Left isolated, systems will tend toward thermodynamic equi- librium. Open systems can be maintained away from equilibrium via the exchange of energy and matter with the environment. In addition, biological systems typically consist of a large num- ber of interacting parts. This paper presents a way of describing these ‘parts’ as morphisms in a category. A category consists of a collection of objects along with morphisms or arrows between objects, obeying certain conditions. We consider time-homogeneous Markov processes as a general framework for modeling various biological and biochemical systems whose dynamical equations are linear. Viewed as morphisms in a category, the ‘open Markov processes’ discussed in this paper provide a framework for describing open systems which can be combined to build larger systems. Intuitively, one can think of a Markov process as specifying the dynamics of a probability or ‘population’ distribution that is spread across a finite set of states. A population distribution is a non-normalized probability distribution, see for example [15]. The population of a particular ∗Email: [email protected] 1 state can be any non-negative real number. The total population in an open Markov process is not constant in time as population can flow in and out through certain boundary states. Part of the utility of Markov processes as models of physical or biological systems stems from the flexibility in choosing the correspondence between the states of the Markov process and the actual system it is to model. For instance, the states of a Markov process could correspond to different internal states of a particular molecule or chemical species. In this case, the transition rates describe the rates at which the molecule transitions among these states. Or, the states of a Markov process could correspond to a molecule’s physical location. In this case, the transition rates encode the rates at which that molecule moves from place to place. This paper is structured as follows. In Section 2 we give some preliminary definitions from the theory of Markov processes and explain the concept of an open Markov process. In Section 3 we introduce a model of membrane transport as a simple example of an open Markov process. In Section 4, we introduce the category DetBalMark. The objects in DetBalMark are finite sets of ‘states’ whose elements are labeled by non-negative real numbers which we call ‘populations’. The morphisms in DetBalMark are Markov processes equipped with a detailed balanced equilibrium distribution as well as maps specifying input and output states. If the outputs of one process match the inputs of another process the two can be composed, yielding a new open Markov process. We refer to the union of the input and output states as the boundary of an open Markov process. In Section 5, we show that if the populations at the boundary of an open detailed balanced Markov process are held fixed, then the non-equilibrium steady states which emerge minimize a quadratic form, which we call the ‘dissipation,’ subject to the constraint on the boundary popula- tions. Depending on the values of the boundary populations these non-equilibrium steady states can exist arbitrarily far from the detailed balanced equilibrium of the underlying Markov process. In recent work [4], Baez, Fong and the author construct a functor (cid:3) : DetBalMark → LinRel from the category of open detailed balanced Markov process to the category of linear relations. Applied to an open detailed balanced Markov process, this functor yields the subset of allowed steady state boundary population-flow pairs, providing an effective ‘black-boxing’ of open detailed balanced Markov processes. In Section 6 we show that, for fixed boundary populations, this princi- ple of minimum dissipation approximates Prigogine’s principle of minimum entropy production in the neighborhood of equilibrium plus a correction term involving only the flow of relative entropy through the boundary of the open Markov process. 2 Open Markov processes In this section we define open Markov processes, describe the detailed balanced condition for equi- libria and define non-equilibrium steady states for Markov processes. An open Markov process, or open continuous time, discrete state Markov chain, is a triple (V, B, H) where V is a finite set of states, B ⊆ V is the subset of boundary states and H : RV → RV is an infinitesimal stochastic Hamiltonian Hij ≥ 0, i 6= j Hij = 0. Xi For each i ∈ V the dynamical variable pi ∈ [0, ∞), i ∈ V, is the population at the ith state. We call the resulting function p : V → [0, ∞) the population distribution. Populations evolve in 2 time according to the open master equation dpi dt = Xj Hij pj, i ∈ V − B pi(t) = bi(t), i ∈ B. The off-diagonal entries Hij , i 6= j are the rates at which population transitions from the jth to the ith state. A steady state distribution is a population distribution which is constant in time: dpi dt = 0 for all i ∈ V. A closed Markov process, or continuous time, discrete state Markov chain, is an open Markov process whose boundary is empty. For a closed Markov process, the open master equation becomes the usual master equation In a closed Markov process the total population is conserved: dp dt = Hp. dpi dt Xi = Xi,j Hijpj = 0, enabling one to talk about the relative probabilities of being in particular states. A steady-state distribution in a closed Markov process is typically called an equilibrium. We say an equilibrium q ∈ [0, ∞)V of a Markov process is detailed balanced if Hij qj = Hjiqi for all i, j ∈ V. An open detailed balanced Markov process is an open Markov process (V, B, H) together with a detailed balanced equilibrium q : V → (0, ∞) on V . Notice that the populations of all states in a detailed balanced equilibrium are non-zero. For a pair of distinct states i, j ∈ V , the term Hij pj is the flow of population from j to i. The net flow of population from the jth state to the ith is Jij (p) = Hijpj − Hjipi. Summing the net flows into a particular state we can define the net inflow Ji(p) ∈ R of a particular state to be Ji(p) = Xj Jij(p) = Xj Hij pj − Hjipi. Since Pj Hjipi = 0, the right side of this equation is the time derivative of the population at the ith state. Writing the master equation in terms of Jij(p) or Ji(p) we have dpi dt = Xj Jij (p) = Ji(p). The net flow between each pair of states vanishes identically in a detailed balanced equilibrium q: Jij(q) = 0. 3 The existence of a detailed balanced equilibrium is equivalent to a condition on the rates of a Markov process due known as Kolmogorov’s criterion [14], namely that Hi1i2 Hi2i3 · · · Hin−1in Hin i1 = Hi1in Hinin−1 · · · Hi3i2 Hi2i1 for any finite sequence of states i1, i2, . . . , in of any length. This condition says that the product of the rates along any cycle is equal to the product of the rates along the same cycle in the reverse direction. A non-equilibrium steady state is a steady state in which the net flow between at least one pair of states is non-zero. Thus there could be population flowing between pairs of states, but in such a way that these flows still yield constant populations at all states. In a closed Markov process the existence of non-equilibrium steady states requires that the rates of the Markov pro- cess violate Kolmogorov’s criterion. We show that open Markov processes with constant boundary populations admit non-equilibrium steady states even when the rates of the process satisfy Kol- mogorov’s criterion. Throughout this paper we use the term equilibrium to mean detailed balanced equilibrium. 3 Membrane diffusion as an open Markov process To illustrate these ideas, we consider a simple model of the diffusion of neutral particles across a membrane as an open detailed balanced Markov process with three states V = {A, B, C}, input A and output C. The states A and C correspond to the each side of the membrane, while B corresponds within the membrane itself, see Figure 1. A B C Figure 1: A simple model for passive diffusion across a membrane. In this model, pA is the number of particles on one side of the membrane, pB the number of particles within the membrane and pC the number of particles on the other side of the membrane. The off-diagonal entires in the Hamiltonian Hij, i 6= j are the rates at which population hops from j to i. For example HAB is the rate at which population moves from B to A, or from inside the membrane to the top of the membrane. Let us assume that the membrane is symmetric in the sense that the rate at which particles hop from outside of the membrane to the interior is the same on either side, i.e. HBA = HBC = Hin and HAB = HCB = Hout. We can draw such an open Markov 4 process as a labeled graph: Hin Hout qA qA qB qC qC Hout Hin The labels on the edges are the corresponding transition rates. The states are labeled by their detailed balanced equilibrium populations, which, up to an overall scaling, are given by qA = qC = HinHout and qB = H 2 in. Suppose the populations pA and pC are externally maintained at constant values, i.e. whenever a particle diffuses from outside the cell into the membrane, the environment around the cell provides another particle and similarly when particles move from inside the membrane to the outside. We call (pA, pC) the boundary populations. Given the values of pA and pC , the steady state population pB compatible with these values is pB = HinpA + HinpC −HBB = Hin Hout pA + pC 2 . In Section 5 we show that this steady state population minimizes the dissipation, subject to the constraints on pA and pC . We thus have a non-equilibrium steady state p = (pA, pB, pC ) with pB given in terms of the boundary populations above. From these values we can compute the boundary flows, JA, JC as and JA = Xj JC = Xj JAj(p) = HoutpB − HinpA JCj(p) = HoutpB − HinpC. Written in terms of the boundary populations this gives and JA = JC = Hin(pC − pA) 2 Hin(pA − pC ) 2 . Note that JA = −JC implying that there is a constant net flow through the open Markov process. As one would expect, if pA > pC there is a positive flow from A to C and vice-versa. Of course, in actual membranes there exist much more complex transport mechanisms than the simple diffusion model presented here. A number of authors have modeled more complicated transport phenomena using the framework of networked master equation systems [20, 30]. In our framework, we call the collection of all boundary population-flows pairs the steady state ‘behavior’ of the open Markov process. The main theorem of [4] constructs a functor from the category of open detailed balanced Markov process to the category of linear relations. Applied to an open detailed balanced Markov process, this functor yields the set of allowed steady state boundary population-flow pairs. One can imagine a situation in which only the populations and 5 flows of boundary states are observable, thus characterizing a process in terms of its behavior. This provides an effective ‘black-boxing’ of open detailed balanced Markov processes. As morphisms in a category, open detailed balanced Markov processes can be composed, thereby building up more complex processes from these open building blocks. The fact that ‘black-boxing’ is accomplished via a functor means that the behavior of a composite Markov process can be built up from the composite behaviors of the open Markov processes from which it is built. In this paper we illustrate how this framework can be utilized to study linear master equation systems far from equilibrium with a particular emphasis on the modeling of biological phenomena. Markovian or master equation systems have a long history of being used to model and under- stand biological systems. We make no attempt to provide a complete review of this line of work. Schnakenberg, in his paper on networked master equation systems, defines the entropy production in a Markov process and shows that a quantity related to entropy serves as a Lyapunov function for master equation systems [29]. His book [30] provides a number of biochemical applications of networked master equation systems. Oster, Perelson and Katchalsky developed a theory of ‘net- worked thermodynamics’ [19], which they went on to apply to the study of biological systems [20]. Following the untimely passing of Katchalsky, Perelson and Oster went on to extend this work into the realm of chemical reactions [21]. Starting in the 1970’s, T. L. Hill spearheaded a line of research focused on what he called ‘free energy transduction’ in biology. A shortened and updated form of his 1977 text on the subject [10] was republished in 2005 [11]. Hill applied various techniques, such as the use of the cycle basis, in the analysis of biological systems. His model of muscle contraction provides one example [12]. One quantity central to the study of non-equilibrium systems is the rate of entropy production [9, 24, 18, 8]. Prigogine’s principle of minimum entropy production [25] asserts that for non- equilibrium steady states that are near equilibrium, entropy production is minimized. This is an approximate principle that is obtained by linearizing the relevant equations about an equilibrium state. In fact, for open Markov processes, non-equilibrium steady states are governed by a different minimum principle that holds exactly, arbitrarily far from equilibrium. We show that for fixed boundary conditions, non-equilibrium steady states minimize a quantity we call ‘dissipation’. If the populations of the non-equilibrium steady state are close to the population of the underlying detailed balanced equilibrium, one can show that dissipation is close to the rate of change of relative entropy plus a boundary term. Dissipation is in fact related to the Glansdorff-Prigogine criterion which states that a non-equilibrium steady state is stable if the second order variation of the entropy production is non-negative [8, 29]. Starting in the 1990’s, the Qians and their collaborators developed a school studying non- equilibrium steady states, publishing a number of articles and books on the topic [13]. More recently, results concerning fluctuations have been extended to master equation systems [1]. In the past two decades, Hong Qian of the University of Washington and collaborators have published numerous results on non-equilibrium thermodynamics, biology and related topics [26, 27, 28]. This paper is part of a larger project which uses category theory to unify a variety of diagram- matic approaches found across the sciences including, but not limited to, electrical circuits, control theory and bond graphs [3, 2]. We hope that the categorical approach will shed new light on each of these subjects as well as their interrelation, particularly as we generalize the results presented in this and recent papers to the more general, non-linear, setting of open chemical reaction networks. 6 4 The category of open detailed balanced Markov processes In this section we describe how open detailed balanced Markov processes are the morphisms in a certain type of symmetric, monoidal, dagger-compact category. In previous work, Baez, Fong and the author [4] used the framework of decorated cospans [7] to construct the category DetBalMark. Here we give an intuitive description of this category and refer to those papers for the mathematical details. An object in DetBalMark is a finite set with populations, i.e. a finite set X together with a map pX : X → [0, ∞) assigning a population pi ∈ [0, ∞) to each element i ∈ X. A morphism M : (X, pX ) → (Y, pY ) consists of an open detailed balanced Markov process together with input and output maps i : (X, pX ) → (V, q) and o : (Y, pY ) → (V, q) which preserve population so that pX = iq and pY = oq. The boundary B ⊆ V of an open Markov process is the union of the images of the input and output maps B = i(X) ∪ o(Y ). One can draw an open detailed balanced Markov process as a labeled directed graph whose vertices are labeled by their equilibrium populations and with specified subsets of the vertices as the input and the output states. Recall our simple model of membrane diffusion as an open detailed balanced Markov process, which we now think of as a morphism from the input X = {A} to the output Y = {C}: X qA i qA qB o qC qC Y HBA HCB HAB HBC This is a morphism in DetBalMark from X to Y where X and Y are finite sets with populations. In this simple example, X and Y both contain a single element, namely A and C respectively. Suppose we had another such membrane as depicted in Figure 2. This is a morphism in DetBalMark from C ′ D E Figure 2: Another layer of membrane whose interior population is labeled by D and whose exterior populations are labeled by C ′ and E. with input Y = {C ′} and output Z = {E}. Two open detailed balanced Markov processes can be composed if the detailed balanced equilibrium populations at the outputs of one match the detailed balanced equilibrium populations at the inputs of the other. This requirement guarantees that the composite of two open detailed balanced Markov process still admits a detailed balanced 7 equilibrium. HBA HCB HDE HC′ D X qA qA qB qC qC Y Y qC ′ qC ′ qD qE qE Z HAB HBC HED HDC′ If qC = qC ′ in our two membrane models we can compose them by identifying C with C ′ to yield an open detailed balanced Markov process modeling the diffusion of neutral particles across membranes arranged in series: HBA HCB HDC HED X qA qA qB qC qD qE qE Z HAB HBC HCD HDE Notice that the states corresponding to C and C ′ in each process have been identified and become internal states in the composite which is a morphism from X = {A} to Z = {E}. This open Markov process can be thought of as modeling the diffusion across two membranes in series, see Figure 3. A B C D E Figure 3: A depiction of two membranes arranged in series. One can ‘black-box’ an open detailed balanced Markov process by converting it into an electrical circuit, applying the already known black-boxing functor for electrical circuits [3] and translating the result back into the language of open Markov processes [4]. The key step in this process is the construction of a quadratic form which we call ‘dissipation’, analogous to power in electrical circuits, which is minimized when the populations of an open Markov process are in a steady state. 8 5 Principle of minimum dissipation Here we show that by externally fixing the populations at boundary states, one induces steady states which minimize a quadratic form which we call ‘dissipation.’ Definition 1. Given an open detailed balanced Markov process we define the dissipation func- tional of a population distribution p to be D(p) = 1 2 Xi,j Hijqj(cid:18) pj qj − pi qi(cid:19)2 . Given boundary populations b ∈ RB, we can minimize this functional over all p which agree on the boundary. Differentiating the dissipation functional with respect to an internal population, we get Multiplying by qn 2 yields ∂D(p) ∂pn = −2Xj Hnj pj qn . qn 2 ∂D(p) ∂pn = −Xj Hnjpj, where we recognize the right-hand side from the open master equation for internal states. We see that for fixed boundary populations, the conditions for p to be a steady state, namely that is equivalent to the condition that dpi dt = 0 for all i ∈ V, ∂D(p) ∂pn = 0 for all n ∈ V − B. Definition 2. We say a population distribution obeys the principle of minimum dissipation with boundary population b if p minimizes D(p) subject to the constraint that pb = b. With this we can state the following theorem: Theorem 3. A population distribution p ∈ RV is a steady state with boundary population b ∈ RB if and only if p obeys the principle of minimum dissipation with boundary population b. Proof. This follows from Theorem 28 in [4]. Given specified boundary populations, one can compute the steady state boundary flows by minimizing the dissipation subject to the boundary conditions. Definition 4. We call a population-flow pair a steady state population-flow pair if the flows arise from a population distribution which obeys the principle of minimum dissipation. Definition 5. The behavior of an open detailed balanced Markov process with boundary B is the set of all steady state population-flow pairs (pB, JB) along the boundary. Indeed, there is a functor (cid:3) : DetBalMark → LinRel which maps open detailed balanced Markov processes to their steady state behaviors. This is the main result of our previous paper [4]. The fact that this is a functor means that the behavior of a composite open detailed balanced Markov process can be computed as the composite of the behaviors. 9 6 Dissipation and Entropy Production In the last section, we saw that non-equilibrium steady states with fixed boundary populations minimize the dissipation. In this section we relate the dissipation to a divergence between population distributions known in various circles as the relative entropy, relative information or the Kullback- Leibler divergence. The relative entropy is not symmetric and violates the triangle inequality, which is why it is called a ‘divergence’ rather than a metric, or distance function. We show that for population distributions near a detailed balanced equilibrium, the rate of change of the relative entropy is approximately equal to the dissipation plus a ‘boundary term’. The relative entropy of two distributions p, q is given by I(p, q) = Xi pi ln(cid:18) pi qi(cid:19) . It is well known that for a closed Markov process admitting a detailed balanced equilibrium, the relative entropy with respect to this detailed balanced equilibrium distribution is monotonically decreasing with time, see for instance [14]. There is an unfortunate sign convention in the definition of relative entropy: while entropy is typically increasing, relative entropy typically decreases. More generally, the relative entropy between any two population distributions is non-increasing in a closed Markov process. In an open Markov process, the sign of the rate of change of relative entropy is indeterminate. Consider an open Markov process (V, B, H). For any two population distributions p(t) and q(t) which obey the open master equation let us introduce the quantities and Dpi Dt = dpi dt − Xj∈V Hijpj Dqi Dt = dqi dt − Xj∈V Hijqj , which measure how much the time derivatives of p(t) and q(t) fail to obey the master equation. Notice that Dpi Dt = 0 for i ∈ V − B, as the populations of internal states evolve according to the master equation. In terms of these quantities, the rate of change of relative entropy for an open Markov process can be written as d dt I(p(t), q(t)) = Xi,j∈V Hijpj(cid:18)ln(cid:18) pi qi(cid:19) − piqj qipj(cid:19) +Xi∈B Dpi Dt ∂I ∂pi + Dqi Dt ∂I ∂qi . The first term is the rate of change of relative entropy for a closed Markov process. This is less than or equal to zero [5, 23]. Thus, the rate of change of relative entropy in an open Markov process satisfies d dt I(p(t), q(t)) ≤ Xi∈B Dpi Dt ∂I ∂pi + Dqi Dt ∂I ∂qi . This inequality tells us that the rate of change of relative entropy in an open Markov processes is bounded by the rate at which relative entropy flows through its boundary. If q is an equilibrium solution of the master equation dq dt = Hq = 0, 10 then the rate of change of relative entropy can be written as d dt I(p(t), q) = Xi,j∈V (Hij pj − Hjipi) ln(cid:18) piqj qipj(cid:19) +Xi∈B Dpi Dt ∂I ∂pi Furthermore, if q satisfies detailed balance we can write this as d dt I(p(t), q) = − 1 2 Xi,j∈V Jij Aij +Xi∈B Dpi Dt ∂I ∂pi ., where is the thermodynamic flux from j to i and Jij(p) = Hij pj − Hjipi Hjipi(cid:19) Aij (p) = ln(cid:18) Hij pj is the conjugate thermodynamic force. This quantity: 1 2 Xi,j∈V Jij Aij is what Schnakenberg calls “the rate of entropy production” [29]. This is always non-negative. Note that due to the sign convention in the definition of relative entropy, in the absence of the boundary term, a positive rate of entropy production corresponds to a decreasing relative entropy. We shall shortly relate the rate of change of relative entropy to the dissipation for open detailed balanced Markov processes, but first let us consider the quantity Aij (p). It is the entropy production per unit flow from j to i. If Jij (p) > 0, i.e. if there is a positive net flow of population from j to i, then Aij (p) > 0. In addition, Jij (p) = 0 implies that Aij (p) = 0. Thus we see that this form of entropy production is, by definition, non-negative. In the realm of population dynamics, we can understand Aij(p) as the force resulting from a difference in chemical potential. Let us elaborate on this point to clarify the relation of our framework to the language of chemical potentials used in non-equilibrium thermodynamics. Suppose that we are dealing with only a single type of molecule or chemical species. The states could correspond to different locations of the molecule, as in our example of membrane transport. Another possibility is that each state correspond to a different internal configuration of the molecule. In this setting the chemical potential µi is related to the concentration of that chemical species in the following way: µi = µo i + T ln(ci), where T is the temperature of the system in units where Boltzmann’s constant is equal to one and µo i is the standard chemical potential. The difference in chemical potential between two states gives the force associated with the flow of population which seeks to reduce this difference in chemical potential µj − µi = µo j − µo i + T ln(cid:18) cj ci(cid:19) . In general the concentration of the ith state is proportional to the population of that chemical species divided by the volume of the system ci = pi In this case, the volumes cancel out in V . 11 the ratio of concentrations and we have this relation between chemical potential differences and population differences: i + T ln(cid:18) pj pi(cid:19) . µj − µi = µo j − µo This potential difference vanishes when pi and pj are in equilibrium and we have 0 = µo j − µo i + T ln(cid:18) qj qi(cid:19) , or that = e− µo j −µo i T . qj qi If q satisfies detailed balance, then this also gives an expression for the ratio of the transition rates Hji in terms of the standard chemical potentials. Thus we can translate between differences in Hij chemical potential and ratios of populations via the relation qjpi(cid:19) , µj − µi = T ln(cid:18) pjqi which if q satisfies detailed balance gives µj − µi = T ln(cid:18) Hij pj Hjipi(cid:19) . We recognize the right hand side as the force Aij(p) times the temperature of the system T : µj − µi T = Aij (p). Let us return to our expression for d dt I(p(t), q) where q is an equilibrium distribution: d dt I(p(t), q) = − 1 2 Xi,j∈V (Hijpj − Hjipi) ln(cid:18) qipj qjpi(cid:19) +Xi∈B Dpi Dt ∂I ∂pi . Consider the situation in which p is near to the equilibrium distribution q in the sense that pi qi = 1 + ǫi where ǫi ∈ R is the deviation in the ratio pi qi denoted by ǫ. Expanding the logarithm to first order in ǫ we have that from unity. We collect these deviations in a vector d dt I(p(t), q) = − 1 2 Xi,j∈V (Hijpj − Hjipi) (ǫj − ǫi) +Xi∈B Dpi Dt ∂I ∂pi + O(ǫ2), which gives d dt I(p(t), q) = − 1 2 Xi,j∈V (Hijpj − Hjipi)(cid:18) pj qj − pi qi(cid:19) +Xi∈B Dpi Dt ∂I ∂pi + O(ǫ2). 12 By O(ǫ2) we mean a sum of terms of order ǫ2 rewrite this quantity as i . When q is a detailed balanced equilibrium we can d dt I(p(t), q) = − 1 2 Xi,j Hijqj(cid:18) pj qj − pi qi(cid:19)2 +Xi∈B Dpi Dt ∂I ∂pi + O(ǫ2). We recognize the first term as the negative of the dissipation D(p) which yields d dt I(p(t), q) = −D(p) +Xi∈B Dpi Dt ∂I ∂pi + O(ǫ2). We see that for open Markov processes, minimizing the dissipation approximately minimizes the rate of decrease of relative entropy plus a term which depends on the boundary populations. In the case that boundary populations are held fixed so that dpi dt = 0, i ∈ B, we have that Dpi Dt = −Xj∈V Hijpj, i ∈ B. In this case, the rate of change of relative entropy can be written as d dt I(p(t), q) = Xi∈V −B pi qi dpi dt + 2Xi∈B Dpi Dt + O(ǫ2). Summarizing the results of this section, we have that for p arbitrarily far from the detailed balanced equilibrium equilibrium q, the rate of relative entropy reduction can be written as dI(p(t), q) dt = − 1 2 Xi,j Jij (p)Aij (p) +Xi∈B Dpi Dt ∂I ∂pi . For p in the vicinity of a detailed balanced equilibrium we have that dI(p(t), q) dt = −D(p) +Xi∈B Dpi Dt ∂I ∂pi + O(ǫ2) where D(p) is the dissipation and ǫi = pi − 1 measures the deviations of the populations pi from qi their equilibrium values. We have seen that in a non-equilibrium steady state with fixed boundary populations, dissipation is minimized. We showed that for steady states near equilibirum, the rate of change of relative entropy is approximately equal to minus the dissipation plus a boundary term. Minimum dissipation coincides with minimum entropy production only in the limit ǫ → 0. 