id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
631a10083940c20533f81566 | 5 | based approaches, which are in general intermediate in accuracy and computational expense between MM and full QM. While these approaches are currently too slow for routine molecular dynamics sampling in the condensed phase, their use for the rapid generation of reference data for parametrizing FFs is appealing, as demonstrated by the recent refinement of GAFF2 parameters against ANI2x torsional scans . |
631a10083940c20533f81566 | 6 | Here, we present OpenFF BespokeFit, an open-source, automated python package for bespoke FF parameter fitting. This first version of BespokeFit specifically aims to derive bespoke torsion parameters for individual molecules but we plan to extend it to additional FF terms in future. BespokeFit is designed for compatibility with the SMIRKS Native Open Force Field (SMIRNOFF) format, hence it uniquely provides users with the opportunity to re-fit torsion parameters, using robust methods that are consistent with the base OpenFF parametrization philosophy. We make use of the unified quantum chemistry (QC) program executor QCEngine to provide simple, resource-agnostic access to a wide range of quantum, semi-empirical and machine learning based reference data, which can be generated on-the-fly. Furthermore, we introduce OpenFF QCSubmit as an open-source tool for curating, submitting and retrieving large QM reference datasets from QCArchive to aid large-scale FF parameter fitting. We demonstrate the utility and easeof-use of the BespokeFit/QCSubmit interface by deriving bespoke torsion parameters for a large dataset of 671 QM torsion scans derived from fragments of drug-like molecules. We further demonstrate the ability of BespokeFit to construct FFs for a congeneric series of inhibitors of the TYK2 protein, and benchmark the accuracy of the resulting FFs by computing protein-ligand binding free energies and comparing against experimental data. |
631a10083940c20533f81566 | 7 | OpenFF BespokeFit is a scalable and extensible framework that automates the optimization of bespoke torsion parameters for SMIRNOFF-style FFs against QC reference data. It is designed with reproducibility and ease-of-use in mind. The process begins by defining a (or retrieving the default) workflow protocol that defines the entire fitting process, which typically involves four stages: 1) fragmentation, 2) SMIRKS generation, 3) QC reference data generation, and 4) parameter optimization. This general workflow can then be applied to a target set of input molecules, producing a series of specific fitting schemas that can be submitted to BespokeFit for processing. The general workflow can also be serialized to file in JSON format for later use and shared with others to ensure reproducibility of the fitting protocol. |
631a10083940c20533f81566 | 8 | The modular design of the code base makes building the workflow straightforward, as users can simply select the module they wish to use for each stage of the workflow and add it to the schema. For example, users can choose between two pre-defined fragmentation modules, which offer rule-or heuristic-based fragmentation . Extending the workflow is also trivial as users can add new modules for any of the above fitting stages by using the plugin framework. This allows for fast prototyping of new modules, such as the addition of new parameter optimization methods, without needing to update the core package, which is critical in keeping up with the developing landscape of QC-based parameter derivation. Next, we discuss the various stages of a typical BespokeFit workflow. |
631a10083940c20533f81566 | 9 | BespokeFit uses the OpenFF Fragmenter package to fragment larger molecules into smaller representative entities. As well as reducing the number of degrees of freedom to minimize at the QM level during a torsion scan, fragmentation can also help to avoid hysteresis by reducing the opportunities for steric clashes in more complex molecules. |
631a10083940c20533f81566 | 10 | Fragmenter aims to preserve the local environment around the targeted torsion while retaining as little of the parent molecule as possible to reduce the computational demands of the calculation. However, oversimplification of the local chemical environment can result in fragments that inaccurately approximate the parent potential energy surface, which may lead to parameters that transfer poorly back to the parent molecule. It has been shown that the Wiberg bond order (WBO) provides a fast and robust measure of whether a torsion profile has been disrupted by fragmentation . BespokeFit then only accepts a proposed fragmentation if the WBO of the parent and proposed fragment agree to within a defined threshold (0.03 e by default). |
631a10083940c20533f81566 | 11 | Having generated the fragments for parametrization, the dihedral angle to be scanned must be assigned a SMIRKS Native Open Force Field (SMIRNOFF) format parameter for incorporation into the FF. When adding any new parameter to a force field, a trade-off between accuracy and transferability must generally be made to avoid the proliferation of parameters and/or atom types. However, in the case of bespoke parametrization, accuracy is favored, and so highly specific encodings are used to ensure the patterns can only be re-used in similar chemical environments, rather than being applied to chemistries that they were not intended to cover. Each fragment is considered to be the minimum electronically decoupled substructure that preserves the local chemical environment of the torsion (due to the conservation of the WBO). Hence, the maximum common substructure between the parent and fragment molecules is embedded into the SMIRKS pattern used to label the scanned dihedral. This SMIRKS pattern is produced by the ChemPer package , and is linked to the generated bespoke torsion parameters, thus facilitating transferability between the fragment and the parent, as well as any other molecules that share this exact substructure (common in studies of congeneric series, as we shall show later). Figure shows an example of a molecule, and the corresponding fragment, alongside the SMIRKS pattern generated for the highlighted torsion scan. The reader does not need to be too familiar with SMIRKS patterns to recognize that every possible attribute has been added to the pattern, making it highly specific to the molecule under study. |
631a10083940c20533f81566 | 12 | An additional decision to make is how many parameters should be used to parametrize a given torsional rotation? Rotatable bonds can have many torsion parameters associated with them, due to the number of unique combinations of atom quartets running through the central bond, and the overall torsional profile is given by the sum of these terms. Using a traditional atom-typed parameter assignment scheme can lead to insufficient flexibility. For example, for the highlighted rotation in aspirin, shown in Figure (right), atom typed schemes would use a single set of torsion parameters, corresponding to the GAFF atom types c-os-ca-ca. |
631a10083940c20533f81566 | 13 | Instead, BespokeFit groups the torsion parameters using symmetry labels as defined by the bonding topology of the molecule. In the case of the highlighted rotatable bond in aspirin, two unique torsions (9-8-11-6 and 9-8-11-12) are identified due to the terminal atoms (labelled 6 and 12) having different local environments (Figure (left)). As SMIRKS patterns can be more expressive than the combination of predefined atom types, we can create a single bespoke pattern that matches all of the torsions with identical symmetry labels (as determined by RDKit or the OpenEye toolkit ). This allows BespokeFit to automatically introduce extra parameter flexibility via the splitting of torsion parameters into unique symmetry types, which has been shown to be essential for accurate reproduction of reference potential energy surfaces . Conversely, a similar analysis is shown in Figure for biphenyl, where symmetry labels avoid the unnecessary splitting into separate torsion types and simplify the parameter optimization step. Thus, the use of symmetry labels and SMIRKS patterns in BespokeFit provides an automated means to optimally select torsion types for fitting. |
631a10083940c20533f81566 | 14 | Finally, once the torsion parameters have been fit, the parameters are added back into the main FF library. Traditionally, in atom-typed FFs, introducing a new torsion parameter would require the creation of new, potentially redundant atom types. However, since OpenFF uses separate substructure searches for each parameter type (along with a separate parameter hierarchy), the introduction of new torsion parameters does not complicate other parts of the FF, nor does it create issues with conflicting or complex atom types. Instead, the new parameter is simply placed at the bottom of the hierarchy to ensure that it takes precedence when the exact substructure is identified in future parameter assignment workflows. |
631a10083940c20533f81566 | 15 | BespokeFit interfaces with the TorsionDrive package to automatically perform 1D torsion scans around the targeted rotatable bond in each fragment molecule. Consistent with the methods used to parametrize the underlying FFs, TorsionDrive schedules a series of geometry optimizations via the geomeTRIC package , with the targeted dihedral angle constrained to values drawn from a regularly spaced grid (from -180 • to 180 • ). TorsionDrive makes use of wavefront propagation to re-seed neighboring grid points with new low energy structures, which helps to avoid hysteresis in scans where multiple rotatable bonds are present . |
631a10083940c20533f81566 | 16 | In order to offer flexibility in the generation of the reference data, BespokeFit makes extensive use of QCEngine , which is a program executor and IO standardizer, offering one unified interface to a plethora of QM, ML, MM and semi-empirical computation backends. Due to this unified interface and the plugin nature of QCEngine, new computational reference methods can be rapidly prototyped and used with Bespok-eFit, with no changes to the source code. For example, as more accurate and faster ML potentials become available in QCEngine, it will be trivial to make them available in BespokeFit for next-generation FF fitting. |
631a10083940c20533f81566 | 17 | The development and benchmarking of the valence terms in modern high quality FFs can require execution and collection of thousands of QM calculations , often conducted via complex error-prone workflows involving multiple scripts, file formats and software with little to no provenance. With increased computational power via modern hyper-threaded CPUs or HPCs, FF developers have the ability to generate QC data at an unprecedented scale. However, it is prohibitively time consuming to manually build datasets of the size that will be required in BespokeFit, for example. Distributed compute and database platforms, such as QCFractal , were designed to overcome these issues and make the orchestration and storage of large scale QM calculations trivial. However, the community is currently lacking robust tools to assist in managing the creation, submission, and collection of large datasets, and handling their interaction with public or private QCFractal instances. |
631a10083940c20533f81566 | 18 | Here, we present OpenFF QCSubmit as an open-source framework to curate and schedule large QC datasets, and retrieve them from any QCFractal instance, including the public QCArchive. In particular, the framework aims to define reproducible workflows for the construction of QC datasets with a range of purposes, including but not limited to single point Hessian calculations, global optimizations and torsion drives. |
631a10083940c20533f81566 | 19 | QCSubmit also provides an extensive set of modular workflow components that can be combined in any order to produce unique dataset creation pipelines starting from large collections of input molecules. These components enable common pre-processing operations such as filtering, state enumeration, fragmentation and conformer generation. Figure shows an example of how a 'TorsiondriveDatasetFactory', of the type required for BespokeFit input, can be constructed using a selection of configurable workflow components in a python script. |
631a10083940c20533f81566 | 20 | QCSubmit also facilitates the aggregation of completed calculations across multiple QCFractal instances and datasets into a single local results collection. In line with the dataset creation workflow, results collections can also be processed with many common filtering components such as net charge, element coverage or SMARTS queries, which allows users to construct customized datasets for training or testing of FFs, or even ML models. Datasets can be serialized to JSON and used as a source of provenance. These are lightweight references to the calculations that store only vital information, such as the SMILES , |
631a10083940c20533f81566 | 21 | InChlKey and QCFractal record identification, but can also be used to quickly pull down the associated records to access the raw data. This enables users to do quick local filtering for specific molecule data, including proprietary molecules, without contacting the QCFractal instance until the raw record data is required. Figure gives a broad overview of the modular plugin-based architecture of QCSubmit, which enables efficient large-scale communication with private or public QCFractal instances. |
631a10083940c20533f81566 | 22 | The BespokeFit workflow concludes with the optimization of the torsion parameters to the QM (or alternative) reference data using an interface with the ForceBalance package . ForceBalance iteratively optimizes the FF parameters related to the torsional potential, which in the case of the class 1 additive FF employed here, is described by the following truncated Fourier series: |
631a10083940c20533f81566 | 23 | for each bespoke torsion are assigned using the base OpenFF FF. Extra degrees of freedom are then introduced by fully expanding the periodicities of the torsion term to include all integers from = 1 to 4, and an initial value of zero is given to the associated parameters of newly introduced terms. |
631a10083940c20533f81566 | 24 | where = 1.0 kcal/mol is a scaling factor, and QM ( ) and MM ( ) represent the relative energies of conformations compared to the QM and MM global minima, respectively. A weight factor ( ) controls the contribution of each grid point to the objective function. It is constant up to a first cutoff (1.0 kcal/mol) and then attenuates to the second hard cutoff (10.0 kcal/mol) after which all weights are zero: |
631a10083940c20533f81566 | 25 | Following the methods used to parametrize the base OpenFF, at each grid point, the four atoms forming the targeted dihedral are held fixed, and all remaining atoms undergo a MM relaxation with a harmonic energy restraint of 1 kcal/mol/ Å 2 . This ensures that the overall conformation of the molecule remains close to the QM minimum, while ensuring that the torsion parameters do not have to compensate for deficiencies in the other terms in the force field, such as overly stiff repulsive LJ interactions with nearby atoms in the molecule. |
631a10083940c20533f81566 | 26 | The L1 regularization penalizes the absolute difference between the optimized torsion parameters and their initialized values, |ΔΦ |. Usually, the regularization term prohibits large deviations from the initial parameters. However, over-fitting is less of a concern during bespoke parametrization, and so the default prior widths ( ) on the parameter restraints are increased (from 1.0 to 6.0) so as to not limit the maximum achievable accuracy of the optimization due to poor initial guesses. |
631a10083940c20533f81566 | 27 | To demonstrate the utility of BespokeFit in deriving accurate, bespoke torsion parameters at scale, we have chosen to parametrize the entire dataset of ligands compiled by Wang et al. . The dataset comprises 199 drug-like molecules taken from 8 congeneric series, with diverse chemical moieties and a range of net charges. The dataset is often used to validate FF accuracy in the context of free energy calculations , and a possible avenue to accuracy improvements is through the bespoke parametrization of the torsion parameters. Here, we will show that the accuracy of the MM potential energy surface about each torsion angle can be substantially improved via bespoke parametrization, compared to the base FF. |
631a10083940c20533f81566 | 28 | First, in order to distribute the required QC calculations across multiple HPCs worldwide and store the calculations for public use, we created an OpenFF QCSubmit workflow to process the molecules and create torsion drive datasets compatible with QCFractal. The workflow processed the multiple ligand SDF files as input and fragmented the molecules using the WBO fragmentation workflow component with default settings. |
