id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
65fb2c4ce9ebbb4db92085fd | 15 | where a i -coefficients of the linear model. Such loss function forces some of MD to be almost zero, and MD with coefficients more than 0.1 are selected (all MD were scaled to the range [0; 1] when this threshold value is applied). We use the implementation of L1-regularized regression from the Smile package (version 2.6.0). |
65fb2c4ce9ebbb4db92085fd | 16 | The Boruta algorithm is based on other algorithms that can provide importance scores of features. In addition to real features, the same number of "fake" features is added. Fake features are made from real ones by random shuffle of rows. A feature is considered "important" if its importance score is better than the best "fake" feature. The final importance is the number of repeats in which this feature was considered as important. Feature importance scores provided by random forest are based on the decrease of the impurity measure when the corresponding variable is used. For the Boruta algorithm , we use our own implementation of the algorithm. We use implementations of random forest from the Smile package (version 2.6.0) with default hyperparameters for initial importance score of features. 80 rounds of the Boruta algorithm were used. |
65fb2c4ce9ebbb4db92085fd | 17 | For the GA method, we use the implementation of the genetic algorithm from the Scikit-learn python package (sklearn-genetic, version 0.6.0). The GeneticSelectionCV function with cv = 5 is used, OLS linear regression is used as a regression estimator, and the coefficient of determination (R 2 ) is used as an error measure. For the PLS_VIP method , we use the implementation of PLS from the Scikitlearn python package (version 1.4.0). In sequential removing (SEQ_REM), at each stage of the algorithm, the random forest model is built and 10 MD with the least values of importance are removed until the required number of MD are remained. In the last step, less than 10 MD are removed if the total number of MD to be removed cannot be divided by 10. |
65fb2c4ce9ebbb4db92085fd | 18 | For SEQ_REM, we also consider MD preselection when MD with a Pearson correlation coefficient r > 0.8 are removed. Before training the models, all MD were scaled to the range [0; 1] and the RI values were divided by 1000. This is necessary to avoid too large coefficients and incorrect operation of the framework. The coefficient values given in this work are given without taking into account scaling (for actual values), unless otherwise indicated. The above-mentioned Smile package is used for OLS and the building of final linear equations. |
65fb2c4ce9ebbb4db92085fd | 19 | In order to characterize the accuracy, the mean absolute error (MAE), median absolute error (MdAE), root mean square error (RMSE) were computed, 10-fold cross-validation is used, unless otherwise specified. For the comparison purpose, the "black box" models created using previously published software are considered. A detailed description of this software and machine learning models is given in our previous work . To evaluate the accuracy of the models, we also used 10fold cross-validation (the "CV" command line option of the above-mentioned software). |
65fb2c4ce9ebbb4db92085fd | 20 | In previous works, QSRR were often used not for practical prediction of RI, but for characterization of SP in order to draw some conclusions about the nature of retention mechanisms (see Fig. ). However, it is not usually tested whether the selection of MD is reproducible with small changes to the data set. If the conclusions characterize the SP in general, then these conclusions should not be altered if the data set (that plays the role of a probe) is slightly changed: for example, several molecules are removed. Thus, the procedure of MD selection was repeated many times and each time Also, the accuracy of the model (and of the method in general) depends on the set of selected features. Typically, in QSRR works, a set of MD is selected only once and then the accuracy of the model is carefully investigated (see Fig. ) using cross-validation or one-leave-out approaches. |
65fb2c4ce9ebbb4db92085fd | 21 | In this work, we apply the modified procedure and after each alteration of the training set we repeat the MD selection (see Fig. ). This approach allows for the evaluation of accuracy of the approach in general, rather than the accuracy of one randomly selected MD set. The evaluation of the reproducibility is made for all three IL, but the detailed results are shown only for Bis4MPyC6, unless otherwise specified. The corresponding data set contains 123 compounds of various classes. All conclusions regarding the comparison of the MD selection methods in terms of reproducibility and accuracy are the same for any of the considered IL. |
65fb2c4ce9ebbb4db92085fd | 22 | After removing a given number of molecules and before the MD selection procedure, a preliminary reduction of the MD set is done. The MD that were constant for all molecules or that coincided with other MD up to a linear dependence were removed. The resulting MD set (before the selection procedure) contained 110 -120 MD (the exact number depends on the exact data set and changes with random removing of molecules). |
65fb2c4ce9ebbb4db92085fd | 23 | Retention data (n-alkane RI and RT) were acquired for 178 compounds (at least for one of the columns with IL) for three columns with IL-based SP. These compounds include 108 aromatic and 70 aliphatic compounds. Among these molecules, 37 are ethers, 49 are phenols, 13 are aldehydes or ketones, and 129 molecules have a hydroxyl group attached to an aliphatic atom. All considered molecules contain carbon, hydrogen, and oxygen. Some of these compounds contain other elements: |
65fb2c4ce9ebbb4db92085fd | 24 | During the acquisition of data for the final data sets, all data for each column were measured within 5 days using an autosampler. RT were extracted from chromatograms using GCMSsolution GCMS Postrun Analysis (version 4.50) software. The RT was recorded at the top of the peak. The fraction of compounds (randomly selected) was remeasured after ~15 days after the end of the first acquisition in order to estimate the reproducibility and measurement error. The mean deviations (the average of deviations for multiple compounds) between the results of the first and later measurements are 0.066, 0.026, and 0.030 min. for the Bis4MPyC6, Bis2MPyC9, and Hex4MPy columns, respectively. The mean percentage deviations are 0.53%, 0.44%, and 0.67% for these three columns, respectively. The Bis4MPyC6 column is the longest and the most polar, so the absolute values of the RT are the largest for this column. Due to this, the absolute mean deviation is the largest, while the relative deviation is not. |
65fb2c4ce9ebbb4db92085fd | 25 | No significant dependence was observed: for 5 compounds, the RT was the same for each of the compounds for an injected volume within the range 0.1 -1 µl. The deviation for successive measurements was not more than 0.01 min. In addition, there is almost no difference whether there was one or multiple different compounds in the solution. Errors of the RI measurement are less than 10 units for almost all compounds. Compounds with RI more than 3500 were not included due to a possible high error in the RI estimation: in this area, peaks of n-alkanes tend to be broad and located closely to each other. The use of RI systems other than those based on n-alkanes can be the scope of further research. The data set containing RT and RI can be downloaded from the Figshare repository . |
65fb2c4ce9ebbb4db92085fd | 26 | The first method of the feature selection employed in this work was "greedy" stepwise selection (SEQ_ADD). MD were added one by one. At each step, the MD that allows achieving the most The prediction accuracy of this approach is demonstrated in Table . Confidence intervals of the accuracy measures are also shown. Standard deviations for various error measures are in the range of 9 -11 units. Such relatively large values of standard deviation show that the comparison of accuracy of prediction methods must be done very carefully, the accuracy varies with random modification of the data set. However, in many works it was done based only on one cross-validation experiment . |
65fb2c4ce9ebbb4db92085fd | 27 | The more molecules we remove from the data set each time we train, the less reproducible the set of selected MD is. Such a dependence is shown in Fig. . The dependence of the average number of the MD selected in both experiments for all pairs of experiments is shown. Even if we remove only one molecule, the MD selection will not be reproducible. In addition, a typical dependence of accuracy on the number of MD is shown in Fig. . The average of 200 repeats is shown, confidence intervals are too narrow to be shown. It can be seen that the prediction error (as expected) decreases with increasing number of MD, but with a further increase in the number it decreases very slowly. Based on this, we decided to select 10 MD. Fig. . A -dependence of the average size of the intersection between sets of selected MD for all pairs of repeats with an altered data set on the number of molecules randomly excluded from the data set during each repeat (1 -without preliminary removal of highly correlated MD; 2 -with preliminary removal of MD with the Pearson correlation coefficient r > 0.8 with any other MD); B -dependence of cross-validation accuracy on the number of MD. |
65fb2c4ce9ebbb4db92085fd | 28 | In Fig. , a heatmap given that shows the Pearson correlation coefficient r between some MD for this data set. It can be seen that MD, which at first glance are almost unrelated to each other, are often correlated. For example, the number of methoxy groups (fr_methoxy) and a topological MD characterizing the contribution of polar atoms to the total surface of the molecule (VSA_EState9) are strongly correlated. For each pair of correlated MD, we can arbitrarily remove one of the two. But if we do this, then our qualitative conclusions based on the set of MD may also change depending on which of them we remove. However, we considered such a reduction of the MD set. We removed from the original set all MD having r > 0.8 with any of the others. There were 49.9 ± 2.0 MD left (confidence interval, p = 0.95, N = 200). With this approach to preselection of the MD, we conducted the same experiments in order to evaluate the reproducibility. excluded each time). Reproducibility was slightly improved, as expected, but this approach includes a virtually random removal of half of the MD (from a pair of correlated ones, we randomly choose which one to remove). We do this in a reproducible way (for the same pair of correlated MD, the same MD is removed each time). However, the reproducibility is still not very good and it is clear that as the number of molecules removed from the data set increases, the reproducibility also decreases. Thus, the problem of the selected MD set not being reproducible across the data set changes cannot be explained solely by the presence of highly correlated MD. |