7 Minimum Dissipation versus Minimum Entropy Produc- tion We return to our simple three-state example of membrane transport to illustrate the difference between populations which minimize dissipation and those which minimize entropy production: X qA qA 1 1 1 1 qB 13 qC qC Y For simplicity, we have set all transition rates equal to one. In this case, the detailed balance equilibrium distribution is uniform. We take qA = qB = qC = 1. If the populations pA and pC are externally fixed, then the population pB which minimizes the dissipation is simply the arithmetic mean of the boundary populations pB = pA + pC . 2 The rate of change of the relative entropy I(p(t), q) where q is the uniform detailed balanced equilibrium is given by d dt I(p(t), q) = −(pA − pB) ln( pA pB ) − (pB − pC) ln( − {z 1 2 Pi,j∈V Jij Aij pB pC ) + (pA − pB)(ln(pA) + 1) + (pC − pB)(ln(pC ) + 1) . } Pi∈B {z Dpi Dt ∂I ∂pi } Differentiating this quantity with respect to pB for fixed pA and pC yields the condition pA + pC 2pB − ln(pB) − 2 = 0. The solution of this equation gives the population pB which extremizes the rate of change of relative entropy, namely pB = pA + pC 2W (cid:16) (pA+pC ) 2 e2(cid:17) , where W (x) is the Lambert W -function or the omega function which satisfies the following relation x = W (x)eW (x). The Lambert W -function is defined for x ≥ −1 e , 0). This sim- ple example illustrates the difference between distributions which minimize dissipation subject to boundary constraints and those which extremize the rate of change of relative entropy. For fixed boundary populations, dissipation is minimized in steady states arbitrarily far from equilibrium. For steady states in the neighborhood of the detailed balanced equilibrium, the rate of change of relative entropy is approximately equal to minus the dissipation plus a boundary term. e and double valued for x ∈ [ −1 8 Discussion Treating Markov processes as morphisms in a category leads naturally to open systems which admit non-equilibrium steady states, even when the transition rates of the underlying process satisfy Kolmogorov’s criterion. Microscopically, all reactions should be reversible with perhaps a large disparity between the forward and reverse rates. Nonetheless, it is clear that biological organisms are capable, at least locally, of storing free energy. This is typically accomplished via the interaction with other systems or the environment. In this paper, the environment served as a reservoir maintaining boundary populations at constant values. Since open Markov processes are morphisms in the category DetBalMark, one can compose these open systems, thereby building up complicated systems in a systematic way. We saw that the non-equilibrium steady states which 14 emerge minimize a quadratic form which depends on the deviation of the steady state populations from the populations of the underlying detailed balanced equilibrium. For steady states in the neighborhood of equilibrium, we saw that the dissipation is in fact the linear approximation of the rate of change of relative entropy with respect to a detailed balanced equilibrium plus a boundary term. In our framework, dissipation appears to be the fundamental quantity as it is minimized for non-equilibrium steady states arbitrarily far from equilibrium. There has been much work examining the regime of validity of Prigogine’s principle of minimum entropy production [16, 17, 6]. In future work, we aim to generalize our framework for composing Markov processes to the non- linear regime of chemical reaction networks with an eye towards incorporating recent interesting results in the area [22]. We anticipate that the perspective achieved by viewing interacting systems as morphisms in a category will bring new insight to the study of living systems far from equilibrium. Acknowledgments The author would like to thank John C. Baez for his help developing the ideas presented in this paper and improving the quality and clarity of their exposition. The author also thanks Brendan Fong for many useful discussions as well as Daniel Cicala for his comments on the draft of this article. The author is grateful to the organizers of the Workshop on Information and Entropy in Biology held at the National Institute for Mathematical and Biological Synthesis (NIMBIOS) in Knoxville, TN as well as to NIMBIOS for its support in attending the workshop. Part of this project was completed during the author’s visit to the Centre for Quantum Technologies (CQT) at the National University of Singapore (NUS) which was supported by the NSF’s East Asia and Pacific Summer Institutes Program (EAPSI) in partnership with the National Research Foundation of Singapore (NRF). References [1] Andrieux D.; Gaspard P. Fluctuation theorem for currents and Schnakenberg network theory. J. Stat. Mech. Theor. Exp. 2006, P01011. 6 [2] Baez, J. C.; Eberle, J. Categories in Control. Theory Appl. Cat. 2015, 30, 836–881. 6 [3] Baez, J. C.; Fong, B. A compositional framework for passive linear networks. Available as http://arxiv.org/abs/1504.05625. 6, 8 [4] Baez, J. C.; Fong, B.; Pollard, B. A compositional framework for open Markov processes. Available as http://arxiv.org/abs/1508.06448. 2, 5, 7, 8, 9 [5] Baez, J. C.; Pollard, B. Relative entropy in biological systems. Available as http://arxiv.org/abs/1512.02742. 10 [6] Bruers, S.; Maes C.; Netocn´y, K. On the validity of entropy production principles for linear electrical circuits. Jour. Stat. Phys. 2007, 129, 725–740. 15 [7] Fong, B. Decorated cospans. Theory Appl. Cat. 2015, 30, 1096–1120. Available at http://www.tac.mta.ca/tac/volumes/30/33/30-33abs.html. 7 15 [8] Glandsorf, P.; Prigogine, I. Thermodynamic Theory of Structure, Stability and Fluctuations; Wiley-Interscience: New York, 1971. 6 [9] de Groot, S. R.; Mazur, P. Non-equilibrium Thermodynamics; North-Holland Publishing Com- pany: Amsterdam, 1962. 6 [10] Hill, T. L. Free Energy Transduction in Biology: The Steady-State Kinetic and Thermodynamic Formalism; Academic Press: New York, 1977. 6 [11] Hill, T. L. Free Energy Transduction and Biochemical Cycle Kinetics; Springer-Verlag: New York, 1989. Republished, Dover: New York, 2005. 6 [12] Hill, T. L.; Eisenberg E. Muscle contraction and free energy transduction in biological systems. Science 1985, 227, 999-1006. 6 [13] Jiang, D.; Qian M.; Qian, M. P. Mathematical Theory of Nonequilibrium Steady States; Springer: Berlin, 2004. 6 [14] Kelly, F. P. Reversibility and Stochastic Networks; Cambridge U. Press: Cambridge, 2011. 4, 10 [15] Kingman, J. F. C. Markov population processes. Jour. Appl. Prob. 1969, 6, 1–18. 1 [16] Landauer, R. Inadequacy of entropy and entropy derivatives in characterizing the steady state. Phys. Rev. A 1975, 12, 636–638. 15 [17] Landauer, R. Stability and entropy production in electrical circuits. Jour. Stat. Phys. 1975, 13, 1–16. 15 [18] Lindblad, G. Non-equilibrium Entropy and Irreversibility; D. Reidel: Dordecht, Holland, 1983. 6 [19] Oster, G.; Perelson, A.; Katchalsky, A. Network thermodynamics. Nature 1971, 234, 393–399. 6 [20] Oster, G.; Perelson, A.; Katchalsky, A. Network thermodynamics: dynamic modeling of bio- physical systems. Quart. Rev. Biophys. 1973, 1, 1–134. 5, 6 [21] Perelson, A; Oster, G. Chemical reaction networks. IEEE Trans. Circ. Sys. 1974, 21, 709–721. 6 [22] Poletinni, M.; Esposito, M. Irreversible thermodynamics of open chemical networks I: Emergent cycles and broken conservation laws. J. Chem. Phys. 2014, 141, 024117. 15 [23] Pollard, B. A Second Law for open Markov processes. Available as arXiv:1410.6531. 10 [24] Prigogine, I. Non-Equilibrium Statistical Mechanics; Interscience Publishers: New York, 1962. 6 [25] Prigogine, I. Etud´e Thermodynamique des ph´enom´enes irr´eversibles; Dunod: Paris and Desoer: Li´ege, 1947. 6 16 [26] Qian, H. Open-system nonequilibrium steady state: statistical thermodynamics, fluctuations, and chemical oscillations. J. Phys. Chem. B 2006, 31, 15063–74. 6 [27] Qian, H.; Beard, D. A. Thermodynamics of stoichiometric biochemical networks in living systems far from equilibrium. Biophys. Chem. 2005, 114, 213–220. 6 [28] Qian, H.; Bishop, L. The chemical master equation approach to nonequilibrium steady-state of open biochemical systems: Linear single-molecule enzyme kinetics and nonlinear biochemical reaction networks. Int. J. Mol. Sci. 2010, 11 (9), 3472–3500. 6 [29] Schnakenberg, J. Network theory of microscopic and macroscopic behavior of master equation systems. Rev. Mod. Phys. 1976, 48, 571–585. 6, 11 [30] Schnakenberg, J. Thermodynamic Network Analysis of Biological Systems; Springer: Berlin, 1981. 5, 6 17
1008.3616	1	1008	2010-08-21T09:03:29	Pore-polymer interaction reveals non-universality in forced polymer translocation	[ "physics.bio-ph" ]	We present a numerical study of forced polymer translocation by using two separate pore models. Both of them have been extensively used in previous forced translocation studies. We show that variations in the pore model affect the forced translocation characteristics significantly in the biologically relevant pore force, i.e. driving force, range. Details of the model are shown to change even the obtained scaling relations, which is a strong indication of strongly out-of-equilibrium dynamics in the computational studies which have not yet succeeded in addressing the characteristics of the forced translocation for biopolymers at realistic length scale.	physics.bio-ph	physics	Pore-polymer interaction reveals non-universality in forced polymer translocation V. V. Lehtola, K. Kaski, and R. P. Linna Department of Biomedical Engineering and Computational Science, Aalto University, P.O. Box 12200, FI-00076 Aalto, Finland (Dated: August 17, 2010) We present a numerical study of forced polymer translocation by using two separate pore models. Both of them have been extensively used in previous forced translocation studies. We show that variations in the pore model affect the forced translocation characteristics significantly in the biologically relevant pore force, i.e. driving force, range. Details of the model are shown to change even the obtained scaling relations, which is a strong indication of strongly out-of-equilibrium dynamics in the computational studies which have not yet succeeded in addressing the characteristics of the forced translocation for biopolymers at realistic length scale. I. INTRODUCTION Polymer translocation has been under intensive research for the past decade due to its relevance for e.g. ultra-fast DNA sequencing [1 -- 5] and many biological processes [6]. Most of the computational research on forced translocation, where the polymer is driven through a nanoscale pore by a poten- tial, has dealt with fairly weak pore potential or force and assumed close-to-equilibrium dynamics. Recently, hydrody- namics has been shown to significantly affect forced translo- cation for experimentally and biologically relevant force mag- nitudes [7, 8]. The pore geometry has been noted to have ef- fect on the experimental translocation process (see e.g. [5]). In addition, the effect of the pore model has been discussed in [9, 10]. However, the significance of the pore model in the forced polymer translocation has not yet been determined. We have recently made a comparison of the unforced and forced translocation [11]. In the first case, the process was seen to be close to equilibrium, as expected, and accordingly its dynamics turned out to be robust against variations in the computational model. However, in the second case the ob- served out-of-equilibrium characteristics for biologically and experimentally relevant pore force magnitudes, made the dy- namics more sensitive to differences in the computational models [8, 12]. Accordingly, we expect the pore model to play a significant role in the forced translocation case [11]. So far, we have consistently used a cylindrical pore imple- mented by a potential symmetrical around the pore axis [8, 11, 12]. However, this differs from the typical pore imple- mentation where the pore is formed by removing one particle from a wall consisting a monolayer of immobile, point-like particles [13 -- 17]. A notable difference is that the latter pore implementa- tion results in a driving potential that is inhomogeneous in the direction of the pore axis. In the case of lattice Boltz- mann simulations these bead-pore models include both the square [16, 17], and cylindrically [18] shaped pores. Typi- cally, the pore-polymer interaction is solely repulsive but the effect of attractive interaction has been investigated in a 2D model [19]. In order to investigate how susceptible translocation dy- namics is to variations in the pore model we have imple- mented in our computational translocation model also the bead pore used in [14]. We compare translocation processes for the bead pore and our original cylindrical pore. The model dependency is expected to be strongest in the pore region un- der strong forcing. We also try to evaluate the effect of the pore model for moderate forcing. When the force bias inside the pore is small, the velocities involved are small enough for the hydrodynamics to be ne- glected [8, 11], and the Langevin dynamics (LD) is a valid choice for the computational model. Assuming that the poly- mer remains close to equilibrium throughout the translocation, the control parameter may be defined as the ratio of the total external force and the solvent friction, ζ = ftot/ξ as done by Luo et al. in [14]. Then the translocation dynamics would depend only on ζ. They reported this to be the case even for a long polymer chain driven through the pore with a signifi- cant value of total pore force ftot. The scaling of the average translocation time with respect to the chain length, τ ∼ N β would then be determined solely by ζ. This is in clear contra- diction with our earlier observation that the scaling exponent β increases with the force due to crowding of the monomers on the trans side [8, 12]. In addition, the effect of the pore model along with other model-dependent factors would in this case be negligible [11]. The increase of β with force cannot hold asymptotically as pointed out in [14], since ultimately very long polymers would translocate faster with a smaller pore force. From Fig. 1 b) in [8] it can be evaluated that this unphysical condition would be seen for N & 106 for the pore force magnitudes 1 and 10. This is comparable to actual DNA lengths of 106 . . . 109 that are far beyond polymer lengths of N . 1000 used in sim- ulations. The translocation dynamics then would have to be different for N > 106, which is in keeping with our earlier qualitative description of the forced translocation [8, 12]. The polymers were seen to be driven out of equilibrium throughout the translocation already for modest pore force magnitudes. The continually increasing drag force on the cis and crowd- ing on the trans side were seen to change the force-balance condition, which is directly reflected on the translocation dy- namics. This force balance is sure to be different for very long polymers. The suggested control parameter ζ, if applicable, would im- ply that the forced translocation could be completely char- acterized through varying it, which again is in stark con- trast to our observation that the forced translocation is a non- universal, out-of-equilibrium process for biologically relevant magnitudes of pore force [8, 12]. Therefore, it is essential to (a) (b) y -z x y x z FIG. 1: (Color online) Schematic depiction of the two pore models. (a) The pores are viewed from the trans side along the z-axis. The small red circle depicts the cylindrical pore of diameter 1.2σ. The bead pore is defined by the eight beads each at distance 1.5σ from the z-axis. The pore beads are drawn with circles using the LJ poten- tial cutoff length 21/6 as their radius. The light blue area in the center of the pore indicates the region where polymer beads have no inter- action with the pore beads. In contrast, the cylinder pore model has a damped-spring-like potential that acts on particles everywhere inside the pore. (b) Side view. The polymer about to translocate (s = 1) is drawn as connected dots. The potentials of the two pore models differ in both the xy-plane and along the z-axis. determine if this contradiction arises from using different pore models. This paper is organized such that we first describe the poly- mer model in Section II and the translocation models in Sec- tion III. The relation between computational and physical pa- rameters is discussed in Section IV. The results are presented and discussed in Section V. We conclude the paper by sum- marizing our findings in Section VI. II. POLYMER MODEL The standard bead-spring chain is used as a coarse-grained polymer model with the Langevin dynamics. In this model adjacent monomers are connected with an-harmonic springs, described by the finitely extensible nonlinear elastic (FENE) potential, UF EN E = − K 2 R2 ln(cid:0)1 − r2 R2(cid:1), (1) where r is the length of an effective bond and R = 1.5 is the maximum bond length. The Lennard-Jones (LJ) potential ULJ = 4ǫ(cid:20)(cid:16) σ r(cid:17)12 −(cid:16) σ r(cid:17)6(cid:21) , r ≤ 21/6σ ULJ = 0, r > 21/6σ, (2) is used between all beads of distance r apart. The parameter values were chosen as ǫ = 1.2, σ = 1.0 and K = 60/σ2. The used LJ potential mimics good solvent. 2 III. TRANSLOCATION MODELS To perform simulations of polymer translocation in 3D we use our translocation model based on Langevin dynam- ics [11]. Hence, the time derivative of the momentum of bead i reads as pi(t) = −ξpi(t) + ηi(t) + f (ri), (3) where ξ, pi(t), ηi(t), and f (ri) are the friction constant, the momentum, random force of the bead i, and the external driv- ing force, respectively. f (ri) is constant but exerted only inside the pore. For unforced translocation f (ri) = 0. ξ and ηi(t) are related by the fluctuation-dissipation theorem. The dynamics was implemented by using velocity Verlet al- gorithm [20]. In our model the wall containing the pore was implemented as a surface on which no-slip boundary conditions are ap- plied to the polymer beads. The two different computational pore models to be compared can be described as follows (see Fig. 1): Cylindrical pore. The pore, aligned with the z-axis, is of diameter 1.2σ(= 1.2b) and length l = nb, where b = 1 is the Kuhn length of the model polymer, and n is either 1 or 3. The pore is implemented by a cylindrically symmetric damped harmonic potential that pulls the beads toward the pore axis passing through the middle of the pore in the z-direction. The pore is nb long, so the total pore force is taken as ftot = nf , where f is the external driving force applied on each polymer bead inside the pore. Bead pore. An octagonal composition of eight immobile particles was used as the bead pore model, as depicted in Fig. 1. This pore model includes a region where a poly- mer bead can reside without interacting with the pore beads, see Fig. 1. The effective pore length is approximately b, so ftot = f . Unlike in our model, in [14] also the wall was made of immobile particles. We verified that this has no effect on the translocation dynamics by reproducing essentially identi- cal results on τ for the force magnitudes that were reported in [14]. In our previous Langevin dynamics simulations with the cylindrical pore [8, 11], parameter values ξ = 0.73, m = 16, and kT = 1 were used for the friction constant, the polymer bead mass, and the temperature in reduced units, respectively. Hence, for long times the one-particle self-diffusion constant was obtained from Einstein's relation as D0 = kT /ξm ≈ 0.086. Time steps of 0.01 and 0.03 were (previously) used in the forced and unforced simulations, respectively. In this pa- per, to compare the two pore models, we have used m = 1, and kT = 1.2 as in [14], unless otherwise noted. Here, the time step is typically 0.001. The number of beads N is odd for polymers initially placed halfway inside the pore and even for polymers having initially only the first bead(s) inside the pore, see Fig. 1. Before al- lowing translocation a polymer is let to relax for longer than its Rouse relaxation time. We register events when a seg- ment s = s0 in the pore is replaced by the segment s0 − 1 or s0 + 1. The polymer is considered translocated, when it has completely exited the pore to the trans side. Exit to either trans or cis side is regarded as an escape from the pore. IV. RELATING THE COMPUTATIONAL AND THE PHYSICAL PORE FORCE The external driving force, called the total pore force, ftot = nf , where n is the number of polymer beads inside the pore and f the z-directional driving force excerted on each bead inside the pore, will be taken as the effective pore force in our model. The thermal energy is set to kT = 1.2 in re- duced units and the length scale is set by the polymer bond length b. The magnitude of the effective pore force is then determined with respect to thermal fluctuations by comparing ftotb with kT . It is noteworthy, however, that since the pore length is of the order of the polymer bond length, there exists an inevitable computational artefact due to the discrete poly- mer model. The external force f is excerted on beads residing in the pore, whose number n does not stay constant during translocation. In addition this number is different for differ- ent pore lengths, so the pore force does not increase strictly linearly with the pore length. These effects are, however, mit- igated by the fact that n averaged over translocation time is constant. Correspondence between computational and physical length scale can be established by taking the polymer bond length b as the Kuhn length for the physical polymer. In SI units the bond length for our FJC model polymer can be ob- tained as b = 2λp, where λp is the persistence length, 40 A for a single-stranded (ss) DNA [21]. The total pore force in SI units, ftot, is then obtained from the dimensionless total pore force, f , by relating ftotb/kB T = ftotb/kT . Here, kB is the Boltzmann constant, and the physical temperature T is taken to be 300 K. Thus the effective pore force of ftot = 1 corresponds to ftot = 1.2 pN for a ssDNA. To relate to experi- ments, in the α-HL pore a typical pore potential of ≈ 120 mV would correspond to ftot ≈ 5 pN when Manning conden- sation leading to drastic charge reduction is taken into ac- count [22, 23]. The used computational pore force magnitudes are thus in the physically relevant range [8]. V. RESULTS A. Forward transfer probability Previously [11], we have shown that during the unforced translocation the polymer remains close to equilibrium so that the forward transition probabilities Pf (s) ∝ (cid:16)1 − 1 s + 1 N − s(cid:17)1−γ (4) derived from the free energy apply in 3D, where γ = 0.69. Including the pore force in the calculation only introduces a prefactor to the above Pf (s) not changing its form, i.e. de- pendence on s. For this to hold, the translocating polymer ) s ( P f 3 f3 f2 f1 1 s/N 1 0.75 0.5 0.5 FIG. 2: (Color online) Forward transfer probabilities Pf as functions of the reaction coordinate normalized with polymer length, s/N. The data is given for the bead (bd) and cylindrical (cyl) pore. Here ξ = 0.7 and N is 255 or 256, depending on the polymer's initial position. The pore force ftot has the following values from top to bottom: f1 = 5.0 (cyl,bd), f2 = 1.17 (bd), f3 = 0.5 (bd). At the bottom are Pf for ftot = 0.1 for the bead (bd, distinct (cid:3) from the solid curve) and the cylindrical (cyl) pore (red) obtained from sim- ulations together with the black solid curve calculated from Eq. (4) for the unforced case. For the ftot values 0.1, 0.5, and 1.17 (f2 up- per curve) the polymer was initially placed halfway through the pore s = (N − 1)/2. For the ftot values 5.0 (both) and 1.17 (f2 lower curve), the polymer started from the cis side s = 1. The shape of the probability curve depends of the pore model (f1), polymer's initial position (f2), and changes with the force. has to remain close to equilbrium. Indeed, except for a small increase due to the force term, the shapes of the measured Pf (s > N/2) for the unforced and forced translocation with the total pore force ftot = 0.1 are seen to be similar, cf. Fig. 2. Pf for the two pore models do not differ appreciably for this small pore force. Hence we conclude that the transition prob- abilities can be obtained from equilibrium framework for pore force magnitudes of the order of 0.1. For the bead pore already for the force ftot = 0.5, the Pf (s) curve differs from the equilibrium shape, as shown in Fig. 2. Hence, the effect of the pore force can no more be taken as a small perturbation to equilibrium dynamics. In keeping with this, the transition probabilities Pf (s > N/2) for ftot = 1.17 are seen to depend on the polymer's initial position, here ei- ther on the cis side or halfway through the pore. In other words, the transition probability from the current state s to the next depends on the path through which this state was ar- rived at. The polymer cannot then have relaxed to equilibrium between previous transitions. Using the cylindrical pore, the probability Pf is seen to be almost constant and close to one for all s with ftot = 5, see Fig. 2. This is in accordance with our previous results that the polymer was seen to almost always translocate with ftot ≃ 3, which means that also in this case Pf (s > 0) would be close to one [8]. However, Pf (s) for the bead pore deviates significantly from Pf (s) for the cylinder pore when ftot = 5, 4 ζ fixed (a) τ 3 10 2 10 10 f1 f2 f3 100 N (b) ) s ( t 8 6 4 2 0 25 50 75 100 125 s FIG. 3: (Color online) Results from 3D Langevin dynamics with a bead pore model. Keeping ζ = ftot/ξ fixed, results for three pairs of parameter values (ftot, ξ): (1.17, 0.7), (5, 3), and (10, 6), are shown. ftot = f1, f2, and f3, respectively. (a) Scaling of the average translocation time τ with respect to the chain length N, τ ∼ N β. The scaling exponent β has values 1.32 ± 0.03 (dotted magenta line), 1.48 ± 0.03 (solid black line) and 1.52 ± 0.03 (dashed blue line) for the parameter pairs, respectively. The exponents are from fits for N ∈ [32, 64, 128, 256]. (b) Average transition times t(s) = t(s − 1 → s) for N = 128. Plots from bottom to top are for ftot = f1, f2, and f3, respectively. The transition time profile changes significantly when increasing the pore force eventhough the proposed contol parameter ζ is kept fixed. see Fig. 2. Thus it is evident that the pore force magnitudes in the two model pores do not have an exact correspondence for large pore force magnitudes. The pore force at which translocation is seen to be a strongly out-of-equilibrium process is surprisingly low for both pore models. It may be compared with the random force in the one-particle Langevin equation sat- isfying fluctuation-dissipation theorem in 3D, for which hf (t)f (t′)i = 6kT ξδ(t − t′). With the parameter values ξ = 0.7 and kT = 1.2 used in our simulations, this yields f ≃ 2.2. Already at fairly moderate pore force values, a local force balance governs the translocation dynamics. This was first proposed by Sakaue [24], and demonstrated in simulations by us [8, 12]. Considering a concept of 'mobile beads' near the pore, we obtained and described qualitatively the scaling rela- tion τ ∼ N 1+ν−σ/ftot, where the parameter σ taking into ac- count the varying number of mobile beads diminishes toward zero with increasing pore force [8, 12]. Sakaue has recently given a fairly quantitative out-of-equilibrium formulation for the scaling τ ∼ N 1+ν/ftot [25]. However, neither description takes into account the crowding of monomers on the trans side shown to be significant for these pore force magnitudes [8]. Here, we present results from simulations of the polymer translocation covering the relevant range of the pore force. When ftotb/kT / 0.1, Pf (s) attains the equilibrium shape and when ftotb/kT ' 5, Pf (s) is close to one. posed control parameter ζ = ftot/ξ fixed. For the parameter pair (ftot, ξ) values (1.17, 0.7), (5, 3), and (10, 6) were used, see Fig. 3 (a). Similar average translocation times τ were ob- tained for large pore force magnitudes, ftot = {5, 10}. How- ever, τ differs appreciably for ftot = 1.17, resulting in a dif- ferent value for the scaling exponent β. The measured average waiting (or state transition) times t(s) = t(s − 1 → s) are compared in Fig. 3 (b). The shown three plots for t(s) are for the above-mentioned val- ues of (ftot, ξ). In the limit ftotb/kT ≪ 1, the average state transition times t(s) were seen to be similar. This is in accord