631a10083940c20533f81566 | 29 | Up to four diverse conformers for each fragment were produced, using OMEGA from OpenEye Table . The performance of Sage and BespokeFit parameters on the fragmented Wang dataset , using the default QC chemistry as the reference method. The final two rows test details of the parameter optimization procedure (with differences in procedure noted in parentheses), as described in the main text. the torsion scans. The workflow resulted in a torsion drive dataset comprising 490 molecular fragments and 671 unique scans. OpenFF QCSubmit was then used to submit the dataset to the public QCArchive instance with two different compute specifications, the default OpenFF QC method (B3LYP-D3BJ/DZVP ), using the PSI4 package , and the GFN2-xTB semi-empirical method (Supporting Information Section S1.1). Once all of the calculations were complete, a local BespokeFit server was set up to generate and optimize bespoke torsion parameters. Each fragment was initially parametrized using the base OpenFF 2.0.0 (Sage), and then torsion parameters were optimized against the chosen reference data (that is, either the default QC or the xTB torsion scans), following the procedure described in Section 2.6. |
631a10083940c20533f81566 | 30 | Once the parameter optimization was complete, we analyzed the accuracy of the new parameters by computing deviations between QM and MM geometries and energy profiles. In particular, starting from the QM optimized geometry for each fragment at each point on the torsion scan, the conformer underwent a full MM relaxation, with only the targeted torsion angle being fixed. The MM energy was recorded at the final relaxed geometry and the root mean square error (RMSE) was computed over the full scan, relative to the QC reference relative energies. The root mean square deviation (RMSD) was also computed between QM and MM relaxed coordinates, and the maximum value across the scan was recorded. |
631a10083940c20533f81566 | 31 | The average RMSE in the energy profiles and RMSD in the relaxed coordinates, for both the base Sage and bespoke FFs, relative to the QC torsion scans are shown in Table . As expected, the bespoke parametrization shows a clear improvement in the energy profiles across the set of 671 torsion scans, with the average RMSE reducing from 1.1 (base Sage) to 0.4 kcal/mol (Sage+BespokeFit). Figure exemplifies the improvement in the potential energy surface using BespokeFit, compared with the QM reference data and base Sage parametrization. It is important to emphasize that we would not expect the energy error using Bespoke-Fit to reach zero, both because of limitations in the FF functional form, and because the default BespokeFit workflow attenuates the contributions of any reference data with relative energy between 1.0-10.0 kcal/mol above the minimum (see Section 2.6). A further twelve representative torsion scans are shown in Figures , and demonstrate a generally very good agreement between BespokeFit and QM torsion scans at low energies, with some larger deviations in the high energy regions. |
631a10083940c20533f81566 | 32 | Finally, Table demonstrates that similar to the base Sage FF, the final relaxed geometries using the Sage+BespokeFit parametrization remain close to the QC relaxed structures after full optimization with all restraints relaxed (with maximum RMSDs around 0.6-0.7 Å). The relaxed MM coordinates are also affected by the other valence and non-bonded terms in the force field, and so again perfect agreement with QM is not expected. |
631a10083940c20533f81566 | 33 | Removing all restraints (Sage+BespokeFit (no restraints)) slightly improves the fit to the QM potential energy surface (0.35 kcal/mol), as might be expected. However, this is at the expense of increasing the distances between QM and MM optimized structures (maximum RMSD between optimized structures increases to 0.75 Å). An alternative scheme that we have implemented allows us to add the RMSD between QM and MM optimized structures directly into the ForceBalance objective function (Supporting Information Section S1.2). This removes the need for weak restraints and adds small improvements in both energetic (RMSE reduced to 0.35 kcal/mol) and geometric (RMSD reduced to 0.57 Å) measures of agreement with QM. For consistency with the base FF, however, we retain the Sage+BespokeFit method, with weak restraints, as the default behavior in BespokeFit. |
631a10083940c20533f81566 | 34 | As we have shown in Case Study 1, the introduction of data generation, curation and sharing tools (QCFractal, QCArchive, QCSubmit), in combination with automated parametrization workflows (BespokeFit), opens the possibility of the routine use of bespoke parameter derivation in applications such as alchemical free energy calculations. In order to validate the use of OpenFF BespokeFit in such drug discovery efforts, we compute here relative binding free energies for a congeneric series of inhibitors of the TYK2 protein parametrized with BespokeFit derived FFs. Furthermore, to highlight the flexibility of the reference data generation via QCEngine, we derive FFs for the ligands from multiple sources, including the xTB semi-empirical method and QC at the OpenFF default specification level (Supporting Information Section S1.1). |
631a10083940c20533f81566 | 35 | The target system is part of the Wang benchmark series discussed in Case Study 1, and so the fragments of each of the 16 ligands have already been processed and reference scans are available in QCArchive . Once again, a BespokeFit optimization server was set up locally and the QCArchive torsion scans were downloaded using the cache update CLI tool. Each ligand was initially parametrized using OpenFF Parsley 1.3.0, and was further processed using the default BespokeFit parametrization workflow as outlined in the Computational Methods. |
631a10083940c20533f81566 | 36 | Due to the congeneric nature of the series, it is easy to see that there is a common core shared between the TYK2 ligands used in this study, and this core is also identified by the fragmentation procedure as being electronically decoupled from the proposed substitutions. To save parametrization time, the internal parameter caching system of BespokeFit was used to process a single molecule (id ejm-31) to generate a set of shared, core parameters. For the remaining molecules, bespoke parameters were only derived for new substitutions, in the presence of the optimized common core parameters. This reduces the parameter optimization time as the three rotatable bonds of the central core, and their corresponding 32 free parameters, do not need to be re-optimized for each of the 16 ligands in the series. geometries (still measured relative to the default QC scans). |
631a10083940c20533f81566 | 37 | Furthermore, rows 3 and 4 in Table correspond to reference data generated using optimized geometries obtained with GFN2-xTB, combined with single point energies used to refine the potential energy surface, using either ANI2x or QM at the B3LYP-D3BJ/DZVP level, respectively. We can see that accuracy gains are possible using this flexible, hybrid approach for reference data generation. In particular, for the B3LYP-D3BJ/DZVP//GFN2-xTB method, the RMSE is significantly reduced from 1.1 to 0.6 kcal/mol, with a total computational cost at a fraction of the full QM torsion drive (relative computational costs of all of these methods have been reported elsewhere ). All of the BespokeFit augmented FFs tested also show a decrease in the maximum RMSD between the QM reference and MM optimized geometries, despite only a small proportion of the total valence terms of each molecule being optimized. The methods shown in Table represent a hierarchy of increasing accuracy as the reference data generation method becomes more expensive. Thus, users are able to optimize the balance between fitting time and accuracy to suit their needs all through a common interface between BespokeFit and QCEngine. |
631a10083940c20533f81566 | 38 | Having established the accuracy of BespokeFit torsion parameters in gas phase scans, we now move on to computing (as a proof-of-principle) the relative binding free energies of the TYK2 series using the base Parsley FF, and two of our bespoke augmented FFs, namely BespokeFit (GFN2-xTB) and BespokeFit (B3LYP-D3BJ/DZVP) from Table . This protein target was chosen as a system for which conventional force fields tend to perform well , and so sampling issues are unlikely to affect the interpretation of the data. The fragment FFs were first combined into a single FF, which can be used to parametrize all molecules in the set using the FF combiner CLI tool of BespokeFit. Relative binding free energies were calculated using a workflow based on pmx , which is described in Supporting Information Section S1.3. |
631a10083940c20533f81566 | 39 | Correlations between computed and experimental binding free energies for the base FF (Parsley) and BespokeFit (B3LYP-D3BJ/DZVP) are shown in Figure along with statistics and 95% confidence intervals as reported using cinnabar (formerly Arsenic) . Reassuringly the base FF, Parsley 1.3.0, performs very competitively achieving the sub 1 kcal/mol accuracy required to efficiently guide a drug discovery campaign. |
631a10083940c20533f81566 | 40 | But the BespokeFit variant of the FF further improves all of the reported statistical measures. In particular, the correlation between the calculated and experimental binding free energy is improved from 0.72 0.87 0.35 to 0.93 0.97 0.84 and the confidence interval is significantly narrowed. These data also indicate that our choice of fragmentation scheme has generated torsion parameters that transfer well from the fragment to the parent, without introducing any irregularities into the FFs. |
631a10083940c20533f81566 | 41 | As a further experiment, we have also re-run the free energy calculations using the BespokeFit FF fit to xTB torsion scans. Interestingly, as we saw in Table , the BespokeFit (GFN2-xTB) FF is intermediate in accuracy, on all measures, between the base FF and the FF fit to the default QC data. For example, the RMS error in binding free energies is 0.64 0.93 0.39 kcal/mol (Figure ). Whether this accuracy hierarchy holds more generally, however, will require further protein-ligand free energy benchmarking. Here, we simply present this case as an example application to protein-ligand binding to show that bespoke torsion parameter fits can be relevant to the accuracy of binding predictions. |
631a10083940c20533f81566 | 42 | Bespoke FF parametrization has the potential to significantly improve the accuracy of binding free energy calculations in drug discovery applications. With increased compute power, resources for data storage and The optimized torsion parameters improve the agreement between our QM reference and MM modelled potential energy surfaces, from 1.1 kcal/mol for the base FF to 0.4 kcal/mol using the default BespokeFit workflow settings. As well as providing a concise base FF, the use of SMIRKS patterns to encode the bespoke torsion terms also means that they are transferable without modification between fragmented molecules and their parents. We have made extensive use of this feature to build bespoke FFs for a congeneric series of inhibitors of TYK2. We find that, while the base Parsley FF (OpenFF 1.3.0) provides competitive accuracy, our BespokeFit FF leads to improvements in all reported free energy statistical measures. |
631a10083940c20533f81566 | 43 | Furthermore, the flexibility of BespokeFit and its interface with QCEngine allows us to investigate the balance between accuracy and speed for a range of QC, semi-empirical and machine learning based reference data generation methods. Further work will be required to investigate the optimal combinations of methods to generate and refine the reference potential energy surfaces , and to determine whether improvements in reference data always lead to improved binding free energy estimates, as we saw here. It seems likely that in some cases sampling, rather than force field quality, may be a limiting factor. |
631a10083940c20533f81566 | 44 | Despite the clear increase in accuracy over the base FF (Table and Figure ), there is still room for improvement in the reproduction of the underlying reference potential energy surfaces. The infrastructure described here will provide a useful resource for experimenting with improved functional forms, such as coupling between valence terms, improved non-bonded models and 1-4 scaling interactions. Finally, the current approach is quite conservative, as the torsion parameters of every rotatable bond are subject to optimization regardless of the initial accuracy of the base FF. Methods to predict the confidence in torsion parameter accuracy may be helpful in determining which angles would benefit from bespoke parametrization to further increase the throughput and efficiency of the workflow. All of the software and data used in the current study are freely available and permissively licensed, and a subset of this BespokeFit workflow focusing on bespoke torsions has also been implemented within the Cresset Flare software . We welcome suggestions from the community for future improvements. |
60c741ea842e654a80db1fa6 | 0 | Structure-based virtual screening (SBVS) uses the 3D structure of a target protein to screen large compound libraries for small molecules likely to bind. "Explicit" SBVS uses protein-ligand docking to predict the binding mode of each compound within the active site, and a scoring function to predict the strength of binding . While it is possible to compute the binding affinity of a compound using more rigorous methods such as free energy perturbation , their computational cost makes them impractical for screening libraries of millions of compounds 7 . To save time, in SBVS, more approximate scoring functions designed to estimate the binding affinity are used. Scoring functions are typically employed for three tasks in SBVS: correctly identifying the binding mode of a ligand (pose prediction or "docking"); classifying molecules as either active or inactive ("virtual screening"); and ranking ligands in order of their binding affinity for a given protein target ("scoring"). Popular protein-ligand docking packages, such as GOLD , Glide 9,10 , ICM 11 , FlexX 12 , Surflex 13 , and the AutoDock family , rely on a single scoring function to perform all three tasks simultaneously. These scoring functions make use of molecular force fields, statistical potentials, or linear combinations of empirical terms to assign a score to a receptor-ligand complex, and are often collectively referred to as 'classical' scoring functions. While docking methods have been used successfully to predict binding modes, and for virtual screening, correctly ranking ligands by their binding affinity for a protein remains extremely challenging . |
60c741ea842e654a80db1fa6 | 1 | Recently, however, the application of machine learning techniques has led to the development of new scoring functions that outperform classical scoring functions in terms of ranking compounds by binding affinity . Like classical scoring functions, these methods use the 3D structure of the protein and an automatically generated binding mode of the ligand to compute structure-based interaction features between the protein and ligand. The binding mode-derived features are then used as inputs to a machine learning algorithm to predict the binding affinity. The features used by such machine learning scoring functions typically focus on capturing interactions between the ligand and the protein, but make limited use of the bulk physical, chemical, and topological properties of the protein and/or the ligand alone. Ligand-based features are widely used to select potential binders in ligand-based virtual screening, and have proven to be very effective in understanding polypharmacological relationships between proteins 24 . |
60c741ea842e654a80db1fa6 | 2 | Although scoring functions such as SFCscore , NNScore 2.0 20 , and those used in AutoDock 4 14 and AutoDock Vina 15 include some features of the ligand, the use of detailed information about the ligand to predict cognate protein-ligand binding affinity remains limited. Perhaps the closest is the field of proteochemometric modeling, in which classification models are constructed using a combination of features describing the protein and the small molecule. This approach has proven to be capable of predicting ligand selectivity 27 and capturing polypharmacology. Motivated by the utility of ligand-based features in virtual screening and proteochemometric modelling, we investigated whether a more detailed representation of the ligand can improve the ability of a scoring function to predict its binding affinity. Using the cheminformatics toolkit RDKit (, accessed 17/05/2019) we computed a diverse set of 1D and 2D ligand molecular descriptors and combined these with the structure-based features used by the machine learning scoring functions RF-Score 19 , RF-Score v3 29 , NNScore 2.0 20 , as well as the empirical scoring function of AutoDock Vina 15 . We show that a Random Forest (RF) regression model using both structure-based and ligand-based features consistently outperforms a model using only structure-based features when benchmarked on three versions of the PDBbind 31 core set, corresponding to the scoring power test of the three Comparative Assessment of Scoring Functions 17, . |