65fb2c4ce9ebbb4db92085fd | 29 | Thus, we can draw the following conclusion. The stepwise algorithm for selecting MD for linear regression is not reproducible when small changes in the data set are made, and no "physicochemical" conclusions can be drawn from the set of once selected MD. Unfortunately, a number of previous works made such conclusions. |
65fb2c4ce9ebbb4db92085fd | 30 | The LASSO regression (L1-regularized linear regression) is an accurate linear regression and (simultaneously) a MD selection method. When a weighted sum of absolute values of coefficients is added to the loss function, the minimization of the loss function leads to zeroing of most of coefficients. We consider a MD to be selected if its coefficient (for scaled to the [0; 1] range value of the MD) is positive and above the threshold value of 0.1. In Fig. , the dependence of the accuracy and number of selected MD on l 1 constant is shown. Smaller values of l 1 result into better accuracy and larger number of important MD. At values l 1 < 1.0, the accuracy decreases very slowly with decrease of l 1 , and l 1 = 1.0 was used for further investigation. It can be clearly seen that the following MD: LabuteASA, BCUT2D_MRHI are the most important (for Bis4MPyC6) and play the largest role. The 5 most important MD and their order can be established reliable. The average number of MD selected in both runs for all pairs of repeats is ~16.0. |
65fb2c4ce9ebbb4db92085fd | 31 | The next considered algorithm of the MD selection is the Boruta algorithm . The reproducibility of the MD selection is quite high in this case. We performed 200 repeats with random exclusions of 25 molecules from the data set and each time we performed 80 rounds of the Boruta algorithm. For Bis4MPyC6, there are 11 MD (SMR_VSA7, LabuteASA, ExactMolWt, BCUT2D_MRHI, FractionCSP3, VSA_EState6, HeavyAtomMolWt, BertzCT, SlogP_VSA6, Ipc, fr_benzene) that are considered as important in all repeats and in all rounds of the Boruta algorithm. |
65fb2c4ce9ebbb4db92085fd | 32 | The average (over 200 repeats) number of rounds of the Boruta algorithm, in each of which the MD is considered as important, is shown. The confidence intervals (N = 200, p = 0.95) are shown. However, the accuracy of the OLS linear regression built using MD selected using the Boruta algorithm is very low (see Table ), much worse than with step-by-step selection. Consequently, this algorithm is not suitable for constructing linear QSRR, although it allows evaluating the importance of the MD in reproducible way. |
65fb2c4ce9ebbb4db92085fd | 33 | The genetic algorithm, as well as the stepwise algorithm, selects MD based on the accuracy of the OLS regression built on this set of MD, while the PLS-VIP and Boruta algorithms select according to criteria that have nothing to do with the accuracy of the OLS regression. Therefore, just as in the case of a step-by-step algorithm, one can expect that the accuracy of the OLS regression built on these MD will be quite high. Indeed, Table shows that the genetic algorithm allows obtaining relatively accurate linear equations. If we compare random pairs of MD sets obtained in different runs, then on average only ~2.2 out of 10 MD are the same for GA. The accuracy of final linear equations for GA is close to that for SEQ_ADD, while the reproducibility of MD selection is significantly worse compared with SEQ_ADD, LASSO and BORUTA methods at least with the used number of generations. Same as for the SEQ_ADD algorithm, the probability to be selected in each repeat is shown in Fig. . |
65fb2c4ce9ebbb4db92085fd | 34 | The last two considered algorithms were PLS_VIP and SEQ_REM. The average importance scores estimated using these methods are shown in Fig. and Fig. , respectively. In case of SEQ_REM, the importance score is estimated using a random forest method based on the decrease of the impurity measure when the corresponding variable is used. As well as the Boruta method, both these methods are reproducible in order to estimate importance of MD but cannot be used for the MD selection for OLS linear regression. The accuracy of PLS regression itself was not investigated in this work. |
65fb2c4ce9ebbb4db92085fd | 35 | Different algorithms select different MD sets. We compared MD sets selected by BORUTA, SEQ_ADD and LASSO methods. In case of SEQ_ADD and LASSO we selected 10 most important MD, in case of BORUTA we selected 11 MD, since each of them is considered as important at all iterations of the Boruta algorithm. The Venn diagram for the resulting MD sets is shown in Fig. . |
65fb2c4ce9ebbb4db92085fd | 36 | Only one MD (BCUT2D_MRHI) is considered as important in all cases for Bis4MPyC6. A total of 4 MD (BCUT2D_MRHI, fr_benzene, LabuteASA, VSA_EState6) are considered as important by at least 2 algorithms simultaneously. The fact that different methods select different sets of MD, and the results of each algorithm are not completely reproducible with minor changes in the data set, shows that in order to draw any "physicochemical" conclusions from such a study, it is necessary to carefully consider issues of reproducibility. Otherwise, one can draw conclusions based on a random (statistically insignificant) result. |
65fb2c4ce9ebbb4db92085fd | 37 | The data sets acquired for three SP are slightly different, because some compounds are not eluted at reasonable temperatures on some SP, compounds with RI > 3500 are not included in the considered data set (low accuracy of the RI determination in these cases) and due to other reasons. The information about the overlap of these data sets is provided in Supplementary material, Fig. . In the previous sections, we demonstrated that even a small difference in a data set severely affects the set of selected MD and their importance. Therefore, the comparison of the sets of selected MD was performed using the "intersection" data set. For each compound from this data set, RI is available for all three IL, as well as for 5%-phenyl-methylpolysiloxane (also denoted as DB-5 for conciseness, after the common column with such SP) and for polyethylene glycol . It should be noted that such incorrect comparison was made in previous works. For example, in Ref. (Journal of Chromatography A), the authors built QSRR for 4 SP having significantly different data sets (different in dozens of compounds). The authors selected MD using a sequential algorithm and commented on the chemical nature of the separation based on the selected MD set. |
65fb2c4ce9ebbb4db92085fd | 38 | Three IL (Bis4MPyC6, Bis2MPyC9, Hex4MPy) and two polymeric SP: polyethylene glycol and 5%phenyl-methylpolysiloxane were considered. MD selected for Bis4MPyC6 and Bis2MPyC9 are very similar to each other. This is consistent with the fact that these IL are very close in their chemical nature. However, Bis4MPyC6 is more polar, consequently the RI values are higher (the data set consisted of polar molecules) and the absolute values of the coefficients in the LASSO regression before MD are higher. Thus, the PEOE_VSA7 descriptor characterizes the accessible surface of atoms which Gasteiger charge is in the range [-0.05; 0]. Such charges typically have aromatic carbon and other atoms in moderately polar groups, while aliphatic carbons are hidden by positive-charged hydrogens. The BCUT2D_LOGPLOW is the lowest eigenvalue of a matrix which diagonal elements contain contributions of atoms to LogP (factor of lipophilicity) and non-diagonal elements contain information about the connectivity between the corresponding atoms. Both MD are topological and related to the polarity of the molecule, and the average coefficients before them increase with the polarity of the molecule. It should be noted that the most influential according to different MD selection methods fr_benzene (the number of benzene rings) and PEOE_VSA7 are not strongly correlated: the Pearson correlation coefficient is ~0.5 for the considered data set. The topological chi1 descriptor is higher for linear molecules and lower for branched ones and characterizes the shape of the molecule. in structure, and the set of selected MD also significantly differs. Thus, despite all the above notes that the MD selection is not reproducible when the data set is changed, it is possible to compare SP using QSRR. Polymeric SP are even more different compared with IL-based SP. For Bis4MPyC6 and Bis2MPyC9, the fr_benzene descriptor was selected with a very high probability by the SEQ_ADD method, it is the MD that is the most correlated with RI. For Hex4MPy, it is much less significant according to the same method. The tendency continues with less polar PEG. As for siloxane, it is absent in the top 10. As expected, the polar and aromatic Bis4MPyC6 and Bis2MPyC9 are the most sensitive to aromatic systems. |
65fb2c4ce9ebbb4db92085fd | 39 | The difference between the results obtained with different MD selection and MD importance estimation methods is much greater than the difference between SP. Fig. -7F show the Venn diagrams for sets of MD selected using the SEQ_ADD, BORUTA, and LASSO methods. In the case of SEQ_ADD and LASSO we selected 10 the most important MD, in the case of BORUTA we selected more than 10 MD, because each of them is considered as an important at all iterations of the Boruta algorithm. |
65fb2c4ce9ebbb4db92085fd | 40 | Finally, we made the same comparison using the full versions of the data sets. The results are shown in Supplementary material, Fig. . It can be clearly seen that the difference between different versions of the data sets is much greater than between different SP using the same data sets. Thus, it can be concluded that QSRR-based comparisons of SP should be made using exactly the same data sets and should be made very carefully. Generally speaking, our results do not confirm the claims that the QSRR with a diverse set of MD (including topological ones) is a truly informative method that allows characterizing SP. In many cases (for example, in the work ) the reproducibility is not checked, the data set is not equal for different SP and it can easily result in misleading conclusions. |
65fb2c4ce9ebbb4db92085fd | 41 | Table shows examples of linear QSRR equations for RI prediction. It can be clearly seen that the use of RI_PEG_LM and RI_PEG_DL descriptors improves the accuracy. The improvement of the accuracy is highest for Hex4MPy and Bis2MPyC9 and lowest for Bis4MPyC6. This is consistent with the fact that Bis4MPyC6 is the most polar and the most different from PEG. These MD are the most significant or are among the most significant for all three IL-based SP and for all MD selection methods. |
65fb2c4ce9ebbb4db92085fd | 42 | It should be noted that neither RI_PEG_LM nor RI_PEG_DL is enough to predict RI on ILbased SP alone without the use of other MD. It means that the selectivity and the retention mechanism for IL-based SP is considerably different from such on polyethylene glycol. Fig. shows the dependence of prediction accuracy on the number of MD when RI_PEG_LM or RI_PEG_DL is used (SEQ_ADD MD selection method, 200 repeats). For both MD in all repeats, these MD are always selected in the first iteration. It is clearly seen that the use of these MD alone does not allow achieving reasonable accuracy and it works well together with other MD. |