with the transition times in the unforced translocation obeying a close-to-Poissonian distribution [11]. When ftotb/kT ≫ 1 the friction of the solvent dominates over the stochastic term in the Langevin equation, Eq. (3). The transition time profiles for the bead pore are qualitatively similar with those obtained for the cylindrical pore with different parameters [12]. From the different scaling of the translocation times ob- tained for large and small pore force magnitudes and the dif- ferent average state transition times we conclude that ζ which was kept fixed cannot be a universal control parameter for forced translocation regardless of the pore model. This is in keeping with our previous finding that the forced translo- cation is a highly out-of-equilibriun, non-universal process for the biologically and experimentally relevant pore force range [8, 12]. B. The proposed control parameter Using the bead pore, we first measured the scaling exponent β for different values of pore force ftot while keeping the pro- Originally, the need to compare the cylindrical and bead pore models arose from the differences in the scaling of the translocation time τ with the polymer length N in forced C. Pore model matters in forced translocation (a) 4 10 τ 3 10 2 10 1 10 10 m=16 ξ=3 ξ=0.7 100 N 5 1.5 β 1.25 10 1 ξ (b) τ 7 10 6 10 5 10 4 10 3 10 2 10 10 100 N 1000 FIG. 4: (Color online) Scaling of the average translocation time with respect to the chain length, τ ∼ N β. The results are from 3D Langevin dynamics simulations with both cylindrical (cyl) and bead (bd) pore models with different pore lengths l = nb. (a) The constant total pore force ftot = 5. The exponents are from fits for N ∈ [32, 64, 128, 256]. For ξ = 0.7, the scaling exponent β = 1.26 ± 0.02((cid:3)), 1.27 ± 0.02(N) and 1.36 ± 0.03((cid:13)), for l = 3 (cyl), l = 1 (cyl) and l = 1 (bd), respectively. For ξ = 3, the scaling exponent β = 1.39 ± 0.02((cid:13)) and 1.48 ± 0.02((cid:3)), for l = 3 (cyl) and l = 1 (bd), respectively. Increasing the bead mass from 1 to 16 increases the absolute value of τ , and β = 1.40 ± 0.03(•) for ξ = 0.7 (cyl). See text for details. (b) For ftot = 0.5, ξ = 0.7 (H: bd) the polymers are initially halfway through the pore. We obtain β = 1.25 ± 0.04 (green dashed line) with N ∈ {31, 63, 127, 255, 511}. For ftot = 0.1, ξ = 0.7 ((cid:3): cyl, N: bd) polymers are also initially halfway through the pore. The absolute values of τ differ slightly for the two the pore models with shorter polymer chains but not so for N ≃ 511 when all chains escape to the trans side. For a very low force ftot = 0.01((cid:13): bd) even the longest (N = 511) chains may escape to the cis side. We obtain β = 2.2 ± 0.1 (solid black line). For reference, results with ftot = 0.1, ξ = 6 (⋄: bd) are shown. Inset: The scaling exponent β as a function of the friction constant ξ, ftot = 5 (bd). For ξ = {0.7, 1.5, 3.0, 6.0}, we have β = 1.36 ± 0.04, 1.44 ± 0.03, 1.48 ± 0.03, and 1.50 ± 0.03, respectively. translocation obtained using these two pore models. The translocation times averaged over at least 1000 and at most 2500 runs are shown in Fig. 4 (a) as functions of N . The total pore force is constant, ftot = 5, and chosen to be within the experimentally relevant range [8]. For the friction parameter ξ three values, 0.7, 3, and 6, are used. For the two separate pore models we find clear differences in both the absolute value of τ and its apparent scaling. Regardless of the value for ξ, we obtain a smaller scaling exponent β for the cylindrical than for the bead pore. In addition, increasing ξ from 0.7 to 6 increases β for both pore models, see inset in Fig. 4 (b). This in part explains the different results obtained with the two pores as the effective friction experienced by a chain is different inside them. This difference is naturally enhanced at higher veloci- ties. Our previous results were obtained with the polymer bead mass m = 16 due to the requirement of the stochastic rotation dynamics that the polymer beads should be heavier than sol- vent beads [8, 12]. Therefore, we checked how the bead mass affects the characteristics of the translocation time. Increas- ing the bead mass from 1 to 16 is seen not only to increase the average translocation time τ but also to change its appar- ent scaling with polymer length τ ∼ N β, see Fig. 4 (a). To be certain, we checked this using two independent Langevin algorithms. In accordance with our previous findings for cylindrical pore, β was seen to increase with pore force also for the bead pore. For ftot = {0.5, 1.17, 5, 10}, the scaling exponent β = 1.25 ± 0.04 (Fig. 4 (b)), β = 1.32 ± 0.03 (Fig. 3 (a)), β = 1.36 ± 0.03 (Fig. 4 (a)), and β = 1.37 ± 0.03 (not shown) were obtained, respectively. Surprisingly, for the cylindri- cal pore β, which increased substantially with increasing ftot when m = 16 [8, 12], did not show such strong tendency when m = 1. Changing m seems then to change even the qualitative characteristics of forced polymer translocation. The change of β in the scaling τ ∼ N β due to a change in ξ is hardly surprising. However, β changing with m is a bit more subtle. It suggests that we are actually looking at the forced translocation in a continually changing or tran- sient stage. This is in accord with our previous finding that for polymers of lengths that are used in simulations, N ≤ 1000, the number of moving polymer beads changes throughout the translocation and thus continually alters the force balance con- dition resulting from the forced translocation taking place out of equilibrium [8]. Also the crowding of the polymer beads on the trans side, whose relaxation toward equilibrium is slower than the translocation rate changes the force balance continu- ally. Hence, the simulated forced translocation processes do not reveal the asymptotic (N > 106) characteristics for ex- perimentally relevant pore force magnitudes. Then reporting scaling exponents for such forced translocation seems unwar- ranted. The remaining question of interest then is, why does the forced translocation exhibit the scaling albeit with varying exponents? D. Translocation at low pore force For simulations at low pore force the chains were initially placed halfway through the pore unlike at large pore force where the polymer was placed so that only its end was inside the pore. At the very low force of ftot = 0.1 the pore model was found to affect only slightly the absolute value of τ, see Fig. 4 (b). For the bead pore model chains of length N ≤ 127 had a finite probability to escape to the cis side. Unlike for the bead pore, even some of the 255 beads long polymers escaped to the cis side for the cylindrical pore. This is probably due to the cylindrical pore aligning the polymer chain toward the pore axis in the middle thus effectively reducing the friction between the polymer and the pore. For the pore force ftot = 0.01 we obtain β = 2.2 ± 0.1, see Fig. 4 (b), which is the expected exponent for unforced translocation, β = 2ν + 1 = 2.2, since ν = 0.6 has been measured for the swelling exponent in our model [12]. Ap- proaching the unforced case makes the translocation dynam- ics increasingly robust to variations in the model, which is a natural consequence of the small pore force not dominating over entropic forces. We regard the pore force magnitude 0.1 already small enough for the translocating polymer to remain close to equilibrium, see Fig. 2. However, we did not obtain a clear scaling for ftot = 0.1 for either pore models, but β close to the unforced translocation value 2ν + 1 was obtained for chains shorter than N = 127 while a lower value of β was obtained for longer chains. This applies for both ξ = 0.7 and ξ = 6, see Fig. 4 (b). We expect the lower value of β to be 1 + ν (see [26]), although a wider range of N would be needed to confirm this. This apparent cross-over behavior for low pore force is outside the scope of the present paper and calls for a separate, more thorough investigation. For the presumably small pore force ftot = 0.5 we ob- tained β = 1.25 ± 0.04. This differs considerably from β = 1.58 ± 0.03 reported in [14]. From Fig. 2 it can be seen that Pf for ftot = 0.5 clearly deviates from the close- to-unforced form of Pf for ftot = 0.1. Together with the low value of β this observation suggests that ftot = 0.5 is large enough to drive the polymer more and more out of equi- librium as the translocation proceeds. The different values for β here and in [14] also support this view, since the two simulations started from different initial positions. When the relaxation time of the polymer toward thermal equilibrium is slower than the translocation time, the process possesses memory, or correlation over time, i.e. the memory function M (t − t0) 6= δ(t − t0). Thus, a polymer starting half-way through the pore is in a different state than a polymer that has arrived at this position and started from another initial posi- tion. The data presented for different pore force magnitudes in Sections V C and V D show that for ftot ≥ 0.5 the value of β increases with increasing pore force [8, 12] also for the bead 6 pore. For very low pore force β is found independent of the pore force. VI. SUMMARY In summary, we have simulated forced polymer translo- cation in 3D by using Langevin dynamics. We have imple- mented two pore models in our algorithm, (i) a pore sur- rounded by eight immobile beads, which we call bead pore and (ii) a cylindrical pore where a damped harmonic poten- tial confines the beads inside the pore region. The present study compares the effect these two pores have on the forced translocation. We measured the forward transition probabilities Pf (s) to determine the range of pore force that would be sufficiently large to include cases where polymer translocation takes place close to and strongly out of equilibrium. We found that the polymer remains close to equilibrium when the total pore force ftotb/kT / 0.1. Here the translocation dynamics was found to be robust to variations in the translocation model, such as the details of the pore. Pf (s) was found to deviate significantly from the close-to-equilibrium form already for a pore force as small as 0.5. The polymer was found to be driven far from equilibrium when ftotb/kT ' 5. For small pore force magnitudes the forced translocation processes are identical for the two pore models. However, the translocation characteristics were found to be increasingly model dependent when the pore force is increased. This is a natural consequence of the dynamics of the forced translo- cation being determined by a continually changing force- balance condition when the pore force is large enough, i.e. (ftotb/kT ' 0.5). Accordingly, it seems that universal ex- ponents for the forced translocation cannot be found in the biologically relevant pore force regime. In addition, attempts to define a control parameter whose magnitude would con- sistently determine these exponents would seem futile, which we showed for one proposed candidate by using the bead pore. Qualitatively the forced translocation exhibited simi- lar characteristics with both pore models, most notably the increase with the pore force of the exponent determining the relation between the average translocation time and the poly- mer length. In our view this is another indication of the contin- ually changing force-balance condition governing the highly non-equilibrium forced translocation process. Acknowledgments One of the authors (V.V.L.) thanks Dr. K. Luo for email exchanges. This work has been supported by the Academy of Finland (Project No. 127766). The computational resources of CSC-IT Centre for Science, Finland, are acknowledged. [1] J. J. Kasianowicz, E. Brandin, D. Branton, and D. W. Deamer, [2] W. Storm et al, Nano Lett. 5, 1193 (2005). Proc. Natl. Acad. Sci. U.S.A. 93, 13770 (1996). 7 [3] J. Li, M. Gershow, D. Stein, E. Brandin, and J. A. Golovchenko, 68006 (2009). Nature Materials 2, 611 (2003). [4] A. Meller, L. Nivon, and D. Branton, Phys. Rev. Lett. 86, 3435 [15] M. G. Gauthier and G. W. Slater, Eur. Phys. J. E 25, 17 (2008). [16] M. Fyta, E. Kaxiras, S. Melchionna, and S. Succi, Comp. Sci. (2001). & Eng. 2, 20 (2008). [5] M. Zwolak and M. Di Ventra, Rev. Mod. Phys. 80, 141 (2008). [6] B. Alberts et al., Molecular Biology of the Cell (Garland Pub- lishing, New York, 1994). [17] M. Fyta, S. Melchionna, E. Kaxiras, and S. Succi, Multiscale Modeling and Simulation 5, 1156 (2006). [18] M. Bernaschi, S. Melchionna, S. Succi, M. Fyta, and E. Kaxi- [7] M. Fyta, S. Melchionna, S. Succi, and E. Kaxiras, Phys. Rev. E ras, Nanoletters 8, 1115 (2008). 78, 036704 (2008). [8] V. V. Lehtola, R. P. Linna, and K. Kaski, EPL 78, 58006 (2009). [9] D. K. Lubensky and D. R. Nelson, Biophys. J. 77, 1924 (1999). [10] E. Slonkina and A. B. Kolomeisky, J. of Chem. Phys. 118, 7112 (2003). [19] K. Luo, T. Ala-Nissila, S.-C. Ying, and A. Bhattacharya, Phys. Rev. Lett. 99, 148102 (2007). [20] W. van Gunsteren and H. Berendsen, Mol. Phys. 34, 1311 (1977). [21] B. Tinland, A. Pluen, J. Sturm, and G. Weill, Macromol. 30, [11] V. V. Lehtola, R. P. Linna, and K. Kaski, Phys. Rev. E 81, 5763 (1997). 031803 (2010). [12] V. V. Lehtola, R. P. Linna, and K. Kaski, Phys. Rev. E 78, 061803 (2008). [13] K. Luo, S. T. T. Ollila, I. Huopaniemi, T. Ala-Nissila, P. Po- morski, M. Karttunen, S.-C. Ying, and A. Bhattacharya, Phys. Rev. E 78, 050901(R) (2008). [14] K. Luo, T. Ala-Nissila, S.-C. Ying, and R. Metzler, EPL 88, [22] A. Meller, J. Phys. Condens. Matter 15, R581 (2003). [23] A. F. Sauer-Budge, J. A. Nyamwanda, D. K. Lubensky, and D. Branton, Phys. Rev. Lett. 90, 238101 (2003). [24] T. Sakaue, Phys. Rev. E 76, 021803 (2007). [25] T. Sakaue, Phys. Rev. E 81, 041808 (2010). [26] Y. Kantor and M. Kardar, Phys. Rev. E 69, 021806 (2004).
1708.01273	1	1708	2017-08-03T18:36:28	Coincidences with the Artificial Axon	[ "physics.bio-ph", "q-bio.NC" ]	The artificial axon is an excitable node built with the basic biomolecular components and supporting action potentials. Here we demonstrate coincidence firing (the AND operation) and other basic electrophysiology features such as increasing firing rates for increasing input currents. We construct the basic unit for a network by connecting two such excitable nodes through an electronic synapse, producing pre/post synaptic behavior in which one axon induces firing in another. We show that the system is well described by the Hodgkin-Huxley model of nerve excitability, and conclude with a brief outlook for realizing large networks of such low voltage "ionics".	physics.bio-ph	physics	Coincidences with the Artificial Axon Hector G. Vasquez and Giovanni Zocchi∗ Department of Physics and Astronomy, University of California - Los Angeles The artificial axon is an excitable node built with the basic biomolecular components and supporting action potentials. Here we demonstrate coincidence firing (the AND operation) and other basic electrophysiology features such as increasing firing rates for increasing input currents. We construct the basic unit for a network by connecting two such excitable nodes through an electronic synapse, producing pre/post synaptic behavior in which one axon induces firing in another. We show that the system is well described by the Hodgkin-Huxley model of nerve excitability, and conclude with a brief outlook for realizing large networks of such low voltage "ionics". 7 1 0 2 g u A 3 ] h p - o i b . s c i s y h p [ 1 v 3 7 2 1 0 . 8 0 7 1 : v i X r a correctly oriented ion channel species was circumvented by the use of what we call the current limited voltage clamp (CLVC) and only one channel species. The system exhibits threshold behavior and signal amplification, the two hallmarks of action potentials. Here we use the CLVC to more specifically render the effect of a second ionic species and channel type, moving our artificial electrophysiology platform closer to a bona fide artificial axon. We inject currents using a current clamp in the traditional electrophysiology way and demonstrate multiple firing. We demonstrate coincidence detection: the AND operation, fundamental for biological neural networks [4]. Finally, we connect two axons with an electronic "synapse" and demonstrate signal propagation and conditioning. We show that the dynamics of the system is captured by the classic Hodgkin - Huxley (HH) model. We conclude with the outlook for an extended network of artificial axons. Reconstituted lipid membranes with and without channels have a history dating from the 1960s [5]; one important use today is to study ion channel dynamics [6 -- 9]. The alternative method is patch clamping of real cells [10], which has also been used to study the dynamical system of a collection of channels [11, 12]. FIG. 1: Sketch of the artificial axon. A lipid bilayer painted on a ∼ 100µm hole in a plastic tube separates two fluid compartments where different KCl concentrations are maintained. The membrane is studded with 102 − 103 voltage-gated potassium channels (KvAP). The current limited voltage clamp (CLVC) holds the system off the equilibrium Nernst potential, fulfilling the role of the sodium ions and channels in the real axon. The electrode for measuring membrane voltage also serves as the output electrode for a current clamp and input for the electronic synapse, which is a current clamp connecting to the next axon. Introduction. Electrical signal conditioning under water is the hallmark of nerve cells. Action potentials - electrical spikes in the neuron - result from two coupled nonlinear relaxation phenomena, involving "molecular transistors" (voltage gated ion channels in the cell membrane) and supported by the non-equilibrium state created by ionic gradients maintained across the cell membrane. The fundamental mechanism of action potentials was elucidated by Hodgkin and Huxley in the 1950s, and the knowledge constructed since then through electrophysiology experiments on live neurons is monumental [1, 2]. And yet, action potentials were never taken out of the live cell into an in vitro set- ting. This most interesting of dynamical systems has remained confined to the living world. We recently produced a similar excitable medium using the biological components in an artificial setting [3]. The difficulty of constructiong a functional membrane with two different, Results. In the real axon, two steady state, opposing gradients of N a+ and K + ions, maintained across the cell membrane by ATP driven molecular pumps, and small but essential leak conductances for both ions, result in a steady state potential difference across the membrane (the "resting potential" Vr) which is neither the equi- N ∼ librium (Nernst) potential for the sodium ions (V N a +60 mV ) nor the Nernst potential for the potassium ions N ∼ −100 mV ). Action potentials are generated when (V K synapses effectively inject a small current into the neuron, raising the intracellular potential V (we refer potentials to the grounded "outside" of the cell). The voltage sen- sitive sodium channels open, forcing V → V N a N . At these positive voltages the potassium channels open, while the sodium channels close (inactivate), pushing V → V K N . Finally, the potassium channels close again, and V → Vr. The width of the spike (a few ms for mammalian neurons [1]) is governed by the interplay of several characteristic 2 N − V K N ∼ 100 mV ). time scales; the amplitude is fixed by the ionic gradients (essentially V N a The artificial axon [3] is a similarly excitable supported phospholipid bilayer patch of size ∼ 100 µm, separating two fluid compartments (Fig. 1). This artificial mem- brane is studded with ∼ 100 oriented voltage gated potas- sium channels (KvAP), and a concentration difference [K +]in ≈ 30 mM , [K +]out ≈ 150 mM is maintained be- tween the inside and outside compartments, correspond- ing to an equilibrium (Nernst) potential: VN = T e ln [K +]out [K +]in ≈ + 40 mV (1) We keep the system out of equilibrium, at the "resting potential" Vr ≈ −110 mV where the KvAP channels are closed, using a voltage clamp (Fig. 1), which thus has the role of the second ionic gradient in a real axon. How- ever, no dynamics can be generated if the voltage is really clamped, so we limit the current that can be sourced by the clamp using a resistance Rc in series with the clamp electrode. Rc is chosen so that the maximum current sourced by this "current limited voltage clamp" (CLVC) is larger than the leak current with channels closed, but smaller than the channel current with channels open. Then the system can still "fire". More details on the electronics and construction of the excitable membrane are found in [3] and will be given in a longer publication in preparation. Fig. 2 shows the response of the system to a sub- threshold stimulus (a) and a stimulus above threshold (b). The blue curve is the actual voltage V (t) of the axon, the yellow curve the current ICLV C(t) through the clamp, and the dotted curve shows the clamp protocol Vc(t) , which is a square pulse to −25 mV . In (a) the ef- fect is simply the charging and discharging of the mem- brane capacitance, with an RC timescale given by the membrane capacitance C ≈ 192 pF and the CLVC resis- tance Rc = 100 M Ω. In (b) the square pulse stimulus is longer (100 ms instead of 50 ms), and the response is completely different: the system "fires". Notice that af- ter the clamp is pulled back down to Vc = −125 mV (at t ≈ 100 ms) the axon voltage V (t) continues to climb, because channels are open and the channel current over- whelms the clamp current. V (t) does not quite reach the Nernst potential VN ≈ +40 mV because of this competi- tion. The subsequent decrease of V (t) is due to channel inactivation. These traces are essentially deterministic and can be repeated many times. Two sub-threshold pulses in close proximity will cause the system to fire, implementing an AND operation or coincidence detector. An example is shown in Fig. 3 where each 86 ms long pulse alone is sub-threshold, but two such pulses 80 ms apart cause firing. We now proceed to build more features into the system. In electrophysiology, action potentials are usually evoked using a current clamp to inject the stimulus [2]. This is an FIG. 2: Response of the artificial axon to a sub threshold (a) and above threshold (b) stimulus. The dotted line represents the input (the CLVC voltage protocol), the blue trace is the measured axon voltage V (t), and the yellow trace is the current through the CLVC. In both (a) and (b) the CLVC protocol is a square pulse from −125 mV to −25 mV , but the pulse width is 50 ms in (a) and 100 ms in (b), the latter being sufficient to induce firing. The CLVC resistance is Rc = 100 M Ω, and from the sub threshold V (t) trace, fitting with eq. (2) with closed channels (pO = 0), we find the values C = 192 pF for the membrane capacitance and R(cid:96) = 1.2 GΩ for the leak resistance. electronic analogue for the synaptic input. Fig. 4 shows a spike train similarly obtained stimulating the artificial axon with a current clamp, injecting a constant current Ic ≈ 100 pA. However, due to the inactivation proper- ties of the KvAP, to obtain this dynamics we must add a "trigger" function to the CLVC, which has the role of the missing second ionic species and channels. Namely, when the axon voltage V (t) exceeds a fixed trigger volt- age VT the CLVC switches from its fixed potential VC1 to a more negative potential VC2, for a fixed time τ , then goes back to VC1. This is necessary in order that the KvAP channels recover from inactivation. The firing rate is a function of the input current: Fig. 5 shows the firing rate measured on spike trains as in Fig. 4, for increasing input current. All other parameters (VT , VC1, VC2 and τ ) are fixed. The firing rate first increases linearly with input current, but will eventually saturate due to the constant width of the falling edge of the "ac- tion potential". This behavior is analogous to firing rate 3 open, VN the Nernst potential, and V (t) the actual axon potential; the leak current I(cid:96) = −N0χ(cid:96)(V − VN ), χ(cid:96) be- ing the leak conductance; the clamp current ICLV C = −(V − VCLV C)/RC where Rc is the resistance limiting the current in the CLVC. These currents charge the mem- brane capacitance C: χ(cid:96) χ (pO+ dV dt = − N0χ C )[V (t)−VN ]− 1 RCC [V (t)−VCLV C] (2) If channel opening and closing was much faster than all other time scales in the system, pO would simply be the equilibrium open probability at the present voltage: pO = p(V ) [3]. However, this is not the case and one has to consider channel dynamics, which in HH is repre- sented by a set of rate equations. The standard model has 3 states with respect to conduction: Closed, Open (probability pO(t)), Inactive (probability pI (t)), and sev- eral "internal" states. Namely, the ion channel being a tetramer, each subunit is assigned two states, cα and cβ. The closed state that is accessible to the open and inac- tive states requires all 4 subunits be in the cα state, asso- ciated with probability p4(t). Probabilities p3(t), p2(t), p1(t) and p0(t) are associated with states inaccessible to the open and inactive states. These probabilities cor- respond to 3, 2, 1, and 0 subunits in the cα state, re- spectively (more details are given in Supplementary Ma- terials). The dynamics of pO is then given by the rate equations [2, 13]:  + pI (t)kIC(V ) + p3(t)α(V ) O(t) = p4(t)kCO(V ) − pO(t)kOC(V ) p(cid:48) I (t) = p4(t)kCI (V ) − pI (t)kIC(V ) p(cid:48) 4(t) = pO(t)kOC(V ) − p4(t)[kCO(V ) + 4β(V ) + kCI (V )] p(cid:48) 3(t) = 4p4(t)β(V ) − p3(t)[1α(V ) + 3β(V )] p(cid:48) 2(t) = 3p3(t)β(V ) − p2(t)[2α(V ) + 2β(V )] p(cid:48) 1(t) = 2p2(t)β(V ) − p1(t)[3α(V ) + 1β(V )] p(cid:48) 0(t) = 1p1(t)β(V ) − 4p0(t)α(V ) p(cid:48) + 2p2(t)α(V ) + 4p0(t)α(V ) + 3p1(t)α(V ) (3) and eqs. (2) and (3) constitute the model. For the volt- age dependence of the rates in (3) one takes exponen- tial (Arrhenius) functions; the corresponding parameters have indeed been measured for KvAP [6]. We use pa- rameters that best fit our system. In Fig. 6, the fit superimposed on the measured voltage trace V1(t) shows an example of reproducing the dynamics seen in the ex- periments using this model. There are two interesting aspects. One is that the model can be used as an en- gineering tool to predict how more complicated configu- rations of artificial axons will behave. The other is the nonlinear dynamics of this interesting complex system, which however, has been studied [1]. FIG. 3: Coincidence detection with the artificial axon. Two sub threshold pulses (dotted line) in close proximity cause the axon to fire. The first pulse induces opening of some channels, but the channel current cannot overwhelm the clamp (which has the role of the N a+ gradient in the real neuron) absent the second pulse. The CLVC protocol is two identical pulses from −123 mV to −23 mV , of width 86 ms and 80 ms apart. The CLVC resistance is Rc = 100 M Ω, and fitting to the model (2) (see text) gives the values C = 210 pF and R(cid:96) = 0.7 GΩ. observations in patch clamp experiments of real neurons with constant current clamp inputs. Next we connect two artificial axons through an elec- tronic "synapse" which is a current clamp injecting into axon 2 a current controlled by the voltage V1(t) in axon 1. It is a simple analogue circuit consisting of 220 op- amps and 640 resistors. In Fig. 6 we show action poten- tial propagation from axon 1 (blue trace), through the synapse, to axon 2 (yellow trace). Axon 1 is excited by a current clamp injecting a constant current I = 90 pA for 0.1 < t < 3.3 s. The hardwired synapse takes as in- put the voltage of axon 1, V1(t), and injects into axon 2 a current Is proportional to the input V1(t) when V1(t) > 0 : Is(t) = α V1(t) θ(V1) with α ≈ 2 pA/mV and θ the step function. There are no triggers in this experiment. We see that as V1(t) crosses zero (at t ≈ 0.6 s), V2(t) starts to rise as the synapse injects current. Axon 2 "fires" (chan- nels open) at t ≈ 1.3 s. At t ≈ 1.5 s V1(t) crosses zero on its way down; even though the synapse stops injecting current, V2(t) remains high because the channel current overwhelms the CLVC current. Eventually channels inac- tivate and the CLVC pulls the membrane potential back down to the resting potential. In short, Fig. 6 demon- strates signal propagation between two axons connected by a synapse: the basic unit for a network. Model. The dynamics of the artificial axon is captured by the classic HH model of action potentials (as well as by simplified versions, discussed in a forthcoming publica- tion). There are three contributions to the current in the system: the ion channel