60c741ea842e654a80db1fa6 | 3 | We find that removal of test-set similar proteins and ligands from the training sets degrades performance but does not abrogate the predictive power of ligand-based features. Furthermore, we show that a model using only ligand-based features appears to be predictive of the mean affinity of a ligand for its binding partners when trained and tested on PDBbind data. We computed the relative importance of the features used by each model and show that when structure-based and ligand-based features are combined, both structure-based and ligand-based features are among the top-ranked features. These results suggest that quickly-computed ligand-based features should be used to improve the ability of a machine learning scoring function to predict proteinligand binding affinity. |
60c741ea842e654a80db1fa6 | 4 | The PDBbind database 31 is a curated set of bound macromolecule structures drawn from the Protein Data Bank (PDB) , each with an experimentally-measured binding affinity for its binding partner. Each release of PDBbind includes a "general set", which contains all the protein-ligand structures in the database; and a "refined set", a subset of protein-ligand complexes satisfying strict criteria concerning structure quality, affinity data reliability, and the nature of the complex. The 2018 release of PDBbind contains 16,151 protein-ligand complexes in the general set, with 4,463 complexes in the refined set. We used the refined set as our primary source of training data; however, it has been reported that including the lower-quality data comprising the remainder of the general set can still improve the performance of machine learning scoring functions , so we repeated our analysis using the general set as our source of training data. |
60c741ea842e654a80db1fa6 | 5 | To validate our models, we used a subset of the PDBbind refined set referred to as the "core set". This is obtained by clustering the proteins in the refined set at 90% sequence identity and selecting three or five (depending on the version of PDBbind) representatives of each cluster for which the corresponding ligands have a broad range of binding affinity values, resulting in a diverse, non-redundant set of protein-ligand complexes. We repeated our tests using the core sets from the 2007, 2013, and 2016 releases of PDBbind. These versions of the core set were used as the "scoring power" benchmark in the Comparative Assessment of Scoring Functions (CASF) exercises: CASF2009 32 , CASF2013 17,18 , and CASF2016 33 respectively, allowing our results to be compared directly to previously published scoring function benchmarks. We used the test sets corresponding to all three CASF benchmarks since the contents of each test set are substantially different to the others and so each set offers a different challenge for a scoring function. Excluding proteins and ligands that could not be parsed by OpenBabel or RDKit resulted in 2007, 2013, and 2016 core sets containing 196, 180, and 276 structures respectively. The full list of structures omitted from the core sets because of parsing failures is included in the Supporting Information. These test sets are relatively small, making it difficult to identify statistically-significant differences in results generated by different models. To partially overcome this shortcoming without deviating from widely-used benchmarks, we combined the structures from each core set into a fourth test set with duplicate strictures removed. This combined test set numbered 525 structures, almost twice the size of the PDBbind 2016 core set. |
60c741ea842e654a80db1fa6 | 6 | PDBbind provides for each complex an experimentally-determined value of the inhibition constant K i , the dissociation constant K d , or the half-maximal inhibitory concentration IC 50 , in decreasing order of preference (e.g. if both K i and K d values are available, PDBbind reports the measurement of K i ). The refined set includes only measurements of K i and K d , while the general set also includes data for which only IC 50 measurements were available. For our purposes, these values are used interchangeably and are collectively denoted by the binding constant, K. We used the negative base-10 logarithm of K, commonly denoted as pK: pK = -log 10 K |
60c741ea842e654a80db1fa6 | 7 | We evaluated each scoring function by computing the Pearson correlation coefficient, ρ p , between its predictions { ŷ } and the experimental values { y } of pK for the complexes in the test set. Confidence intervals were estimated at the 95% significance level using 10,000 bootstrapped samples of the predictions. The Mann-Whitney U test was used to compare the distribution of bootstrapped ρ p values to assess the significance of the differences in correlation coefficient between two scoring functions. We also performed a permutation test with 10,000 samples to assess the possibility that correlations arose by random chance. |
60c741ea842e654a80db1fa6 | 8 | To represent the ligand in our models we used a diverse set of molecular descriptors computed using the cheminformatics toolkit RDKit. Using the Descriptors module of the Python RDKit package version 2018.03, we computed a set of 200 molecular descriptors for each ligand. These descriptors are conformation-independent and may be categorized as either (computed) experimental properties (e.g. molar refractivity, logP) or theoretical descriptors derived from a symbolic representation of the molecule. The theoretical descriptors may be further categorized according to the dimensionality of the representation of the molecule from which they are derived. The conformer-independent descriptors we consider are either 1-D compositional properties (e.g. heavy atom counts, bonds counts, and molecular weight) or 2-D topological properties (e.g. fragment counts, topological polar surface area, and connectivity index). Any features with zero variance across the data set, or that were null-valued (i.e. infinite or not computable) within the data set were excluded. We removed the Ipc index (an information theory-derived descriptor) as it produced extreme numerical values for larger molecules (too large to be represented as 32-bit floats). In total, 185 RDKit descriptors were retained. We refer to this set of ligand-based features as "RDKit descriptors" throughout this work. |
60c741ea842e654a80db1fa6 | 9 | To investigate the effects of augmentation with ligand molecular descriptors, we considered the features of several publicly-available machine learning scoring functions, namely RF-Score 19 , RF-Score v3 , and NNScore 2.0 20 . Both RF-Score v3 and NNScore 2.0 include the six terms used by the AutoDock Vina scoring function 15 . We therefore considered the AutoDock Vina terms separately to examine the effect of combining ligand molecular descriptors with just the terms used by a classical empirical scoring function. We computed the features of each of these scoring functions using the implementations provided by the Open Drug Discovery Toolkit (ODDT) version 0.6 . |
60c741ea842e654a80db1fa6 | 10 | For each of the four scoring functions considered (AutoDock Vina, RF-Score, RF-Score v3, and NNScore 2.0) we constructed two sets of features. The first used only the original features of the scoring functions, while the second used the original features of the scoring function, plus the 183 RDKit descriptors of the ligand. Since AutoDock Vina (and hence RF-Score v3 and NNScore 2.0) already use the number of rotatable bonds of the ligand, we dropped this from the set of RDKit descriptors added to avoid including the same feature twice. Finally, to examine whether there is any signal in the RDKit descriptors independent of the structure-based features, we constructed models using only the RDKit descriptors as a separate feature set, resulting in a total of nine different sets of features. |
60c741ea842e654a80db1fa6 | 11 | For each set of features and each training set, we built a scoring function by using Random Forest 30 (RF) regression to fit an estimator for the pK of a protein-ligand complex. We used the implementation of RF in the Python machine learning library scikit-learn . Although RF is generally robust with respect to hyperparameter choice, we tested the effect of varying the number of trees in the forest (n estimators) and the maximum number of features considered at each split (max features). We chose to set n estimators=500 and max features=0.33 as these values yielded optimal out-of-bag performance (Supporting Information Fig. ). We tested several other ML algorithms but found that RF consistently achieved the best cross-validation scores (Supporting Information Fig. ; 6; 7). |
60c741ea842e654a80db1fa6 | 12 | We investigated the affinity predictions for ligands that are found in multiple structures in the PDBbind database. We clustered the structures of the PDBbind 2018 general set by the three-character chemical ID of the ligand as specified in the PDB, and selected all ligands found in at least three structures. Holding out each ligand in turn as a test case, we trained a RF regression model using only the RDKit descriptors on all structures not containing that ligand, and used the resulting model to predict the affinity of that ligand. We repeated the above process for each ligand by removing from the training set all ligands with a Tanimoto similarity to any test set ligand above a defined threshold. We then trained a new RF regression model and predicted the affinity of the test ligand. |
60c741ea842e654a80db1fa6 | 13 | Ligand-Based Features Improve ML Scoring Functions Vina terms and RDKit descriptors. This combination results in predictive performance comparable to that of the RF-Score, RF-Score v3, and NNScore 2.0 features, suggesting that a more complex and detailed set of features describing protein-ligand interactions is not necessarily more predictive than a comparatively simple set of forcefield-like terms (AutoDock Vina terms) and molecular descriptors of the ligand (RDKit descriptors). |
60c741ea842e654a80db1fa6 | 14 | The performance of the scoring functions using each feature set when trained on different versions of the PDBbind refined set with no data removed is shown in Fig. , while Fig. shows corresponding results when structures whose ligands have a high Tanimoto similarity to any ligand in the test set are removed from the training set. In both of these cases there is no consistent improvement in performance when larger, more recent versions of the refined set are used. This is contrary to the results of Li et al. and may suggest that, at least when using RF to train an estimator, there is an element of 'learning saturation', beyond which there is limited benefit to using additional training data. However, for a given release of PDBbind, a scoring function trained on the general set outperforms the same scoring function trained on the refined set (Supporting Information Fig. -13), consistent with the findings of Li et al.. This difference in performance vanishes for all scoring functions when structures with testset similar ligands are excluded from the training set, and is greatly reduced when structures with 90% protein sequence identity to those in the test set are excluded. This suggests that the increase in performance when training on the general set can be attributed to increased representation of the core set proteins and ligands in the training data. |
60c741ea842e654a80db1fa6 | 15 | Exclusion from the training set of test-set similar proteins results in significantly reduced scoring function performance (Fig. ). There is a significant drop in the performance of all scoring functions even when a sequence identity cutoff of 100% is imposed, i.e. when only proteins with identical sequence to those in the test set are excluded from the training set. Reducing the sequence identity cutoff from 90% to 50% has a smaller impact on performance than the initial imposition of a 100% cutoff. |
60c741ea842e654a80db1fa6 | 16 | Further reducing the cutoff from 50% to 30% has a more apparent effect. Fig. shows similar behaviour when structures with test-set similar ligands are also excluded from the training set. As in the case where similar proteins are not excluded (Fig. ), the exclusion of structures containing test-set similar ligands has a consistently deleterious effect on the performance of all scoring functions, regardless of whether they make use of the RDKit descriptors. The performance of each scoring function on each of the 2007, 2013, and 2016 core sets, as well as bootstrapped 95%confidence intervals for the value of ρ p when training on the PDBbind 2018 refined set are included in the Supporting Information (SI Fig. ). For all scoring functions under all training and test scenarios, we reject the null hypothesis that the correlation between predicted and experimental affinity was due to random chance (permutation test p < 0.05). |
60c741ea842e654a80db1fa6 | 17 | Using a RF with only RDKit descriptors consistently results in a scoring function with greater performance than a RF with the AutoDock Vina terms, and is often close to or even exceeds the performance of RF-Score, even when test-set similar ligands are excluded from the training data. This is surprising, since we would not expect to be able to predict protein-ligand binding affinity across a diverse set of proteinligand complexes without knowing which protein the ligand was assayed with. We discuss the possible source and interpretation of this signal next, but this observation may suggest that the redundancy between the PDBbind core set and the remainder of the database makes this particular approach to training and test set construction inappropriate for validating and benchmarking machine learning scoring functions for predicting protein-ligand binding affinity. |
60c741ea842e654a80db1fa6 | 18 | Given the success of the RDKit descriptors alone (Fig. ) we examined the affinity predictions for ligands that bind to proteins in multiple structures in the PDBbind database. Our RDKit RF model will produce only one value so it must be "incorrect" for many of the protein-ligand complexes. When the RDKit RF model was tested on a previously-unseen ligand, having been trained on all other data in the PDBbind 2017 general set, the score was found to be strongly correlated with the mean experimental pK of that ligand for its targets across the PDBbind 2017 general set (ρ p = 0.72, Fig. ). For the most common ligands in the PDBbind 2017 general set, the reported affinity values can span several orders of magnitude, so the RDKit RF model is not simply predicting a single "correct" value for a ligand that happens to have many similar affinity measurements. Fig. shows the predictions of the RDKit RF model for the most common ligands in the PDBbind 2017 general set (markers) against the pK values reported by PDBbind (swarm plots). With the clear exception of biotin (BTN), the marker is often close to the centre of the swarm plot, indicating that the RDKit RF model appears to be predicting the mean affinity of the ligand even when the experimental data have a range of several pK units. We verified that many of the structures for biotin were biotin-streptavidin or biotin-avidin complexes, explaining the Further, we found that when nearly-identical ligands (Tanimoto similarity > 0.9) were excluded from the training data, this correlation is actually stronger (ρ p = 0.75) (Supporting Information Fig. ). Reducing the Tanimoto similarity above which ligands are excluded from the training set results in gradually weakening correlation (Supporting Information Fig. ), suggesting that the RDKit RF model is not simply memorizing the average affinity of highly similar training ligands. fore expect these features to capture useful information about the ability of a small molecule to bind to a charged, buried active site. Both these properties are also used to characterize drugs and log P is also used to predict bioavailability . It is possible that the predictive power of these features can in part be attributed to systematic bias in favour of crystallising complexes featuring high-affinity engineered compounds. However, we found that there was no trivial correlation between either of these fea- whose contribution to the model might also be through capturing the shape and surface area of the molecule or some aspect of conformational entropy. We also found that when ligand-based features were combined with structure-based features in our other models, both ligand-based and structure-based features were ranked highly, and that the same ligand-based features were consistently found to be important regardless of which structure-based features were used (Supporting Information Fig. ). This suggests that the ligand-based features are consistently capturing useful information that is not present in the structure-based features, beyond the count of rotatable bonds in the ligand. |