65fb2c4ce9ebbb4db92085fd | 43 | Supplementary material, section S2 shows the linear model that was used in order to calculate the RI_PEG_LM descriptor in explicit form. It should also be noted that when this model was trained, the training set did not contain the molecules that are contained in the data sets for IL-based SP. This way we ensured that there was no "data leak" and the molecules used for testing were not seen by the model at any stage of training. |
65fb2c4ce9ebbb4db92085fd | 44 | A complex two-stage method was recently developed that allows using all benefits of deep learning for RI prediction using training sets with ~100-200 compounds. The idea of this software is similar to the considered above: deep learning models predict, using a molecule structure, RI for multiple common SP (siloxanes, polyethylene glycol), and then these predicted RI are used as input features for a new model for the given SP and data set. Together with these features (RI for a set of SP), other MD are also used. The difference with the approach considered in this work is that this set of features is fed to a linear support vector regression model (with a non-linear kernel) with predefined parameters without any MD preselection. This software (we call it SVEKLA) allows creating a machine learning model for any SP easily, but these models are not interpretable and use an excessive set of features. RI_PEG_LM is calculated using a linear model. The use of this MD is the most simple and interpretable way to accurately predict RI for IL. A graphical comparison of the accuracy of several different approaches is shown in Fig. . |
65fb2c4ce9ebbb4db92085fd | 45 | We have developed the interactive software for QSRR studies in GC and called it CHERESHNYA, the example of a screenshot is shown in Fig. . This software allows the interactive MD generation (2D MD supported by RDKit and CDK packages), MD selection, building of linear (OLS) models for QSRR in GC. The newly developed RI_PEG_LM and RI_PEG_DL descriptors are also supported, as well as similar MD for polydimethylsiloxane, 5%-phenyl-methylpolysiloxane, 94%dimethyl-6%-cyanopropyl-phenyl-polylsiloxane. All MD selection methods listed in Table and described in section 2.4.2 are implemented in this software. The software is written in the Java programming language, Smile framework is used. PLS-VIP and GA methods are implemented using Scikit-learn package. The molecular editor JSME is integrated into the software for interactive MD computation and RI prediction. The figures (heatmaps, bar plots) shown in this article are generated using this software. The reproducibility study can be automatically provided using it. |
65fb2c4ce9ebbb4db92085fd | 46 | Methods for selecting descriptors for constructing linear quantitative structure-retention relationships are not reproducible with respect to changes in the data set. Different selection methods give different results. Conclusions about the retention mechanism and comparison of stationary phases based on such quantitative relationships must be made with extreme caution. Some previous works did not carry out any checks on the reproducibility of the selection of descriptors, but qualitative conclusions were drawn from the fact which descriptors were selected. Such conclusions are unreliable and should be avoided. |
65fb2c4ce9ebbb4db92085fd | 47 | The selectivity of the considered stationary phases significantly differs from the selectivity of polyethylene glycol. The retention on ionic liquids cannot be directly computed using only the retention index on polyethylene glycol. However, the retention index on polyethylene glycol predicted using a machine learning model (trained on other, non-overlapping data) is a very good descriptor for predicting retention indices on ionic liquids. Sufficiently accurate linear models for retention index prediction were developed for these stationary phases. |
62b17d89b7bbed765689cf80 | 0 | Strength and toughness are mutually exclusive properties of materials in general and ceramics in particular. Designs found in biological structural materials such as nacre may provide useful guidelines to alleviate this trade-off . Nacre-like alumina, made of brittle constituents only, combines the hardness and strength of standard alumina materials, and comparatively high toughness. This toughness (damage resistance) is achieved mostly by the deviation of cracks along the platelets interface in their brick-and-mortar microstructure . While the deviation of cracks in brick-and-mortar microstructures is well documented, several other toughening mechanisms can exist in biological or bio-inspired materials . Discrete elements modeling (DEM) has been implemented to understand the toughening mechanisms of nacre-like alumina and provide guidelines of microstructural features to implement to improve the damage resistance of these composites . Establishing quantitative links between nano-to microscopic structures and bulk mechanical properties determined in standardized fracture tests requires additional mechanical modeling at various scales and methods that enable imaging the effects of microstructures on the mechanical behavior. For instance, structural defects (such as local misalignments, variable platelet size) of real microstructures spanning few tens of micrometers can be modeled by DEM, which suggests that they may have a significant effect on crack propagation . To validate these results, we developed in situ spectroscopic and scanning electron microscopy (SEM) imaging of stress field and microstructure during a bending test on nacre-like alumina. |
62b17d89b7bbed765689cf80 | 1 | Samples were prepared as standard single-edge notched beams (SENB), and mirror polished before testing . 3 × 6 × 36 mm samples were used for fluorescence spectroscopy and SEM characterization. The fluorescence of trivalent chromium (Cr 3+ ), a ubiquitous impurity in alumina-based polycrystalline ceramics, was used to map stresses in the composite . Even in very low quantities (< 0.5 wt%), Cr 3+ causes an intense fluorescence characterized by two sharp and intense lines, noted R 1 and R 2 , whose energy shifts with applied stress . |
62b17d89b7bbed765689cf80 | 2 | We used a Horiba™ LabRam HRevolution Raman spectrometer equipped with a motorized stage for mapping the fluorescence signal excited with a 532-nm solid-state diode laser from SENB submitted to a four-point bending test. The laser is focused through a microscope equipped with a long-working distance x50 Mitutoyo™ objective down to a spot of about 1µm. The high fluorescence signal allows a typical recording time τ of the order of one millisecond or less to obtain an intense enough fluorescence spectrum at each point and determine peak position and stress. X-Y maps of the sample lateral surface were obtained in the SWIFT™ mode where the sample is scanned at constant speed in the X direction (perpendicular to the notch, see Figure left) under the laser, and signalprocessed through the CCD detector to record individual points every τ. Spatial resolution is determined by the product of the scanning speed and reading time τ in the X direction, and by step-bystep motion in the Y direction (along notch). Identical steps were chosen in both X and Y directions to avoid potential distortion in the image, and set to 5 and 2 µm for low-and high-resolution images, respectively. Low-resolution images covering a 1 × 2 mm 2 field of view and high-resolution images covering 300 x 750 µm 2 were acquired in about one minute each around the crack. One image with steps of 0.5 µm over a 400 × 150 µm 2 area was acquired in 40 minutes at the end of the test (corresponding to a long crack propagation but without macroscopic failure of the sample) for comparison with post mortem imaging by SEM. The stress was maintained at a set value at each step while maps were acquired. With the optical setting and characteristics of the laser-material interaction, the actual resolution is better than 2 µm. |
62b17d89b7bbed765689cf80 | 3 | -1 is the average piezospectroscopic coefficient perpendicular to the c direction of corundum lattice . In the nacre-like microstructure the macroscopic applied tensile stress is parallel to the basal plane of most alumina platelets (or corundum structure), but the local stress field near the notch is not uniaxial. The average value of 2.5 cm -1 |
62b17d89b7bbed765689cf80 | 4 | for the piezospectroscopic coefficient was used here, which is sufficient for imaging and following the spatial and temporal changes in stress. Stress fluctuations in homogeneous areas yield an estimated uncertainty on relative values of stress of 40 MPa from pixel to pixel, and less than 15 MPa for 3 × 3 pixel squares, i.e. over about 15 microns for low resolution images, and 6 or 3 microns for higher resolution images. |
62b17d89b7bbed765689cf80 | 5 | The sample tested in situ with fluorescence spectroscopy was characterized post operando with a SUPRA 55VP SEM. Due to the elastic spring back, the crack was not open anymore and its exact trajectory in the complex microstructure was determined by comparison with the high-resolution fluorescence image. A complementary SEM in situ analysis was performed on another SENB sample. |
62b17d89b7bbed765689cf80 | 6 | Starting from an initial state with negligible stress, the evolution of the stress map calculated from large fluorescence maps (Fig. ) shows, the tensile stress field that develops around the crack tip as applied force increases (Fig. ). Tensile stresses are detected while the sample is still in the elastic domain, and well before the crack initiation. The stress field is asymmetric with the side at higher stress (left of the notch in Figure ) eventually fracturing with a crack initiating at about 70° of the notch plane for an applied force of 110 N. Before the crack can be observed at the surface of the sample, the area under tension remains constant and does not increase in size, only the value of the calculated stress increases with the loading force to exceed locally 300 MPa. While the crack extends, the tip of the uncracked ligament is subjected to important tensile stresses, with average values up to 250 MPa. The tensile nature of the stress field generated at the bottom of the notch is coherent with the SENB configuration At an applied force of 110 N, the first high-resolution image recorded (Fig. ) shows a stress accumulation pinned on a mechanical asperity along the crack (point 1). The force was increased to 114 N, and the sample was then maintained at a constant force for several hours. High-resolution maps display the evolution of the stress field, which highlight mechanical asperity effects during subcritical crack propagation. The crack first propagated towards a second mechanical asperity (point 2) |
62b17d89b7bbed765689cf80 | 7 | after the first low-resolution map was recorded at 114 N (about 10 minutes), and did not evolve with the force maintained at 114 N for at least 1.5 hours. The stress concentration zone then transferred to point 3 after 3 hours (Fig. ), due to progressive cracking of mechanical bridges that hindered crack propagation at point 2. The microstructure did not evolve when the last map was recorded after 4 hours at 114 N (Fig. ). The separation between each mechanical asperity is roughly 100 µm, from defect 0 (notch tip) to defect 3. |