current IKv = −N0χpO(V −VN ) where N0 is the number of channels, χ the open chan- nel conductance, pO the probability that channels are 4 FIG. 4: Spike train obtained under constant input current conditions. A current clamp injects Ic = 100 pA into the axon. The CLVC parameters are VT = 0 mV , VC1 = −364 mV , VC2 = −636 mV , τ = 1 s (see text). This particular prep has a relatively large leak current. FIG. 6: Signal propagation from axon 1, through the synapse, to axon 2. The synapse is a hard wired current clamp controlled by V1(t) and injecting current into axon 2. Also shown is a fit of the axon 1 potential V1(t) using the model (2), (3). It returns the parameter values N0 = 461, C = 175 pF , χ = (6GΩ)−1, χ(cid:96) = (68GΩ)−1/N0, VN = 56 mV . The CLVC resistance was Rc = 2 GΩ. size can be reduced and much faster channels than the KvAP are available, so the above are by no means unre- alistic conditions), the capacitance is C ≈ 10−13 F and it takes only the transfer of ∼ 105 charges to step the volt- age by ∼ 100 mV , giving a dissipation of ∼ 104 eV over 10 ms, or 0.1 pW . One can certainly imagine a ∼ 10 cm size chip containing ∼ 108 membrane patches. For com- parison, our brain has ∼ 1011 neurons and consumes ∼ 30 W , or 300 pW per neuron, though only a fraction of this dissipation is due to action potentials. A flip-flop circuit in a CPU (consisting of ∼ 10 transistors) con- sumes ∼ 100 µW at 1 GHz or ∼ 10 pW at 100 Hz. This factor 100 in power dissipated compared to our hypothet- ical membrane patch comes essentially from the factor 10 reduction in operating voltage: 1 V → 100 mV between transistors and voltage gated channels. The biological channels are the much sought for low voltage transistors! Indeed, this mismatch is the main problem while inter- facing "bare" electronic components with the artificial axon; for instance, in order to make a one-way synapse one cannot simply connect two axons through a diode, because of the voltage drop across the diode. Now on the negative side, for a viable network there are problems to be solved requiring altogether new inven- tions. The first is scaling up the system, introducing two channel species, and obtaining the necessary robustness. Another challenge is creating an artificial electrochemical synapse which uses only low voltage components. In the next decades, we look forward to progress being made on both fronts. FIG. 5: Measured firing rate vs input current. Each data point is obtained from a spike train as in Fig. 4. The fixed CLVC parameters are: VT = 14 mV , VC1 = −125 mV , VC2 = −459 mV , τ = 500 ms. Discussion. The artificial axon represents an exper- imentally controlled, synthetic realization of this most interesting of nonlinear dynamical systems, the excitable cell. It is to our knowledge the first artificial system based on the bio-components that generates something similar to action potentials. Even in its present simple form it can serve as a breadboard for electrophysiology measure- ments which does not involve lab animals. For example, the electrophysiology of real neurons is extremely tem- perature dependent [14, 15]; future experiments with the artificial axon may elucidate the causes. Networks of excitable nodes are intensively studied com- putationally [16]. What is the outlook for a network of artificial axons? On the positive side there are power and space considerations. Reasoning on a hypothetical A = 100 µm2 membrane patch with N = 100 channels, responding on a time scale ∆t ∼ 10 ms (our present axon is larger: A ∼ 104 µm2, and slower: ∆t ∼ 100 ms, but ACKNOWLEDGEMENTS We thank Roderick MacKinnon for the original gift of the KvAP plasmid, and Amila Ariyaratne for invaluable advice and support. This work was supported by NSF grant DMR-1404400. ∗ [email protected] [1] Christof Koch, Biophysics of Computation (Oxford Univ. Press, 1999). [2] Bertil Hille, Ion channels of excitable membranes (3rd edition, ISBN-10: 0878933212). [3] Amila Ariyaratne and Giovanni Zocchi, "Toward a mini- mal artificial axon," J. Phys. Chem. B 120, 6255 (2016). [4] Ofer Feinerman, Assaf Rotem, and Elisha Moses, "Reli- able neuronal logic devices from patterned hippocampal cultures," Nature Physics 4, 967 -- 973 (2008). [5] Paul Mueller, Donald O. Rudin, H. Ti Tien, and William C. Wescott, "Reconstitution of cell membrane structure in vitro and its transformation into an excitable system," Nature 194, 979 -- 980 (1962). [6] Daniel Schmidt, Samuel R. Cross, and Roderick Mackin- non, "A gating model for the archeal voltage-dependent k+ channel KvAP in DPhPC and POPE:POPG decane lipid bilayers," J. Mol. Biol. 390 (2009). [7] Sophie Aimon, John Manzi, Daniel Schmidt, Jose An- tonio Poveda Larrosa, Patricia Bassereau, and Gilman 5 E. S. Toombes, "Functional reconstitution of a voltage- gated potassium channel in giant unilamellar vesicles," PLoS ONE 6, e25529 (2011). [8] Amila Ariyaratne and Giovanni Zocchi, "Nonlinearity of a voltage gated potassium channel revealed by the me- chanical susceptibility," Phys. Rev. X 3, 011010 (2013). [9] Vanessa Ruta, Youxing Jiang, Alice Lee, Jiayun Chen, and Roderick MacKinnon, "Functional analysis of an archaebacterial voltage-dependent K + channel," Nature 422, 180 -- 185 (2003). [10] B. Sakmann and E. Neher, "Patch clamp techniques for studying ionic channels in excitable membranes," Ann. Rev. Physiol. 46, 455 -- 72 (1984). [11] Hanna Salman, Yoav Soen, and Erez Braun, "Voltage fluctuations and collective effects in ion-channel protein ensembles," Phys. Rev. Lett. 77, 4458 (1996). [12] Hanna Salman and Erez Braun, "Voltage dynamics of single-type voltage-gated ion-channel protein ensembles," Phys. Rev. E 56, 852 (1997). [13] Sakmann and Neher, Single-Channel Recording (Springer, 2nd edition, ISBN: 9781441912305, 2009). [14] L. Hodgkin and B. Katz, "The effect of temperature on the electrical activity of the giant axon of the squid," J. Physiol. 109, 240 -- 49 (1949). [15] Thompson SM, Masukawa LM, and Prince DA, "Tem- perature dependence of intrinsic membrane properties and synaptic potentials in hippocampal ca1 neurons in vitro," Neurosci. 5, 817 -- 24 (1985). [16] Daniel B. Larremore, Woodrow L. Shew, Edward Ott, Francesco Sorrentino, and Juan G. Restrepo, "Inhibi- tion causes ceaseless dynamics in networks of excitable nodes," Phys. Rev. Lett. 112, 138103 (2014). Supplemental Materials 6 MATERIALS AND METHODS We use a voltage gated potassium ion channel from the thermophilic bacterium Aeropyrum pernix (KvAP). The channel is closed at large negative voltages; the midpoint of the open probability curve is at V ≈ −20 mV , and the open channel conductance is χ ≈ 10 pA/60 mV [8, 9]. KvAP has a slow dynamics of inactivation; to recover from inactivation, it must be held at voltages below ∼ −100 mV . Vesicles and decane lipid membranes are made with the phospholipid DPhPC. A. Expression and Reconstitution KvAP protein expression, purification and reconstitu- tion mostly follow the protocols in [9], and are described at length in [8]. Briefly, the KvAP gene in vector pQE60 is expressed in E. coli cells (Aligent XL-1 Blue). Vector transformation into E. coli is achieved with a 45 second, 42 degree heat pulse. Cells with KvAP protein are re- suspended in buffer and lysed in a homogenizer (Avestin EmulsiFlex-C3). Protein is extracted from the lysate with decylmaltoside detergent (DM, Anatrace) and cen- trifugation. Talon nickel bead resin (Clonetech) is added to the his-tagged ion channel solution for binding and pu- rification. The mixture is passed through a Talon affinity column then eluted. The elution is treated with thrombin to remove the his-tag from the ion channel. Thrombin and his-tag are removed from the sample by size exclu- sion chromatography (GE Healthcare, Superdex-200). DPhPC is suspended in buffer and sonicated to produce uni-lamellar vesicles. The protein sample is added to the vesicle solution then passed through 3 spin desalting columns (Thermo Scientific, Zeba) to remove most DM from solution. DM in complexes with lipid are removed with detergent absorbing bio beads (Bio-Rad). After DM removal, the vesicle-channel solution is aliquoted and flash frozen in an ethanol-dry ice bath, then stored at -80 C. support for a decane lipid membrane. The membrane separates two chambers of different K + concentrations. One chamber is grounded, through an Ag/AgCl elec- trode. The other contains two Ag/AgCl electrodes FIG. 7: Gating scheme and rate constant fit values for channel behavior. (see Fig. 1), one for the current limited voltage clamp (CLVC), the other connecting to: 1) an amplifier for the measurement of the axon voltage V (t); 2) an indepen- dent current clamp used to inject a stimulus current; 3) the input of the electronic synapse feeding into axon 2 (for the two-axons experiment). For the CLVC, we use the resistor values Rc = 100M Ω and 2GΩ for different situations. For the two-axon measurements, both wells are connected to independent CVLCs. All electronics are made in-house. The electronic synapse connecting two axons is a current clamp. The membrane voltage V1(t) of axon 1 is the command voltage for the synapse, effectively assigning axon 1 the role of the pre-synaptic neuron. The synapse can inject current into well 2 only when V1(t) > 0. This switch behavior is achieved with a solid state relay [3] that is switched to open/close with an op-amp. Positive/negative V1(t) controls the polarity of op-amp saturation, which in turn switches the relay to the open/close position. Design figures and detailed descriptions are to be provided in later publications. B. Electrophysiology Setup C. Modeling Overall setup, including headstage and voltage amplifiers are described at length in publications [3] and [8], and data acquisition is described in publication [3]. Setup and components will be described briefly. A ∼ 100µm aperture on a plastic cup is used as the The scheme for KvAP gating used to fit the channel be- havior is from [6]. We used as a guideline the rate con- stants in [6] for DPhPc lipid membranes, with modifica- tions to fit our data. All rate constants have the form k = k0e−zV . The gating scheme and k0, z values are provided in Supplemental Material Figure 1.
1007.3411	1	1007	2010-07-20T13:03:18	The phase diagram of random Boolean networks with nested canalizing functions	[ "physics.bio-ph", "cond-mat.dis-nn" ]	We obtain the phase diagram of random Boolean networks with nested canalizing functions. Using the annealed approximation, we obtain the evolution of the number $b_t$ of nodes with value one, and the network sensitivity $\lambda$, and we compare with numerical simulations of quenched networks. We find that, contrary to what was reported by Kauffman et al. [Proc. Natl. Acad. Sci. 2004 101 49 17102-7], these networks have a rich phase diagram, were both the "chaotic" and frozen phases are present, as well as an oscillatory regime of the value of $b_t$. We argue that the presence of only the frozen phase in the work of Kauffman et al. was due simply to the specific parametrization used, and is not an inherent feature of this class of functions. However, these networks are significantly more stable than the variants where all possible Boolean functions are allowed.	physics.bio-ph	physics	EPJ manuscript No. (will be inserted by the editor) 0 1 0 2 l u J 0 2 ] h p - o i b . s c i s y h p [ 1 v 1 1 4 3 . 7 0 0 1 : v i X r a The phase diagram of random Boolean networks with nested canalizing functions Tiago P. Peixoto Institut für Festkörperphysik, TU Darmstadt, Hochschulstrasse 6, 64289 Darmstadt, Germany e-mail: [email protected] September 13, 2018 Abstract. We obtain the phase diagram of random Boolean networks with nested canalizing functions. Using the annealed approximation, we obtain the evolution of the number bt of nodes with value one, and the network sensitivity λ, and we compare with numerical simulations of quenched networks. We find that, contrary to what was reported by Kauffman et al. [Proc. Natl. Acad. Sci. 2004 101 49 17102-7], these networks have a rich phase diagram, were both the "chaotic" and frozen phases are present, as well as an oscillatory regime of the value of bt. We argue that the presence of only the frozen phase in the work of Kauffman et al. was due simply to the specific parametrization used, and is not an inherent feature of this class of functions. However, these networks are significantly more stable than the variants where all possible Boolean functions are allowed. 1 Introduction Boolean networks (BN) were introduced by Kauffman [1,2] as a simple model of gene regulation. In this model the transcription states of the genes are described by Boolean variables, and their dependency to other genes by Boolean functions. In Kauffman's original model, which is usu- ally called a Random Boolean Network (RBN) [3], both the functions and their inputs are randomly distributed among all possible choices. Since this clearly discards any possible structure which may be selected by the evolu- tionary process, these networks are null models of gene regulation, from which general features may be derived, which are independent of the missing details [4]. Indeed this simple model already shows an emergent behaviour which may be applicable to real system, namely the ex- istence of two dynamical phases: Frozen and "chaotic". In the frozen phase, small perturbations have a limited prop- agation, and eventually stop. In the chaotic phase, small perturbations propagate exponentially fast, often reach- ing a finite portion of the system. It has been argued [1] that real systems may share features with RBNs which lie exactly at the interface between these two phases -- the so-called critical networks. In this point of the con- figuration space, small perturbations propagate only lin- early; thus the system retains some stability of the frozen phase, as well as some of the excitability of the chaotic phase, which may be necessary for the system to respond to external signals. Although this a very interesting fea- ture, a more plausible comparison with real gene regula- tion can only be made if more realistic properties are in- cluded, such as more realistic topologies [5,6,7,8,9,10,11] or update functions [12,13,14], for instance. In this paper we will consider RBNs with nested canalizing functions (NCFs), introduced in [15,16]. These functions are a nat- ural extension of the concept of canalization often present in biological systems [17,18]. A function with a canalizing input is such that if this input is at its canalizing value (either 1 or 0), then the output of the function is automat- ically defined, for any combination of the remaining input values. It has been observed that many real functions have a canalizing input [18]. If this concept is carried out to the remaining inputs, such that a hierarchy of canalization is present, the resulting function is a nested canalizing func- tion. Interestingly, the majority of the real functions stud- ied in [18] are also NCFs [15]. In this work we will obtain the phase diagram of RBNs with NCFs. Contrary to what was claimed in [16], such networks possess a rich phase diagram, where both the chaotic and frozen phases oc- cupy sizable portions of the configuration space. We also observe oscillations in the number of values of 1's in the network, for a portion of the parameter space. This paper is divided as follows. In Sec. 2 we define the model, as well as nested canalizing functions, and review some known facts. In Sec. 3 we obtain the evolution of fraction bt of 1's, and in Sec. 4 we obtain the network sensitivity λ and the phase diagrams. We then conclude in Sec. 5, and provide some final considerations. 2 The Model A BN is defined as a directed network of N nodes repre- senting Boolean variables σ ∈ {1, 0}N, which are subject to a dynamical update rule, (1) σi(t + 1) = fi (σ(t)) 2 Tiago P. Peixoto: The phase diagram of random Boolean networks with nested canalizing functions if σ0 = c0 if σ0 (cid:54)= c0 and σ1 = c1 if σ0 (cid:54)= c0 and σ1 (cid:54)= c1 and σ2 = c2 f ({σi}) =  s0 s1 s2 ... sk−1 where fi is the update function assigned to node i, which depends exclusively on the states of its inputs. In this work we consider that all nodes are updated in parallel. All nodes have the same number of inputs k which are chosen randomly. Starting from a random configuration, the dynamics of the system evolves and eventually settles on an attractor, after a transient time. We will characterize the properties of the system after this transient time by the fraction bt of the number of 1's in the network, and by the network sen- sitivity λ, which will differentiate between the dynamical phases. 2.1 Nested canalizing functions As introduced in [15], a nested canalizing function f ({σi}), with inputs {σi}, i ∈ [0..k − 1] , is defined as sd if σ0 (cid:54)= c0 and . . . and σk−1 = ck−1 otherwise, (2) where ci ∈ [0, 1] is the canalizing value of input i and, and si ∈ [0, 1] is the output value associated with input i. The value sd is the default output of the function, when no input is at its canalizing value. It is usually assumed that sd = 1 − sk−1, so that every input is sensitive. However, in this paper we will also consider the situation where sd is a free parameter. We will call an input with si = 1 an activator, and a deactivator otherwise; and if an input is at its canalizing value, we will say it is canalized. We will consider RBNs where the functions are chosen randomly from all possible Nested Canalizing functions, which are weighted according to the following parameters: a, the probability that an input is an activator (i.e. si = 1), c, the probability that the canalizing value of an input is 1, and d, the probability that the default output is 1. In the situation where sd = 1 − sk−1 this last parameter is omitted. In [18] it was shown that many eukaryotic genes seem to be regulated by canalizing functions, and in [15] it was further identified that all but 6 of the 139 genes studied in [18] are in fact regulated by NCFs. There is a simple interpretation for the existence of nested canalizing func- tions in real gene regulation networks: Genes are used to encode mRNA via a protein called RNA Polymerase (RNAP). This protein binds to a region of the DNA called the promoter region, which starts shortly before the gene itself. Genes which serve as inputs for other genes encode proteins which are called transcription factors (TF). These proteins bind to the upstream region of the gene, i.e. the region preceding (and including) the promoter region. The presence of a TF close to the promoter region may increase or decrease the probability that the RNAP will bind, and initiate the transcription. It is easy to imagine a situation where an hierarchy of TFs exists, where a TF which binds closer to the promoter region has more relevance, and in- creases or decreases the binding probability of RNAP by such a factor that it overrides the TFs which are bound further away, giving rise to a (nested) canalizing input. Let us appreciate how much of a deviation is the oc- currence of NCFs, in comparison to canalizing functions on only one input, as well as to all possible functions, by considering the total number of functions belonging to each class. Nested canalizing functions are identical to the unate cascade functions known in computer science [19], which have optimal properties regarding their computa- tion time via binary decision diagrams [20], and for which many properties are known. According to [21], the number of different NCFs with k inputs scales as Nnc ∼ αk!2k(1−log2 ln 2), for k (cid:29) 1, where α is some constant. For comparison, con- sider the number of functions which are canalizing on at least one input [22], Nc ∼ 4k22k−1. Although the frac- tion of both these classes relative to the total number 22k of functions with k inputs vanishes for large k, using the Stirling approximation for k!, we can easily see that the fraction of NCFs is smaller by a factor of Nnc/Nc ∼ 1/22k−1. Thus the presence of nested canalization is a much stronger deviation from a random distribution than single-input canalization. 3 The evolution of number of 1's, bt. In order to obtain the fraction bt of 1's in the network at time t, we will employ the annealed approximation [23]. This is a mean-field approximation, which assumes that the inputs of each function are randomly chosen at each time step. By construction, this forbids local correlations from arising, which makes the analysis easier. Since a quenched disorder should be indistinguishable from an an- nealed one, in the limit of large networks, this approxima- tion is expected to be exact for large RBNs. We begin by defining the probability γ(bt) that a ran- dom input in the network is at its canalizing value, γ(bt) = btc + (1 − bt)(1 − c). (3) Thus, the probability that the output of a randomly cho- sen function will be 1 at the next time step is given simply by the probability that at least one input is canalized and is an activator, or that the default output value is 1, bt+1 = (1 − (1 − γ(bt))k)a + (1 − γ(bt))kd = a + (d − a)(1 − γ(bt))k. (4) The situation where sd = 1 − sk−1 can be obtained sim- ply by setting d = 1 − a. The equilibrium value b∗ = b∞ can then be obtained by solving the above equation for bt+1 = bt = b∗. We note that limk→∞ b∗ = a, except for (a, c) ∈ {(0, 1), (1, 0)}, in which case b∗ = d. In Fig. 1 it is shown some of the solutions as a function of a, compared Tiago P. Peixoto: The phase diagram of random Boolean networks with nested canalizing functions 3 period-2 oscillation to a fixed point is through a pitch- fork bifurcation, where the fixed point becomes unstable (dbt+1/dbt(b∗) > 1) and gives rise to the oscillation (as shown in the bottom of Fig.1). 4 The network sensitivity, λ. The response of the system to small perturbations is char- acterized by the network sensitivity λ [24,25], which is defined as k times the probability that the output of ran- domly selected function will change if a randomly selected input is flipped. Thus, for large networks, the average number of nodes flipped after some small time t (cid:28) ln N should be λt. Therefore if λ < 1 the number of affected nodes will tend to zero after some time, and the dynamics is said to be in the frozen phase. If λ > 1, the number of affected nodes will increase exponentially, and the dy- namics is said to be in the "chaotic" phase. For the special value of λ = 1, the average number of affected nodes in- creases linearly with time, and the dynamics is said to be in the critical line between the two phases. We can obtain the value of λ for RBNs with NCFs with the annealed approximation, as we did for the values of b∗. We start by defining the probability η that two consecutive inputs have the same canalized output, η = a2 + (1 − a)2, (5) and the probability η0 that the last input has the same canalized output as the default output, η0 = ad + (1 − a)(1 − d). (6) Fig. 1. Fraction b∗ of the number of 1's in the network af- ter the transient time, as a function of the fraction a of ac- tivators. On top, for c = 1 and d = 1, we have b∗ > 0 only for a > a∗, where a∗ = 1/k. On the bottom it is shown the bifurcation diagrams, from an oscillatory regime to fixed point, for the parameters indicated in the legend. For both figures, the symbols represent averages over 10 indepen- dent numerical realizations of the dynamics, for networks with N = 105, and the solid lines are solutions of Eq. 4. with numerical simulations. There are two interesting be- haviors: 1. If c = 1 and d = 0 (or symmetrically if c = 0 and d = 1, not shown), there is a second-order transition of the value of b∗ at a = 1/k (or a = 1− 1/k), below which b∗ = 0 and above which b∗ > 0. For other values of c and d, this behaviour is replaced by a continuous variation of b∗ (not shown); 2. If d − c is small enough, the value of bt shows an oscillatory behaviour after a the transient time, between two values of b∗. These oscillations hap- pen when most of the default output values correspond to the canalizing values of a large portion of the inputs, and most canalized outputs are different from the canal- izing values. In such situation, the canalized inputs tend to deactivate themselves, which in turn will increase the default activations, and so on. The transition from this Using this, we can write the probability λi that the output will be flipped if input i is flipped, λi = (1 − γ(b∗ ))i(cid:2)(1 − (1 − γ(b∗ ))k−i−1)(1 − η) + (7) ))k−i−1(1 − η0)(cid:3) , (1 − γ(b∗ value λ =(cid:80) which accounts for the probability that all the inputs j < i are not canalized, and that the output of i is different than any input j > i which may be canalized. The sensitivity i λi amounts then simply to, 1 − (1 − γ(b∗))k +k(1−γ(b∗ γ(b∗) ))k−1(η−η0). (8) λ = (1−η) The situation where sd = 1 − sk−1 can be obtained by setting d = 1 − a, as before, and making the substitution η0 → η0(1 − δi,k−1) in Eq. 7, which yields λs = (1 − η) (1 − γ(b∗ ))k−1(cid:2)k(η − 1) + (1 − η0)(k − 1) + 1(cid:3) . (9) We note that λs ≥ λ for small values of k, but for k (cid:29) 1 we have λs ∼ λ, and this value approaches limk→∞ λ = (1−η)/γ(b∗), for γ(b∗) > 0. This means that in the limit of 1 − (1 − γ(b∗))k γ(b∗) + 0.00.20.40.60.81.0a0.00.20.40.60.81.0b∗k=2k=3k=4k=5k=100.000.050.100.150.20a0.00.20.40.60.81.0b∗c=1,d=0.3,k=10c=1,d=0.4,k=10c=1,d=0.5,k=10c=1,d=0.7,k=10c=1,d=1,k=104 Tiago P. Peixoto: The phase diagram of random Boolean networks with nested canalizing functions Fig. 2. Network sensitivity λ (Eq. 8) for several parameter values, as indicated in the labels and axes. The critical line λ = 1 is indicated by the solid lines. The regions in white correspond to parameter regions where the system displays oscillations, and thus the value of λ is not well defined. The points marked with stars ((cid:63)) were verified empirically in Fig. 4. Fig. 3. Network sensitivity λs (Eq. 9) for the case were sd = 1−sk−1, for several parameter values, as indicated in the labels and axes. The critical line λs = 1 is indicated by the solid lines. The regions in white correspond to parameter regions where the system displays oscillations, and thus the value of λs is not well defined. The points marked with stars ((cid:63)) were verified empirically in Fig. 4. large k, NCFs tend to be much more stable than randomly chosen functions, which have λ = k/2. In Figs. 2 and 3 are the phase diagrams for several parameter values. We note that these diagrams are sym- metric in respect to the appropriate parameter transfor- mations, such as (c, d) ↔ (1−c, 1−d). For d = 0 and c = 1, we observe that the critical line is composed of the a = 1/2 line plus the a = 1/k line, which corresponds to the criti- cal values of a where b∗ > 0, as discussed previously. We note that, since λ ∼= (1 − η)/γ(b∗) for k (cid:29) 1, there is a universal critical line at a = 1/2 for which η = 1/2 and γ(b∗) ∼= 1/2, and thus λ ∼= 1, independent of c and d. This is interesting, since it means that the most entropic situation a = c = d = 1/2 leads to a critical network, if k is not too small, and either changing c or d leaves the value of λ unmodified. Thus criticality can be attained for a large number of possible configurations, which is maybe a reason why this class of functions are favoured biologi- cally. However, there are large portion of the configuration space where λ > 1, and the dynamics finds itself in the chaotic phase. Additionally, there are significant regions where the fixed point of b∗ is unstable ans is replaced by a period-2 oscillation, as discussed previously. In this sit- uation (which are marked in white in Figs.2 and 3), the value of λ is not meaningful, since it supposes that b∗ is a stable fixed point, and the networks are neither in the frozen nor in the chaotic phase. The values of λ and λs in Eqs. 8 and 9 can be verified empirically, by constructing RBNs and obtaining the so- called Derrida plots [26], as shown in Fig. 4. These plots show the normalized hamming distance rt+1 at time t + 1 between two identical copies of the network, after only one of them had a random fraction of rt nodes flipped at time t. As can be seen on the top of Fig. 4, the different 0.00.20.40.60.81.0a100101102103kc=1,d=00.00.20.40.60.81.01.21.41.61.8λ0.00.20.40.60.81.0a100101102103kc=0.8,d=00.000.150.300.450.600.750.901.05λ0.00.20.40.60.81.0a100101102103kc=0.5,d=00.00.10.20.30.40.50.60.70.80.91.0λ0.00.20.40.60.81.0a100101102103kc=0,d=10.00.20.40.60.81.01.21.41.61.8λ0.00.20.40.60.81.0a100101102103kc=0.5,d=0.50.00.10.20.30.40.50.60.70.80.91.0λ0.00.20.40.60.81.0a100101102103kc=1,d=10.30.60.91.21.51.82.12.4λ0.00.20.40.60.81.0a100101102103kc=10.30.60.91.21.51.82.12.4λs0.00.20.40.60.81.0a100101102103kc=0.80.000.150.300.450.600.750.901.051.20λs0.00.20.40.60.81.0a100101102103kc=0.50.00.10.20.30.40.50.60.70.80.91.0λsTiago P. Peixoto: The phase diagram of random Boolean networks with nested canalizing functions 5 which have shown an excellent agreement. We have seen that the steady state value b∗ = limt→∞ bt displays two interesting features for some parameter combinations: 1. For a probability c = 1 that the canalizing value of a ran- dom input is one, and a probability d = 0 that the default output of a random input is one (and symmetrically for d = 1 and c = 0), there is a second-order transition from b∗ = 0 to b∗ > 0 at critical fraction a = 1/k of inputs which are activators; and 2. If d − c is small enough, bt will oscillate between two values of b∗, up to a value of a for which the fixed point of bt will become stable again. A similar oscillation is also observed in RBNs with threshold functions [27]. We have also observed that the phase diagram has large regions where λ > 1, and thus the system is in the chaotic phase. This contradicts the claim by Kauff- man et al. [16] that this class of functions always leads to λ < 1. The results in [16] were obtained for specific in-degree distributions, and a selection of the canalizing values according to specific probabilities. These probabili- ties were chosen so that they match the experimental data they analysed, but for which no other explanation was of- fered. This distribution was then extrapolated to obtain the entire phase diagram, which resulted only in values of λ < 1. In this work, the functions were chosen with differ- ent probabilities, according to the simple parameters a, c and d, which resulted in a phase diagram with large por- tions where λ > 1. Therefore, since there seems to be no reason to believe the specific parametrization in [16] has a general character, it should not be concluded that the exis- tence of NCFs always leads to λ < 1. On the other hand, the values of λ for RBNs with NCFs are much smaller than RBNs with all functions chosen with equal probabil- ity. Thus the conclusion in [16] that NCFs convey more stability to the system seem well justified, and it may in- deed be a reason why these functions are observed in real biological systems. This work has been supported by the DFG under Con- tract No. Dr300/5-1. References 1. S. Kauffman, Nature 224(5215), 177 (1969), ISSN 0028- 0836 2. S.A. Kauffman, Journal of Theoretical Biology 22(3), 437 (1969), ISSN 0022-5193 3. B. Drossel, in Reviews of Nonlinear Dynamics and Com- plexity, edited by H.G. Schuster (Wiley, 2008), Vol. 1, ISBN 3527407294 4. S. Bornholdt, Science 310, 449 (2005) 5. J.J. Fox, C.C. Hill, Chaos: An Interdisciplinary Journal of Nonlinear Science 11(4), 809 (2001), ISSN 10541500 6. M. Aldana, P. Cluzel, Proceedings of the National Academy of Sciences of the United States of America 100(15), 8710 (2003) 7. M. Aldana, Physica D: Nonlinear Phenomena 185(1), 45 (2003), ISSN 0167-2789 8. A.C. e Silva, J.K.L. da Silva, J.F.F. Mendes, Physical Re- view E 70(6), 066140 (2004) Fig. 4. Fraction of perturbed nodes rt+1 after one time step, as a function of an initial fraction of random per- turbed nodes rt, for parameter values marked as in Figs.2 and 3, and shown in the legend. The solid lines are slopes of the form rt+1 = λrt, with λ or λs calculated according to Eqs.8 or 9, respectively. networks (even those with the same value of λ) have a different perturbation propagation. However, if we only consider small values of rt, these curves are matched quite exactly by slopes of type rt+1 = λrt, as can be seen on the bottom of Fig. 4. 