60c741ea842e654a80db1fa6 | 19 | Varying the composition of the training set chronologically, by restricting to data available only up until a particular year, had little effect on affinity predictions. This suggests an element of learning saturation for the targets tested with the data currently available. We showed that, in contrast, excluding proteins from the training set that are sequence-similar to those in the test set has a deleterious effect on affinity predictions and that even excluding only those proteins with identical sequence to those in the test set leads to significantly reduced scoring function performance. We also showed that even when ligands with high Tanimoto similarity to those in the test set were excluded from the training set, the predictive power of the scoring functions was still increased by including ligand-based features. |
60c741ea842e654a80db1fa6 | 20 | Given the power of the ligand-based features, we investigated their predictive ability for ligands that bind to multiple targets and found that the predicted binding affinity of a model using only ligand-based features was strongly correlated with the mean of the experimental protein-ligand binding affinity of a ligand for its binding partners. This correlation remained strong when ligands with a Tanimoto similarity of greater than 0.9 to the test ligand were excluded from the training data. This correlation gradually weakened when progressively less similar ligands were also excluded, suggesting that while the model's predictions are not reliant upon memorization of previously-seen highly-similar ligands, it does not extrapolate well to completely novel ligands. |
60c741ea842e654a80db1fa6 | 21 | We found that when structure-based and ligand-based features are combined, both structure-based and ligand-based features were ranked highly, and that the same ligandbased features are ranked highly regardless of which structure-based features they were combined with. This suggests that the same information is consistently extracted from the ligand-based features and that this information is not redundant with that provided by structure-based features. Our results suggest that even under stringent validation, the addition of a diverse, quickly-computed set of ligand-based features to a scoring function yields improved predictions of binding affinity. |
6399f5ab963bf319949964da | 0 | Tetrathiafulvalene (TTF) and its numerous derivatives are among the most versatile electroactive organic molecules due to their remarkable redox properties, their use as building blocks for the development of molecular conductors, and diverse applications in organic electronics. TTF derivatives have also been widely explored as molecular switches, sensors, rectifiers, transistors, and for nonlinear optics, among other uses. More recently and since the pioneering study reporting the palladium-catalysed direct C-H arylation, tetraarylated TTFs have been employed for the construction of functional porous frameworks such as Metal-Organic Frameworks (MOFs), Covalent Organic Frameworks (COFs) or Hydrogen-bonded Organic Frameworks (HOFs). However, direct C-H arylation has been limited to TTF derivatives, and the catalytic peripheral modification of other π-extended TTF cores is encouraged for the construction of electroactive frameworks exhibiting new electronic properties and functions. |
6399f5ab963bf319949964da | 1 | Among TTF derivatives, dithiophene-tetrathiafulvalene (DT-TTF) (Scheme 1) received considerable attention as an active material for organic field-effect transistors (OFETs) due to its high charge carrier mobility. In contrast to neutral TTF which exhibits a twisted boat-like conformation, the DT-TTF molecule is completely planar in the solid state, and its oxidation potential (E°1 ox = 0.78 V, E°2 ox = 0.96 V vs SCE) is higher than that of TTF leading to improved ambient stability. In order to optimise the OFET performance, one strategy consists of adjusting the electronic structure of the electroactive molecules to align their HOMO energy level with the work function of the electrode (e.g. 5.1 eV for gold). In this sense, the incorporation of electron-withdrawing or electron-donor groups is an efficient strategy to fine-tune the HOMO and LUMO energy levels. However, so far, no studies on the direct functionalization of DT-TTF building blocks allowing the modulation of its electronic structure have been reported. Scheme 1. Molecular structures of TTF and DT-TTF. Given the importance of TTF and DT-TTF based systems in organic electronics, the study of their organisation in thin films such as H-bond based self-assembled molecular networks (SAMNs) adsorbed on solid substrates is desired. The use of scanning tunnelling microscopy (STM) allows the direct visualisation of such SAMNs, enabling the study of their domain size, presence of defects as well as intermolecular bonding landscape. In particular, the self-assembly of TTF derivatives on conductive substrates such as highly oriented pyrolytic graphite (HOPG) has been studied previously. Herein we report the synthesis and characterisation of a family of arylated DT-TTF derivatives 2-5 (Scheme 2) obtained via palladium-catalysed direct C-H arylation. Monoarylation of DT-TTF introducing -NO 2 (2) or -NMe 2 (3) functional groups results in a decrease and increase of HOMO energy levels, respectively. Furthermore, the corresponding tetraarylated DT-TTFs bearing ester (4) or acid (5) groups have also been synthesised, both of them displaying a non-planar configuration as revealed by density functional theory (DFT) calculations. STM studies revealed that the DT-TTF-tetrabenzoic acid derivative 5 forms densely packed 2D self-assembled networks at the heptanoic-acid/graphite interface. Furthermore, molecular models based on experimentally measured unit cell parameters suggest that 5 lies flat on the graphite surface due to van der Waals interactions with the surface as well as H-bonding with neighbouring 5 building blocks. |
6399f5ab963bf319949964da | 2 | DT-TTF was synthesised according to previously described procedures. This six-step synthetic route provides a higher yield than those previously reported (see ESI †). Arylated DT-TTF derivatives were obtained by optimising a previously reported synthetic procedure based on the palladium-catalysed direct C-H arylation of aromatic compounds with aryl halides. In this study, the aryl bromides used in the reaction were functionalised with electron-withdrawing -NO 2 (2) or electron-donating -NMe 2 (3) groups to evaluate their influence on the DT-TTF electronic properties (HOMO and LUMO energy levels). The reaction of DT-TTF (1) (1 equiv.) with aryl bromides (1 equiv.) in the presence of cesium carbonate, tri-tert-butylphosphonium tetrafluoroborate and palladium acetate (5 mol%) in THF under reflux yielded the monoarylated derivatives 2 and 3 in 13% and 85% yields, respectively (Scheme 3 and ESI † for experimental details). The lower yield in the case of using electron-withdrawing aryl bromides such as Br-Ar-NO 2 has also been observed in other direct C-H arylation reactions under similar conditions. 33 |
6399f5ab963bf319949964da | 3 | The electrochemical properties of DT-TTF derivatives 1-4 were investigated by cyclic voltammetry (CV) using n-Bu 4 NPF 6 (0.15 M) as electrolyte in CH 2 Cl 2 at room temperature (Fig. ). Cyclic voltammogram of unsubstituted DT-TTF shows two redox processes assigned to the oxidation to the radical cation DT-TTF •+ (E ox1 = +0.29 V vs Fc/Fc + ) and dication DT-TTF 2+ (E ox2 = +0.74 V vs Fc/Fc + ) species. As expected, the first (E ox1 = +0.38 V vs Fc/Fc + ) and second (E ox2 = +0.83 V vs Fc/Fc + ) oxidation potentials increase in 2 in comparison to the unsubstituted DT-TTF 1 due to the incorporation of electronwithdrawing -NO 2 functional group. On the contrary, oxidation potentials of 3 (E ox1 = +0.27 V; E ox2 = +0.73 V vs Fc/Fc + ) slightly decrease when introducing electron-donor groups indicating the increase of the HOMO level. Oxidation potentials of 4 also show positive shifts relative to those of 1 which is consistent with the presence of four ester groups. The HOMO energy levels of 1-4 were estimated from the onset of the first oxidation redox processes (see Table ). CV of 5 using n-Bu 4 NPF 6 (0.15 M) as electrolyte in DMF (Figure , ESI †) shows two weak redox peaks at positive potentials which decrease in intensity after consecutive cycles, possibly due to the irreversible decarboxylation process upon oxidation originated by the Kolbe reaction, as it has been previously reported for similar TTF derivatives bearing carboxylic groups. UV-vis spectra of DT-TTF derivatives 1-4 were measured in CHCl 3 (0.05 mM) at room temperature (Fig. ). Whereas absorption spectra of 1 and 3 remain similar, the absorption bands of 2 and 4 are redshifted due to the electron-withdrawing character of the nitro and ester substituents. The optical gaps for 1-4 were calculated from the onset of the most red-shifted band of the UV-vis spectra and were, in turn, used to estimate the energy of the LUMOs (Table ). |
6399f5ab963bf319949964da | 4 | All compounds were systematically explored with DFT models at the B3LYP-6-31G(d,p) level revealing that the DT-TTF moiety adopts a boat conformation with a low stabilization energy with respect to its planar counterparts of circa 1 kcal/mol for 1 and 2 kcal/mol for 2 and 3. This low energy matches the fact that 1 is flat in the solid state while in vacuum and in CH 2 Cl 2 continuum the presence of imaginary frequencies indicates this is not a minimum (Fig. ). For 4 and 5, the more extensive functionalization implies a larger energetic penalty, 18 kcal/mol, when compared to planar conformations. This is due to the increased number of functional groups and the larger steric hindrance from the extra methyl groups. The HOMO and LUMO energy levels and optical gap were computed at the B3LYP-6-311+G(2d,p)-CH 2 Cl 2 /B3LYP-6-31G(d,p)-CH 2 Cl 2 and give trends in reasonable agreement with the experimental observations (Table ). DFT results on both conformations of 1 show how the effect on the electronic properties on conformation is far from negligible. In addition, 2, 4, and 5 show a strongly reduced density on the TTF DT-TTF moiety (Fig. ) which accounts for the large stabilization estimated experimentally. The HOMOs densities are more homogenous with 1, 2, 4, and 5 showing very similar densities localized on the DT-TTF moiety and only 3 showing a different density. |
6399f5ab963bf319949964da | 5 | Compound 5 was selected for characterization using STM due to the presence of carboxylic groups on the DT-TTF backbone which are known to facilitate the formation of hydrogen-bonded selfassembled networks. The self-assembly of 5 was studied using STM at the heptanoic acid/graphite interface. 5 forms long-range ordered crystalline networks at the heptanoic acid /graphite interface with domains extending over 150 nm. Large scale STM images show wellordered bright features ordered in a pseudo hexagonal arrangement (Fig. and S21, ESI †). Smaller scale STM image presented in Fig. shows a well-ordered arrangement of bright features which can be assigned to the electron rich DT-TTF core of 5. Unit cell parameters extracted from graphite calibrated STM images are a = 1.9 ± 0.1 nm, b = 2.3 ± 0.1 nm, γ = 62° ± 1°, which suggests the existence of a distorted hexagonal network. A molecular model built using calibrated STM data is presented in Fig. . The molecular model reveals that 5 forms one-dimensional hydrogen bonded chains using two diagonally opposite carboxyl groups present on the DT-TTF backbone. The remaining two carboxyl groups are thus not engaged in hydrogen bonding with neighbouring DT-TTF molecules and most likely interact with partially adsorbed molecules of heptanoic acid which was used as a solvent. The molecular model also reveals substantial empty areas between the hydrogen bonded 1D arrays of 5. These areas are possibly filled with partially/transiently adsorbed heptanoic acid molecules which are mobile on the time scale of STM measurements. The observation of partial hydrogen bonding between the molecules of 5 is not completely surprising. It is well known that molecular materials, especially those forming porous networks do not always scale with the size of the building blocks. This is because increased pore sizes are thermodynamically disfavoured and go against the natural tendency of close packing. If the molecules of 5 were to form a 2D network based on hydrogen bonding using its all four carboxyl groups, the resultant cavities would be too big to be stable. Thus, the system must make a compromise between forming strong two-fold hydrogen bonds between the remaining carboxyl groups and close packing. Under the experimental conditions used, the latter seems to be the choice. Such scalability study has been previously reported for self-assembled networks of aromatic tricarboxylic acids. Considering that intermolecular H-bonds formed between carboxylic acid groups are highly directional, it is likely that 5 reaches a planar geometry to form such SAMNs, as supported by the molecular model shown in Fig. as well as in further calculations shown in Fig. ). This planarization is the result of van der Waals interactions between the HOPG surface and the DT-TTF core extended with the phenyl groups. |
6399f5ab963bf319949964da | 6 | General procedure for the synthesis of 5. A 50 mL flask was charged with DT-TTF-(PhMeCOOEt) 4 (4) (100 mg, 0.10 mmol) and subjected to three cycles of evacuation and refilling with N 2 . Degassed MeOH (4 mL) and anhydrous THF (4 mL) were added to generate a suspension. In a separate flask, NaOH (50 mg, 1.1 mmol) was dissolved in degassed water (4 mL). The sodium hydroxide solution was added to 4 under nitrogen and the reaction was heated to reflux for 12 hours. The reaction was then cooled to room temperature and the volatiles were removed in vacuum. A 1M solution of HCl (5 mL) was added to afford an orange precipitate, which was collected by filtration and washed with water (20 mL) and CH 2 Cl 2 (20 mL). The product was collected and dried under vacuum for 12 hours to afford 5 as an orange solid (70 mg, 94 % yield). Characterisation: |
6399f5ab963bf319949964da | 7 | All reagents and solvents were of high purity grade and were purchased from Sigma-Aldrich Co., and TCI. H liquid NMR spectra were recorded on a Bruker AVANCE 300 spectrometer (300 MHz). Dimethylsulfoxide-d 6 (DMSO-d 6 ) was used as solvent. Tetramethylsilane was used as the internal reference. Chemical shifts (δ) are quoted in ppm and the coupling constants (J) in Hz. Infrared spectra were recorded on an ATR FT-IR GALAXY SERIES FT-IR 7000 (Mattson Instruments) spectrometer in the 4000-400 cm -1 range using powdered samples. UV/visible absorption spectra were recorded on a Jasco V-780 spectrometer. TLC analyses were performed on commercial glass plates bearing a 0.25-mm layer of Merck Silica gel 60F254. Preparative separations were performed by silica gel chromatography (Wako gel C-200). The electrochemical experiments were performed on an Autolab electrochemical workstation (PGSTAT302N with FRA32M Module) connected to a personal computer that uses Nova 2.1 electrochemical software. A typical three-electrode experimental cell equipped with a glassy carbon as the working electrode, platinum wire as the counter electrode, and silver wire as the pseudoreference electrode was used for the electrochemical characterization. The electrochemical properties were studied by measuring CV under an inert (N 2 ) atmosphere at different scan rates in previously N 2 -purged 0.1 M TBAPF 6 /CH 2 Cl 2 solutions. Ferrocene (Fc) was added as an internal standard upon completion of each experiment. All potentials are reported in V versus Fc/Fc + . DFT calculations. DFT calculations were performed with Gaussian 09. For this, different molecular structures with different symmetry groups were generated and minimized at the B3LYP-6-31G(d,p) and the B3LYP-6-31G(d,p)-CH 2 Cl 2 levels. All minima were checked for the presence of imaginary frequencies. In addition, selected structures were also reoptimized at the M06-2X-6-31G(d,p)-CH 2 Cl 2 giving similar results. The electronic structure, HOMO and LUMO levels, and TD-DFT calculations were performed at the B3LYP-6-311+G(2d,p)-CH 2 Cl 2 level. Vertical ionization potentials and electron affinity calculations were also performed yielding very similar values to the HOMO and LUMO energies reported. |