62b17d89b7bbed765689cf80 | 8 | Progressively, the stress transfer along the crack is accompanied by a decrease of the tensile stress field around the crack tip, with final values around 250 MPa. Further propagation of the crack would require building a sufficient stress field at the tip of the crack to break the asperity on which it is now pinned. The sub-critical propagation is promoted by the initial high tensile stress surrounding the crack tip at point 1. When the loading force is maintained constant, and as long as the stress at the crack tip exceeds the fracture properties of the interface, the crack propagates by jumping from one defect to the other. It took about 2 hours to jump from point 2 to point 3. |
62b17d89b7bbed765689cf80 | 9 | The crack path up to its tip (point 3) can be located from a sharp contrast between the mechanically relaxed areas (yellowish tones) above the crack and those under tensile stress below it. A major crack is outlined by yellowish tones next to reddish ones, and secondary cracks branch on it and separate reddish areas from bluish ones. Between the notch and the crack tip, the points where crack pinning and stress accumulation was noted during the test (Fig. ) correspond to specific defects in the microstructure as observed from the surface. At point 1, the crack had to bypass a cluster of misaligned platelets, resulting in a diverted trajectory and stress contrast that contours both sides of the cluster. At point 2, branching is observed with a secondary crack seemingly appearing below a larger-than-average alumina platelet. At point 3, the crack is again pinned on a cluster of misoriented alumina platelets, and a very local stress concentration is observed before the experiment was stopped after 4 hours at a force of 114 N. |
62b17d89b7bbed765689cf80 | 10 | The nature of microstructural defects involved in crack deviation and branching was further characterized using in situ SEM imaging of another SENB sample under stress. Similar behaviors and features were observed with fast crack propagation steps separated by arrests. With in situ imaging, the active cracks are open and can be directly visualized (Fig. ) whereas in a posteriori characterization, cracks are closed and their exact location in the complex contrasts of the microstructure is more difficult to determine (Fig. ). Although the sample used for in situ SEM was different from the one used in fluorescence mapping, similar microstructural features are associated to crack deviation and the microstructure are likely to play a role because platelets cannot be broken along a mechanically strong direction of the platelets. Thus the crack has to deviate along an easier path, the weaker aluminosilicate interphase (Figs. and). This is consistent with the results of numerical simulations that used the observed microstructures to show cracks propagate at higher stress around bundles of misoriented platelets, thus contributing to toughening the composite . These defects likely contribute to the increasing toughness (R curve) of the nacre-like alumina. High-stress gradients are clearly associated with defects, in spite of the intrinsic limitation that the observation is performed on the surface of a 3D sample. |
62b17d89b7bbed765689cf80 | 11 | In situ fluorescence mapping of stress and SEM imaging during mechanical tests prove useful to identify local mechanisms and microstructural features that contribute to toughening in these bioinspired composites. Future improvements will focus on measurements that can be combined on the same sample and use the orientation information provided by EBSD to locally analyze the stress from the scale of the individual components (here the alumina platelets) to the scale of the aggregate (platelets bundles) and of the bulk sample. Besides using fluorescence, stress can also be characterized using Raman spectroscopy in non-luminescent materials. This approach could be applied to other advanced ceramics like zirconia, or natural ceramics like apatite and calcium carbonates . |
638a2c797b7c91020ce029df | 0 | Transition metal dichalcogenides (TMDCs), such as MoS2, are two-dimensional materials with exciting perspectives for application in (opto-)electronic devices or catalysis. Within the three known polytypes (1T, 2H, and 3R) of this material, 2H-MoS2 is of particular interest due to its semiconducting nature and thickness-dependent band gap. Apart from the established synthetic pathways toward 2H-MoS2 with defined layer thickness, including mechanical or liquid based exfoliation and chemical as well as physical vapor deposition, wet-chemical techniques affording colloidally stable TMDCs have recently emerged, which are based on principles originally developed for the synthesis of colloidal quantum dots. Some advantages of wet-chemically synthesized TMDCs are their scalability, relative ease of purification and solution-processability, which render them particularly suitable for application in LEDs or solar cells. For these applications, detailed knowledge of the electronic structure, such as the exact position of the band edges, is crucial to design efficient device stacks, in which the layers of several materials, e.g. electron-/hole-conducting and blocking layers as well as the active layer are electronically matched. Standard electrochemical investigations, such as cyclic voltammetry (CV) or differential pulse voltammetry (DPV), can provide such information, which are much easier and usually faster to perform compared to ultraviolet photoelectron spectroscopy or scanning tunnelling spectroscopy investigations. However, a short-coming of standard electrochemical techniques is the lack of differentiation between band edges and trap states, which is only afforded by combining electrochemistry with optical spectroscopy. Such spectro-electrochemistry detects changes in the absorption or luminescence of a sample upon oxidation or reduction. This way, the band edges or energy levels involved in a specific optical transition can be identified by determining the potential at which this transition bleaches, either due to electron injection into the final state (reduction) or a depletion of electrons in the initial state (oxidation). In contrast, reducing or oxidizing trap states will have a minor effect on the absorption spectrum, allowing for the distinction between band edges and trap states. |
638a2c797b7c91020ce029df | 1 | While the limited sensitivity of standard spectroelectrochemistry may be problematic for the investigation of thin films with few nanometer thickness, Hickey et al. introduced potential-modulated absorption spectroscopy (EMAS), which is sensitive enough to study monolayers of CdSe quantum dots. Using EMAS, DPV and electrochemical gating (ECG) experiments, we present here the electronic structure of colloidal 2H-MoS2 nanosheets with a thickness of 1-2 monolayers and lateral size of 20-25 nm. We demonstrate that the novel wet-chemical pathway affords 2H-MoS2 nanosheets of the same quality and unique electronic features as those previously described for 2H-MoS2 films obtained by liquid exfoliation, namely a direct bandgap, intrinsic n-doping as well as strong bandgap renormalization upon doping. We determine the exact energetic positions of the band edges involved in the direct transition as well as those constituting the indirect transition in bilayers. We find strong indications for two distinct band nesting transitions and report the energetic position of the points in the conduction band edge involved in these. While oxidation of the 2H-MoS2 is irreversible, we observe a mostly reversible reduction even for potentials far above the conduction band edge, rendering this material suitable for reductive electrocatalysis, e.g., for electrocatalytic hydrogen evolution. |
638a2c797b7c91020ce029df | 2 | Compared to standard spectroelectrochemistry, the lock-in amplifier-based EMAS technique achieves an excellent sensitivity allowing to study thin films of 4 nm thickness or less. A more in-depth description of this measurement principle can be found elsewhere. In short, a constant potential, located in the band gap at the open circuit potential (OCP), is applied to the transparent working electrode coated with the sample. This constant potential is modulated with periodic rectangular square pulses, such that a periodic redox potential is applied to the sample given by the amplitude of the square pulse. Simultaneously, the transmission of monochromatic, To facilitate this assignment, it is often useful to compare the EMAS spectrum with the steadystate absorption spectrum. Figure depicts the steady-state absorption spectrum of colloidal 2H-MoS2 nanosheets with a lateral size of 20-25 nm and a thickness of 1-2 monolayers deposited on a fluorine doped tin oxide (FTO) substrate. For better visibility of the main optical transitions, we also plot the second derivative of the data. The A and B excitonic transitions are visible at 658 nm (1.88 eV) and 610 nm (2.03 eV), respectively. The prominent C exciton (or band nesting transition) occurs as a broad signal at 433 nm (2.86 eV) and a shoulder at 388 nm (3.20 eV). Utilizing the empirically found correlation between the number of monolayers per sheet N and A exciton wavelength λA, N = 2.3⋅10 36 ⋅exp(-54888/λA), we anticipate an average layer thickness of 1.36 layers, which is in accordance with the TEM measurements in from the OCP at -0.5 V vs. the Ferrocene/Ferrocenium redox couple (Fc/Fc + ) until a minimum potential of -1.1 V, -1.7 V and -2.9 V vs. Fc/Fc + , respectively. (From here on, all potentials are reported against Fc/Fc + ). It should be noted that the FTO working electrode exhibits a potentialdependent ∆A itself (most notably at λ > 750 nm), but its magnitude is negligible compared to the EMAS signal of thick 2H-MoS2 nanosheet films (cf. Figure ). We find a bleach of the A and B exciton as soon as the potential is increased above the OCP towards more reductive potentials (Fig. ), where the bleach of the B exciton appears more pronounced than that of the A exciton. Further notable features for this reductive scan include the bleach of the C exciton and its shoulder (390 -430 nm), an induced absorption at 700 nm as well as a strong induced absorption between the C and A/B excitonic transitions. The maximum of this last signal shifts towards higher energies with more reductive potentials. Upon increasing the reductive potential Above |-1.2 V|, the formerly clearly separated bleaches of the A/B exciton broaden and dissolve into a single feature (Fig. ). We find that the change in ∆A is non-monotonous, which is why we depict the potential-dependent ∆A for selected wavelengths in Figure . Most notably, the bleach of the A-and B-exciton exhibits two relative maxima at roughly -1.5 V and -2.4 V. |
638a2c797b7c91020ce029df | 3 | The corresponding data of 78 nm and 4 nm films is shown in the Supporting Information in Figure . Similar observations hold true for the bleach of the C exciton as well as the induced absorption between the C and B excitons. The induced absorption at around 700 nm reaches a maximum at -1.1 V, after which it decreases again. We observe a small broad bleach with a minimum at 735-740 nm for potentials of -1.3 V to -1.7 V, which transitions into a weak induced absorption for reductive potentials > |-2.0 V|. |