5 Conclusion We have obtained the phase diagram of random Boolean networks with nested canalizing functions. Using the an- nealed approximation, we have analytically calculated the fraction bt of nodes with value one, and the sensitivity λ of the network to small perturbations. We compared the results with numerical realizations of quenched networks, 0.00.20.40.60.81.0rt0.00.20.40.60.8rt+1k=10,a=0.5,c=1,d=0k=20,a=0.2,c=1,d=0k=3,a=0.333333,c=1,d=0k=4,a=0.25,c=1,d=0k=10,a=0.1,c=0.5,d=0.5k=20,a=0.11,c=1,d=1k=5,a=0.3,c=1,sd=1−sk−1k=4,a=0.2,c=0.8,sd=1−sk−10.0000.0020.0040.0060.0080.010rt0.0000.0050.0100.0150.0200.025rt+1k=10,a=0.5,c=1,d=0,λ'1k=20,a=0.2,c=1,d=0,λ'1.6k=3,a=0.333333,c=1,d=0,λ'1k=4,a=0.25,c=1,d=0,λ'1k=10,a=0.1,c=0.5,d=0.5,λ'0.37k=20,a=0.11,c=1,d=1,λ'2k=5,a=0.3,c=1,sd=1−s4,λs'1.3k=4,a=0.2,c=0.8,sd=1−s3,λs'1.26 Tiago P. Peixoto: The phase diagram of random Boolean networks with nested canalizing functions 9. R. Serra, M. Villani, L. Agostini, Physica A: Statistical Mechanics and its Applications 339(3-4), 665 (2004), ISSN 0378-4371 10. S. ichi Kinoshita, K. Iguchi, H.S. Yamada, AIP Conference Proceedings 982(1), 768 (2008) 11. B. Drossel, F. Greil, Physical Review E 80(2), 026102 (2009) 12. T. Rohlf, S. Bornholdt, Physica A: Statistical Mechanics and its Applications 310(1-2), 245 (2002), ISSN 0378-4371 13. A.A. Moreira, L.A.N. Amaral, Physical Review Letters 94(21), 218702 (2005) 14. A. Szejka, T. Mihaljev, B. Drossel, New Journal of Physics 10(6), 063009 (2008), ISSN 1367-2630 15. S. Kauffman, C. Peterson, B. Samuelsson, C. Troein, Pro- ceedings of the National Academy of Sciences of the United States of America 100(25), 14796 (2003) 16. S. Kauffman, C. Peterson, B. Samuelsson, C. Troein, Proc. Nat. Ac. Sci. 101, 17102 (2004) 17. C.H. Waddington, Nature 150(3811), 563 (1942), ISSN 0028-0836 18. S.E. Harris, B.K. Sawhill, A. Wuensche, S. Kauffman, Complexity 7(4), 23 (2002) 19. A.S. Jarrah, B. Raposa, R. Laubenbacher, Physica D: Non- linear Phenomena 233(2), 167 (2007), ISSN 0167-2789 20. J.T. Butler, T. Sasao, M. Matsuura, IEEE Transactions on Computers 54(9), 1041 (2005), ISSN 0018-9340 21. E. Bender, J. Butler, Computers, IEEE Transactions on C-27(12), 1180 (1978), ISSN 0018-9340 22. W. Just, I. Shmulevich, J. Konvalina, Physica D: Nonlinear Phenomena 197(3-4), 211 (2004), ISSN 0167-2789 23. B. Derrida, Y. Pomeau, Europhys. Lett 1(2), 45 -- 49 (1986) 24. B. Luque, R.V. Solé, Physica A: Statistical Mechanics and its Applications 284(1-4), 33 (2000), ISSN 0378-4371 25. I. Shmulevich, S.A. Kauffman, Physical Review Letters 26. B. Derrida, G. Weisbuch, Journal de Physique 47(8), 7 93(4), 048701 (2004) (1986) 27. Greil, Drossel, Eur. Phys. J. B 57, 109 (2007)
1204.4721	2	1204	2012-09-17T09:23:34	Optimal number of pigments in photosynthetic complexes	[ "physics.bio-ph", "quant-ph" ]	We study excitation energy transfer in a simple model of photosynthetic complex. The model, described by Lindblad equation, consists of pigments interacting via dipole-dipole interaction. Overlapping of pigments induces an on-site energy disorder, providing a mechanism for blocking the excitation transfer. Based on the average efficiency as well as robustness of random configurations of pigments, we calculate the optimal number of pigments that should be enclosed in a pigment-protein complex of a given size. The results suggest that a large fraction of pigment configurations are efficient as well as robust if the number of pigments is properly chosen. We compare optimal results of the model to the structure of pigment-protein complexes as found in nature, finding good agreement.	physics.bio-ph	physics	Optimal number of pigments in photosynthetic complexes Simon Jesenko and Marko Žnidarič Faculty of Mathematics and Physics, University of Ljubljana, Slovenia Abstract. We study excitation energy transfer in a simple model of photosynthetic complex. The model, described by Lindblad equation, consists of pigments interacting via dipole-dipole interaction. Overlapping of pigments induces an on-site energy disorder, providing a mechanism for blocking the excitation transfer. Based on the average efficiency as well as robustness of random configurations of pigments, we calculate the optimal number of pigments that should be enclosed in a pigment- protein complex of a given size. The results suggest that a large fraction of pigment configurations are efficient as well as robust if the number of pigments is properly chosen. We compare optimal results of the model to the structure of pigment-protein complexes as found in nature, finding good agreement. 1. Introduction Photosynthesis is the main natural process for harvesting Sun's energy on Earth, providing a food source for a great variety of organisms ranging from highly evolved photosynthetic systems in higher plants to simpler bacteria [1]. In spite of such diversity, basic underlying principles are shared among majority of light harvesting organisms. The initial stage of photosynthesis usually involves multiple pigment protein complexes (PPC) that consist of a number of pigment molecules (i.e., chromophores) held in place by a protein cage. PPCs are employed to absorb incoming photons as well as to transport the resulting excitation to the reaction center, where the excitation is used to initiate chemical reactions. The absorption of light and in particular channeling of the absorbed energy to the reaction center is known to achieve high efficiency [1]. One of the most studied PPCs is the Fenna-Matthews-Olson (FMO) complex that is found in the photosynthetic apparatus of green sulfur bacteria. It is the first PPC for which the atomic structure has been determined [2]. FMO is composed of a large protein envelope that encloses a tightly packed group of 7 pigment molecules‡ called bacteriochlorophylls (BChl). In FMO the main role of BChl pigments is actually not to absorb light but instead to transport electronic excitations from the input pigment, which is close to the antenna complex, towards a "sink" pigment channeling ‡ Recent structural analysis [3] suggests also the presence of an eighth BChl that is weakly bound to each monomeric unit as an additional input site. Its distance from the core 7 pigments is quite large and therefore we do not take it into account in our study. Optimal number of pigments in photosynthetic complexes 2 excitation to the reaction center. Recently discovered long-lasting quantum coherence in FMO complex [4] gave an additional boost to studies of excitation transport in PPCs. Previously it was namely believed that the transport in PPCs is predominantly of a classical nature. Many aspects of time dependence have been studied [5, 6, 7], including the functional role of quantum coherences [8, 9, 10, 7], as well as the possibility of transport enhancement via environmental interaction [11, 12, 13, 14]. Also important is the question of structural characteristics of PPCs that enable efficient excitation transfer, for instance, why has a given complex precisely the shape found in nature. Positions and orientations of pigments, known with high precision from crystallographic measurements, on casual inspection show no clear ordering or organization that would enable an easy classification of efficient configurations. It is also not clear, how special efficient configurations are - whether efficient configurations are a result of a long-lasting process of evolutionary improvement, or, can efficient configurations be readily achieved probing few random configurations of pigments. Prior to the availability of crystallographic structural data, the average minimal distance between pigment molecules was estimated based on the comparison between florescence yield of in vivo and in vitro chlorophyll solutions [15]. Recently, the efficiency of random configurations within simplified models of PPC was inspected [16, 17, 18, 19], suggesting that the efficient configurations are relatively probable. This complies with the results obtained for the Photosystem I from plants and cyanobacteria [20, 21], where random orientations of pigment molecules were probed, and high efficiencies of excitation energy transfer (EET) were obtained irrespectively of the pigment orientation. Also, the PPC configuration was shown to be robust to the removal of a pigment from the complex [20]. Fundamental, yet still unanswered question that we address in present work is, why has a particular photosynthetic complex exactly the specified number of pigments. In other words, what is the optimal number of pigments for a given size of the complex§, or, equivalently, what is the optimal size of the complex for a given number of pigments. For instance, why do we find exactly 7 pigments in the FMO complex and not more or less. Using a simple model whose parameters are taken from experiments, that is without any fitting parameters, we calculate the optimal number of photosynthetic pigment molecules for different complex sizes and compare these theoretical predictions with the actual number of pigments found in naturally occurring complexes. We find a very good agreement for a variety of PPCs in different organisms. To judge the optimality we use two criteria: (i) the average efficiency of the excitation transfer, where the averaging is performed over random configurations of pigment molecules, and (ii) the robustness of the efficiency to small variations of pigment's locations. A rationale behind these two choices is that "good" PPCs should have high efficiency but at the same time also be robust. A specially "tuned" configuration, that has a very high efficiency which though is very fragile, will obviously not work in a natural environment with its changing § Note that the size of the PPC might be fixed by external factors like for instance the membrane thickness. Optimal number of pigments in photosynthetic complexes 3 conditions. Additionally, from an evolutionary perspective, the efficiency should be stable with respect to different foldings of the protein cage. It is advantageous to have a PPC with such a number of pigments that will results in high average efficiency, i.e., in many close-to-optimal pigment configurations. Thereby, a small change in the environment, be it of a chemical origin or for instance a genetic mutation, will still result in a functional PPC. Robustness of quantum coherence to structural changes in the PPCs has been also found experimentally [22]. The model we use to describe the excitation transfer across the PPC consists of a Lindblad master equation describing a dipole-coupled pigments with an on-site excitonic energies being determined by the distances between disc-like pigments. If two discs come too close, i.e., if they overlap, this effectively rescales their energies, introducing a disorder. We should say that the optimal number of pigments that we predict is quite insensitive to details of the underlying model. Agreement between predictions of our model and naturally-occurring PPCs shows that Nature has optimized PPCs by using just the right number of pigments so that the resulting PPCs are highly efficient and robust at the same time. 2. The model To calculate the efficiency of the excitation transfer in the PPC, we need equations of motion describing dynamics of excitation on multiple chromophores, coupled to the environment. It is the ratio of chromophore-chromophore interaction strength to the chromophore-environment coupling that determines the applicability of various models. When interaction between chromophores is small compared to the environmental coupling the Förster theory [23] is applicable, leading to a picture of incoherent hopping of excitation between chromophores. In the opposite limit of strong inter-chromophore interaction and weak coupling to the environment, the excitation dynamics can be described by quantum master equation, either the Redfield equation or, employing a secular approximation, the Lindblad equation [24]. For nonperturbative parameter ranges, more advanced methods have been developed [25, 26, 27], usually at the expense of higher computational complexity. Our main optimality criterion, namely the transfer efficiency, is relatively insensitive to details of excitation time-dependence. Thus, we are going to use the simplest description of EET with the Lindblad equation, which retains coherent nature of transport, while still taking environmental interactions into account. We expect that more exact descriptions, which in general enhance excitonic oscillations, lead to similar results due to our averaging procedure. Also, these oscillations appear on the time scales of few 100 fs, which is much shorter than the time scale of excitonic transfer. In the following, we will introduce the Lindblad equation for the overlapping disc model, used for the description of excitation dynamics in PPCs. Optimality criteria for the efficiency of PPC configurations as used in latter sections will also be presented. Optimal number of pigments in photosynthetic complexes 4 2.1. Lindblad master equation Internal dynamics of the system of N chromophores within a single-excitation manifold is determined by the Hamiltonian of the form [28] H = nnihn + Vmn(mihn + nihm), (1) NX n=1 NX n6=m=1 dm· dn − 3(dm· rmn)(dn · rmn) ! where a state ni represents an excitation on the n-th chromophore site, i.e., the electronic state of the n-th chromophore being in the 1st excited state. Because EET is sufficiently fast, events with two excitations being present at the same time are rare and it is sufficient to consider only zero and single-excitation subspace [9]. The coupling Vmn is due to dipole-dipole interaction between chromophores of the form Vmn = 1 4πε0 , r5 mn r3 mn (2) where rmn = xm − xn is a vector, connecting the m-th and n-th chromophores, dn is a transition dipole moment between the ground and the 1st excited state of the n-th chromophore. Because the system of chromophores is coupled to the protein and nuclear degrees of freedom it is described by a reduced density matrix ρ. Decoherence due to environmental interaction, recombination of excitation to the ground state, and transfer of excitation to the sink, are modeled by Lindblad superoperators that augment the von Neumann equation for the time evolution of density matrix, ρ = −i[H, ρ] + Ldeph(ρ) + Lsink(ρ) + Lrecomb(ρ). (3) To model the effects of the environment, we have taken a simplified picture of purely dephasing Lindblad superoperators (i.e. Haken-Strobl model), which is believed to capture the basic environmental effects and was used in various previous studies [29, 11, 13, 12]. For longer time, relevant for the efficiency of PPC, it has been shown that the description with the Lindblad equation accounts for the main features of the dynamics [30, 31]. Dephasing Lindblad superoperator destroys a phase coherence of any coherent superposition of excitations at different chromophores, and is given by Ldeph(ρ) = 2γ 2{nihn, ρ} (4) where a site-independent dephasing rate is given by γ and { , } represents the anticommutator. Irreversible transfer of excitation from the N-th chromophore to the sink si is modeled by Lindblad superoperator sihN ρNihs − 1 (5) where κ denotes the sink rate. The irreversible loss of excitation due to recombination is given by an analogous term Lrecomb(ρ) = 2Γ (cid:19) 2{NihN, ρ} 0ihn ρnih0 − 1 Lsink(ρ) = 2κ NX (cid:18) (cid:18) (cid:19) (6) , 2{nihn, ρ} , n=1 (cid:18) nihn ρnihn − 1 NX n=1 (cid:19) , Optimal number of pigments in photosynthetic complexes 5 with a site-independent recombination rate given by Γ. The ground state of a chromophore system (state without any excitation) is represented as 0i. Note that Lsink and Lrecomb can be equivalently represented by an antihermitian Hamiltonian at the expense of non-conserved density matrix probability [12], avoiding the need for an additional sink and the ground state. Relevant environmental parameters going into Lindblad equation (3) are dephasing strength γ, recombination rate Γ and the sink rate κ. We use standard values inferred from experiments and used before [12], sink rate κ = 1 ps−1, recombination rate Γ = 1 ns−1 and dephasing rate at room temperature γ = 300 cm−1k. Dephasing rate, being a product of temperature and the derivative of the spectral density, can be estimated by using experimentally determined parameters of the spectral density (as done e.g. in reference [12]). This value approximately agrees with the optimal dephasing rate at which transfer is most efficient [12, 11]. We note that the results shown depend very weakly on the actual value of the dephasing rate as long as it is of the same order of magnitude as γ = 300 cm−1. In Appendix A we show that the values of γ = 150 cm−1 and 600 cm−1 give almost the same optimal PPC size. 2.2. Overlapping discs model Because we want to study the dependence of efficency on the size of PPC, keeping the number of pigments fixed, we have to account for the size-dependence of the Hamiltonian. On-site energies and interaction strengths in equation (1) will be determined from the geometry of pigments configuration. Changing the PPC's size two gross effects are at play. First, as the distance between pigments is reduced, the dipole- dipole interaction between pigments gets larger, enhancing the transfer of excitation among them; this effect is already taken into account by the ∼ 1/r3 dependence of Vnm in equation (2). Secondly, as chromophores get even closer together, approaching the distances comparable to the extent of the chromophore electronic orbitals, the Hamiltonian (1) is not sufficient for the description of EET anymore because the effects due to electronic orbital overlap become important. A detailed analysis of processes that take place as chromophores get close together would require advanced quantum chemistry methods and is out of scope of this paper. However, the main effect can be effectively taken into account by appropriately rescaling parameters of equation (1). Because the pigment molecules will be deformed, their excitation energies will also change. As the on-site energies n, being of the order of few eV, are about ∼ 100 times larger than Vnm, even a small relative change in n can have a large effect. Effectively, close or even overlapping chromophores will therefore result in widely different values of on-site energies n at different sites, i.e., in a disorder. Thus there are two competing factors that determine the optimal size of PPCs: reducing inter- k The values entering the Lindblad equation (3) should be in units of frequency, e.g. s−1. The conversion from inverse centimeters cm−1 as used traditionally in spectroscopy is given by ω[s−1] = 200πcν, where c = 3 × 108 and ν is value in [cm−1]. Optimal number of pigments in photosynthetic complexes 6 Figure 1. Illustration of the simple model of PPC. (a) BChl molecule with omitted phytyl tail, enclosed by the cylinder as used in simulations (a = 1 Å, r = 4 Å). (b) Discs, randomly positioned in sphere, prior to the reduction of radii of overlapping discs. Green discs at the top / bottom represent the input site / output site. (c) Same sample as in (b) after the shrinking of overlapping discs. chromophore distance increases EET, while the overlapping of chromophores introduces a disorder that effectively suppresses EET. On-site energy has an additional contribution due to local environment (e.g. because of pigment-protein interactions), which is however of the order of ∼ 102 cm−1 and is usually much smaller that the on-site disorder due to pigment overlap, which is proportional to the unperturbed on-site energy of ∼ 104 cm−1. Therefore, the effects of pigment-protein interaction were neglected when obtaining the results presented in the main text. To verify whether neglecting of on-site disorder due to protein interactions is justifiable, we have also calculated the optimal size for N = 7 pigments with random on- site disorder added to each random sample of chromophores. The results (see Appendix B) show that disorder of such magnitudes indeed has no gross effect on the results. Each chromophore in our model is represented as a disc -- a thin cylinder -- of radius r and height a. Each disc is supposed to represent an approximate size of the electronic cloud of the orbitals involved in the EET (highest occupied electronic orbital, lowest unoccupied electronic orbital). We have estimated the height to be a = 1 Å, while the cylinder radius is taken as r = 4 Å. Size of this cylinder in comparison to a BChl pigment molecule can be seen in figure 1a. Radius r = 4 Å is chosen so that it contains 16 closest non-hydrogen atoms to the Mg atom located in the center of the pyridine ring. For given locations and orientations of discs, we then determine if there are any overlaps between discs. If two discs overlap, for an example see figure 1b, we rescale the radius of one of them to the new radius rn so that they do not overlap anymore but (a)(b)(c)Optimal number of pigments in photosynthetic complexes 7 instead only touch. After eliminating all overlaps we end up with disc's radii rn, for an example see figure 1c. Provided the radius of the n-th disc rn is different from the non-overlapping size r, we have to appropriately rescale the on-site energy n. If the effective size of the electronic cloud is reduced from r to rn, the kinetic energy of electron increases by a factor r2/r2 n. We therefore estimate that the energy of excitation on a resized disc will also scale quadratically with its size, giving the on-site energy dependence (7) (cid:19)2 − 1 n (0) = (cid:18) r rn where (0) is the excitation energy of non-deformed pigments, i.e., the energy difference In FMO (0) is approximately (0) ≈ of two lowest electronic states on a pigment. 12 300 cm−1. The overall offset of on-site energies is irrelevant for the dynamics in the model, therefore we shift all energies by (0). Such quadratic on-site energy scaling can be rigorously shown under an assumption that the electronic eigenfunctions of the rescaled pigment are just the rescaled eigenfunctions of the original pigment of radius r. Let orbitals ψi be the eigenfunctions of the Hamiltonian H(x) = T(x) + U(x), where T(x) is kinetic energy operator and U(x) is a confining potential. The on-site energy of a given chromophore n is the difference between the energy of ground and 1st excited state, n = E2 − E1. Assuming that eigenstates ψi are just scaled to a smaller volume, ψ∗ i (x) = λ3/2ψi(x/λ), the scaling of on-site energies from equation (7) is obtained by comparison of eigenvalue equations for the original eigenstate Hψi = Eiψi and the i , where the scaled confining potential in H∗ has to be scaled eigenstate H∗ψ∗ U∗(x) = U(x/λ)/λ2. i = E∗ i ψ∗ Dipole strength of the chromophores is similarly scaled linearly with the radius rn of the cylinder, dn = rn r d, (8) where d is the bare transition dipole moment of the original chromophore of size r, and dn is the scaled dipole strength of the resized disc. This can be justified on the same grounds as the scaling of on-site energies, by inserting the rescaled wavefunction ψ∗ i (x) into the expression for transition dipole matrix of relevant chromophore transition, d = hψgexψei. There are different possibilities of how to precisely resize discs in order to avoid overlaps. While different procedures lead to different on-site energies, the determined optimal complex size changes by little. Results in the main text were obtained by sequentially inspecting each pair of discs, resizing only the disc having greater radius afterwards, while keeping the other disc intact. We have verified other resizing procedures, for instance, resizing both discs in pair to the same size. Such resizing effectively reduces disorder of on-site energies as even strongly overlapping pigments will have identical on-site energies. Nevertheless, for such resizing procedure, the determined optimal radius of PPCs are within 2 Å of the values obtained by the resizing procedure used throughout the paper, and are thus within the error bounds of the model. Optimal number of pigments in photosynthetic complexes 8 To summarize, in our overlapping disc model the matrix elements of H are calculated for given PPC configuration (positions, as well as disc and dipole orientations) by first resizing all overlapping discs, obtaining rescaled radii rn and then scaling dipole strengths and on-site energies according to equations (7) and (8). 