6399f5ab963bf319949964da | 8 | Scanning tunnelling microscopy experiments were carried out employing a PicoLE (Keysight) or Molecular Imaging (Agilent Technologies) STM operating in constant-current mode at room temperature. STM tips were mechanically cut using Pt/Ir wire (80/20, diameter 0.25 mm, Advent Research Materials). Prior to every STM experiment at the solid/liquid interface, highly oriented pyrolytic graphite substrate (grade ZYB, Momentive Performance Materials) was rinsed with acetone and freshly cleaved using Scotch tape. After this, the solution was drop casted on the substrate and started scanning at room temperature for a designated time ranging from 3-8 h. 1 mg of solid was weighed and added to 1 mL solvent i. e. asreceived heptanoic acid (Merck). Sonication was used to dissolve the materials which resulted in a saturated solution. After this, a 10% dilution was created in the same solvent. STM imaging parameters such as I set and U bias are shown in each figure and correspond to the tunnelling current and applied substrate bias. STM images were graphite-calibrated using Scanning Probe Imaging Processor software (SPIP, Image Metrology). Molecular models were created using HyperChem™ software (version 8.0.1) as described in the Electronic Supplementary Information (ESI). |
6399f5ab963bf319949964da | 9 | We have reported a simple synthetic procedure based on Pdcatalysed direct C-H arylation to obtain a new family of DT-TTF derivatives bearing different functional groups (2-5). The electronic structure of DT-TTF can be modulated by introducing electron-withdrawing (2) or electron-donor (3) groups to decrease or increase the HOMO energy level, respectively, becoming very attractive for organic electronics. In addition, long-range ordered crystalline hydrogen-bonded networks of DT-TTF derivative 5 were observed at the graphite/liquid interface as revealed by STM studies. Interestingly, molecule 5 can reach a planar geometry on the graphite surface due to van der Waals interactions with the surface as well as H-bonding with neighbouring molecules, as supported by molecular models based on experimentally measured unit cell parameters. This work will also open the way to explore the design of new extended electroactive frameworks such as MOFs or COFs based on DT-TTF building blocks, being of great interest due to their promising electronic properties. |
66cbb4baa4e53c48769f3dfd | 0 | Nanotechnology fields are focusing to innovate new performance nanomaterials for several engineering fields to meet the mankind needs. Such fields concern materials engineering and catalysis as they are known as central sciences, which deeply contribute to the nation building and development through nanotechnology growth. As known the protection and cleaning our environment from both inorganic and organic wastes is highly needed . the development of intelligent nanomaterials are so regarded in fuel cells technologies , conversion of both harmful and harmless products , in hydrogen production , and so fourth. Nowadays, many hybrid materials are elaborated and characterized to allow several applications,such as in engineering technologies, biomedical applications, medicinal applications to treat such as cancer, diabetes, virus, and so on. These nanomaterials include carbons, transition metals, supported transition metals, alloys, minerals in combination with other metals, core-shell materials, metal organic frameworks, zeolites, with diverse method of preparations. As known also, many elements from the rare-earth materials might possess advanced properties such as magnetic characteristic (Neodymium) , lighting features (La), constituent of an exceptional sort of glass (Praseodymium), MRI, X-ray scanning method, and television screens (Gadolinium), catalytic at high temperature (Cerium/La), optical displays in devices (Yttrium, terbium, europium). So, in recent years researchers worldwide are exploring this class of intelligent materials in order to create an advanced nanomaterials for diverse technologies . Moreover, in the past few years, a new natural material has been introduced, this material is coming from a meteorite component and it is named as tetrataenite which has a chemical formula of (NiFe), it is an alloy. This new material has properties as the transition rare-earth materials, it possesses a huge magnetical properties. This material may possess advanced properties,which could be advantages for future catalysis, engineering, and technologies applications. Interesting composite materials may be engineered in combination with other materials as we know synergistic effect is playing interesting role in materials sciences applications. Therefore, we dedicate here a brief review in the recent applications of nanomaterials designed from the inspiration of tetrataenite features in the engineering fields. |
66cbb4baa4e53c48769f3dfd | 1 | As demonstrated in the litterature, tetrataenite is a favorable materials coming from a mineral meteorite, NiFe is his chemical composition. First, in 1988 Wasilewski, P did a characterization of the magnetic mineral tetrataenite and its distinction with isochemical taenite. So, it is described that taenite with [48-57] % of nickel is magnetically soft, however when changed through atomic ordering below 320 °C it becomes the novel magnetic mineral tetrataenite along with appreciable magnetic hardness and other singular magnetic properties. Also, it noted that tetrataenite is omnipresent in chondrite meteorites, happen as discrete grain, as a distinct taenite rim, and in the etched nebulous zone. So, here Haikal,R.R., et al, reported an approach to synthesized hybrid ultramicroporous material (HUM) precursor to graphene-embbeded tetrataenite as superb highly active and durable non-precious group (NPG) metals for the electrocatalytic oxygen evolution reaction (OER). Indeed, the authors well demonstrated the improvement in the properties of the designed materials. So, after Raman spectroscopy analysis, it was founded that the ratio (ID/IG) is about 1.562 of the graphitic carbon in the designed materials while normally the unmodified graphene or ordered graphite has (ID/IG) ratio of around [1-1.2]. thus this indicate a higher degree of disorder in the synthesized catalyst. In fact, the elaborated materials (NiFe@G) shows outstanding properties for the oxygen evolution reaction (OER), such as overpotential (0.21 V vs RHE), low OER onset potential (1.44 V vs RHE and a reduced overpotential at 10 mA/cm 2 (ƞ10 = 266 mV). The electrochemical characterization of the evaluated materials in 1.0 M KOH media are shown in Fig. , which shows a huge different in the considered materials with greater performance of NiFe@G for the OER. Also, the X-Ray photoelectrons Spectroscopy analysis of NiFe@G revealed the well incorporation of all considered elements in the catalyst structure as fully shown in the Fig. . Moreover, both atomic force microscopy (AFM) and transmission electron microscopy (TEM) were considered, the TEM analysis exhibits that there is a continuous and homogenous coatings of graphene (G) with the alloy (NiFe). It seems as plates of the NiFe deposited on the graphene surface. Recently, Wu, H., et al studied an amorphous Fex(NiCu)3-x-MOFs and proposed it as a potential candidate catalyst for the OER in alkaline media. So, the optimal ternary FeNiCu-MOF exhibited higher current density of 10 mA/cm 2 merely at low overpotential as low as 260 mV with a Tafel slope of 61 mV/dec along with superior durability over 24 h. In particular, it is suggested that DFT calculation proved that the iron (Fe) sites are the active sites for the OER in the the optimal proposed catalyst. This material (FeNiCu-MOF) surpass the material proposed by Haikal and his colleague by -6 mV in terms of overpotential. This might become to the copper contributions and as well as the MOF structure. This shows that the synergistic effect between catalysts' constituent is more interesting in material sciences and engineering technologies. 2 showing the stability of the NiFe@G catalyst in 1 M KOH at a 100 mV/s for CVs and 10 mV/s for the LSVs, Reprinted with permission from copyright @ 2024, Royal Society of Chemistry. |
66cbb4baa4e53c48769f3dfd | 2 | In 2015, Baidya, T., et al reported a novel Ni-based mixed oxide nanomaterials (Ni-Fe-CaO) to catalyze tar removal from producer gas. So, it is shown that 1.6% (Ni40Fe20Ca40) composition exhibits a stable conversion of about 80% at the end of 50 h, while 1% (Ni65Fe35)/Carbo HSP catalyst shows much lower yields as compared to the results of 1.6% (Ni40Fe20Ca40). The singular role of CaO and Fe in promoting Ni for improving catalysts (Ni-Fe-CaO) features was fully analyzed. So, it is proved that both Fe and CaO promoted nickel in the considered catalyst mixed oxides (Ni-Fe-CaO). It found that FeNi3 alloy phase was generated in the catalyst (Ni-), as a result, Ni-Fe-CaO achieved higher TOF value for toluene conversion as compared to single Ni-Ca or Ni-Fe. The catalyst (Ni-Fe-CaO) presented the highest selectivity with Ni/Fe ratios between 2 and 3, which consequently leads to least release by coke. Moreover, Rao, Q et al described a Ni-Fe alloy supported over La0.8Sr0.2FeO3 for the light-enhanced reverse water gas shift (RWGS) process to convert CO2 to CO at low temperature with amazing rates. So, it is shown that the apparent activation energy of RWGS reaction decreases by closely 50% under the cordial light irradiation, and the CO generation rate is enhanced by two times. However, the generation of (Fe, Ni) phase is contributing to the high reactivity of the catalyst during enough operation time. Furthermore, Smirnov, A., et al reported the abiotic NH4 + generation in the existence of Ni-Fe metals and alloys << Awaruite (Ni80Fe20), and tetrataenite (Ni50Fe50) >> and its involvements for the hadean nitrogen cycle. In fact, it is noted after investigation that alloys were less reactive than the metals toward all examined nitrogen species. Also, they noted that nitrite and nitrate were fast converted to ammonium as compared to dinitrogen, and the reduction activity had a strong temperature dependency. Ni-Fe NPs composition influence was investigated for Abiotic carbon dioxide conversion to early prebiotic organics by Beyazay, T., et al's research paper . Indeed, both Mono and bimetallic Ni-Fe NPs with various Ni/Fe ratios revealed to transform carbon dioxide and hydrogen into intermediates and outcome of the acetyl-coenzyme A pathway-formate, acetate, and as well as pyruvate-in mM range under moderate hydrothermal conditions. Also, the Ni-Fe catalysts reached to transform CO2 the same products with the use of water instead of molecular hydrogen. It is mentioned that the two metals play complemental roles for carbon dioxide capture. In particular, due to the remarkable benefit of perovskite-type oxide, Steiger, P., et al investigated the segregation of Ni/Fe bimetallic particles from Lanthanum doped Strontium Titanates to boosts the Sulfur stability of Solid Oxide Fuel Cell Anodes. Indeed, the proposed catalyst tested in a water gas shift reaction, the effect of the additional metal on the catalytic activity and sulfur stability of the Ni catalyst was fully analyzed. As a result, the addition of iron and nickel perovskite catalyst was found to boost the sulfur tolerance. Considerable physicochemical analysis were taken characterize this Ni/Fe segregated perovskite material. An ultra-small FeNi nanoparticles (NPs) was anchored on hierarchically mesoporous carbon by Gao, Y., et al's team for the oxygen evolution reaction . So, the synthesized (fd-FeNi/C) catalyst with 4 nm of particles sizes demonstrated excellent OER performance in alkaline media. As a results, it exhibited in 1.0 M KOH solution 301 mV an overpotential at a current density of 10 mA/cm 2 . Also a small Tafel slope was noted at least 64.5 mV/dec along with higher stability. As concluded, it is taught that this study provides an easy, scalable, and template-free route to transform biomass into powerful active electrocatalysts, which might possess great benefit for future technologies. Moreover, it is shown from Suksumrit's group that Ni-based catalysts are potential candidate for the catalytic methanation of carbon dioxide. So, they presented the possibility to use siderite ore as a catalyst or matrix material for Nibased catalysts. Also, MgO matrix material was as taken support of Ni-based catalysts, to compare their performance as |
66cbb4baa4e53c48769f3dfd | 3 | MgO is better for this issue. As a result, at 648 K the catalytic activity was comparable, but the proposed catalyst exhibited limited catalytic selectivity to yield methane. Fortunately, when MgO incorporated to the proposed catalyst both CO2 conversion and selectivity to output methane increases significantly. Finally, the (Ni/Siderite-reduced/MgO) catalyst achieved almost 99% of methane selectivity and the CO2 recyclable were close to the thermodynamic equilibrium. Hence, such synthesis might be interesting to look for. Recently, Insyani, R., et al notified that TiO2-anchored Ni catalysts are very promising materials which could be taken as candidates to achieve biomass valorization leading to the selective hydrodeoxygenation (HDO). So, the catalyst they prepared with low iron content about (1.5 wt%) in the Ni/TiO2 highly promoted the formation of cyclohexanol(ch) from guaiacol with 183 ( molch/molNi.h), 85% of rate and selectivity respectively. So, the formed interfacial sites of the fully dispersed NiFe-FeOx core-shell structures over TiO2 matrix promote the demethoxylation of reactant to selectively make cyclic alcohols. Khairani's group studied the synthesis of magnetic (FeNi) bimetallic alloy nanoparticles by the utilisation of laser ablation method in liquids. Their results confirmed that the structure, size, phase, and oxidation of (FeNi) NPs generated by pulsed laser ablation in liquids may be altered by changing the solvent or just removing the water impurities in the organic solvent. But the author thought that the elaboration of ironnickel (FeNi) NPs with control over the generated phases, surface chemistry, particle size, and as well as oxidation state remains a challenge that block its applications. Raimundo's research group presented an elaboration and characterisation of NiFe-carbon fibers (NiFe/CFs) by solution blow spinning and its usages for the oxygen evolution reaction (OER). So, the physicochemical characterization showed that the synthesized material consist of sphere-like NiFe NPs with a diameters ranging from 4 to 12 nm with uniform dispersion over the matrix carbon fibrillar. It has a typical ferromagnetic properties and its catalytic features toward the OER in alkaline media was : 296 mV vs RHE of an overpotential at 10 mA/cm 2 of current density. In addition, over 15 h of electrolysis at 10 mA/cm 2 , the (NiFe/CFs) catalyst was electrochemically very stable. Also, the above Author , studied the prepared of NiFe alloy nanoparticles by utilizing the proteic sol-gel route, followed by a reduction at 700 and 500 °C in H2 atmosphere. So, both magnetical and morphology, structural properties were fully tuned through the thermal treatment in H2 environment. Indeed, the prepared catalysts showed different electrocatalytic properties when applied in the oxygen evolution reaction. At 10 mA/cm 2 , the NiFe-700 exhibited the lower overpotential (307 mV) while the NiFe-500 has an overpotential of 319 mV. Hence, indicating superior electrochemical features of the FCC NiFe alloy. Their activity were even beyond to the RuO2 & IrO2 the state-of-art catalysts. Upon 15 h of electrochemical stability analysis by chrono potentiometric investigations, both catalysts revealed to be very stable. Hence, tetrataenite is best. |
66cbb4baa4e53c48769f3dfd | 4 | To pioneer new performance nanomaterials for diverse engineering areas in order to meet the human beings of necessity. Researchers and engineers from the world are focusing on the search of those materials and to make easy ways of designing and synthesizing those expected materials. Indeed, from an inspiration of a material coming from a natural mineral (meteorite), it is demonstrated here that tetrataenite (NiFe) is a new technological materials for many applications with a properties very similar to the rare-earth materials. So, as a result of our search, it is noted that many composite materials might be designed and synthesized from this natural material with a satisfactory features. Those nanomaterials can be applied for several reactions as oxygen evolution reaction (OER), (nitrite, nitrate, nitrogen) conversion to ammonia (NH3), carbon dioxide (CO2) transformation to high value-added products, and so fourth. Therefore, the prospection of tetrataenite would be advantages for considerable engineering technologies and it may solve many issues in engineering fields. |