638a2c797b7c91020ce029df | 4 | In Figure , we scan a similar 138 nm thick 2H-MoS2 nanosheet film on FTO in 0.1 M n-Bu4NPF6/CH3CN into the oxidative direction from the OCP at -0.5 V towards +1.8 V and display the EMAS signal in the same manner as for the reductive direction in Figure . The most striking result is that the qualitative EMAS response is mostly inverse to the observations during the reductive scan. With increasing potential, slightly blue-shifted (∆λ = 2 nm) induced absorptions of equal intensity are seen for the A and B excitons. The C exciton shows a redshifted induced absorption at 434 nm, which is less pronounced compared to the signal in reductive direction. Further notable features are a broad bleach with two components between the C and B exciton, a strong bleach at 690 nm as well as another bleach between the A-and B-exciton. At wavelengths around 400 nm and > 830 nm, the underlying signal of the FTO (cf. |
638a2c797b7c91020ce029df | 5 | In Figure , we display the corresponding linecuts of ∆A vs. the applied oxidative potential at the wavelength maxima of the A-and B-exciton. The trends are characteristic for all other features described above as well as for films of varying thickness (see SI Figures ). ∆A changes immediately upon leaving the OCP and reaches a maximum at +1.0 V after which it monotonously decreases. |
638a2c797b7c91020ce029df | 6 | To aid the interpretation of the EMAS experiments, we perform differential pulse voltammetry (DPV) and electrochemical gating (ECG) measurements on the same colloidal 2H-MoS2 nanosheets. DPV gives insights into the exact position of electronic states within the material, including conduction/valence band and trap states. ECG allows drawing conclusions about the type of states (distinguishes between trap states and band edges) by measuring simultaneously the electrochemical density of states and the conductivity as a function of the applied potential. |
638a2c797b7c91020ce029df | 7 | Figure depicts DPV of the nanosheets in 0.1 M n-Bu4NPF6/CH3CN on a Pt disc electrode (blue curves) against the background of the bare Pt electrode (orange curves). All measurements are initiated at the OCP of the uncoated electrode at -0.2 V and the electrochemical measurement window for this electrolyte is +1.8 V to -2.8 V. In both scan directions, the current already starts to increase/decrease slightly in the vicinity of the OCP. In the oxidative direction, we observe a broad peak at 0.77 V, including a shoulder at 0.37 V. At even more oxidative potentials, the current increases further, however without a distinguishable feature. In the reductive direction, we find a small peak at -1.18 V and a prominent peak at -1.83 V. The intensity of the second peak continuously drops towards more reductive potentials, but with a much flatter flank compared to the onset, suggesting that further reductive processes are buried within this signal. |
638a2c797b7c91020ce029df | 8 | ECG determines the potential-dependent conductivity, the differential capacitance, and the total injected charge. By measuring the amount of charge injected or withdrawn from the film within each single potential step, the differential capacitance contains information about the density of states. At the end of each potential step, the steady-state conductivity is measured, which allows the distinction between mobile or trapped charge carriers. The accumulated charge is determined by integrating the differential capacitance over the whole potential scan range, providing information on the reversibility of the charge injection. graph. The respective backward scans starting from the reductive or oxidative potential and going back to the OCP, are presented as dotted lines. For oxidative scans, we observe that both, the differential capacitance and the steady-state conductivity, change immediately. The differential capacitance starts to increase directly (Figure ), indicating an immediate removal of electrons, while the conductivity decreases (Figure ). At a potential of 0.25 V, the differential capacitance passes through a maximum and the conductivity drops more sharply. |
638a2c797b7c91020ce029df | 9 | The differential capacitance continues to drop steadily up to higher potentials, while the conductivity falls almost to zero above a potential of about 0.8 V. During the back scan, we see clear signatures of an irreversible electrochemical process, specifically the negligible conductivity (Fig. , dotted line), a sluggish decrease in the number of accumulated charge (Fig. , dotted line) and an unsymmetric differential capacitance compared to the forward scan (Fig. ). |
638a2c797b7c91020ce029df | 10 | We find a fundamentally different behavior in the reductive direction. For maximum potentials below |-2.08 V|, a comparison of forward and backward scans suggests full electrochemical reversibility (Fig. , blue curve). Scanning to even more reductive potentials shifts the drop of the differential capacitance to higher potentials with each subsequent scan (see the shoulder forming in the red curve, which is a measurement that was performed after the film was reduced below -2.08 V several times), which is mirrored by the same behavior in the respective back scans (Fig. ). In addition, the measurement at highly reductive potentials exhibits a shoulder at approx. -2 V in the forward and backward scans. We find comparable signs for electrochemical reversibility until -2.08 V in the steady-state conductivity (Fig. ). |
638a2c797b7c91020ce029df | 11 | Identical to the oxidative scan direction, the samples exhibit an intrinsic conductivity at the OCP, which continuously increases with increasing reductive potential. Beyond -2.08 V, similar shifts of the onsets occur as in Fig. in conjunction with a notable overall increase in the conductivity by approximately doubling it. Analogous to the course of the differential capacitance, a shoulder arises at approx. -2 V vs. Fc/Fc + . During the back scans (dotted lines), the conductivity is smaller and no longer mirrors the forward scan. The accumulated charge (Fig. ) also reflects this analysis. For reductive potentials below |-2.08 V|, the forward and back scans are identical, while a hysteresis occurs for more extended reductive scan windows. |
638a2c797b7c91020ce029df | 12 | Comparable oxidative and reductive ECG scans of uncoated Pt interdigitated electrodes measured in 0.1 M n-Bu4NPF6/CH3CN can be found in the supporting information in Figure . Our comprehensive (spectro-)electrochemical analysis of colloidal 2H-MoS2 nanosheets with a thickness of 1-2 layers and a lateral size of 20-25 nm affords the electronic structure shown in Figure . To approximate the corresponding energy values with respect to the vacuum level, the reference value for the ferrocene/ferrocenium redox couple needs to be subtracted from the reported potentials. While different suggestions for this reference level exist, we recommend the widely used value of -5.1 eV. |
638a2c797b7c91020ce029df | 13 | Determination of the conduction band edge. We identify the potential of the conduction band edge with -1.2 V. This is inferred from the inflection point in EMAS at this potential (Fig. ) for the traces of the A-and B-exciton as well as the first reductive peak in the DPV scan at -1.18V (Fig. ). Utilizing points of inflection in EMAS for band edge localizations has been discussed before for CdSe quantum dots, and the relatively weak DPV signal at the conduction band edge can be understood in terms of the low DOS at the K point of the conduction band shown in previous theoretical calculations for monolayers. Our assignment of the conduction band edge is also supported by the differential capacitance which exhibits reducible states at around -1.2 V (Fig. ). The fusion of the formerly clearly resolved A-and B-exciton signals into a single broad band at -1.2 V (Fig. ) is further evidence for reaching the band edge as a result of enhanced coulomb scattering of the excitons with free charge carriers. Doping and bandgap renormalization. We note that the position of the OCP at -0.5 V, that is, only 0.7 V below the conduction band edge, indicates substantial n-doping of the nanosheets, which is supported by the immediate decrease of the steady-state conductivity upon oxidation (Figure ). We stress that bleaches of the A, B and C excitons are instantaneously formed in the reductive EMAS scans upon departing from the OCP (Figure ) even for potentials far below the conduction band edge. The bleaching below -1.2 V is therefore not caused by band filling but rather a sign of bandgap renormalization, previously observed in TA measurements , by chemical doping and by spectroelectrochemistry of 3-layer, 150 nm large 2H-MoS2 nanosheets produced by liquid based exfoliation. Briefly, upon electron injection into 2H-MoS2 nanosheets the self-energy of the electron and hole states decreases due to many-body effects, and this decrease is typically larger for the electron states. If the concomitant decrease in excitonic binding energy (due to increased charge screening) is smaller than this band gap renormalization, a net red shift in absorption results from increased n-doping. Likewise, a net blueshift is expected for lowering the Fermi level. We therefore attribute the blue-shifted induced absorptions near the A/B-excitonic transitions during oxidation in Figure to the same bandgap renormalization effect, given that it occurs immediately upon departing from the OCP. |
638a2c797b7c91020ce029df | 14 | Localizing the potential of the valence band edge is more challenging in view of the irreversible oxidation found for this material, which prevents an unambiguous determination of an inflection point in EMAS (Fig. ). Irreversible oxidation of 2H-MoS2 has previously been reported, and is most likely due to oxidation of Mo 4+ to Mo 6+ . We interpret the pronounced oxidation peak in the DPV measurement at 0.77 V (Fig. ) as the valence band edge. This agrees with the potential of 0.8 V during the oxidative ECG scan (Figure ) at which the conductivity vanishes almost completely. This finding can be rationalized with a compensation of the n-dopants as well as the oxidative degradation once the valence band edge is reached. It is also in reasonable agreement with the EMAS traces in Fig. , considering that upon reaching the valence band, one would expect a competition between induced absorption (due to band gap renormalization) and bleaching of the excitonic transitions (due to band filling). |
638a2c797b7c91020ce029df | 15 | Excitonic binding energy. From these considerations, we infer an electrochemical bandgap of ~1.95 eV, which is less than 0.1 eV larger than the optical gap. Although this excitonic binding energy appears low for MoS2, it is in line with the strong dependence of the excitonic binding energy on the dielectric constant of the surrounding medium. In the present case, we may approximate the environment with pure acetonitrile, which exhibits a very large dielectric constant of ~35 and, thus, exerts strong dielectric screening for the electron-hole pair, which decreases the binding energy. |