2.3. Optimality criteria We have already introduced equations of motion that govern the dynamics of excitation on chromophores, as well the overlapping disc model that allows us to determine the Hamiltonian for a given configuration of chromophores. What is left are criteria that will enable us to determine whether a given configuration of chromophores is efficient in terms of EET. The efficiency of the PPC complex is characterized by the probability that the excitation, initially localized on the input site, will be funneled to the reaction center trough the output site. For an example of time evolution see Appendix C. The efficiency in the model is not unity because the excitation can be lost. The probability that the excitation will be transported to the reaction center can be expressed as η = 2κ dt ρN N(t), (9) which will be used as our main efficiency criterion. Closely related is the average transfer time, which signifies the speed of transfer of excitation to the reaction center, and is expressed as 0 Z ∞ Z ∞ 0 τ = 2κ η dt t ρN N(t), (10) with smaller transfer times being better. As an additional viability criterion of PPC, robustness of efficiency to static disorder will be also inspected. Dynamic disorder due to thermal motion is already effectively described by the dephasing Lindblad terms in equation (3). Static disorder due to structural changes of PPC, for instance due to changes in biological environment, like temperature, electric charges, etc., should be treated separately. A given configuration of pigments in PPC is robust to the static disorder if random displacements of pigments from the original locations do not induce large changes in PPC's efficiency η (or equivalently, the average transfer time τ). To put it on a more quantitative ground, we define the pigment configuration robustness ση(x) for a given configuration of pigments¶ with positions x = (x1, x2, ..., xN), as a standard deviation of efficiency η when pigment coordinates are varied in the neighborhood of original positions, σ2 η(x) = (η2(x + y)− ¯η(x)2)w(y) dy, η(x + y)w(y) dy, (11) Probability density w(y) defines the neighborhood of a given configuration, and is localized around the original location of the pigments. The most straight-forward ¶ We omitted the disc and dipole orientations from the definition of pigment configuration robustness to simplify the expressions. However, no qualitative differences are to be expected if orientations are also varied when inspecting the robustness. ¯η(x) = Z Z Optimal number of pigments in photosynthetic complexes 9 Figure 2. Probability density pR,N(η) for N = 7 for a range of sphere radii R (density plot in the background of the left plot; contours connecting equal values of pR,N are also shown). Solid blue curve is the average efficiency hη(R)i. On the right a close-up of hηi and the average transfer time hτi (dotted curve, right axis) around the maximum of hηi is plotted. (cid:18) y· y (cid:19) choice for the distribution w(y) is a product of uncorrelated normal distributions at each pigment location, w(y) = (2πσ2)−N/2 exp (12) where σ defines the size of neighborhood in which the robustness is being probed. With given probability distribution, the robustness ση is a function of original pigment locations x and size of deviations from original locations σ. In the limit of small deviations, σ → 0, the expression can be simplified to 2σ2 , 3NX i=1 (cid:12)(cid:12)(cid:12)x0 σ2 η(x) = σ2 ηi(xi)2, ηi(xi) = ∂η ∂x0 i i=xi , (13) where the sum goes over all components of pigment coordinates. The robustness ση in the limit of small pigment displacements is thus proportional to the amplitude of pigment displacements. 510152025303540R[A]0.00.20.40.60.81.0η0.010.100.101.001.001.0010.0010121416182022R[A]0.890.900.910.920.930.940.953040506070hτi[ps]Optimal number of pigments in photosynthetic complexes 10 Figure 3. (a) The optimal enclosing radius Ropt(N) of the sphere for different number of chromophores N, based on the maximum of the average efficiency hη(R)i. The width of the shading (i.e. ≈ ±1Å) denotes a range of R for which the average efficiency is within 1% of maximal value. Vertical dashed line marks the case of FMO with N = 7 chromophores. (b) Structure of FMO complex as determined from spectroscopic data, enclosed into the optimal sphere of the radius R ≈ 16 Å. The structural data was obtained from PDB entry 3EOJ [3] 3. Optimal number of pigments: the case of Fenna-Matthews-Olson complex In this section, the efficiency and robustness of random configurations within the model will be considered on an example of FMO complex. FMO consists of N = 7 BChl pigments. On-site energy for BChl was chosen to be (0) = 12 300 cm−1, which is within the range of on-site energies for BChls in FMO as determined in the literature [32, 33, 34]. The strength of transition dipole moment d = d was taken as d2 = 26 D2 + (note that published values for d from calculations and experimental data vary considerably [35]). On-site energy (0) and dipole strength d used hold for BChl pigments in general and therefore the results presented are expected to be valid also for other PPCs containing BChls, not just for the FMO complex. To determine the optimal size of PPC for a given number of chromophores (or equivalently, optimal number of chromophores for a given size), we considered two criteria based on overall behavior of efficiencies and robustness of random configurations. In the following, the motivation for choosing such optimality criteria will be given, and the results for the case of FMO will be presented. + Dipole moment is given in units of debye (D). Conversion of interaction Vmn to units of cm−1 is obtained by conversion D2 4πε0 = 5030 cm−1Å3. 34567891011121314N810121416182022Ropt[A](a)(b)Optimal number of pigments in photosynthetic complexes 11 3.1. Average efficiency We shall use the average efficiency hηi, averaged over random positions and orientations of pigments enclosed in a predefined volume. The reason to use random averaging with uniform distribution is twofold: first, high average EET efficiency under uniform averaging will mean that there are many different configurations that have high efficiency, i.e., high efficiency is globally robust. Choosing averaging over random configurations therefore offers insight in how special efficient configurations of chromophores are within the space of all configurations. Second reason is that we have a priori no knowledge what would be the appropriate measure for possible pigment configurations under say different protein cage foldings due to for instance mutations. A uniform measure represents in this case a "least-information" distribution. ∗ Using configurations sampled according to a uniform distribution over chromophore positions within a ball of radius R and random orientations of dipoles and discs, the average efficiency is calculated. Formally, it can be written as Z hη(R)i = R η(X)wconf(X) dX, (14) where X contains positions and orientations of chromophore discs and dipoles (apart from the positions of input and output sites which are fixed on the poles of the sphere), and wconf(X) ∝ 1 signifies a uniform distribution of chromophores inside the sphere. Observing the dependence of the average efficiency hη(R)i for different number of chromophores and different radii R of the enclosing sphere, we can determine the optimal number of chromophores for a given radius R, or equivalently, the optimal radius Ropt for a given number of chromophores. pR,N(η) = R To obtain a more detailed information about efficiencies of random configuration, we also observed the probability distribution over efficiencies pR,N(η), defined as R δ(η(X) − η)wconf(X)dX. For the number of chromophores as found in FMO (N = 7), the probability distribution over efficiencies pR,7(η) is shown in figure 2. When going from large radii R to smaller, configurations tend to get more efficient, which is expected as the chromophores are closer to each other, thus increasing the dipole coupling. However, as R is reduced even further, overlapping of chromophores gets more probable, causing an on-site energy disorder. This leads to the localization of excitation on chromophores not connected to the sink site. Such configurations have low efficiency. Therefore, as R gets smaller the distribution pR,N becomes bimodal, with lower efficiency mode due to overlapping configurations and high efficiency mode for non-overlapping configurations. Low efficiency of overlapping configurations therefore leads to a maximum in the average transfer efficiency hη(R)i at the optimal radius Ropt. The average transfer efficiencies at the optimal radius are rather high, e.g. for N = 7 in figure 2 it is ∗ We have to note that we also checked other distributions, for instance a uniform distribution on the surface of a sphere of radius R, and obtained practically the same results. For instance, the difference in the position of the maximum in figure 2 was within our error estimate of 1 Å (seen as an "error" band in figure 3). Optimal number of pigments in photosynthetic complexes 12 hη(Ropt)i ≈ 0.95 with a large fraction of configurations having even larger efficiency than the average. Thus within the model, high efficiency is not due to finely tuned pigment positions and orientations, but occurs for majority of pigment configurations for parameters estimated to be relevant in PPCs. For the FMO case with N = 7, the optimal radius was estimated to Ropt ≈ 16 Å, which fits the actual configuration of pigments very well (see figure 3). The average transfer time hτi is also minimal at R = Ropt (see figure 2). Optimal average transfer time of ∼ 30 ps is so large due to the contribution of very inefficient configurations of chromophores. Looking at the average transfer time of the 5% of most efficient configurations, we get the value of 5 ps, which is comparable to the transfer times as determined using different models of the FMO in references [12, 13, 14, 36]. The estimated optimal radius is quite insensitive to small variations of input parameters, e.g. dipole moment d, chromophore disc radius r or its thickness a, or the scaling of on-site energies and dipole strengths of resized discs. For instance, decreasing disc radius to r = 3.5 Å decreases Ropt by ≈ 2 Å, changing disc thickness to a = 0.5 Å or 1.5 Å changes Ropt by ≈ ∓2 Å, while changing quadratic energy scaling to a linear or cubic one again changes Ropt by ≈ ∓2 Å. Similarly, changing the dephasing rate γ by a factor of 2 changes Ropt by ≈ 2 Å, see Appendix A. Details of the disc resizing procedure also change Ropt for less than 2 Å, as the extreme case of resizing each overlapping disc pair to the same size reduces Ropt by ≈ −2 Å. The optimal radius of the enclosing sphere was obtained from the average efficiency over all random configurations within a sphere. However, even though the evolutionary drive to more efficient configurations might not be very strong if majority of configurations are already efficient, still some optimization is to be expected. Thus one might argue that the optimal enclosing volume of the natural PPCs should be determined considering only the ensemble of more optimal configurations. We will denote such averages with hηip where p specifies a portion of most efficient configurations that should be taken into account when calculating the average (e.g. hηi0.05 is the average of η over 5% of most efficient configurations as shown in figure 4). As p is reduced, the overall value of average efficiency hηip will increase. The increase will be more pronounced in the region of R < Ropt, where the distribution is bimodal. The location of the maximum of hηip will be thus moved to smaller values of R, indicating more densely packed chromophore configurations. However, as we will see in next subsection, robustness of such densely packed configurations deteriorates very quickly, supporting our choice of estimator for the Ropt. Note that the overlaps between pigments and protein cage are not considered If overlaps with protein cage would be taken into account, explicitly in the model. Ropt would represent the size of a protein cage, whereas in our model without pigment- protein overlaps, Ropt is the size of a sphere that contains all pigment centers. For instance, looking at figure 3b, we can see that the sphere with Ropt contains all pigment centers, while parts of few pigments still protrude the bounding sphere. If overlaps of pigments with the protein cage would be taken into account explicitly, Ropt would be Optimal number of pigments in photosynthetic complexes 13 Figure 4. The average robustness of top 5% of most efficient configurations hσηi0.05 (solid blue curve) as a function of the enclosing sphere radius R. Dash-dotted green curve represents the average robustness of the top 15 % of most efficient configurations hσηi0.15. Dotted red curve is the average efficiency over the top 5% of configurations hηi0.05 (right axis). approximately by a disc radius r = 4 Å larger, i.e. in corresponding figure, the bounding sphere would enclose all pigments completely. 3.2. Robustness Robustness of PPC configurations to static disorder should also be taken into account when determining whether a given configuration of pigments is feasible, as the conditions in which PPCs operate are subject to constant environmental changes. In previous subsection, we have inspected the probability distribution of efficiencies η over random configurations, showing that majority of random configurations achieve relatively high efficiency when the enclosing volume is optimal. In this subsection, we will present analogous analysis of the robustness of random configurations, in particular of those with high η. We shall show that highly efficient configurations in small enclosing R are very fragile. We have defined robustness of efficiency ση in equation (11). In simulations we have displaced the pigment positions according to normal distribution with a width of σ = 0.1 Å, which is small enough to quantify the robustness in the neighborhood of specific configuration, while larger than displacements due to thermal vibrations, which are already effectively described by the Lindblad equation. We are specifically focusing 46810121416182022R[A]10−310−210−1100hσηip0.9800.9850.9900.995hηipOptimal number of pigments in photosynthetic complexes 14 on a subset of the most optimal configurations in terms of η. The average robustness of the subset of optimal configuration is denoted as hσηip, where p specifies a fraction of most optimal configurations in terms of EET efficiency η. As an example, we will consider robustness of top 5% of efficient configurations hσηi0.05. The dependence of the average robustness on the radius of the enclosing sphere R is shown in figure 4. The average efficiency of optimal configurations hηi0.05 is also shown in the figure. While the average efficiency of top 5% of optimal configurations continues to rise as the enclosing sphere radius R is reduced, we can see that the average robustness hσηi worsens very quickly as the R drops below Ropt. Quick worsening of EET robustness with reducing sphere radius suggests that even if the PPC configurations occurring in nature are indeed optimized in terms of pigment positions and orientations, the excessive stacking of pigments is not favored as it makes PPC configurations very sensitive to any displacements of pigments. The transition from robust to non-robust regime takes place at a radius comparable to Ropt at which the average efficiency hηi has a maximum. This is not surprising as both, the efficiency of configurations and robustness of configurations, are strongly influenced by the overlapping of pigments which gets more pronounced for R (cid:46) Ropt. We have presented results for the robustness ση of the 5% of most efficient configurations, with pigment displacements σ = 0.1 Å. General characteristics of hσηip however do not quantitatively change for different p (the case of p = 0.15 is also shown in figure 4) or displacements σ. Most importantly, the radius R at which the robustness of configurations drops significantly takes place at approximately the value of Ropt. Same behavior of robustness is observed also in the limit of infinitesimal robustness from equation (13) where σ → 0. 4. Optimal pigment numbers in other PPCs In previous section we have calculated the optimal size R or the optimal number of pigments for the FMO complex. We also demonstrated that a large portion of chromophore configurations has high efficiency when the enclosing volume is properly chosen (∼ Ropt). Additionally, robustness of configurations to chromophore displacements starts to deteriorate quickly once the enclosing volume is reduced below Ropt. Based on these two observations, we argue that the enclosing volume of PPCs occurring in nature should be close to the optimal volume as determined by our simple model. In this section we will present similar results for the PPCs containing chlorophyll (Chl) chromophores. We compare results of the model to the structure of PPCs from the Photosystem II (PSII) [37], found in cyanobacteria, algae and plants. PSII consists of multiple functional units, which are either part of the outer light-harvesting antenna or the inner core, to which excitations are funneled. In the light-harvesting antenna we will consider the light harvesting complex II (LHCII), while in the core we will focus on the PC43 and PC47 complexes that funnel excitations to the reaction center and thus have a Optimal number of pigments in photosynthetic complexes 15 similar role as the FMO complex in bacteria. A monomeric unit of LHCII contains 14 chlorophyll molecules (8 Chl-a and 6 Chl-b), while CP43 and CP47 contain 13 and 16 Chls respectively. Model parameters for the sink rate, dephasing and recombination are kept the same as in the FMO case, while the transition dipole strength and on-site energies are different for Chl molecules. Transition dipole moment of Chl molecules is chosen as d2 = 15 D2 and the on-site energy (0) = 15 300 cm−1, where values were taken according to reference [38] (we take the average between values for Chl-a and Chl- b). For the CP43 and CP47 complexes we have simulated random configurations of 13 and 16 chromophores enclosed into a sphere as the actual chromophore positions are distributed relatively uniformly in all directions. The shape of LHCII is however significantly elongated in one direction. We therefore choose the cylindrical volume, having only one additional parameter that has to be provided, i.e. the ratio between the cylinder radius Rc and cylinder height A. Based on positions of the LHCII chromophores we have estimated the ratio of the two to be Rc/A = 0.34. The CP43 and CP47 primarily play a role of an exciton wire, making the model with input site and output site at the opposite sides of the sphere applicable. The optimal radius as predicted by the model is R ≈ 18 Å for CP43 and R ≈ 20 Å for CP47. As the LHCII also has to transport excitations from adjacent complexes, we have also determined the optimal shape of LHCII with input and output sites located at the opposite sides of the enclosing cylinder. With the ratio Rc/A fixed, we have varied the height A of the cylinder and determined the optimal height to be Aopt ≈ 43 Å. In addition to acting as an excitation wire, CP43 and CP47 complexes are also directly involved in the absorption of photons, in which case the role of the input site can be taken by any chromophore site. This is even more common scenario in the LHCII complex, whose primary role is the absorption of photons. To verify whether the findings about optimal enclosing volume are also valid when the main purpose of PPC is the absorption of photons, we randomly placed the input site inside the enclosing geometry for each configuration in random ensemble. General characteristics of the distribution over efficiencies pR(η) do not change considerably, however, the distribution is somewhat shifted to the higher efficiencies because in many random configurations the input site is considerably closer to the output site than the diameter of the enclosing volume. This results in the optimal size of enclosing volume being somewhat larger, Ropt ≈ 20Å for CP43 and Ropt ≈ 22Å for CP47. For the LHCII we have moved the output site to the midpoint on the side between top and bottom of the cylinder, where the actual output site is supposedly located [41]. For such geometry and previously used Rc/A = 0.34, we have obtained the optimal height of the enclosing cylinder at A ≈ 47Å. The optimal enclosing values obtained from the model (averaged between the case for fixed input site and random input site) were compared to the actual configurations of pigments as obtained from spectroscopic data, and are shown in figure 4a-c, showing good agreement. For the spherical geometries, we have centered the sphere of optimal radius Ropt to the arithmetic mean of locations of BChl/Chl centers. For the LHCII, Optimal number of pigments in photosynthetic complexes 16 Figure 5. (a) - (c) Various pigment-protein complexes (PPCs) enclosed in optimal geometries as described in the main text. The structural data was obtained from PDB entries 1RWT (LHCII) [39] and 3ARC (CP43, CP47) [40]. In (c) enclosing geometries of two additional monomeric units of the LHCII trimer are also shown. (d) Plot showing dependence of the average efficiency hηi on the size of the enclosing geometry. The upper panel shows the case of CP43 (green) and CP47 (blue), while the lower panel shows the case of LHCII. Solid curves are for the fixed input site, while dotted line for the randomly placed input site. Vertical dashed lines mark the sizes of enclosing volumes as used in subfigures (a) - (c). such that PN where cylinder was used as the enclosing geometry, the cylinder axis was determined i r2⊥i was minimal, where r⊥i is the distance from the cylinder axis to the position of i-th Chl center. Interestingly, three cylinder axes do not lie in a plane but are instead tilted at an angle 15◦ to the plane containing three cylinder centers. It is (a)CP43,Ropt=19A(b)CP47,Ropt=21A(c)LHCII,Aopt=45A101214161820222426280.700.750.800.850.90hηiR[A]303540455055600.760.780.800.820.840.860.880.90hηiA[A](d)Optimal number of pigments in photosynthetic complexes 17 not known if this plays any functional role. 5. Conclusion We have studied the efficiency of excitation energy transfer in protein-pigment complexes for random configurations of pigments. The Hamiltonian part of Lindblad master equation is determined from the geometry of pigment configurations. If pigments are too close, so that they overlap, this introduces a disorder in on-site energies, effectively inhibiting excitation transport. Fixing the enclosing volume in which pigment molecules are located we have calculated the average efficiency over random pigment configurations as well as robustness of efficiency to variations of pigment locations. Doing this we have determined the optimal number of pigments for a given size of the complex. Even though the model is an oversimplification of actual processes that take place in nature, statistical predictions obtained from the model are robust to its variations. Comparing theoretically predicted optimal number of pigments with several naturally-occurring complexes we find good agreement. This might indicate that PPCs are not optimized just to have the highest possible efficiency -- in fact, efficient configurations are quite common -- but instead to be robust to variations in pigment locations. Namely, it turns out that configurations optimized for the highest efficiency, that is those with specially tuned positions and dipole orientations, are very sensitive to small perturbations. The number of pigments in nature is therefore chosen in such a way that the probability of having efficient configurations that are at the same time also robust is the highest. The presented findings could be in principle verified experimentally by modifying the structure of known PPCs and probing the efficiency of excitation transfer. For the FMO complex the structure was already changed by mutation of genes encoding the structure of BChls, as well as by substituting the carbon 12C atoms with 13C [22]. Comparison of excitonic spectra revealed no distinctive differences in the dynamics of excitations, complying with the hypothesis that configurations are not highly tuned but instead very robust. An additional intriguing possibility would be also to inspect the characteristics of FMO with mutated protein cage, modifying positions and orientations of pigment molecules. One could also compare our predictions for the optimal sizes (e.g. figure 3a) with other complexes occurring in nature. Appendix A. Dependence on the dephasing rate γ In the simplified model used, environmental interaction is described by dephasing rate γ, having the same value for all chromophore sites. In principle, environmental interaction requires more involved equations of motion (e.g. HEOM [7]), taking into account the spectral density of the environmental modes. However, due to crude nature of the model, simplified description is expected to account for main environmental effects influencing [30, 31] for more detailed comparison the efficiency of exitation transfer (see e.g. Optimal number of pigments in photosynthetic complexes 18 Figure A1. 150 cm−1, 300 cm−1, 600 cm−1. random on-site energies disorder σrand. (a) Average efficiencies hηi for different values of dephasing γ = (b) Average efficiencies hηi for different values of of approaches). The adequacy of simple Lindblad-type description of dynamics for determination of optimal size is also justified due to the high robustness of the results to the actual choice of dephasing value γ, as seen in figure A1a, where Ropt only changes for ∼ ±2Å as dephasing rate γ is changed by a factor of 2. Optimal size Ropt of the PPC is somewhat smaller as dephasing rate gets stronger, which is expected as larger dephasing rate enables transfer across the sites with greater on-site energy mismatch, getting increasingly common in more compact configurations of chromophores. Appendix B. Random on-site disorder To verify whether effects of the local chromophore environment due to e.g. pigment- protein interaction can affect the findings about the optimal PPCs sizes, we have of the amended the Hamiltonian in equation (1) with random on-site disorder rand magnitudes as found in naturally occurring PPCs (i.e. on-site energy differences in the order of ∼ 100 cm−1). The values of disorder for each realization of random PPC were calculated according to Gaussian distribution with variance σrand. Results are shown in figure A1b. In the region R < Ropt, where average transfer efficiency is strongly affected by disc overlaps, an addition of random on-site energy disorder has no noticeable effect. The effect is more pronounced for R > Ropt where overlapping of discs is not the limiting factor of transfer efficiency any more. The estimated optimal size Ropt however is not changed considerably by an addition of random on-site disorder. n Appendix C. Time evolution of site populations To provide some insight into the temporal dynamics of the excitation transport, we present the time evolution of site populations for two different realizations of PPC within the Lindblad model. In figure C1a, time evolution for the FMO Hamiltonian 1012141618202224R[A]0.800.850.900.951.00ηγ=600,300,150cm−1(a)1012141618202224R[A]0.800.850.900.951.00ησrand=200,100,50,0cm−1(b)Optimal number of pigments in photosynthetic complexes 19 Figure C1. (a) Time evolution of site populations hnρni, calculated using Lindblad model for the FMO Hamiltonian from reference [6], resulting in efficiency of η = 0.99 and average transfer time of τ = 6.2 ps. At t = 0, excitation is localized on site 1. Sink is connected to the site 3. Dotted line represents the population of the sink. (b) Time evolution for a random sample, generated for R = 18 Å, with efficiency η = 0.90 and average transfer time τ = 45 ps. Blue line is the population of the input site that is initially excited. Red line is the population of the output site, connected to the sink. Dotted line represents the population of the sink. from reference [6] is shown, and in figure C1b, the time evolution of randomly generated PPC of radius R = 18 Å. The values of dephasing, sink rate and recombination rate are the same as used in the main text. References [1] Blankenship R E 2002 Molecular Mechanisms of Photosynthesis (Blackwell Science Ltd) [2] Matthews B W, Fenna R E, Bolognesi M C, Schmid M F and Olson J M 1979 J. Mol. Biol. 131 [3] Tronrud D E, Wen J, Gay L and Blankenship R E 2009 Photosynth. Res. 100 79 -- 87 [4] Engel G S, Calhoun T R, Read E L, Ahn T K, Mancal T, Cheng Y C, Blankenship R E and Fleming G R 2007 Nature 446 782 -- 6 [5] Brixner T, Stenger J, Vaswani H M, Cho M, Blankenship R E and Fleming G R 2005 Nature 434 [6] Cho M, Vaswani H M, Brixner T, Stenger J and Fleming G R 2005 J. Phys. Chem. B 109 10542 -- 56 [7] Ishizaki A and Fleming G R 2009 Proc. Natl. Acad. Sci. U. S. A. 106 17255 -- 60 [8] Rebentrost P, Mohseni M and Aspuru-Guzik A 2009 J. Phys. Chem. B 113 9942 -- 7 [9] Ishizaki A, Calhoun T R, Schlau-Cohen G S and Fleming G R 2010 Phys. Chem. Chem. Phys. 12 259 -- 285 625 -- 8 7319 -- 37 [10] Pachón L A and Brumer P 2011 J. Phys. Chem. Lett. 2 2728 -- 2732 [11] Plenio M B and Huelga S F 2008 New J. Phys. 10 113019 [12] Rebentrost P, Mohseni M, Kassal I, Lloyd S and Aspuru-Guzik A 2009 New J. Phys. 11 033003 [13] Mohseni M, Rebentrost P, Lloyd S and Aspuru-Guzik A 2008 J. Chem. Phys. 129 174106 [14] Caruso F, Chin A W, Datta A, Huelga S F and Plenio M B 2009 J. Chem. Phys. 131 105106 [15] Beddard G S and Porter G 1976 Nature 260 366 -- 367 [16] Scholak T, Wellens T and Buchleitner A 2011 Europhys. Lett. 96 10001 [17] Scholak T, de Melo F, Wellens T, Mintert F and Buchleitner A 2011 Phys. Rev. E 83 021912 0200040006000800010000t[fs]0.00.20.40.60.81.01234567(a)0200040006000800010000t[fs]0.00.20.40.60.81.0inputoutput(b)Optimal number of pigments in photosynthetic complexes 20 [18] Scholak T, Wellens T and Buchleitner A 2011 J. Phys. B: At. Mol. Opt. Phys. 44 184012 [19] Mohseni M, Shabani A, Lloyd S and Rabitz H 2011 arXiv:1104.4812 [20] Sener M K, Lu D, Ritz T, Park S, Fromme P and Schulten K 2002 J. Phys. Chem. B 106 7948 -- 7960 [21] Sener M K, Jolley C, Ben-Shem A, Fromme P, Nelson N, Croce R and Schulten K 2005 Biophys. [22] Hayes D, Wen J, Panitchayangkoon G, Blankenship R E and Engel G S 2011 Faraday Discuss. J. 89 1630 -- 42 150 459 (Wiley-VCH) 93 -- 98 [23] Förster T 1948 Ann. Phys. 437 55 -- 75 [24] Ishizaki A and Fleming G R 2009 J. Chem. Phys. 130 234110 [25] Ishizaki A and Fleming G R 2009 J. Chem. Phys. 130 234111 [26] Nalbach P, Braun D and Thorwart M 2011 Phys. Rev. E 84 041926 [27] Ritschel G, Roden J, Strunz W T and Eisfeld A 2011 New J. Phys. 13 113034 [28] May V and Kühn O 2011 Charge and Energy Transfer Dynamics in Molecular Systems 3rd ed [29] Cao J and Silbey R J 2009 J. Phys. Chem. A 113 13825 -- 38 [30] Wu J, Liu F, Shen Y, Cao J and Silbey R J 2010 New Journal of Physics 12 105012 [31] Wu J, Liu F, Ma J, Silbey R J and Cao J 2011 arXiv:1109.5769 [32] Adolphs J and Renger T 2006 Biophys. J. 91 2778 -- 97 [33] Schmidt am Busch M, Müh F, El-Amine Madjet M and Renger T 2011 J. Phys. Chem. Lett. 2 [34] Olbrich C, Jansen T l C, Liebers J, Aghtar M, Strumpfer J, Schulten K, Knoester J and Kleinekathoefer U 2011 J. Phys. Chem. B 115 8609 -- 21 [35] Milder M T W, Brüggemann B, van Grondelle R and Herek J L 2010 Photosynth. Res. 104 257 -- 74 [36] Kreisbeck C, Kramer T, Rodríguez M and Hein B 2011 Journal of Chemical Theory and Computation 7 2166 -- 2174 [37] Croce R and van Amerongen H 2011 J. Photochem. Photobiol. B: Biol. 104 142 -- 53 [38] Novoderezhkin V, Marin A and van Grondelle R 2011 Phys. Chem. Chem. Phys. 13 17093 -- 103 [39] Liu Z, Yan H, Wang K, Kuang T, Zhang J, Gui L, An X and Chang W 2004 Nature 428 287 -- 92 [40] Umena Y, Kawakami K, Shen J R and Kamiya N 2011 Nature 473 55 -- 60 [41] Renger T 2011 Procedia Chemistry 3 236 -- 247