619400b658bc4b931a95b35c | 0 | The acrylate was also reported in catalysis studies of bifunctional 2D Cd(II)-based metalorganic framework . Conjugate addition of perfluoroarenes to t-butyl 4methylphenylcyanoacrylate was enabled by an alkoxide-hydrosilane system . t-Butyl cyanoacrylate polymers have found multiple applications like water-based ink compositions , preparation of polymer microspheres , and positive working radiation-sensitive composition . |
619400b658bc4b931a95b35c | 1 | All t-butyl phenylcyanoacrylates (TBPCA) compounds were synthesized by Knoevenagel condensation The preparation procedure was essentially the same for all the monomers. In a typical synthesis, equimolar amounts of t-butyl cyanoacetate and an appropriate benzaldehyde were mixed in equimolar ratio in a 20 mL vial. A few drops of piperidine were added with stirring. The product of the reaction was isolated by filtration and purified by crystallization from 2-propanol. The condensation reaction proceeded smoothly, yielding products, which were purified by conventional techniques. The compounds were characterized by IR, 1 H and 13 C NMR spectroscopies. No stereochemical analysis of the novel alkyl ring-substituted TBPCA was performed since no stereoisomers (E or/and Z) of known configuration were available. |
619400b658bc4b931a95b35c | 2 | Copolymers of the ST and the TBPCA compounds, P(ST-co-TBPCA) were prepared in 25-mL glass screw cap vials at ST/TBPCA = 3 (mol) the monomer feed using 0.12 mol/L of ABCN at an overall monomer concentration 2.44 mol/L in 10 mL of toluene. The copolymerization was conducted at 70ºC. After a predetermined time, the mixture was cooled to room temperature, and precipitated dropwise in methanol. The composition of the copolymers was determined based on the nitrogen content (cyano group in TBPCA). The novel synthesized TBPCA compounds copolymerized readily with ST under free-radical conditions (Scheme 2) forming white flaky precipitates when their solutions were poured into methanol. The conversion of the copolymers was kept between 10 and 20% to minimize compositional drift (Table ). Nitrogen elemental analysis showed that between 11.1 and 29.8 mol% of TBPCA is present in the copolymers prepared at ST/TBPCA = 3 (mol), which is indicative of relatively high reactivity of the TBPCA monomers towards ST radical which is typical of alkyl ring-substituted TBPCA . Since TBPCA monomers do not homopolymerize, the most likely structure of the copolymers would be isolated TBPCA monomer units alternating with short ST sequences (Scheme 2). |
64fb43c5b6ab98a41c162c1e | 0 | The structure of sperm chromatin differs significantly from that of chromatin in somatic cells. In somatic cells, the base unit of chromatin is the nucleosome consisting of DNA wrapped around an octamer of histone proteins to form a "bead on a string" structure. In mature sperm cells, the vast majority (>90%) of somatic histones ultimately get replaced during spermatogenesis by cationic peptides known as protamines. These protamines are short (4-8 kDa), highly basic peptides that contain 50-70% arginine residues. The high number of cationic arginine residues enables a strong binding to negatively charged DNA phosphates resulting in a near-crystalline condensation of the DNA in sperm chromatin to a volume roughly 6 times more compact than is achieved by mitotic chromosomes. Sperm chromatin, like viruses, therefore represent some of the tightest packaging of DNA in vivo. Within the sperm nuclei, protamine-DNA forms toroidal structures with DNA helices inside the toroid arranged parallel to one another in a hexagonal array. This unique structure and composition of sperm chromatin is believed to be critical for facilitating both the transport of sperm DNA and protecting its genetic intergrity. During maturation, protamines are initially phosphorylated and subsequently dephosphorylated. It has long been thought that this phosphorylation serves an important function to weaken interactions, enabling the correct binding of protamines to DNA and neighboring protamines during the late phases of spermatogenesis. Improper sperm chromatin remodeling is known to correlate with infertility, DNA fragmentation, and poor reproductive outcomes. Male germ cells, including sperm, are known to produce reactive oxygen species (ROS). While a certain level of ROS is necessary for proper physiological function, an overproduction can negatively affect sperm development. Due to the tight packaging of DNA in sperm chromatin, all DNA repair functions are inactive, making sperm DNA particularly sensitive to oxidative damage. One of the key consequences of exessive oxidative stress in sperm is DNA damage including DNA base oxidation, single and double strand breaks, and DNA fragmentation. Sperm DNA fragmentation has long been considered indicators of male fertility. Recent studies have highlighted the significance of both single and double strand breaks, linking them to reproductive complications such as fertilization failure, increased miscarriage rates, and potentially harmful mutations in offspring. It is believed that DNA damage susceptibility is typically greater in histone-rich regions of the sperm chromatin due to their lower DNA packaging density compared to protamine-rich regions. This has been supported by recent studies on mouse sperm nuclei, which found that histone-rich peripheral and basal regions are particularly prone to oxidation. However, recent findings show that protamines can carry multiple post-translational modifications (PTMs) that would also likely decrease local DNA compaction, potentially resulting in increased DNA damage in these regions. While the common understanding has been that tighter packaging in condensed DNA provides greater protection against ROS damage, the precise relationship between DNA packaging density and susceptibility to oxidative damage in condensed DNA remains unclear and warrants further investigation. Polycations, with a charge equal to or greater than +3, can condense DNA into well-ordered nanoparticles termed polyplexes. These polycation-DNA complexes, including reconstituted protamine-DNA, have also been shown to be organized into a hexagonal array of DNA helices and form toroidal shapes, making them ideal models for studying protamine-mediated DNA packaging in sperm. Previous research has shown that the equilibrium spacing between DNA helices within these polyplexes depends on both the length and composition of the polycation. Polyplexes have potential applications in fields such as nonviral gene delivery and nucleotide-based vaccines. However, their utility is constrained by two factors: their long-term stability in solution, and the extent to which they can protect the encapsulated gene of interest. Improving our understanding of these factors will be crucial for the future exploitation of polyplexes in these fields. In this study, we used reconstituted systems made with DNA and various polycationic peptides to directly examine the influence of DNA packaging on its protection against free radical damage. UV-Vis spectroscopy and dynamic light scattering (DLS) experiments were carried out to assess DNA precipitation capability of the polycations and characterize the colloidal properties of the resultant polyplexes. DNA packaging densities within all polyplexes were quantified using small-angle X-ray scattering (SAXS). To mimic oxidative stress conditions, free radicals were generated from the thermal decomposition of 2,2'-Azobis(2amidinopropane) dihydrochloride (AAPH) that produces peroxyl radicals in the presence of oxygen. The free radicalinduced damage of supercoiled plasmid DNA in the condensed state was then evaluated by gel electrophoresis. Our results indicate that the extent of DNA protection in the condensed state depends on both the packaging density within the polyplex condensate and the length, or equivalently charge, of the polycation. Using hexaarginine peptides as protamine mimics, the effect of a phosphorylated serine residue was also studied. To better understand the potential mechanisms involved, we conducted molecular dynamics (MD) simulations to study the intrinsic dynamic behavior of condensing agents of different lengths within the polyplexes. These simulations revealed that shorter peptides exhibit a higher degree of motion within the polyplexes, which may account for the heightened sensitivity of oligocation polyplexes to free radical damage. |
64fb43c5b6ab98a41c162c1e | 1 | Here, the x-axis shows the nitrogen-to-phosphate (N/P) charge ratio between polycation amines (N) and DNA phosphates (P) and the y-axis shows the fraction of soluble DNA remaining after centrifugation determined by UV-Vis absorbance at 260 nm. All data shown are reported as mean ± standard deviation (n = 3). precipitation curves plotted as the normalized absorbance at 260 nm as a function of increasing N/P charge ratio, or the ratio of cationic amine (N) groups from the polycations to negatively charged nucleic acid phosphate (P) groups. For all polycations studied except R3, noticeable DNA condensation is observed with increasing polycation concentrations from N/P 0.25 to full condensation. All cationic systems studied fully condensed DNA at N/P charge ratios between 1-2. The shortest peptide studied, R3, required the highest N/P charge ratio (N/P ~ 2) to fully precipitate DNA. The longest polycations (PolyR, polyK and protamine) are observed to fully condense DNA at N/P ~ 1.0 -1.2 while the intermediate length hexaarginine peptides (R6 and R3pSR3) fully condense DNA at N/P ~1.5. The extent of precipitation also varied depending on the polycation length and charge. The three longest polycations condensed in a linear fashion maintaining an ~1:1 ratio. For example, at N/P 0.5, where sufficient polycation has been added to compensate for half of the negatively charged DNA phosphates, we observe ~50% of the DNA is precipitated. This trend continues until complete DNA complexation is observed at N/P ~1. The precipitation of DNA by the trivalent cation proceeds in a very different manner. As shown in Figure , R3 did not show appreciable DNA precipitation until N/P ~1.2. However, at higher N/P ratios, DNA precipitation of DNA by R3 proceeds in a cationdependent manner with full precipitation observed at N/P ~2. The hexaarginine peptides (R6 and R3pSR3) show intermediate behavior with full condensation observed only at N/P ≥ 1.5. Although DNA precipitation is continuous at all N/P ratios < 1.5, less DNA is precipitated by these two peptides when compared to salmon protamine, polyR, or polyK. At N/P 1, for example, only ~60% of the total DNA is precipiated with both hexaarginine peptides. X-ray diffraction experiments were used to determine the DNA packaging densities within the fully condensed polyplexes. All polycations studied were able to spontaneously condense DNA resulting in a hexagonal packaging of DNA helices within the polyplexes as depicted in were condensed at an N/P charge ratio of 3 to ensure complete DNA condensation as observed in Figure . All polyplexes studied resulted in the formation of tightlypacked, colloidal nanoparticles with hydrodynamic radii ranging from ~70-200 nm. All DLS results were performed in triplicate, and hydrodynamic radius results are reported as mean ± standard deviation. |
64fb43c5b6ab98a41c162c1e | 2 | AAPH-induced DNA damage in uncondensed DNA and polyplexes. The DNA cleavage activity of AAPH was studied using gel electrophoresis of supercoiled pBR322 plasmid DNA. In all gel lanes, the lower band is due to supercoiled (SC) plasmid DNA while the upper band is due to open circular DNA (OC). It is believed that a single strand break is sufficient to convert SC DNA to the more relaxed OC DNA. When both strands are cleaved, a linear (L) DNA band appears between the SC and OC bands. Previous work has shown that AAPH, a water-soluble azo compound, generates peroxyl and alkoxyl free radicals upon degradation capable of inducing DNA oxidative damage. We first determined the effect of temperature on the generation of free radicals by AAPH. Figure shows the effect of incubating 200 ng of uncondensed pBR322 plasmid in 10 mM Tris-HCl buffer (pH 7.5) with varying concentrations of AAPH for 1 hr at either 37 ºC or 60 ºC. Representative agarose gels obtained are shown in Figure . In the absence of AAPH, pBR322 plasmid DNA at both temperatures is primarily in the supercoiled (SC) form (Lane 1). At both temperatures, we observe a dose-dependent increase in the appearance of damaged (OC and L) forms of pBR322 plasmid when incubated with AAPH. To quantify the damage, we also plot the percentage of intact SC DNA as a function of increasing AAPH concentration for triplicate gels in Figure . Significantly greater DNA damage is observed at higher temperature for equivalent AAPH concentrations. Incubation of pBR322 plasmid at 60 ºC in the absence of AAPH (lane 1) does not result in the loss of DNA supercoiling compared to 37 ºC, However, 50 µM AAPH at 60 ºC is sufficient to convert all supercoiled DNA to linear and open coil forms. In contrast, AAPH concentrations greater than 1 mM are required to completely eliminate the supercoiled band when plasmid DNA is treated with AAPH at 37 ºC. Having demonstrated the capacity to induce significant DNA strand breaks at low (10-200 µM) AAPH concentrations at 60 ºC, these conditions were used to examine the effect of condensation on the protection of DNA against free radical damage. Next, we evaluated the protective effect of protamine condensation on DNA damage by AAPH-generated free radicals. The results for pBR322 plasmid DNA in the absence or presence of salmon protamine are shown in Figure for three different AAPH concentrations. The percent undamaged (SC) DNA is plotted in Figure . In the absence of AAPH (Fig. , lane 1), pBR322 is primarily in the supercoiled form. As expected, increasing the AAPH concentration results in a significant decrease in SC DNA for uncondensed DNA. After incubation with 20 µM AAPH for 1 hr, uncondensed plasmid was readily oxidized by radicals generated by AAPH, resulting in the vast majority (>70%) of the DNA being in the open coiled form (Figure , lane 2). By 500 µM AAPH, a clear linear DNA band is observed giving evidence for double-strand breaks due to oxidation. Condensation by salmon protamine conferred significant protection to DNA from free radical damage. For these studies, all protamine-DNA polyplexes were formed at N/P ratio 3 to ensure complete condensation of the DNA. After mixing, polyplexes were incubated for 30 min at room temperature prior to treatment with AAPH. After AAPH treatment for 1 hr, polyplexes were decondensed by competitive dissociation with dextran sulfate and immediately subjected to agarose electrophoresis. At 20 µM AAPH, the percent SC DNA in protamine-condensed plasmid DNA is not significantly different (p = 0.569) from the control DNA that was not exposed to AAPH. A dose dependent increase in DNA damage was observed for protamine-condensed samples. However, at all AAPH concentrations studied, significantly higher amounts of supercoiled DNA are found for protamine-condensed DNA than those of naked DNA. For example, after 1 hr of incubation with 100 µM AAPH, less than 10% of the uncondensed DNA is intact, compared to ~73% of the DNA remaining SC when condensed with protamine. Results in Figure were averaged from five gels and results are given as mean ± standard deviation. Higher concentrations of protamine, or higher N/P ratios, did not significantly alter the degree of damage observed in condensed samples. for the DNA against AAPH-induced damage. Protamine and polyK condensed DNA is protected similarly to R6-DNA. Statistical analysis of the results in Figure shows that polyR-condensed DNA shows the highest levels of DNA protection that is significantly greater (P < 0.001) than that observed for any other polycation studied. Although undamaged DNA from polylysine and protamine condensed DNA is observed to be higher than that of R6 condensed DNA, the observed difference is not significantly greater at P < 0.001 level. Atomistic molecular dynamics simulations. To examine the molecular-level behavior that explains the relationship between a polyarginine chain's length and the protection that it offers to DNA, we performed atomistic molecular dynamics simulations of a DNA helix in the presence of either R3, R6, or R20, where the R20 chain served as a model for polyR. A sufficient number of polyarginine chains were added at random locations in each simulation box to result in an N/P ratio of 3. Many peptide chains spontaneously moved toward and became bound to the DNA as shown in Figure . However, in each simulation, some arginine chains remained free in solution, suggesting the DNA helices are saturated with the arginine oligomers in the simulations. We calculated the average number of polyarginine residues and chains that were in direct contact (within 3 Å) with DNA during the last 50 ns of each simulation (Table ). In all three simulations, enough arginine residues are in direct contact with DNA to neutralize at least 90% of the DNA charge. Thus, the polyplex has a net positive charge when considering the total charge of the chains bound to DNA. Chain length does have a modest impact on the number of residues bound to DNA, with R3 having fewer residues in contact with DNA (36 ± 3) than R6 (40 ± 2) and R20 (41 ± 1). This result also correlates with the solvent-accessible surface area (SASA) of the DNA in the three simulations. A similar number of arginine residues in the R6 and R20 simulations are bound to DNA, and the DNA SASA is similar for these systems. In the R3 simulation, however, fewer arginine residues are in contact with the DNA, and the SASA of the DNA is larger, potentially increasing the risk of DNA damage from exposure to damaging radicals in solution. |