638a2c797b7c91020ce029df | 16 | Location and splitting of the C-exciton. We find that the electrochemically induced bandgap renormalization probed for the A-and B-exciton (Fig. , red and green trace) is also reflected in the potential-dependent absorption of the C-exciton (Fig. , blue trace). While this may be fundamentally expected, is has not been previously demonstrated, and we speculate that its observation is facilitated by the much higher absorption intensity of the C transitions compared to the A and B transitions in the laterally smaller 20-25 nm large 2H-MoS2 nanosheets used in our experiments (e.g. compared to ref. 25). Moreover, we find clear signs of two separate bleaches for the C-exciton between 390 and 430 nm (Fig. ), which merge into a broad bleach at roughly -1.2 V. This is evidence for the theoretical prediction of two distinct band nesting features in MoS2 , which has previously been confirmed for bulk 2H-MoS2 , but, to the best of our knowledge, not for monolayer 2H-MoS2. First insights in higher energy transitions of various monolayer TMDCs are reported by Hong et al., but no distinction between different contributions to the C band in 2H-MoS2 was given. Although Fig. and the seemingly identical inflection points for the bleaching of the A/B-vs. the C-excitons at -1.2 V may suggest that the involved electron states are identical, this is fundamentally impossible. The A/B excitons require a point in the Brillouin zone for which ∇ 𝑘 𝐸 𝐶 = 0, while the definition of band nesting is ∇ 𝑘 𝐸 𝐶 ≠ 0 (and ∇ 𝑘 𝐸 𝐶 = ∇ 𝑘 𝐸 𝑉 ). Therefore, the involved electron states of the A/B excitons vs. the C-excitons must be located at different points in the Brillouin zone, but at roughly the same energy. Note that the corresponding hole state of the C excitons would be expected at a potential of > +1.7 V, which prevents its determination here due to the rapid degradation at such potentials. |
638a2c797b7c91020ce029df | 17 | The analysis of the weak bleach at 735 -740 nm in the potential range of -1.3 V to -1.7 V (Fig. ) is complicated due to the adjacent induced absorption caused by the red-shifted A-exciton and overlapping induced absorption of FTO at wavelengths >750 nm. The most likely explanation for this EMAS band is the indirect transition inherent to 2H-MoS2 with more than one monolayer. Since we determined the average thickness of the nanosheets studied here with 1-2 monolayers, it is reasonable to assume that a significant portion of the sample will exhibit such an indirect transition. Computational studies have shown that for 2 layers of 2H-MoS2, the electron state involved in the indirect transition is the same as that occupied by the A/B excitonic transitions. The identical reductive potential at which we observe the bleaching of the A/B excitons and the transition at 730-740 nm therefore supports the assignment of the latter as the (Figure ). However, a phase transition to the metallic 1T allotrope would also explain the increase in conductivity. |
638a2c797b7c91020ce029df | 18 | We have determined the electronic structure near the band gap of colloidal 2H-MoS2 nanosheets with a thickness of 1-2 monolayers by a combination of potential-modulated absorption spectroscopy (EMAS), electrochemical gating (ECG) and differential pulse voltammetry (DPV). We identified the edges of the conduction and valence band of the direct transition in the monolayers at -1.18 V and 0.77 V vs. Fc/Fc + respectively, yielding a band gap of 1.95 eV, i.e. an excitonic binding energy of less than 0.1 eV in the high dielectric environment of 0.1 M n-Bu4NPF6/CH3CN. For the bilayers, we found an indirect band gap of 1.55 eV with the valence band edge located at 0.37 V. The as-synthesized 2H-MoS2 nanosheets are n-doped and show strong bandgap renormalization effects upon electrochemical oxidation and reduction. We identified two separate band nesting transitions that utilize points in the conduction band with roughly the same energy as the conduction band edge (-1.2 eV). While the oxidation of the valence band is irreversible, we find the reduction of the conduction band to be mostly reversible over a large potential range. These results indicate that colloidal chemistry affords high-quality 2H-MoS2 nanosheets with a well-preserved electronic structure and potential applications in reductive electrocatalysis, e.g., for the electrocatalytic hydrogen evolution reaction. |
638a2c797b7c91020ce029df | 19 | We synthesized colloidal 2H-MoS2 nanosheets based on the previously reported protocol to obtain mono-and bilayers with a lateral dimension of approximately 20-25 nm. The Moprecursor was prepared by dissolving molybdenum(V)-chloride in a mixture of oleylamine and oleic acid of 10:1 by stirring inside a glovebox for 2 days. The concentration of the precursor was set to 240 mM. For the nanosheet synthesis, elemental sulfur (87.5 mg, 2.73 mmol) and 17 mL oleylamine were added to a three-neck flask and degassed under vacuum from an oilpump for 30 min at 85°C. During this step, the sulfur dissolved, and the solution turned dark. |
638a2c797b7c91020ce029df | 20 | After a constant temperature was achieved, the molybdenum precursor (1.15 mL, 276 µmol) was added over 30 min using a syringe pump. After the first drops of precursor were added, a final color change to opaque black was immediately visible. The synthesis was continued at the same temperature for 30 minutes after the addition of the precursor and then quickly cooled to room temperature. After the reaction, the synthesized MoS2 nanosheets were precipitated by addition of 16 mL hexane and centrifugation at 3300 rcf for 10 minutes. The precipitate was redispersed in hexane and the centrifugation step was repeated two times. The entire precipitation was done under inert conditions and the sample was then stored dispersed in hexane and under nitrogen. |
638a2c797b7c91020ce029df | 21 | All chemicals used for electrochemical experiments were stored and handled under inert atmosphere. Dry and degassed CH3CN was obtained by three times distillation of HPLC-grade solvent over P4O10 followed by three freeze-pump-thaw cycles. Afterwards the acetonitrile was stored over 3 Å molecular sieve, which was activated at 220°C and 2-3 mbar for 48 h beforehand. Prior to use, the electrolyte n-Bu4NPF6 (98%, Alfa Aesar) was recrystallized five times from 3:1 EtOH/H2O followed by drying at 105°C and 2-3 mbar for 6 days. Purity was checked by 1 H-, C-, F-and 31 P-NMR spectroscopy. Before preparing fresh electrolyte solutions for each experiment, the stored CH3CN was run over a column of neutral alumina, priorly activated at 220°C and 2-3 mbar for 6 days. AgClO4 (≥ 97%, anhydrous, Alfa Aesar) and Ferrocene (98%, Acros Organics) were used without further purification. |
638a2c797b7c91020ce029df | 22 | Experiments were performed in a glovebox under nitrogen atmosphere and additionally, the used full-glass gas tight electrochemical cell was placed in a faraday cage for better electrical shielding. A three-electrode arrangement was employed: A 3 mm diameter Pt disc electrode (Metrohm part no. 6.1204.310) as working electrode (WE), a 1 mm diameter coiled platinum wire as counter electrode (CE) and a Haber-Luggin double-reference electrode (RE) composed of a Ag/Ag + system (1 mm diameter silver wire in a 0.01 M AgClO4 solution in 0.1 M n-Bu4NPF6/CH3CN) capacitively coupled (10 nF) to a Pt wire which is immersed into the electrolyte next to the capillary opening. To avoid any contamination of the electrolyte within the sample chamber by the reference electrode solution, or vice versa, an additional frit filled with 0.1 M n-Bu4NPF6/CH3CN was inserted between the Haber-Luggin capillary and the reference electrode compartment. |
638a2c797b7c91020ce029df | 23 | Measurements were carried out starting at the open circuit potential of the bare Pt electrode of -0.2 V vs. Ag/Ag + to -2.8 V in reductive and to 2 V in oxidative direction. Scans were performed with a potential step width of 4 mV, 50 mV amplitude, 0.06 s pulse width, 0.02 s sampling width and a 0.5 s pulse period at a measurement sensitivity of 10 µA/V. DPV measurements are iR-compensated by the integrated positive feedback function of the CHI760E. |
638a2c797b7c91020ce029df | 24 | The sample chamber of the electrochemical cell was filled with 12 mL of 0.1 M n-Bu4NPF6/CH3CN. At first, background electrolyte scans were performed with a freshly polished Pt disc electrode. Afterwards 30 µL of a 10 mM 2H-MoS2 solution in hexane were drop casted on the working electrode, dipped in a 3% ethane-1,2-dithiol solution in acetonitrile for a few minutes and afterwards washed with acetonitrile. Then sample measurements were performed five times in a row with the same parameters as for the respective background scan. |
638a2c797b7c91020ce029df | 25 | Before film preparation, both substrates used for ECG and EMAS were cleaned as follows: Pt interdigitated electrodes (IDE) with a gap width of 5 µm (Metrohm G-IDEPT5) were rinsed several times with H2O and MeOH. Fluorine doped tin oxide (FTO) coated glass (Solaronix TCO22-15) was sonicated for 15 min each in acetone, hexane, 10% solution of Extran MA01 (Merck) and H2O. From here on, both substrates were treated identically. To ensure good contacting of the substrate, not the entire substrate surfaces were coated, but mainly the part that is in contact with the electrolyte later on. (3-mercaptopropyl) trimethoxysilane (MPTMS) treatment was performed by placing the substrates in a stirred 3% MPTMS toluene solution at 50°C for 90 min followed by rinsing with toluene. Afterwards, the substrates are placed in a 4 mM 2H-MoS2 nanosheet solution in hexane for 22 h, which was sonicated for 15 min beforehand. Film preparation happens by precipitation of the long term unstable colloidal solution. For the thinner EMAS samples the film preparation times were 30 min, 5 min and 30 s. |
638a2c797b7c91020ce029df | 26 | Measurements were performed under the same inert conditions in the same full-glass-cell with the same electrolyte solution, CE and RE, only the WE was different: A beforehand mentioned Pt interdigitated electrode with a gap width of 5 µm between the electrode arrays and total channel length of 3.373 m (Metrohm G-IDEPT5) was used. Both arrays are independently controlled by the two working electrode channels of the CHI760E bipotentiostat. All measurements were initiated at the OCP at -0.4 V vs. Fc/Fc + . Before the 2H-MoS2 nanosheet coated IDEs were measured, an uncoated sample was investigated with the same parameters to check whether there were any contaminations in the electrolyte solution. |
638a2c797b7c91020ce029df | 27 | Differential capacitance was determined via chronoamperometry (CA) by measuring in 20 mV steps towards the respective reductive or oxidative direction. At each of these steps, which were applied to both working electrodes simultaneously, the current was detected over 5 seconds and the amount of flowed charge was determined via integration. The capacitive current was extracted from the total measured current by background correction (assuming the Faraday current exhibits Cottrell behavior). At the end of such a 20 mV step, the final potential was held and the steady-state conductivity was measured by means of Cyclic Voltammetry (CV). The potential of first WE was held constant while the potential of the second WE was scanned ±10 mV around this constant potential. In case a conductive channel is present between both WE, the current detected on both working electrode channels have the same magnitude, but different signs. The steady-state conductivity was calculated from the slopes of the iV-curves. |