1510.02940	1	1510	2015-10-10T14:12:08	Thermodynamic Free Energy Methods to Investigate Shape Transitions In Bilayer Membranes	[ "physics.bio-ph", "cond-mat.soft", "cond-mat.stat-mech" ]	The conformational free energy landscape of a system is a fundamental thermodynamic quantity of importance particularly in the study of soft matter and biological systems, in which the entropic contributions play a dominant role. While computational methods to delineate the free energy landscape are routinely used to analyze the relative stability of conformational states, to determine phase boundaries, and to compute ligand-receptor binding energies its use in problems involving the cell membrane is limited. Here, we present an overview of four different free energy methods to study morphological transitions in bilayer membranes, induced either by the action of curvature remodeling proteins or due to the application of external forces. Using a triangulated surface as a model for the cell membrane and using the framework of dynamical triangulation Monte Carlo, we have focused on the methods of Widom insertion, thermodynamic integration, Bennett acceptance scheme, and umbrella sampling and weighted histogram analysis. We have demonstrated how these methods can be employed in a variety of problems involving the cell membrane. Specifically, we have shown that the chemical potential, computed using Widom insertion, and the relative free energies, computed using thermodynamic integration and Bennett acceptance method, are excellent measures to study the transition from curvature sensing to curvature inducing behavior of membrane associated proteins. The umbrella sampling and WHAM analysis has been used to study the thermodynamics of tether formation in cell membranes and the quantitative predictions of the computational model are in excellent agreement with experimental measurements. Furthermore, we also present a method based on WHAM and thermodynamic integration to handle problems related to end-point-catastrophe that are common in most free energy methods	physics.bio-ph	physics	Thermodynamic Free Energy Methods to Investigate Shape Transitions In To appear in Int. J. Adv. Eng. Sci. & Appl. Math. Bilayer Membranes N. Ramakrishnan∗ Department of Bioengineering, University of Pennsylvania, Philadelphia, PA, 19104∗ Richard W. Tourdot Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, PA, 19104 † Ravi Radhakrishnan Department of Bioengineering, Department of Chemical and Biomolecular Engineering, Department of Biochemistry Biophysics, University of Pennsylvania, Philadelphia, PA, 19104 ‡ (Dated: October 2, 2018) Abstract The conformational free energy landscape of a system is a fundamental thermodynamic quantity of importance particularly in the study of soft matter and biological systems, in which the entropic contributions play a dominant role. While computational methods to delineate the free energy landscape are routinely used to analyze the relative stability of conformational states, to determine phase boundaries, and to compute ligand-receptor binding energies its use in problems involving the cell membrane is limited. Here, we present an overview of four different free energy methods to study morphological transitions in bilayer membranes, induced either by the action of curvature remodeling proteins or due to the application of external forces. Using a triangulated surface as a model for the cell membrane and using the framework of dynamical triangulation Monte Carlo, we have focused on the methods of Widom insertion, thermodynamic integration, Bennett acceptance scheme, and umbrella sampling and weighted histogram analysis. We have demonstrated how these methods can be employed in a variety of problems involving the cell membrane. Specifically, we have shown that the chemical potential, computed using Widom insertion, and the relative free energies, computed using thermodynamic integration and Bennett acceptance method, are excellent measures to study the transition from curvature sensing to curvature inducing behavior of membrane associated proteins. The umbrella sampling and WHAM analysis has been used to study the thermodynamics of tether formation in cell membranes and the quantitative predictions of the computational model are in excellent agreement with experimental measurements. Furthermore, we also present a method based on WHAM and thermodynamic integration to handle problems related to end-point-catastrophe that are common in most free energy methods. PACS numbers: 87.16.-b, 87.17.-d 5 1 0 2 t c O 0 1 ] h p - o i b . s c i s y h p [ 1 v 0 4 9 2 0 . 0 1 5 1 : v i X r a ∗ [email protected] † [email protected] ‡ [email protected] 1 I. INTRODUCTION Surfactant molecules self assemble into mesoscale structures (characteristic lengths are in the order of hundreds of nanometer) when their concentration in an aqueous solvent exceeds a threshold value, gener- ally called the critical micelle concentration (CMC). Examples of these mesoscale entities include simple structures like a monolayer of surfactants at the air-water/air-oil interface or more complex structures like a micelle and a bilayer of surfactants in the bulk. The stability of a given mesoscale structure is in turn is governed by the geometry and chemistry of the individual surfactant molecules [1]. Characteristic energies of a self assembled surfactant interface are comparable to the thermal energy kBT , where kB is the Boltzmann constant and T is the equilibrium temperature, and a result the spatial organization of the molecules, which is characterized at the mesoscale by the morphology and topology of the interface, is susceptible to thermal fluctuations in the solvent. A similar but a more complex system that is of importance to cell biology is the lipid bilayer membrane, formed by the self assembly of lipid molecules, which defines the outer boundaries of most mammalian cells and their organelles. Lipid molecules are fatty acids synthesized within the cell and like a surfactant molecule they also have a hydrophilic head group and a hydrophobic tail -- commonly occurring lipids include glycerol based lipids such as DOPC, DOPS and DOPE, sterol based lipids like cholesterol, and ceramide based lipids like sphingomyelin [2]. The cell membrane is formed by the self assembly of these different types of lipid molecules along with other constituents namely proteins and carbohydrates, and the composition of these building blocks differ across different cell membranes [3 -- 5]. Being the interface of the cell, the lipid membrane plays a dominant role in a number of biophysical processes either by virtue of its surface chemistry at the molecular scale or through modulations in its physical properties at the mesoscale: the most obvious examples of the latter include inter- and intra-cellular trafficking [6 -- 9], membrane mediated aggregation of cell signaling molecules [10, 11] and cell motility [12 -- 14]. Hence, it is natural to expect an inherent feedback between the physical properties of the cell membrane and the biophysical processes it mediates. The primary aim of this article is to review theoretical and compu- tational approaches at the mesoscale that can be used to develop an understanding of this feedback. In particular, our focus is to show how thermodynamic free energy methods employed in a variety in a contexts in condensed matter physics can be applied to the theoretical models for membranes at the mesoscale. In equilibrium statistical mechanics, the ground state of a system whose intensive or extensive vari- ables are coupled to the environment, and hence can exchange for instance heat or area or volume or number with the bath, is governed by its thermodynamic potential which is also called the free energy of the system. The various thermodynamic observables can be determined by measuring the suitable thermodynamic potential that depends on the ensemble in which the system is defined [15]. Excellent introduction to the implementations and applications of the various free energy methods for molecular systems is provided by Frenkel and Smit [16]. II. CONTINUUM MODELS FOR CELL MEMBRANES The spatial and temporal resolution of the various biophysical processes observed in cell membranes can be classified into two broad classes, namely (a) biochemical processes in which the dynamics of the system is primarily determined by the chemistry of the constituent molecules and (b) biophysical processes where collective phenomena and macroscopic physics govern the behavior of the membrane. These two class of processes have disparate time and length scales. The large separation in the time and length scales allows one to decouple the slower degrees of freedom from the faster ones and this feature can be exploited in constructing physical models at multiple scales for the cell membrane. Molecular scale models such as all-atom or coarse grained molecular dynamics are faithful to the underlying chemistry and are hence more appropriate for investigating membrane processes in the sub cellular length and nanoscopic time scales. In the other limit, phenomenology based field theoretic models neglect the membrane dynamics at the nanoscale and instead focus on how the collective effects of these molecular motions manifest at length and time scales comparable to those accessed in conventional experiments like light microscopy and mechanical measurements of cells. More rigorous discussions on the formulation of multiscale models for membranes can found in a number of review articles on this topics [17 -- 24]. In this article, we will use the thermodynamic formalism of membrane biophysics to demonstrate how free energy methods can extended to the study of diverse class of problems involving the cell membrane at the mesoscale. 2 The phenomenology based approach focuses primarily on the conformational states of the bilayer membrane at length scales (> 100 nm) that are large compared to the thickness of the membrane (∼ 5 nm). In this approach the membrane is treated as a thin elastic sheet of a highly viscous fluid with nearly constant surface area (the number of lipids under consideration is assumed to be constant). This sheet is representative of the neutral surface of a membrane bilayer: it is defined as the cross sectional surface in which the in-plane strains are zero upon a bending transformation, see references [17, 23, 24] for details. The thermodynamic weights of the conformational state of the membrane is governed by the well known Canham-Helfrich energy functional [25, 26] commonly written as, (cid:110) κ (cid:90) S (cid:111) (cid:90) + V H = dS (2H − H0)2 + κGG + σ 2 dV ∆p. (1) If c1 and c2 are the principal curvatures at every point on the membrane surface S then H = (c1 + c2)/2 and G = c1c2 are its mean and Gaussian curvatures respectively. The elastic moduli κ and κG are the isotropic and deviatoric bending moduli. Experimental measurements on lipid and cell membranes have estimated their bending stiffness κ to be in the range 10 − 100kBT , with the lower values corresponding to model membrane structures like giant uni-lamellar vesicles. The deviatoric modulus is normally taken to κG = −κ but the Gaussian energy term can be neglected, by virtue of the Gauss-Bonnet theorem [27] if the topology of the membrane does not change during the analysis. The surface area of the membrane A and the volume V are coupled to their respective conjugate variables namely the surface tension σ and osmotic pressure ∆p. Reported values of membrane surface tension (combined contributions from both the lipids and the underlying cytoskeleton) varies between 3 -- 300 µN/m depending on the cell type [28, 29]. The osmotic pressure difference ∆p is a function of the difference in the osmolyte concentration between the inside and outside of the cell. The spontaneous curvature H0 denotes an induced curvature which can arise in a number of context such as defects in lipid packing, the presence of intrinsic degrees of freedom in the constituent lipids, interactions of the membrane with non-lipid molecules like proteins or nanoparticles and also due to the coupling of the membrane with the underlying cytoskeleton. Since the spontaneous curvature is an important parameter in most of our discussions later, it is important to have a closer look at how its impacts the conformational states of the membrane. When H0 = 0, the energy given by the first term in eqn. (1) is quadratic in the mean curvature H and hence the probability of finding a membrane conformation with a given curvature H∗ is a Gaussian peaked around H = 0, with its width being proportional to the bending stiffness κ [30]. On the other hand, when the membrane has a non-zero spontaneous curvature the peak of the probability distribution now shifts to a value H = H0 and as a result highly curved membrane regions are observed with much larger probabilities. For purposes of computer simulations, a number of discretizations based on eqn. (1) have been intro- duced in the literature. The free energy methods for membranes presented in the later sections are based on the Dynamical Triangulation Monte Carlo technique which has been reviewed in brief below. The two dimensional membrane surface is discretized into an interconnected set of T triangles that intersect at N vertices (nodes) forming L independent links. The values of N , T , and L define the topology of the membrane surface in terms of Euler characteristic as χ = N + T − L. The degrees of freedom of the discretized membrane are the position vectors of the N vertices given by { (cid:126)X} = [(cid:126)x1 ··· (cid:126)xN ] and the triangulation map given by {T } = [T1 ··· TT ]. The discrete form of the elastic Hamiltonian is thus a sum over the curvature energies at every vertex in the triangulated surface given by, (cid:110) κ N(cid:88) v=1 (cid:111) H = Av (c1,v + c2,v − H0,v)2 + σ 2 + ∆pV. (2) The index v denotes a vertex on the triangulated surface and c1,v and c2,v are respectively its principal curvatures, H0,v is the local spontaneous curvature, and Av denotes the surface area associated with the vertex. The principal curvatures are computed using the methods introduced by Ramakrishnan et. al. [31]. The spontaneous curvature at a vertex is expressed using the general form: H0,v = C0D(v, v(cid:48)), (3) N(cid:88) v=1 3 with C0 being the magnitude of the induced curvature and D(v, v(cid:48)) the functional form of the curvature contribution at vertex v(cid:48) due to a curvature field at vertex v. The various forms of D(v, v(cid:48)) relevant in different contexts have been discussed in references [23, 32 -- 35]. In this article, we limit our discussions on curvature induced membrane remodeling to protein that have isotropic curvature fields with a Gaussian profile: (cid:18) (cid:19) D(v, v(cid:48)) = exp −xv − xv(cid:48)2 22 . (4) Here 2 denotes the range of a curvature field, i.e. a curvature field defined at a vertex v can induce a non- zero spontaneous curvature at a far vertex v(cid:48). We denote the set of all protein fields as {φ} = [φ1 ··· φN ]. In addition to the elastic potential given in eqn. (2) the membrane vertices are also subjected to a re- pulsive hard-sphere potential in order to enforce self avoidance. If the vertices of the membrane are taken to spheres of diameter a0 then the length E of the links, connecting any two vertices, obey the constraint a0 ≤ E < √3a0. Detailed discussions on this topic can be found in reference [cite Elsevier chapter]. The conformational state of the triangulated membrane is given by η = [{ (cid:126)X},{T },{φ}] and the various state are sampled using a set of three Monte Carlo moves: (i) a vertex move in which a randomly chosen vertex is displaced to new location that leads to change in state [{ (cid:126)X},{T },{φ}] → [{ (cid:126)X(cid:48)},{T },{φ}], (ii) a link flip wherein two previously unconnected nodes of a randomly chosen quadrilateral on the triangulated surface are connected to form a new set of triangulation leading to [{ (cid:126)X},{T },{φ}] → [{ (cid:126)X},{T (cid:48)},{φ}], and (iii) a field exchange move to simulate diffusion of the protein field in which the protein field at vertex v is exchanged with that at vertex v(cid:48) that leads to a change in state [{ (cid:126)X},{T },{φ}] → [{ (cid:126)X},{T },{φ(cid:48)}] . The various Monte Carlo moves are accepted using Metropolis scheme [36] given by , where β = 1/kBT denotes the temperature of the system. Pacc = min More details on the implementation and usage of Dynamical Triangulation Monte Carlo techniques can be found in reference [23]. (cid:16)H (η(cid:48)) − H (η) (cid:17)(cid:17)(cid:111) (cid:110) 1, exp −β (cid:16) III. AN OVERVIEW OF FREE ENERGY METHODS TO STUDY MEMBRANE DEFORMA- TIONS As noted in the introduction, the thermodynamic free energy of a system is a fundamental quantity in equilibrium statistical mechanics since it contains all the information about the thermodynamic variables of the system. However, the free energy landscape of many body systems is a very complex quantity and the complexity arises primarily from the large number of degrees of freedom associated with such systems -- for example an N particle system in one dimensions has a conformational free energy landscape that is N dimensional. In most problems in condensed matter physics, computational biology, and computational chemistry the free energy landscape in the conformational space of the atoms/molecules is an over representation of the system and hence the problem of large number of dimensions can be overcome through coarse graining or representing the system in terms of a fewer macroscopic variables generally called order parameters. The use of the computational methods to delineate free energy landscape is highly optimal in such coarser representations. The problem of protein induced curvature remodeling of membranes is one such problem that is amenable to the use to free energy methods. The partition function of the triangulated membrane surface with n protein fields is defined in terms of the thermodynamic state η as, (cid:88) (cid:90) (cid:88) Qn = φ∈{φ} T ∈{T } d{X} exp (−βH (x, T , φ)) . (5) The partition trace in eqn. (5) is performed over all vertex positions and triangulations of the discrete surface and also over all possible configurations of the n proteins on the membrane. The absolute free energy of the membrane with n proteins is thus: Computing the absolute free energy Fn requires the calculation of the partition function Qn which is a Fn = −kBT ln Qn. (6) 4 problem that requires extensive sampling of the infinitely large conformational states of the membrane- protein system -- Metropolis Monte Carlo is not suitable for such purposes since it can only sample the configurational states close to the free energy minimum and the higher energy states can instead be sampled using rare event Monte Carlo techniques which are described later in the context of umbrella sampling. On the other hand, the relative free energy of state m with respect to state n, i.e. the quantity Fm − Fn, does not require the knowledge of Qn and can be computed with less expensive computations. In the remainder of this section, we will describe three different methods namely the Widom insertion, thermodynamic integration (TI), and Bennett Acceptance method (BAM) to essentially perform the same calculation -- to compute ∆F = Fn+1 − Fn, the free energy difference between a membrane with n and n + 1 proteins. A. Widom insertion technique The Widom particle or test-particle insertion method is a computational technique to probe the chem- ical potential of a system [37]. It is known from statistical mechanics that the total chemical potential of a system is defined as the change in its free energy in response to a change in the system size. In the case of membrane-protein systems, the total chemical potential of the membrane with n proteins (µP ) is essentially the required free energy difference, i.e. µP = ∆F/∆n. The total chemical potential can be separated into an ideal and an excess part such that µP = µid P + µex P . (7) The excess part of the chemical potential µex P can be computed using Widom insertion technique in which a virtual test(ghost) protein field is inserted at a randomly chosen location on the membrane. The configurational component of the ideal part can be shown to be µid P = kBT ln ρ, where ρ is the protein density. If ∆H be the change in the elastic energy due to the insertion of a test protein field then the excess chemical potential is given by, (cid:90) µex P = −kBT ln (cid:104)exp(−β∆H )(cid:105) Puniform(sn+1)dsn+1. (8) The ensemble average (cid:104)·(cid:105) is taken over the configurational space of the partition function, see eqn. (5). Here, sn+1 = xp, with p = n + 1, is the position of the n + 1th protein field on the membrane surface, and Puniform(sn+1) denotes a uniform probability distribution from which the coordinates of the n + 1th particle/field is drawn. The integral over sn+1 amounts to the sum over all Widom test particle/field insertion trials, and Puniform(sn+1) equals the reciprocal of the total number of trials. For conciseness, we represent the right-hand-side term in equation (8) as −kBT ln(cid:104)exp (−β∆H )(cid:105)n and here the subscript n denotes that the ensemble average is taken on a membrane with n proteins. The form of the excess chemical potential given in eqn. (8) has been derived by treating all insertion sites on the membrane to be homogeneous, while the local excess chemical potential which depends on the spatial location x on the membrane surface is given by, P (x) = −kBT ln(cid:104)exp (−β∆H (x))(cid:105)n. µex (9) In eqn. (9), ∆H (x) denotes the change in energy due to the insertion of a protein field at spatial location x. The inhomogeneous chemical potential and the spatially uniform chemical potential µex P can be used to determine the inhomogeneous, scaled spatial density of the proteins using the relation: ρ(x) = exp (βµP ) exp (−βµex P (x)) , (10) B. Thermodynamic integration (TI) method Thermodynamic integration is perturbative technique that can be used to determine the relative free energy difference between any two thermodynamic states A and B, provided there exists a continuous path C that connects state A to B; in the context of protein induced membrane remodeling the states A and B correspond to a membrane with n and n + 1 proteins respectively. If Hn and Hn+1 be the 5 energies of these two states then the various intermediate states of the membrane along the path C can be obtained by varying the coupling parameter 0 ≤ λ ≤ 1 such that the energy of any intermediate state is given by H (λ), subject to the boundary condition H(0) = Hn and H (1) = Hn+1. The energy of an intermediate state with a given value of λ can be expressed in terms of the energies of the end states as [16]: The free-energy change along this path can expressed in its integral form as, H (λ) = (1 − λ)Hn + λHn+1. ∆FTI = Fn+1 − Fn = ∂F(λ) ∂λ dλ. (cid:90) 1 0 (11) (12) Using the definitions of the partition function and the free energy, given in eqns. (5) and (6), in the above equation the relative free energy ∆FTI can be shown to be (cid:90) 1 (cid:28) ∂H (λ) (cid:29) ∆FTI = 0 ∂λ dλ. (13) In practice, the integrand in eqn. (13) is estimated from independent simulations of the system at pre- determined, sufficiently small intervals of λ and the relative free energy ∆FTI is then estimated through numerical integration of eqn. (13). C. Bennett acceptance ratio method (BAM) Bennett acceptance method is another perturbative technique that can be used to approximate the free-energy difference between two states close to each other in their conformational space [38]. The microscopic reversibility for the transition of the membrane-protein system between the two states with n and n + 1 proteins, also called detailed balance condition, can be stated as M (Hn+1 − Hn) exp(−βHn) = M (Hn − Hn+1) exp(−βHn+1), (14) where M is some function that defines the distribution of the acceptance probability for a transition of the membrane from a state with n proteins to a state with n+1 proteins and vice versa. First, integrating both sides of eqn. (14) over the entire conformational space over which the partition trace of eqn. (5) is defined, and then multiplying and dividing both sides with their corresponding partition functions the above equation can be rewritten as, (cid:82) dη M (Hn+1 − Hn) exp(−βHn) Qn Qn (cid:82) dη M (Hn − Hn+1) exp(−βHn+1) Qn+1 Qn+1 = , (15) Here, for conciseness we use the state variable η to denote integration over all the states of the triangulated surface. The above equation reduces to the form Qn+1 Qn ≡ exp −∆FBAM kBT = (cid:104)M (Hn+1 − Hn)(cid:105)n (cid:104)M (Hn − Hn+1)(cid:105)n+1 , A→B (16) which gives an exact expression for the relative free energy difference denoted as ∆FBAM, which can be exactly computed if the form the the transition function M is known. A common choice of M is the Metropolis function M (x) = min(1, exp(−βx)), which defines the acceptance probability according to a Boltzmann distribution. In eqn. (16), (cid:104)·(cid:105)m represents the ensemble average of M for the transition from state m → m + 1 membrane in state m to state m(cid:48). 