64fb43c5b6ab98a41c162c1e | 3 | To examine the dynamics of polyarginine-DNA interactions, we determined when each of the polyarginine chains was in contact with DNA during the last 50 ns of each simulation (Figure ). In all three simulations, there are peptide chains that form a long-lasting interaction with the DNA duplex as well as chains that do not interact with the DNA. In the case of R20 (Figure ), four of the six R20 chains are always in contact with the DNA during this time period, while the other two chains remain free in solution. In the case of the R6 and, especially, the R3 simulations, interactions between the DNA and the peptide chains are more dynamic, and DNApeptide interactions continue to break and/or form. Such changes in DNA-polyarginine interactions were rare in the R6 simulation, as a single R6 chain that had a long-lasting interaction with DNA broke away approximately 90 ns into the simulation. An additional R6 chain made contact with DNA during only a single snapshot relatively late in the simulation. Dynamic interactions between DNA and R3 chains (Figure ) were more common, with seven peptide chains breaking away from the DNA after being in contact for more than ~5 ns. The relationship between interaction stability and polyarginine chain length can be explained by considering the number of residues of each peptide that interact with DNA. Of the R3 chains bound to DNA, only ~2 residues were in contact with the DNA at any given time. In contrast, ~5 arginine residues of each R6 chain and ~10 residues of each R20 chain were in contact with the DNA. Therefore, the release of an R6 or R20 chain that was bound to DNA would typically require the disruption of more DNAarginine residue interactions than the release of an R3 chain. |
64fb43c5b6ab98a41c162c1e | 4 | In addition to impacting binding between DNA and the entire polyarginine chain, chain length also impacted the dynamics of interactions between DNA and individual arginine residues. We calculated the contact fraction (i.e., the fraction of simulation snapshots in which the residue was in contact with DNA) for each arginine residue (Figure ). For each type of peptide, ~15 residues form stable interactions with the DNA and remain in contact with the DNA throughout the simulation (contact fraction > 0.99). About 50 additional residues contact the DNA at some point during the simulation (contact fraction between 0.01 and 0.99). For the ~50 arginine residues most commonly in contact with the DNA, there appears to be a relationship between the contact fraction and chain length, as the long R20 tends to have more residues that make a higher contact fraction (fraction greater than 0.5) than the other two systems. |
64fb43c5b6ab98a41c162c1e | 5 | In this study, we characterized the influence of polycation chemistry and valency on DNA protection in the condensed state against free radical oxidation. Potential mechanisms of DNA protection could include DNA compaction by polycations, cation binding to DNA, and/or radical scavenging by polycations. The latter scenario would result in the formation of radical polycations, thereby sparing the DNA from damage. However, there is also evidence that peptide radicals may be a route to transfer damage to DNA in the compacted state. Although the peptide radical model was not extensively tested in this study, our previous in vitro antioxidant scavenging experiments (unpublished) have indicated negligible antioxidant activities for protamines. This suggests that radical scavenging does not adequately account for the observed DNA protection in these polyplexes. This hypothesis is further supported by our findings that higher nitrogen-to-phosphate (N/P) ratios for protamine-DNA and other polyplexes, where more unbound polycation chains are available to interact with free radicals, did not significantly modify DNA damage levels within the condensed phase. We propose that DNA condensation could protect polyplex DNA by limiting the access of free radicals to the DNA helices within the condensate. To investigate potential steric effects, we employed different polycations to achieve variable DNA packaging densities and quantified DNA damage directly within the condensed phase. |
64fb43c5b6ab98a41c162c1e | 6 | When DNA is introduced to polycations (valency ≥ +3), it spontaneously condenses in vitro, a process driven by electrostatic interactions between positively charged polycations and the negatively charged phosphate backbone of DNA. It has been established that DNA condensation occurs when approximately 90% of the DNA charge is neutralized by the condensing agent. In our study, before assessing damage, we initially characterized the DNA precipitation behavior with each polycation. Figure illustrates that DNA precipitation, as a function of the N/P charge ratio, depends on the length, or equivalently, the charge of the polycation. For the longest polycations used in this study, salmon protamine, polyR, and polyK, DNA precipitation occurs in a continuous manner upon addition of the peptide. For example, at N/P 0.5, where sufficient polycation is added to charge compensate 50% of the DNA phosphates, we see 50% of the total DNA condensed for these three systems. At N/P ~1, where there are sufficient cation amines to fully compensate for DNA phosphate charges, salmon protamine, polyR, and polyK all fully condense DNA. Decreasing the polycation length affects DNA precipitation. We studied triarginine (R3) and two hexaarginine peptides (R6 and R3pSR3), chosen as protamine mimics, to explore the potential effects of residual phosphates on sperm chromatin. Compared to longer polycations, the two hexaarginine peptides condensed less total DNA for a given N/P ratio. For example, at N/P 0.5 and 1, only approximately 20% and 60% of total DNA were precipitated. Interestingly, the presence of the phosphate did not significantly alter the DNA precipitation behavior in the hexaarginine peptides, although it had a substantial impact on the resulting DNA packaging density. Complete condensation by these two peptides only occurs at N/P charge ratios >1.5. The precipitation behavior of R3 is very different. As shown in Figure , DNA precipitation by R3 does not occur below a critical polycation concentration. Only above this concentration does DNA precipitation take place, consistent with previous studies on trivalent cations such as cobalt hexammine and spermidine. For example, Raspaud et. al. |
64fb43c5b6ab98a41c162c1e | 7 | showed that 9 µM lambda DNA solution was condensed by trivalent spermidine only at concentrations greater than ~2 mM. Similarly in our experiments, DNA precipitation only begins at R3 concentrations resulting in N/P ~1.2 with full condensation not observed until N/P 2. Consistent with previous studies on condensed chicken blood or calf thymus DNA, the interhelical spacing between DNA helices (Dint) within these polyplexes depends on both the the length and composition of the polycation. While DNA-DNA spacings can change depending on equilibration time, our measurements of Dint values in polyplexes after one hour (as shown in Figure ) align with those of previous samples equilibrated for two weeks. The shorter equilibration time used here corresponds more closely to those in subsequent DNA damage assays. When comparing peptides, arginines package DNA more compactly than lysine peptides of similar lengths. For example, polyR-DNA displays Dint values over 2 Å smaller than those for polyK-DNA, suggesting a 20% reduction in packaging volume within the polyR condensates. Biological systems exploit this more efficient packaging by arginines. Histones, which use lysines and arginines, package somatic chromatin, while mature sperm cells primarily use argininerich protamines for condensing DNA. This compact packaging in sperm chromatin effectively silences it transcriptionally until fertilization. Polycation length significantly influences packaging density, with a clear decrease in Dint and an increase in DNA packaging density correlating with greater valency as observed when comparing R3, R6, and polyR. Using osmotic stress measurements, we previously showed that this tighter packaging is due to increased attractions in the more highly charged peptides. Of the arginine peptides, all can condense DNA more tightly than polyK, with the exception of R3, which produces a similar Dint value as polyK. Salmon protamine, due to the presence of uncharged amino acids that increase repulsive forces within the condensate, results in a DNA packaging comparable to that of DNA condensed with R5 or R6. Protamine phosphorylation and subsequent dephosphorylation during spermiogenesis is thought to modulate protamine-DNA interactions enabling proper chromatin remodeling. Recently, protamine phosphorylation was also shown to initate parental genome reprogamming consistent with this electrostatic model. Although having little effect on precipitation behavior, a negatively charged phosphoserine in the hexaarginine peptide significantly reduces the resulting DNA packaging density. For example, R3pSR3-DNA has a Dint roughly 4 Å larger than that of R6-DNA, indicating a volume change of roughly 50% within the condensed phase. This marked reduction in packaging induced by the presence of a phosphoserine group could hold biological relevance for the maturation of spermatozoa. It is likely that looser packaging, due to residual phosphates or other post-translational modifications on protamines in sperm chromatin, leaves DNA more vulnerable to free radical insult. Polycation-induced DNA condensation in dilute solutions, similar to our damage assays, leads to the formation of wellordered polyplex nanoparticles (NP). DLS measurements of the resulting hydrodynamic radii (Rh) of the NPs were measured for all polyplexes at N/P 3 (Figure ). Interestingly, we found that the size of the resulting polyplex particles did not correspond to the densities of the DNA packaging. Instead, they mainly depended on the length of the polycation used. The smallest Rh values were recorded for polyplexes created with polyK, polyR, and salmon protamine. Conversely, R3-DNA formed the largest particles, even though its DNA-DNA interhelical spacing was similar to that of polyK-DNA. Both hexaarginine peptides yielded similar Rh values for the polyplexes, despite significantly larger Dint values for R3pSR3-DNA compared to R6-DNA. Notably, the size of the particles in our polyplexes did not correlate with the extent of DNA damage observed. The protective role of DNA compaction is well established. DNA condensed by histones exhibits increased resistance to damage than uncondensed DNA, with heterochromatin showing superior protection over euchromatin. Previous studies have also demonstrated that DNA condensed by polycations is highly protective against damage induced by ionizing radiation. In our study, we also found that DNA condensation offers significant protection within polyplexes (Figure ). Even at low concentrations of AAPH (20 µM), uncondensed DNA undergoes significant oxidation, while the % supercoiled DNA in protamine-DNA is nearly identical to plasmid DNA in the absence of AAPH. Higher free radical concentrations result in more DNA oxidation in both condensed and uncondensed states; however, protamine-DNA displays notable resistance across all concentrations of AAPH. Taking into account the steric effects, it is normally hypothesized that tighter DNA packaging would result in greater protection within polyplexes. Although most condensing agents were able to protect DNA from free radical oxidation (as illustrated in Figure ), neither packaging density nor particle size alone could explain the observed damage trends in polyplexes. Our results suggest that both the packaging density and the length of the polycations play a role in DNA protection. For example, DNA condensed by polyR, which results in tighter DNA packaging, shows less damage from free radicals compared to polyK-DNA. Similarly, R6-DNA exhibits less damage compared to R3pSR3-DNA, presumably due to the tighter DNA packaging by R6. Yet, R3-DNA, despite having a comparable DNA packaging density to polyK-DNA, shows little to no protective effect against free radical insult, implying that steric shielding alone does not explain the protection disparity. Differences in DNA protection may arise from the mobility of polycations within the polyplexes. Although cationic chain exchanges between polyplex-bound and unbound chains have been reported, the internal mobility of polycations remains largely unexplored. To investigate these potential effects, we conducted molecular dynamic (MD) simulations using arginine peptides of varying lengths. Molecular dynamics (MD) simulations revealed several effects of chain length on the interactions between polyarginine and DNA. These insights may elucidate the increased resistance to AAPH-induced DNA damage when associated with longer polyarginine chains. First, our simulations revealed that a greater number of residues from longer chains interact with the DNA at any given time. Consequently, these longer chains occupy a larger surface area on DNA, potentially shielding the DNA from radical attack more effectively. Second, chain length also impacts the stability of polyR-DNA interactions. Shorter polyarginine chains, particularly R3, demonstrated less stable interactions and exhibited highly dynamic behavior within the condensates. Despite similarities in the average number of interactions between R20-DNA and R6-DNA, R6 residues generally showed lower contact fractions compared to R20, indicating slight differences in the interaction dynamics. Additionally, the longer polyarginine chains exhibited a greater number of direct contacts with DNA. As a result, disrupting a DNA-polyarginine interaction would require breaking more residue-residue contacts for longer chains. This finding suggests that longer polyarginine chains may be more strongly bound to DNA, which could enhance their protective effect. In conclusion, we have investigated packaging and DNA damage in polyplexes formed with polycations of different compositions and chain lengths. Using a combination of experimental and MD simulations, we probed the relationship between the polycation chain length, DNA packaging density, and accumulation of damage from free radical oxidation. Our findings suggest that both steric effects and polymer dynamic mechanisms can contribute to DNA protection. Tighter DNA packaging achieved through the use of specific polycations enhances the protective effect, while the length of the polycation also plays a significant role. Simulation results shed light on the interactions between arginine chains of varying length and DNA, highlighting their role in DNA protection through increased surface coverage and more stable interactions. These findings advance our understanding of the factors influencing DNA protection in condensed states and have implications for the design of polycation-based systems for gene delivery and other applications involving DNA condensation. |
64fb43c5b6ab98a41c162c1e | 8 | Salmon protamine chloride, calf thymus DNA, 2,2'-Azobis(2-amidinopropane) dihydrochloride (AAPH), poly-L-arginine hydrochloride (MW 5-15 kDa), poly-L-lysine hydrochloride (MW 15-30 kDa), ethidium bromide and dextran sulfate sodium salt from Leuconostoc spp. (MW 9-20 kDa) were purchased from Sigma Aldrich. Agarose was purchased from VWR while pBR322 plasmid DNA was purchased from New England Biolabs. Both CT-DNA and plasmid DNA purity was verified by measuring the ratio of absorbance at 260 and 280 nm of DNA solutions and found to be satisfactory with values exceeding 1.8. All chemicals were used without further purification. |