638a2c797b7c91020ce029df | 28 | Then, the next 20 mV step was applied to both WE by means of CA and the differential capacitance was calculated again., and so on. This combination of CA and CV is done until the final reductive or oxidative potential was reached. Then, the scan direction was reversed and measurements back to the OCP are performed. The accumulated charge was calculated by integrating the differential capacitances until the potential of interest. |
638a2c797b7c91020ce029df | 29 | CA measurements were performed in 20 mV steps with a 5 s pulse width and 1 ms sample interval. CV measurements were conducted with a scan rate of 10 mV/s and a 1 mV sample interval. The measurement sensitivity of both techniques was set to 10 µA/V. As for DPV, all CA and CV measurements were iR-compensated, and the potentials are converted to the Fc/Fc + redox couple as described above. |
638a2c797b7c91020ce029df | 30 | For EMAS measurements a home-built spectroelectrochemical transmission cell, equipped with a 1 mm diameter Ag wire pseudoreference electrode, a 1 mm diameter coiled platinum wire counter electrode and a fluorine doped tin oxide (FTO) coated glass as working electrode was used. The working electrode was either coated with 2H-MoS2 nanosheets as described above, or uncoated for reference measurements. The measurement cell was filled with 6 mL 0.1 M n-Bu4NPF6/CH3CN and assembled within a glovebox. For the actual measurements, it was afterwards transferred outside of the box and placed in a faraday cage. |
638a2c797b7c91020ce029df | 31 | Using the CHI760E bipotentiostat, a constant potential at the OCP of approximately -0.5 V vs. Fc/Fc + was applied to the working electrode. Then, a rectangular shaped modulated potential, produced by a Waveform generator (PSG9080, Joy-IT), with a modulation frequency of 37 Hz and a variable amplitude as an offset was superimposed. Meanwhile, the working electrode is transilluminated by monochromatic light generated by an Apex2 QTH lamp (Oriel Instruments) and selected using a Cornerstone 130 (Oriel Instruments) monochromator. Above 600 nm, a 570 nm long-pass filter was introduced into the beam path to block components with a higher diffraction order. The transmitted light was detected at a DET36A Si biased detector (Thorlabs), and the signal was passed to a MFLI lock-in amplifier (Zurich Instruments). In parallel, the lock-in amplifier received the modulated potential generated by the signal generator as a reference signal and searched for a modulated component in the optical signal that oscillates at the same frequency. Considering the phase relationship between the modulated potential and the modulated component in the optical signal, it was determined whether the measured signal is an induced absorption or a bleach (see Figure and the accompanied text). |
638a2c797b7c91020ce029df | 32 | During the EMAS measurement, the modulated potential component was applied to the sample in such a way, that the total potential signal always oscillated between OCP and a corresponding reductive or oxidative potential. At the beginning of the measurement, the peak-to-peak amplitude was 60 mV. With an appropriate offset, the signal is then shifted to positive or negative values, so that the potential then modulated between OCP and OCP ±60 mV. At this potential, a complete spectrum from 350 to 860 nm was scanned in 3 nm steps. Afterwards, the peak-to-peak amplitude was increased by 60 mV, the signal offset was adjusted accordingly, and a complete spectrum was recorded again. This was done until the final corresponding oxidative or reductive potential was reached and a complete data set was recorded. |
638a2c797b7c91020ce029df | 33 | During the EMAS measurement, iR-compensated by the integrated positive feedback function of the CHI760E was activated and the measurement sensitivity of the bipotentiostat was set to 0.1 mA/V. The AC coupled signal input of the lock-in amplifier was set to 10 MΩ with an input range of 10 mV. Incoming signals were processed by a 6 th order low pass filter having a time constant of 100 ms. The resolution of the monochromator was set to 3-4 nm by adjusting the input and output slit width. Potentials of the measured data were also converted to the formal potential of the external redox couple Ferrocene/Ferrocenium (Fc/Fc + ), which was measured at 0.32 V vs. Ag pseudoreference electrode within the same setup. |
638a2c797b7c91020ce029df | 34 | UV-vis spectra were recorded using a Cary 5000 UV-Vis-NIR spectrophotometer from Agilent Technologies. As samples, the 2H-MoS2 nanosheet coated FTO windows, previously employed for EMAS measurements, were used. The same samples were also used for SEM measurements that were performed using a Hitachi SU 8030. Information about the height of the 2H-MoS2 nanosheet films on FTO glass and on Pt-IDE substrates was obtained using a Bruker Dektak XT-A stylus profilometer measuring over the edge of the film coating over a distance 1000 µm using a tipforce of 1 mg. For identifying the layer thickness and the lateral size of the 2H-MoS2 nanosheets, the colloidal solution was dropcast onto a carbon-coated copper grid (Quantifoil) and measured by TEM with a Tecnai G2 F20 TMP microscope (FEI) with an acceleration voltage of 200 kV. |
638a2c797b7c91020ce029df | 35 | Assuming we have a rectangular shaped potential that modulates in reductive direction between 0 V and -1 V having a period of 2π as displayed for the red curve in Figure . The lock-in amplifier is fed with the same waveform as reference signal and sets the reference phase of 0° to the falling potential at position 0π. Let us further assume that an applied voltage of 0 V is located within the bandgap of the investigated material and -1 V is sufficient negative to reduce the conduction band edge. Therefore, the conduction band is half of the time occupied and half of the time unoccupied. As a result, the absorption behavior of the material under investigation also changes over time. Two cases can be distinguished: |
638a2c797b7c91020ce029df | 36 | If the absorption of optical transition from the valence band to the conduction band is blocked, the absorption of the material under investigation at this specific wavelength is reduced so that the measured Δphotovoltages at the detector increases (blue rectangular curves). In the case that the reduction of the material is instantaneous, and the absorption consequently also decreases directly, while the measured photovoltage ideally follows the course of the dark blue |
625530ddebac3a4918d21843 | 0 | The technological innovations in drug discovery have led to the development of larger, more complex and more hydrophobic compounds as lead candidates and active pharmaceutical ingredients. According to an estimate, 40% of the drugs in market and 90% of the new chemical entities are considered hydrophobic . These molecules require specialized processing and safe vehicles for solubilisation, including problems at various levels from discovery to processing and formulations development . However, many vehicles available do either not have an ideal safety profile, e.g. Cremophor EL (CrEL) for paclitaxel (PTX) administration or are not as good solubilizers as would be desired. Even with the premedication including corticosteroids and anti-histamines, CrEL induced hypersensitivity reactions are commonly observed and most vehicles/excipients do not allow for solubilization of more than 20 wt. -25 wt.% and the overall increase in apparent solubility is limited. Thus, for the delivery of these agents, there is clear need to find alternative formulations and vehicles. |
625530ddebac3a4918d21843 | 1 | They are nanoscopic architectures formed by the self-assembly of amphiphilic polymers . During (or after) the self-assembly process, PMs are capable to encapsulate/solubilize additional hydrophobic molecules. Several PMs formulations are already in clinical use or have been in clinical trials, for example NK911 and Genexol-PM . However, PMs often suffer from intrinsic issues such as low drug loading, instability and poor permeability across cell membranes or other biological barriers . The loading capacity (LC) and stability can directly be correlated to intrinsic properties of polymers, drugs or/and their compatibility with each other . In the past decades, poly(2-oxazoline)s (POx) and poly(2-oxazine)s (POzi) have gained significant attention for their biomedical and pharmaceutical applications . Unlike many other systems, the POx/POzi based ABA triblock copolymer (A and B, hydrophilic and hydrophobic block, respectively) have shown particular potential for development of ultra-high drug loaded micellar formulations . In particular, ABA triblock copolymers featuring poly(2-methyl-2-oxazoline) (pMeOx) as hydrophilic A block and moderately hydrophobic poly(2-n-butyl-2-oxazoline) (pBuOx), poly(2-n-butyl-2-oxazine) (pBuOzi) or poly(2-n-propyl-2-oxazoline) (pPrOx) show very promising properties such as high drug loading, excellent cytocompatibility and are very well tolerated upon in vivo administration . |
625530ddebac3a4918d21843 | 2 | For the development of a successful PMs system, the primary mission is to select a polymer which is not only compatible with the cargo but also provide colloidal stability even at high relative and absolute cargo concentrations. There are a variety of physicochemical techniques available to determine the compatibility between polymer and drugs, but most of these are time-consuming, expensive and laborious . For the development of cost-effective and efficient drug delivery system, it is usually desirable to predict theoretically the extent of compatibility between polymer and drugs before any experimentation. For many years, Hansen solubility parameters (HSPs) were routinely employed for estimation of compatibility between solute and solvent especially in paint/polymer industry . More recently, HSPs are used in other areas like cosmetics , drugs , oligomers or gel formation and nanoparticles . To guide formulation development, Allen and the co-workers calculated HSPs to estimate the compatibility between model drug ellipticine and a variety of chemically diverse polymers. Overall, a good correlation was observed between theoretical compatibility profile and experimental results of the formulation . Importantly, as the HSPs are calculated by various group contribution methods so the polymers which are closely related or structural isomers would yield the same HSPs values and in turn represent the similar compatibility for the drugs. In contrary, Lübtow et al. observed the distinct polymerdrug specificities in POx/POzi based micellar formulation, although the polymers differed by only one methylene unit in the side chain or the backbone (structural isomers) of hydrophobic block . |