6 (cid:18) (cid:19) D. Umbrella Sampling and weighted histogram analysis method In a number of scenarios it is desirable to determine the statistical weight and the associated free energy of a particular state of a system. Canonical simulation techniques based on equilibrium Monte Carlo or Molecular dynamics are not well suited for such purposes if the desired state of the system has a large energy barrier with respect to its equilibrium; Arrhenius' law predicts negligibly small transition rates across this energy barrier and hence one would require infinitely long trajectories of the system in order to generate extensive samples of the state with higher energy. Such at transition event is called a rare event and there are a number of techniques, such as Rosenbluth sampling, Wang-Landau sampling, and umbrella sampling, that can be used to access the rare states of the system within acceptable simulation times. In this article, we only focus on the umbrella sampling technique along with the Weighted Histogram Analysis Method (WHAM) [39] to study the thermodynamics of large deformations in cell membranes. In general, let ζ denote an atomistic or a molecular or a continuum or a collective variable of a system ∗ with an equilibrium probability distribution P(ζ) which is peaked around the equilibrium value ζ = ζ -- conventionally the variable ζ is called a reaction coordinate. Umbrella sampling involves the simulation of the system in NB different windows in the presence of an additional harmonic biasing potential Bi(ζ) = kbias 2 (ζ − ζi)2 , (17) such that ζi denotes the preferred value of ζ in the ith window and Pi(ζ) its probability distribution. kbias is the strength of the biasing spring which is chosen such that the probability distributions from neighboring windows show considerable overlap. The probability distributions Pi(ζ) computed across multiple simulations windows can be combined together using the Weighted Histogram Analysis method to estimate the free energies of all intermediate states, with respect to the first window. The free energy is computed by self-consistently solving the two WHAM equations, for the unknowns P(ζ) and Fi, given by: P(ζ) = i=1Ni exp (−β (Bi(ζ) − Fi)) , (18) NB(cid:80) i=1NiPi(ζ) NB(cid:80) nbins(cid:88) j=1 where Ni is the number of samples in the ith window and P(ζj) exp(cid:0) −βBi(ζj)(cid:1) + C. (19) Fi = −kBT ln Here nbins denotes the number of used over which the free energy is discretized and C is an arbitrary constant. IV. PREDICTING TRANSITION FROM CURVATURE SENSING TO CURVATURE INDUC- ING BEHAVIOR USING WIDOM INSERTION An important question in the area of protein driven curvature remodeling of membranes is "when does a cluster of proteins behave in a cooperative manner?" In vitro experiments on a number of curvature inducing proteins such as BAR domains, ENTH domains, and Exo70 domains have shown that these proteins when at low concentrations localize to high curvature regions on the membrane generated by thermal undulations -- commonly known as the curvature sensing behavior -- while at high concentrations they aggregate into clusters and spontaneously generate membrane curvature to form highly curved membrane morphologies such as tubules and blebs -- a characteristic of curvature inducing behavior. Delineating this transition regime is a challenge in experiments but computational models based on free energy methods are well suited for this purpose. The relative energy free energy of the membrane is an excellent marker for the curvature sensing to transition behavior. It was pointed out 7 in Sec. III that the relative free energy to introduce the n + 1th protein in a membrane with n proteins can be computed using Widom insertion or thermodynamic integration or Bennett-acceptance-method. Here we use the computationally less expensive Widom insertion technique to determine the relative free energies to insert a protein on a membrane with n = 0 (i.e. a pure lipid membrane). In the continuum description a protein is represented as a curvature field with a Gaussian profile that is parameterized using two variables namely the maximum spontaneous curvature (C0) and the extent of the curvature field (2), see eqns. (3) and (4). We express both C0 and 2 in units of a0, which represents the hard sphere diameter associated with a vertex of the triangulated surface. The excess P as a function of 2 (for fixed values of C0) and C0 (for fixed values of 2) are chemical potentials µex shown in Figs. 1(a) and 1(b) respectively. It can be seen that µex P is negative for small values of C0 and 2 which indicates that the free energy of the system is reduced upon introduction of the protein. Assuming the entropic contribution to be negligible, this implies that the total bending potential given by eqn. (2) is smaller in the presence of the protein which is only possible when H ≈ C0. Since the Widom test particle only probes the membrane curvature and does not deforms the membrane it is clear that the equilibrium curvature profile of the membrane matches that of the inserted protein field which is characteristic of curvature sensing behavior. On the other hand when C0 and 2 are larger (C0 > H) µex P become large and positive and since such states are thermodynamically unstable any such proteins associated with the membrane would tend to to generate local curvatures that match their intrinsic curvature profile and this regime where the protein induces curvature. The results presented in Fig. 1 only focus on the thermodynamic stability of a single protein but even weakly curving proteins can transition from curvature sensing to inducing behavior, when their concentration exceeds a critical value. The effect of the cooperative behavior, due to the self- and membrane-mediated interactions of the proteins, on the curvature inducing properties of membrane associated proteins has been recently studied in the context of ENTH domains [35, 40]. Excess chemical potential, in units of kBT , to insert a protein field with maximum spontaneous Figure 1. curvature C0 and extent of curvature 2 on a membrane with zero proteins -- both C0 and 2 are expressed in units of a0. (a) µex P as a function of C0 for fixed values of 2 = 2.3a2 P as a function of 2 for fixed values of C0 = 0.4a −1 −1 0 , and 0.8a 0 0, 4.3a2 0, and 6.3a2 0 −1 0 , 0.6a and (b) µex V. COMPARING PREDICTIONS FROM WIDOM INSERTION, TI, AND BAM The excess chemical potential is reflective of the underlying free energy landscape as shown in Fig. 1 and in section we compare these predictions to the corresponding relative free energy levels obtained using thermodynamic integration and Bennett-acceptance-method. The free energies determined using TI and BAM (see eqns. (13) and (16)) are related to the total chemical potential µP = µid P as: P (ρ) + µex ∆FTI = ∆FBAM = µP . (20) 8 234567892(ina20)−6−4−2024681012µexP(inkBT)(a)C0=0.4C0=0.6C0=0.80.30.40.50.60.70.80.91.01.1C0(ina−10)−505101520(b)2=2.32=4.32=6.3Since the Widom insertion technique can only be used to determine the excess part of the chemical potential the various free energies can be compared only if the total chemical potential can be determined. µid P (ρ) is the entropic configurational component of µP and depends on the number of conformational states visited by a single particle or protein-field. The number of conformational states accessible to the n + 1th protein on a membrane with n proteins can in turn be determined using the trajectories obtained using TI. In TI, the additional protein field is grown from a non-existent entity to a full-existent object by varying the parameter λ from 0 to 1. The degree of localization changes with change in λ and when λ → 1 the protein does not explore all the conformational states but is instead confined to the minimum of the free energy well, and this minimum in the triangulated surface model corresponds to a subset of vertices on the surface. The required correction µid P (ρ) can be calculated from the number of unique vertices Nψ visited by a protein field in various TI simulations with λ ∼ 1 as: (cid:18) 2σψ (cid:19) µid P (ρ) = −kBT ln N . (21) Here N is the total number of vertices on the triangulated surface and the standard deviation σψ of the distribution of unique vertices is computed as,  Nψ(cid:88) υ=1  Nψ(cid:88) υ=1 2 1 2 σψ = υ2Pυ − υPυ , (22) with Pυ being the probability of a protein to visit the υth unique vertex. (a) Distribution of the number of unique vertices P (υ), visited in a TI simulation with λ ∼ 1, for Figure 2. four different values of C0. The points shown alongside each curve correspond to the standard deviation σψ. (b) Comparison of the relative free energies to add one protein to a membrane with zero proteins computed using TI, BAM, and Widom insertion. The probability distribution of the number of unique vertices, for four different values of C0, is shown in Fig. 2(a) and the symbols shown alongside correspond to the value of σψ. The total chemical potential that compares the entropic configurational part is compared against the relative free energies computed using TI and BAM in Fig. 2(b) . It can be seen from Fig. 2(b) that all three methods agree well for low values of C0 while Widom insertion deviates from the other methods above C0 > 0.6a−1 0 . The deviation of TI and Widom insertion methods at high C0 is well known since efficient sampling of µex P suffers for large perturbations in energy or higher densities. 9 020406080100120υ01020304050P(υ)(a)0.8a−100.6a−100.4a−100.2a−100.00.20.40.60.81.0C0(a−10)−50510∆F(inkBT)(b)∆FBAM∆FTIµPVI. IN SILICO TETHER PULLING EXPERIMENTS Extraction of cylindrical protrusions (tethers) from the surface of a cell membrane, using optical tweez- ers or functionalized AFM tips or through attachment of magnetic beads, is a useful method to charac- terize its mechanical properties such as the bending stiffness, surface tension, and degree of cytoskeletal pinning. A tether is characterized by its radius Rtether, its length Ltether and the force required for its extraction denoted by Ftether, as shown in the illustration in Fig. 3. In order to clearly delineate the role of the various parameters characterizing a cell membrane in a typical in vivo tether extraction assay, it is essential to develop physical models that allow us to gain an understanding at a fundamental level. In this section, we present an in silico tether extraction assay by combining the triangulated surface membrane with umbrella sampling techniques and the weighted histogram analysis method. In order to stabilize a membrane tether of length Ltether, we apply a umbrella sampling biasing potential on a set of macroscopic variables which are defined as follows. The tip of the tubular region is represented by a set of pre-determined vertices {X}T with center of mass RT and the base of the tether is represented by another set of of vertices {X}B with center of mass RB, such that Ltether = RT − RB and each vertex in {X}B with position vector xB obeys the constraint RT − xB ≤ 1.5Ltether. The macroscopic positions of the tip and base of the membrane tether RT and RB define an order parameter which is subjected to a harmonic biasing potential in the nth window given as: Bn(RT , RB) = kbias 2 (RT − RB − L ∗ tether)2 (23) ∗ tether. ∗ kbias is the strength of the biasing potential and L tether denotes the preferred tether length. The conformational state of the membrane patch is evolved using the Dynamical Triangulation Monte Carlo technique with the total potential Htot = H + Bn. The conformations of a membrane, with fixed values ∗ of κ = 20kBT , L = 510nm, and Aex = 10%, in five different biasing windows with L tether = 4, 32, 64, 96, and 128 nm are shown in the top panel of Fig. 4. The positions of the center of masses corresponding to the biasing vertices, RT and RB respectively, are also shown alongside and it can be seen that tether like structures are readily formed at larger values of L The probability distribution of the tether length Ltether in 32 different sampling windows, with a window size of 4 nm, are shown in the bottom panel of Fig. 4. Pn(Ltether) shows a normal distribution in each of the 32 windows and the peak of the distribution shifts to a higher value of Ltether with increasing ∗ tether and the strength of the biasing potential kbias was chosen so that distributions from adjacent L windows show a good overlap as seen in Fig. 4. It can also be seen that Pn(Ltether) becomes narrower ∗ following the formation of the tether at L tether ≈ 96nm. The potential of mean force W(Ltether) which denotes the energy required to extract a tether of length Ltether, computed by combining the histograms in Fig. 4 using WHAM, is shown in the top panel of Fig. 5. The PMF shows three distinct regimes which are also shown alongside the PMF in Fig. 5: (i) an initial weakly linear regime (∝ Ltether), (ii) an intermediate quadratic regime (∝ L2 tether), and (iii) a final linear regime (∝ Ltether). These three regimes have a significance in the formation and stabilization of the membrane tether. When a force is applied to an undulating membrane the short wavelength undulations in the membrane conformation are suppressed in the linear response regime and this response characterizes the initial linear regime. The tubular structures nucleate and grow in the intermediate quadratic regime until all the undulations in the membrane are ironed out and are drawn into the tubular region -- the extent of the quadratic regime changes with change in the membrane excess area which sets the intensities of the characteristic undulations in the membrane [41]. In the final linear regime the tether does not extends considerably and all the applied force (i.e. the biasing potential in the current context) is primarily used to stabilize the length of membrane tether and this leads to reduced undulations in the ∗ tether length which is seen in the narrow distribution of Ltether at large values of L tether. The force required to extract a tether can be determined from the PMF as f (Ltether) = −∇LtetherW. Numerical differentiation of the PMF can lead to large errors in the estimates for the tether force and hence we use an alternate method where we utilize the scaling relations to determine f (Ltether). The tether force determined using the scaling relations are shown in the bottom panel of Fig. 5 and the constant force in the final regime is taken to be the tether force that can be compared to that obtained in tether pulling experiments. Estimates for force and the radius of the membrane tether computed using the continuum in silico assay described here are in excellent agreement with those reported in the literature for cells with similar mechanical properties, and this assay has been used by Ramakrishnan et al. to determine 10 Figure 3. (left panel) A snapshot of a tether extracted from a patch of a planar membrane. (right panel) A closer view of the tubular region of length Ltether and radius Rtether with the colors denoting the mean curvature of the surface -- the tubular region has a positive mean curvature and the neck region has a negative mean curvature. The tip and base regions on the membrane tether are represented by the marked vertices on the tubular membrane along with the position of their center of mass. the excess area in membrane regions between cytoskeletal pinning points. VII. BRIDGING TECHNIQUES USING WHAM AND TI The PMF of a system determined using WHAM analysis is accurate only upto an additive constant, as shown in eqn. (19), and hence can only be used to determine the relative energy differences between the various states. Furthermore, the umbrella sampling technique and WHAM also suffer from the problem of end-point-catastrophe, which is a well known phenomenon in alchemical energy methods commonly used to study the free energy landscapes of biomolecular systems. In such systems, the free energy estimates close to the end points of the order parameter are erroneous due to the numerical instabilities arising from the divergence of the interaction potential -- a case point being the free energy calculation involving two atoms, interacting via a Lennard-Jones potential, with their separation r → 0. WHAM also suffers from such shortcomings at the end points albeit for a different reason. An end point is defined as the value of the order parameter beyond which the potential of mean force vanishes since the various underlying interaction potentials vanish. In this regime, where the strength of the PMF is very weak (compared to the strength of the fluctuations), the signal to noise ratio is very small and hence conventional sampling techniques do not yield the correct probability distribution for the order parameter in the windows close to the end point. As a result the PMF computed using WHAM does not accurately capture the energy landscape of the system close to the end point. The problem of end-point-catastrophe can be overcome either by generating infinitely long samples or by performing the simulations using a stronger biasing potential with much smaller window sizes and both these methods leads to large computational costs. A similar method has been discussed in the context of solvation free energies by Souaille and Roux [42]. In certain class of problems, the thermodynamic integration technique can be used to overcome the end- 11 PLANAR REGIONTUBULAR REGIONBIASED VERTICESCENTER OF MASSTIPBASEFigure 4. (top panel) Snapshots of a membrane subject to a biasing potential, along with the positions of the center of masses, for five different values of the preferred tether length L∗ tether = 4, 32, 64, 96, and 128 nm. (bottom panel) Pn(Ltether), the probability distribution of the tether length Ltether in 32 different biasing windows with a window size of 4 nm. The curve corresponding to the snapshots in the top panel are shown as solid colored lines along with the corresponding values of the preferred tether length. point-catastrophe mentioned above and also fix the absolute energy levels for the PMF. We demonstrate the idea of bridging the free energies computed using WHAM and TI using the case of a functionalized nanocarrier interacting with receptor molecules expressed on the surface of a flat membrane. We follow the model previously described by Liu et al [43, 44] to construct the free energy landscape for the interaction of an anti-ICAM functionalized nanocarrier with a membrane surface expressing ICAM receptors. In brief, the nanocarrier is a sphere of radius 50 nm, discretized into 162 vertices, and is functionalized with antibodies, that are represented as radial vectors of length 15 nm. The membrane is represented as a planar substrate on which the receptors molecules are modeled as cylinders of length 19 nm and radius 1.5 nm. The interaction between the tip of an antibody and the tip of a surface receptor is modeled as a Bell bond with a potential: G(dij) = G0 + k 2 d2 ij. (24) G0 is the activation energy gained by the system when a bond is formed, k is the strength of receptor- antibody bond, dij is the distance between the tips of the antibody and receptor molecules. d0 denotes the maximum extension of the antibody-receptor bonds and the bonds break when dij > d0. Typical values for the anti ICAM-ICAM interactions have been shown to be G0 = −19.4kBT , k = 1N/m, d0 ∼ 0.4 nm, and the receptor density being 2000 ICAM/µm2 [43]. A functionalized nanocarrier can form multiple simultaneous antibody-receptor bonds and the degree of bonding varies with position of the nanocarrier (RN C) with respect to the base of the planar membrane (RM ) and hence the relative distance RN C − RM is a suitable choice for the reaction coordinate along the potential of mean force is to be evaluated. For the system considered here, it should be noted that the nanocarrier cannot form bonds when RN C − RM >84.4 nm[45] and hence the potential of mean force is zero beyond this value of the reaction coordinate. Hence we take the region around RN C − RM =84.4 nm to represent the regime corresponding to the end point catastrophe. The conformations of a nanocarrier with one, two, and three simultaneous bonds obtained at different values of the reaction coordinates are shown in Fig. 6(a) (the unbound antibodies and receptors are not shown for clarity). Wbond, the PMF computed using the umbrella sampling/WHAM techniques, through 12 L⇤tether=4nmL⇤tether=32nmL⇤tether=64nmL⇤tether=96nmL⇤tether=128nm010203040506070Ltether0.000.050.100.150.200.25Pn(Ltether)L⇤tether=4nmL⇤tether=32nmL⇤tether=64nmL⇤tether=96nmL⇤tether=128nm(top panel) Potential of mean force (PMF) Wtether, in units of kBT , as a function of the tether length Figure 5. Ltether. The PMF shows three distinct scaling regimes -- an initial linear regime followed by a quadratic regime which crosses over to a final linear regime -- and the scaling relations are also shown alongside. (bottom panel) The force Ftether, in units of pN, required to extract the tether. the application of a biasing potential on the reaction coordinate, is shown in Fig. 6(b) for four different ensembles of the nanocarrier-membrane system. It can be seen that the potential of mean force obtained from different ensembles are shifted upto an arbitrary constant but the relative energy levels between the various states remain nearly constant, which is reflective of the first shortcoming discussed at the introduction. A closer look at the region between 84.0 and 84.4 nm reveals that the harmonic potential, characteristic of a single antibody-receptor bond with a well depth of 19kBT , is only partially captured in the WHAM analysis and this is a clear signature of the end-point catastrophe. On the other hand, as shown in Fig. 6(c), the free energy difference close to the end point region com- puted using TI coupled with WHAM precisely captures both the non-bonded region (RN C − RM >84.4 nm) and also the depth of potential well corresponding to a single bond. These results were obtained using short TI calculations performed in the end point region with an window interval of 0.1 nm. Since the bonded and unbonded states are clearly defined the relative free energy computed using TI can be combined with PMF computed using WHAM to fix the end-point-catastrophe and also represent the † PMF in an absolute scale. The absolute potential of mean force (W bond) obtained by shifting Wbond with respect to ∆Fbond through a linear regression fit is shown in Fig. 6(d). The absolute PMF obtained by combining WHAM and TI is in excellent agreement with the reported values of the PMF obtained by grafting the end points using an analytic function [43]. The bridging technique presented here is very generic and can be applied to even more complex scenarios where the exact form of the analytic function as a function of the order parameter is not known. VIII. CONCLUSIONS The analysis of the thermodynamic free energy landscape can yield unprecedented levels of insight into the behavior of complex systems. However, it is a challenge in the study of these systems to formulate computational methods to delineate their free energy landscape using physically relevant order 13 050100150200250300W(Ltether)(inkBT)∝Ltether∝L2tether∝LtethermeanPMF−10010203040506070Ltether(innm)−10010203040f(Ltether)(inpN)supressundulationstetherformationstabilizationoftethermeantetherforceFigure 6. (a) Snapshots of a functionalized nanocarrier forming one, two, and three simultaneous bonds with the receptor molecules expressed on the membrane surface. RN C and RM denote the center of mass position of the nanocarrier and membrane respectively, (b) The potential of mean force Wbond as a function of RN C -RM for four different ensembles computed using umbrella sampling and WHAM, (c) the relative free energy difference ∆Fbond between a nanocarrier with zero and one antigen-antibody bonds, and (d) the absolute free energy of the system obtained by combining the PMF obtained using WHAM the relative free energy obtained using TI. parameters. The primary focus of this article is to demonstrate how conventional free energy methods can be adopted to in problems related to morphological transitions in cell membrane. Quantitative predictions based on the relative free energies obtained using these simple but elegant methods reproduce many of the emergent behaviors observed in experiments. Since these methods provide a powerful framework to interpret experimental findings it is essential to develop free energy based models and methods which help in understanding the system at a more fundamental level. ACKNOWLEDGMENTS This work was supported in part by the US National Science Foundation Grants DMR-1120901, and CBET-1244507. The research leading to these results has received funding from the European Commis- sion Grant FP7-ICT-2011-9-600841, US NIH U01-EB016027, and NIH 1U54CA193417. Computational resources were provided in part by the National Partnership for Advanced Computational Infrastructure under Grant No. MCB060006 from XSEDE. [1] J. N. Israelachvili, Intermolecular and Surface Forces, third edition ed. (Academic Press, Boston, 2011). [2] P. V. Escrib´a, J. M. Gonz´alez-Ros, F. M. Goni, P. K. J. Kinnunen, L. Vigh, L. S´anchez-Magraner, A. M. Fern´andez, X. Busquets, I. Horv´ath, and G. Barcel´o-Coblijn, J Cellular Mol Med 12, 829 (2008). [3] S. J. Singer and G. L. Nicolson, Science 175, 720 (1972). [4] M. Edidin, Nat. Rev. Mol. Cell Biol. 4, 414 (2003). [5] D. M. Engelman, Nature Cell Biology 438, 578 (2005). [6] S. D. Conner and S. L. Schmid, Nature 422, 37 (2003). [7] G. J. Doherty and H. T. McMahon, Annu. Rev. Biochem. 78, 857 (2009). [8] H. Ewers and A. Helenius, Cold Spring Harbor Perspectives in Biology 3, a004721 (2011). [9] I. Canton and G. Battaglia, Chem. Soc. Rev. 41, 2718 (2012). 14 (a)RNCRMmonovalentdivalenttrivalent7678808284RNC−RM−15−10−505101520WbondestimatefromWHAM(b)meanPMF83.083.584.084.585.0RNC−RM−15−10−505∆FbondestimatefromTI(c)7678808284RNC−RM−40−35−30−25−20−15−10−505W†bondabsolutePMF(d)shiftedmeanPMF[10] B. N. Kholodenko, Nature 7, 165 (2006). [11] A. Sorkin and M. von Zastrow, Nat. Rev. Mol. Cell Biol. 10, 609 (2009). [12] M. P. Sheetz, Nat. Rev. Mol. Cell Biol. 2, 392 (2001). [13] R. Ananthakrishnan and A. Ehrlicher, Int. J. Biol. Sci. 3, 303 (2007). [14] K. Keren, Eur Biophys J 40, 1013 (2011). [15] P. M. Chaikin and T. C. Lubensky, Principles of Condensed Matter Physics (Cambridge University Press, 2000). [16] D. Frenkel and B. Smit, Understanding Molecular Simulation : From Algorithms to Applications, 2nd ed. (Academic Press, 2001). [17] U. Seifert, Advances in Physics 46, 13 (1997). [18] D. P. Tieleman, S.-J. Marrink, and H. J. Berendsen, Biochim Biophys Acta (BBA) -Reviews on Biomem- branes 1331, 235 (1997). [19] M. Venturoli, M. Maddalena Sperotto, M. Kranenburg, and B. Smit, Physics Reports 437, 1 (2006). [20] G. S. Ayton and G. A. Voth, Seminars in Cell and Developmental Biology 21, 357 (2010). [21] W. Shinoda, R. DeVane, and M. L. Klein, Current Opinion in Structural Biology 22, 175 (2012). [22] R. P. Bradley and R. Radhakrishnan, Polymers 5, 890 (2013). [23] N. Ramakrishnan, P. B. Sunil Kumar, and R. Radhakrishnan, Physics Reports 543, 1 (2014). [24] M. Deserno, Chemistry and Physics of Lipids (2014), 10.1016/j.chemphyslip.2014.05.001. [25] P. B. Canham, J. Theor. Biol. 26, 61 (1970). [26] W. Helfrich, Z. Naturforsch. C 28, 693 (1973). [27] M. P. do Carmo, Differential geometry of curves and surfaces (Prentice Hall, Engelwood Cliffs, New Jersey, 1976). [28] Z. Shi and T. Baumgart, Nat Comms 6, 5974 (2015). [29] A. Diz-Munoz, D. A. Fletcher, and O. D. Weiner, Trends in Cell Biology 23, 47 (2013) . [30] D. R. Nelson and T. Piran, Statistical Mechanics of Membranes and Surfaces (World Scientific, 2004) . [31] N. Ramakrishnan, P. B. Sunil Kumar, and J. H. Ipsen, Phys. Rev. E 81, 041922 (2010). [32] N. J. Agrawal, J. Nukpezah, and R. Radhakrishnan, PLoS Comput Biol 6, e1000926 (2010). [33] V. Ramanan, N. J. Agrawal, J. Liu, S. Engles, R. Toy, and R. Radhakrishnan, Integr. Biol. 3, 803 (2011). [34] J. Liu, R. W. Tourdot, V. Ramanan, N. J. Agrawal, and R. Radhakrishanan, Molecular Physics 110, 1127 (2012). [35] R. W. Tourdot, N. Ramakrishnan, and R. Radhakrishnan, Phys. Rev. E 90, 022717 (2014). [36] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, J. Chem. Phys. 21, 1087 (1953). [37] B. Widom, J. Chem. Phys. 39, 2808 (1963). [38] C. H. Bennett, Journal of Computational Physics 22, 245 (1976). [39] B. Roux, Computer Physics Communications 91, 275 (1995). [40] R. W. Tourdot, R. P. Bradley, N. Ramakrishnan, and R. Radhakrishnan, IET Systems Biology 8, 198 (2014). [41] N. Ramakrishnan, D. M. Eckmann, P. S. Ayyaswamy, V. M. Weaver, and R. Radhakrishnan, Under Review. [42] M. Souaille and B. Roux, Computer Physics Communications 135, 40 (2001). [43] J. Liu, G. E. Weller, B. Zern, P. S. Ayyaswamy, D. M. Eckmann, V. R. Muzykantov, and R. Radhakrishnan, Proc. Natl. Acad. Sci. U.S.A. 107, 16530 (2010). [44] J. Liu, N. J. Agrawal, A. Calderon, P. S. Ayyaswamy, D. M. Eckmann, and R. Radhakrishnan, Biophys. J. 101, 319 (2011). [45] Calculated as (nanocarrier radius + length of the antibody + length of the receptor + d0). 15

Subsets and Splits