64fb43c5b6ab98a41c162c1e | 9 | DNA precipitation was monitored by UV-Vis measurements using a Thermo Evolution 201 UV-Vis spectrophotometer. Polycation were incubated with 100 µg calf thymus DNA solution in 10 mM Tris-Cl buffer (pH 7.5) at increasing cationic amine (N) to anionic nucleic acid phosphate (P) ratio. After achieving the desired nitrogen-to-phosphate (N/P) charge ratio, all samples were diluted to a final volume of 300 µL with buffer, incubated for 10 min at room temperature then centrifuged at 11000 x g for 15 minutes before measurement. Then 250 µL of supernatant solution was placed in a 1 mm pathlength cuvette. UV-Vis spectra was obtained in the range of 200-400 nm. DNA precipitation was followed by measuring the change in absorbance of the supernatant at 260 nm as a function of N/P charge ratio of the resulting polyplexes. |
64fb43c5b6ab98a41c162c1e | 10 | Polycation-DNA complexes for x-ray scattering experiments was prepared using calf-thymus DNA (CT-DNA) as follows. Polycation solutions (1 mg/ml) was added to ~250 µg of CT-DNA (1 mg/mL) in 10 mM Tris-Cl (pH 7.5) for each x-ray experiment. A sufficient amount of polycation was added stepwise to ensure complete condensation. The resulting fibrous polyplex samples were thoroughly mixed by vortexing and spun down. Condensates were then transferred to 10 mM Tris-Cl containing ∼ 10 µM excess polycation solution. Samples were measured approximately 1 hr after preparation. X-ray scattering experiments were performed using a graded-multilayer focused Cu Kα radiation (1.54 Å) from a Nonius FR-591 rotating anode fine-focus X-ray generator operating at 45 kV and 20 mA. Collimation of the primary beam used a fine rear aperture beam tunnel. Samples were sealed with a bath of buffer solution (10 mM Tris-Cl) in the sample cell and then mounted into a custom sample holder at room temperature. Sample to detector flight path was filled with helium gas to minimize air scatter. Diffraction patterns were recorded with a PT-135 CCD detector, with phosphor optimized for Cu Kα radiation. Images were analyzed with Fit2d and OriginPro software. Calibration of the SAXS sample-to-detector distance was performed using silver behenate powder and found to be 23.2 cm. Bragg scattering peaks were used to determine interaxial DNA-DNA spacings. For a hexagonal lattice, the relationship between the Bragg spacing and the actual interaxial distance between helices (Dint) is calculated as Dint = (2/√3)DBr. For different samples equilibrated under the same conditions, Dint values were reproducible to within ∼ ±0.1 Å. No significant sample degradation was observed due to Xray exposure. Typical exposure times were 120 s. |
64fb43c5b6ab98a41c162c1e | 11 | The hydrodynamic radii (Rh) of different DNA polyplexes were measured using a Brookhaven Zeta PALS Zeta Potential Analyzer in the Multi-Angle Particle Size Option at a scattering angle of 90º at 25 ºC. Before preparation of the particles, all solutions were filtered through a Whatman 25 mm GD/X filter unit with 0.45 µm pore size to exclude colloidal contamination. Samples were prepared by mixing 1 µg of pBR322 DNA with each polycation (at N/P 3) in 1000 µL of 10 mM Tris. Samples were thereafter incubated at room temperature for 30 minutes, then measured. Samples were prepared in triplicates with the diameter obtained in each replicate being the average of 5 scattered intensity weighted scans taken over 150 seconds. |
64fb43c5b6ab98a41c162c1e | 12 | Cleavage studies of supercoiled pBR322 DNA by AAPH was performed by agarose gel electrophoresis experiments. DNA damage was measured by the conversion of supercoiled plasmid DNA to open circular and linear forms through single strand nicks and double strand breaks. Radicals for inducing plasmid damage were generated by thermal degradation of AAPH in aqueous solution incubated at 60 ºC for 1 hour. Supercoiled pBR322 plasmid DNA (200 ng) was treated with varying concentrations of AAPH in buffer diluted to a final volume of 10 µL. DNA Samples were vortexed and heated for 1 hr at the indicated temperature to induce DNA damage. After incubation with AAPH, samples were mixed with 2 µL of 6x loading dye and loaded on a 0.8% agarose gel and electrophoresis for 1 hr at 85 V for ~2 hr in 1x Tris-acetate-EDTA (TAE) buffer. For DNA polyplex samples, polplexes were mixed to the desired N/P charge ratio, vortexed and incubated at RT for 10 min. Pre-heated AAPH and 10 mM Tris buffer was added to bring final volume of samples to 10 µL and incubated for 1 hr at the desired temperature. After incubation with AAPH, polyplexes were then decondensed by adding 100 µg of dextran sulfate to each tube and incubated for 10 min at RT before loading samples onto the 0.8% agarose gel. After electrophoresis, gels were stained in ethidium bromide (EtBr, 2.5 ug/mL) for 45 min, then destained in 1x TAE buffer for 30 min. Gel images were captured by UV transillumination with a BioRad ChemiDoc XRS system. Band intensities on gels were quantified using Image Lab 6.0 software (Bio-Rad). To account for the decreased ability of ethidium bromide to intercalate into supercoiled DNA versus nicked DNA and linear DNA, a correction factor of 1.52 was applied to supercoiled DNA. The fraction of each form of DNA in a lane was calculated by dividing the intensity of each band by the total intensities of all bands in the lane. |
64fb43c5b6ab98a41c162c1e | 13 | Atomistic molecular dynamics (MD) simulations of a DNA double helix in the presence of either R3, R6 or R20 were performed using the Gromacs 5.0.7 software package. An effictively infinite dsDNA helix was made by placing a helix with sequence d(CG)10 in a simulation box of approximate dimensions 150 Å × 150 Å × 68 Å and binding the top and bottom of the helix across the periodic boundary condition along the z-axis. Either 40 R3, 20 R6, or 6 R20 chains were randomly added to respective simulation boxes, resulting in an N/P ratio of 3.0 in each system. Systems were solvated with ~ 48,000 TIP3P water molecules, and then, neutralized by randomly replacing water molecules with 80 Cl -ions. The parameters for DNA, arginine, water, and ions used in all simulations were the AMBER99 forcefield (AMBER99bsc0 for DNA and AMBER99sb-ildn-phi for arginine) subject to NBFIX corrections 65 that were downloaded from . Equilibration was performed using 5000 steps of steepest descent energy minimization followed by 100 ps of NVT simulation in which the system was heated from 0 to 300 K and 100 ps of NPT simulation at a pressure of 1 atm. The simulations were then continued for 100 ns of production simlation using an integration time step of 2 fs. Temperature and pressure were maintained at 300 K and 1 atm using the Nose-Hoover and Parrinello-Rahman schemes, respectively, with coupling constants of 2 ps. Long-range electrostatic interactions were treated with the particle mesh Ewald method using a 12 Å cut-off and 1.5 Å resolution grid. Interactions between DNA and arginine were analyzed during the final 50 ns of the simulations using analysis tools available in Gromacs . Arginine residues were defined as being in contact with the DNA when any atom of the residue was within 3 Å of any DNA atom, and a polyarginine chain was defined as being in contact with the DNA when any of its residues was in contact with the DNA. The Solvent Accessible Surface Area (SASA) of DNA was measured using the sasa command of Gromacs. We also defined a contact fraction for each arginine residue by dividing the number snapshots in which the residue was in contact with the DNA by the total number of snapshots analyzed. Simulation systems were visualized using VMD 1.9.3 |
67126dc812ff75c3a1cc63fb | 0 | The concept of ionic liquids (ILs) has evolved significantly over the past 40 years. The considerable amount of data obtained from the studies of compounds belonging to this group and the property models created from them have made it possible to verify early generalizations about ILs, that they were supposedly safe, non-toxic, thermally stable and "green" compounds, particularly in comparison to conventional organic solvents. When the potential problems of recycling and disposal of ILs were taken into account, as well as the effects of their introduction into the environment, the search for new sources of ILs ions, and safer methods of their synthesis became important. However, the possibility of almost free adaptation of ILs' structure to the desired properties (so-called "designability")including in the context of their toxicityand the related multiplicity of their potential applications in various industries remains an unquestionable advantage. As a result of the development of specific models on the basis of collected experimental data, methods based on machine learning are now becoming helpful in achieving favorable results and predicting the properties of designed ILs. However, the desired effect can be achieved not only by selecting the right combination of cation and anion or by slight structural modifications to one of the ions (e.g., altering the length of an alkyl chain), but also by creating entirely new ionic systems as a result of mixing 2 or more ILs. Such an approach offers an additional degree of freedom in the choice of cation and anion, does not require the design of a new synthesis process (as in the case of anion exchange or other structural modifications), and the resulting IL mixtures with potentially non-ideal characteristics can have new and unique properties that none of the constituent ionic liquids possess. To distinguish such systems from simple mixtures, the name double salt ionic liquids (DSILs) has been proposed for them. Analogous to deep eutectic solvents (DESs), numerous properties of DSILs result from chemical interactions between the constituent ions, and a change in the molar ratio of parent ILs can result in a significant change in the properties of a given system, which justifies investigating the physical, physicochemical and biological properties of various DSIL systems. Moreover, in some cases DSILs can become the basis for new DESs. In view of the advantages offered by the liquid, modular form of the active ingredient, including the ability to avoid the problems associated with the use of a solid form of the active ingredient (e.g., low bioavailability, polymorphism) and at the same time increase activity against the target organism through structural modifications or the use of an appropriate combination of ions, adherence to ILs (and, more recently, DESs ) concept is a legitimate strategy for designing new biologically active substances. In agrochemical treatments, it is popular to use more than one herbicide, since such a measure broadens the spectrum of action of the herbicide formulation, and can also counteract the acquisition of resistance by undesirable vegetation. It is also known that the transformation of the herbicide active ingredient into an IL may result in a significant increase in the biological activity of the new substance against weeds compared to reference formulations. At the same time, it should be emphasized that the observed increase in efficacy is mainly due to the favorable surface properties of the IL solution applied, which improves the penetration and action of the herbicide anion. Thus, it becomes reasonable to use a DSIL strategy and combine the surface-active cation with two or more active anions originating from known herbicides (e.g., 4-chloro-2methylphenoxyacetic acid (MCPA) or 3,6-dichloro-2-methoxybenzoic acid (dicamba)). |
67126dc812ff75c3a1cc63fb | 1 | Given that the properties of DSILs are strictly dependent on the molar ratio between constituents, in this study we have attempted to further elucidate the phenomenon of synergistic action for DSILs for agrochemical applications and to find the optimal molar ratio of the parent ILs which will provide the most favorable properties. We have also provided a detailed spectral and physicochemical characterization of the new DSILs as well as preliminary evaluation of their aqueous toxicity toward model freshwater algae -Chlorella vulgaris. |
67126dc812ff75c3a1cc63fb | 2 | 1 H NMR spectra were acquired using Varian VNMR-S 400 MHz spectrometer (Palo Alto, USA) with TMS as the internal standard, using deuterated dimethylsulfoxide (DMSO-d6) as a solvent. Samples of 0.1 ±0.001 g were dissolved in 0.6 cm 3 of DMSO-d6 to obtain comparable spectra. 13 C NMR spectra were obtained with the same instrument at 100 MHz. The FT-IR spectra were recorded on IFS 66v/S spectrometer (Bruker Optics, Ettlinger, Germany). The data were sampled from 4000 to 400 cm -1 and visualized using Spectragryph 1.2.13 software. |
67126dc812ff75c3a1cc63fb | 3 | The water content in all obtained products was measured with a TitroLine 7500 KF trace apparatus (SI Analytics, Germany) using the Karl Fischer titration method according to the previously described procedure. Differential scanning calorimeter (DSC) was performed on DSC 209 F1 Phoenix unit (Netzsch, Germany. Samples between 5 and 20 mg were placed in platinum pans and were heated from 25 to 120 °C at a heating rate of 10 °C min -1 and cooled at a cooling rate of 10 °C min -1 to -80 °C. After that, samples were heated again to 120 °C and subsequently cooled to 25 °C. |
67126dc812ff75c3a1cc63fb | 4 | To obtain parent ILs with MCPA and dicamba anions, a quaternization of 2-[2-(dimethylamino)ethoxy]ethanol with 1-bromohexadecane in acetonitrile was conducted analogously to the previously described procedure using a 100 cm 3 EasyMax reactor equipped with ReactIR ® 15 spectrometer, a reflux condenser, a stir bar and a temperature sensor. The FT-IR spectra acquired in situ were subsequently normalized using Pearson's correction on the entire dataset, and the optimal reaction time was established based on the change in relative absorbance of the band occurring at 919 cm -1 . The obtained hexadecyl[2-(2hydroxyethoxy)ethyl]dimethylammonium bromide was then subjected to the two-step anion exchange reaction according to the previously described procedure. In the first stage, 40 cm 3 of ethanolic solution of hexadecyl ethyl]dimethylammonium bromide (0.1 mol) was introduced into a suspension of 80 cm 3 of the AmberTec™ UP550 OH anionic resin in 100 cm 3 of ethanol. The mixture was stirred for 1 h at 25 °C, and in the next step, the resin was filtered and washed three times with small volumes (15 cm 3 ) of ethanol. The solution of hexadecyl[2-(2-hydroxyethoxy)ethyl]dimethylammonium hydroxide was subsequently neutralized with an equimolar amount of the herbicidal acid (MCPA or dicamba). The solvent was subsequently evaporated using a rotary evaporator. To remove the water formed during the neutralization step for analytical purposes, the ILs were subjected to two-step azeotropic distillation: first, they were dissolved in 20 cm 3 of 2-propanol and the 2-propanol-water binary azeotropewas distilled off before evaporating the excess of pure 2-propanol. To remove the residual 2-propanol, the ILs were additionally dissolved in 20 cm 3 of anhydrous methanol and subjected to another step of evaporation. Finally, the obtained ILs were dried under reduced pressure (5 mbar) at 75 °C for 48 h. Both synthesized ILs were stored under reduced pressure over P4O10 to reduce water absorption. |
67126dc812ff75c3a1cc63fb | 5 | A series of DSILs with varying molar fraction of dicamba anion was prepared by mixing the ethanolic solutions of single-ion ILs. First, 0.05 mol of each parent IL (weighed with the accuracy of ±0.0001 g) was dissolved in ethanol to obtain 50 cm 3 of solution, and then both solutions were combined in such volumes as to obtain DSILs with the following MCPAdicamba molar ratios: 9:1, 8:2, 7:3, 6:4, 5:5, 4:6, 3:7, 2:8, and 1:9. To ensure full reproducibility of the results for all DSILs, the same single-anion ILs solutions of a well-defined concentration (1 ±10 -5 mol dm -3 ) were used to obtain the solutions, which were dispensed using the same precalibrated automatic pipette. The obtained mixtures were vigorously stirred for 5 min, and then the solvent was evaporated. The products were pre-dried under reduced pressure conditions (5 mbar, 75 °C) analogously as for single-anion ILs. For analytical purposes, a significant amount of water was removed from the obtained products by two-step azeotropic distillation. The obtained product was dissolved in 30 cm 3 of 2-propanol, after which the azeotrope and the remaining 2-propanol were evaporated under reduced pressure. The residue was dried using a Schlenk line (pressure: 2•10 -5 bar) while heating to 65 °C for at least 6 hours. Subsequently, the residual 2-propanol was removed by performing azeotropic distillation with methanol and drying the residue as in the previous step. The obtained DSILs were stored analogously as in the case of their parent ILs. After drying, using a dicamba-derived single-anion IL and a DSIL containing 0.9 molar fraction of dicamba anion, 2 additional DSILs with molar ratios of MCPA and dicamba anions of 5:95 and 1:99 were also obtained for additional spectral studies according to an analogous method. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.