625530ddebac3a4918d21843 | 3 | In order to investigate the effect of increasing side chain length (rendering the system more hydrophobic) on drug loading and to check the applicability of HSPs, previously a library of 18 different POx/POzi based ABA triblock copolymers was synthesized, comprising pMeOx as A, while varying the hydrophobic B block from series of linear , branched (aliphatic POx with varying side chain lengths; C4-C9) or an aromatic side chains . The polymers with shorter side chain length (C3 and C4) appeared to be the best solubilizer for various drugs, However, no clear trend was predicted by HSPs. Thus, the main aim of the present study is to establish an effective, rapid and simple method for the rational design of drug formulations. Therefore, the goal was to develop a formulation library with high drug loading as well as the comparison of calculated HSPs, to identify if they can be used as means to predict the polymer drug compatibility in these systems. Accordingly, from the previously synthesized library, we selected the best performing polymers and tested their solubilizing capacity with a library of 21 different hydrophobic drugs. The HSPs were also estimated theoretically and experimentally to guide formulation development and to check their applicability for such systems. |
625530ddebac3a4918d21843 | 4 | The polymerization and work-up procedures were carried out as described previously. Briefly, 1 eq. of initiator (MeOtF) was added to a dried and argon flushed flask and dissolved in the respective amount of solvent (PhCN). The monomer was added and the reaction mixture was heated to 110°C (POx) or 120°C (POzi) for approximately 4-6 hours. Reaction progress was controlled by 1 H-NMR spectroscopy. After complete consumption of the monomer, termination was carried out by the addition of 3 eq. of Boc-pip and the mixture was stirred at 50°C for additional 4 hours. PhCN was removed under reduced pressure. In the case of four hydrophobic homopolymers (pBuOx, pPentOx, pPrOzi and pBuOzi), after PhCN removal, the thick polymer mass was dissolved in chloroform and extracted three times against water followed by chloroform removal under reduced pressure. |
625530ddebac3a4918d21843 | 5 | While in the case of pMeOx, hydrophilic homopolymer, the thick polymer mass was dissolved in 1:1 mixture of chloroform and methanol. The polymer solution was added dropwise into the ice cold diethyl ether under continuous stirring. After polymer precipitation, the diethyl ether was decanted. The residual diethyl ether was removed at rotary evaporator. The polymer mass was dissolved in deionized (DI) water and lyophilized. |
625530ddebac3a4918d21843 | 6 | The theoretical solubility parameters were calculated by group contribution methods (GCM) from Hoftyzer Van-Krevelen's (HnV) and Yamamoto Molecular break method (YMB) by commercially available HSPiP software (5 th edition, version 5.0.05). In case of GCM (HnV), each functional or structural group in a compound contribute towards certain molar attractive force and hydrogen bonding energy. In contrast to one dimensional Hildebrand solubility parameters, the Hansen solubility parameters (HSPs) divide the total solubility parameter (δT) into three individual components arising from dispersion (δd), polar (δp) and hydrogen bonding contributions (δh). The δd, δp and δh can be calculated by HnV method by using the following equations: |
625530ddebac3a4918d21843 | 7 | Where Fdi, Fpi and Ehi are the molar dispersion, polar attraction constant and hydrogen bonding energy, respectively and Vi is the group specific volume. Each structural group in the molecule contribute towards the Fdi, Fpi and Ehi and the respective values were obtained from literature . Where, V is the molar volume of drug calculated by Fedor's method . |
625530ddebac3a4918d21843 | 8 | Accordingly, each functional group contribute towards the molar volume and can be calculated by following equation 4. The δdrug and δpolymer are representing the δT for drug and polymer, respectively. The δT is the sum of dispersion (δd), polar (δp) and hydrogen bonding contributions (δh) and can be calculated by following equations: |
625530ddebac3a4918d21843 | 9 | The second method used for HSPs calculation is YMB. This is a sub-program of commercially available HSPiP software. In comparison to HnV, this is relatively simpler method to obtain HSPs. The HSPs can be obtained by inserting SMILES code into the software. The SMILES codes were obtained by drawing the structures in the ChemDraw Professional, version 20.0.0.41. |
625530ddebac3a4918d21843 | 10 | The software HSPiP also provides a method to semi-empirically determine the HSPs values by dissolving the compound in a wide range of different solvents. Therefore, the solubility profiles of homo-/triblock copolymers were checked in 31 different solvents at concentration of 20 g/L by shaking at 25°C for 24 hours. The results were recorded as binary score of 0 and 1, corresponding to insoluble and soluble, respectively. These results were further incorporated into HSPiP software and experimental HSPs values were obtained. |
625530ddebac3a4918d21843 | 11 | Depending upon solubility of drugs in organic solvents, polymer and drug stock solutions were mixed in desired ratios. After complete removal of the solvent at 50°C under a mild stream of argon, the films were further dried in vacuo (≤ 0.2 mbar) for at least 30 minutes to remove the traces of organic solvent (if any). Subsequently, preheated (37°C) DI water was added to obtain desired final polymer (10 g/L) and drug concentrations. To ensure complete solubilisation, the solutions were shaken at 55°C for 15 to 30 min at 1250 rpm with a Thermomixer comfort (Eppendorf AG, Hamburg, Germany). Non-solubilized drug was removed by centrifugation for 5 min at 10.000 rpm with a 3-Speed micro centrifuge, (neoLab, Heidelberg, Germany). Solubilisation experiments were performed in 3 individually prepared samples and results are presented as mean ± standard deviation (SD). |
625530ddebac3a4918d21843 | 12 | HPLC analysis was carried out on a LC-20A Prominence HPLC (Shimadzu, Duisburg, Germany) equipped with a system controller CBM-20A, a solvent delivery unit LC-20 AT (double plunger), an on-line degassing unit DGU-20A, an auto-sampler SIL-20AC, a photo diode array detector SPD-M20A. As stationary phase, a ZORBAX Eclipse Plus (Agilent, Santa Clara, CA, USA) C18 column (4.6 x 100 mm; 3.5 μm) was used. Quantification of drugs was performed with a stepwise gradient using acetonitrile and water with 0.05% TFA. Initially the HPLC methods were developed for individual drugs. After that, the series of standard solutions of different known concentrations were measured to obtain the standard curve followed by the drug quantification at their respective lambda max. For all the HPLC data, please refer to Figure S8-S28. |
625530ddebac3a4918d21843 | 13 | Clofazimine (CFZ) quantification was performed by UV-Vis absorption on a BioTek Microplate spectrophotometer, Thermo Fischer Scientific (MA, USA) using a calibration curve with the known amounts of CFZ (Figure ). Samples were prepared Rotilabo F-Type 96 well plates, Carl Roth GmbH and Co. KG (Karlsruhe, Germany) with the constant volume of 200 µl. Spectra were recorded from 300 to 600 nm at 298 k. CFZ absorption was detected at 450 nm. Prior to UV-Vis absorption measurements, the aqueous formulations were diluted with ethanol. |
625530ddebac3a4918d21843 | 14 | From a previously investigated library of 18 different ABA triblock copolymers , we selected 3 best performing polymers with respect to drug solublization. In all the selected polymers the hydrophilic block A is always poly(2-methyl-2-oxazoline) and comprising moderately hydrophobic block B i.e. poly(2-n-butyl-2-oxazoline) (pBuOx), poly(2-n-propyl-2oxazine) (pPrOzi) and poly(2-n-butyl-2-oxazine) (pBuOzi). All of these polymers were previously investigated for drug formulation and have shown promising results . As all the triblock copolymers comprised of pMeOx as A, so they are represented according to hydrophobic block i.e. A-pBuOx-A, A-pPrOzi-A, and A-BuOzi-A. To this mini-library, we added a fourth polymer (less commonly explored) comprised of hydrophobic poly(2-n-pentyl-2oxazoline) (pPentOx) block. Recently, the polymer A-pPentOx-A was reported to have ultrahigh LC of 47 wt.% for new investigational molecule i.e. BT-44 . The targeted block length for each block in triblock copolymer is A35-B20-A35. For synthesis and characterization of four triblock copolymers, the readers are referred to previous reports . These four polymers represent two pairs of structural isomers. i.e. they differ from each other by one methylene unit which is shifted from polymer backbone to the side of hydrophobic block and vice versa (A-pBuOx-A/A-pPrOzi-A and A-PentOx-A/A-pBuOzi-A). |
625530ddebac3a4918d21843 | 15 | To gain insights into theoretical and experimental solubility parameters (SPs) while keeping in mind the general assumption that mainly hydrophobic block is responsible for drug loading, initially, we synthesized five different homopolymers which collectively comprised our triblock copolymers represented as hydrophobic pBuOx, pPentOx, pPrOzi, pBuOzi and hydrophilic pMeOx (Figure ). All of the homopolymers were synthesized by living cationic ring opening polymerization (LCROP) as previously described . The homopolymers were characterized by 1 H-NMR and GPC. The end-group analysis via 1 H-NMR gave a degree of polymerization of around 20 and 35 for hydrophobic and hydrophilic block, respectively (Figure to S5) which is very close to targeted block length. GPC elugram showed that purified homopolymers appeared essentially 10onomodal with reasonably low dispersity (Ɖ < 1.1) (Figure ) |
625530ddebac3a4918d21843 | 16 | After successful homopolymer and triblock copolymers synthesis , we determined the theoretical and experimental Hansen solubility parameters (HSPs). Initially, the molar volume was calculated by Fedor's method (Table ) followed by theoretical SPs calculations by Hoftyzer Van-Krevelen (HnV) (Table -S24) and Yamamoto molecular break method (YMB) (Table ). The SPs values obtained by HnV method were not affected by degree of polymerization (DP), in contrast, in case of YMB, with the increasing DP, the SPs of selected homopolymer showed exponential increase and became physically unreasonable . For the sake of comparison with HnV method, a single repeat unit was used for YMB method. |
625530ddebac3a4918d21843 | 17 | HnV is a group contribution method, where every single group in the chemical structure contributes towards the SPs of the compound. The δd, δp, δh, δT and molar volume V were calculated by using equation 1 to 6 (Table and Table -S24). Expectedly, and as reported for a library of 18 different triblock copolymers , the HnV could not differentiate between the structural isomers i.e. pBuOx/pPrOzi and pPentOx/pBuOzi resulting in same δd, δp, δh, δT (Figure , Table ) and molar volume values (Figure ). Generally, with the exception of δd value, the δp, δh and δT values of hydrophilic pMeOx remained significantly higher in comparison to all four hydrophobic blocks (Figure ). Overall, the SPs values for all the hydrophobic blocks by HnV method remained in close range. Interestingly, the YMB method was able to differentiate between the structural isomers with respect to δd, δp, δh values. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.