Discovery of potent inhibitors of α-synuclein aggregation using structure-based iterative learning.

Machine learning methods hold the promise to reduce the costs and the failure rates of conventional drug discovery pipelines. This issue is especially pressing for neurodegenerative diseases, where the development of disease-modifying drugs has been particularly challenging. To address this problem, we describe here a machine learning approach to identify small molecule inhibitors of α-synuclein aggregation, a process implicated in Parkinson's disease and other synucleinopathies. Because the proliferation of α-synuclein aggregates takes place through autocatalytic secondary nucleation, we aim to identify compounds that bind the catalytic sites on the surface of the aggregates. To achieve this goal, we use structure-based machine learning in an iterative manner to first identify and then progressively optimize secondary nucleation inhibitors. Our results demonstrate that this approach leads to the facile identification of compounds two orders of magnitude more potent than previously reported ones.

Cryo-electron Microscopy and Molecular Modeling Methods to Characterize the Dynamics of Tau Bound to Microtubules.

The electron microscopy metainference integrative structural biology method enables the combination of cryo-electron microscopy electron density maps with molecular modeling techniques such as molecular dynamics to unveil the atomistic biomolecular structural ensemble and the error in the map data in an efficient manner. Here we illustrate the electron microscopy metainference protocol and analysis used to elucidate the atomistic structural ensemble of the microtubule-associated protein tau bound to microtubules by using state-of-the-art molecular mechanic force field and the electron density map.

Using Generative Modeling to Endow with Potency Initially Inert Compounds with Good Bioavailability and Low Toxicity.

In the early stages of drug development, large chemical libraries are typically screened to identify compounds of promising potency against the chosen targets. Often, however, the resulting hit compounds tend to have poor drug metabolism and pharmacokinetics (DMPK), with negative developability features that may be difficult to eliminate. Therefore, starting the drug discovery process with a "null library", compounds that have highly desirable DMPK properties but no potency against the chosen targets, could be advantageous. Here, we explore the opportunities offered by machine learning to realize this strategy in the case of the inhibition of α-synuclein aggregation, a process associated with Parkinson's disease. We apply MolDQN, a generative machine learning method, to build an inhibitory activity against α-synuclein aggregation into an initial inactive compound with good DMPK properties. Our results illustrate how generative modeling can be used to endow initially inert compounds with desirable developability properties.

Optimization of a small molecule inhibitor of secondary nucleation in α-synuclein aggregation.

Parkinson's disease is characterised by the deposition in the brain of amyloid aggregates of α-synuclein. The surfaces of these amyloid aggregates can catalyse the formation of new aggregates, giving rise to a positive feedback mechanism responsible for the rapid proliferation of α-synuclein deposits. We report a procedure to enhance the potency of a small molecule to inhibit the aggregate proliferation process using a combination of in silico and in vitro methods. The optimized small molecule shows potency already at a compound:protein stoichiometry of 1:20. These results illustrate a strategy to accelerate the optimisation of small molecules against α-synuclein aggregation by targeting secondary nucleation.

Secondary Processes Dominate the Quiescent, Spontaneous Aggregation of α-Synuclein at Physiological pH with Sodium Salts.

The accurate recapitulation in an in vitro assay of the aggregation process of α-synuclein in Parkinson's disease has been a significant challenge. As α-synuclein does not aggregate spontaneously in most currently used in vitro assays, primary nucleation is triggered by the presence of surfaces such as lipid membranes or interfaces created by shaking, to achieve aggregation on accessible time scales. In addition, secondary nucleation is typically only observed by lowering the pH below 5.8. Here we investigated assay conditions that enables spontaneous primary nucleation and secondary nucleation at pH 7.4. Using 400 mM sodium phosphate, we observed quiescent spontaneous aggregation of α-synuclein and established that this aggregation is dominated by secondary processes. Furthermore, the presence of potassium ions enhanced the reproducibility of quiescent α-synuclein aggregation. This work provides a framework for the study of spontaneous α-synuclein aggregation at physiological pH.

Optimizing molecular potential models by imposing kinetic constraints with path reweighting.

Empirical force fields employed in molecular dynamics simulations of complex systems are often optimized to reproduce experimentally determined structural and thermodynamic properties. In contrast, experimental knowledge about the interconversion rates between metastable states in such systems is hardly ever incorporated in a force field due to a lack of an efficient approach. Here, we introduce such a framework based on the relationship between dynamical observables, such as rate constants, and the underlying molecular model parameters using the statistical mechanics of trajectories. Given a prior ensemble of molecular dynamics trajectories produced with imperfect force field parameters, the approach allows for the optimal adaption of these parameters such that the imposed constraint of equally predicted and experimental rate constant is obeyed. To do so, the method combines the continuum path ensemble maximum caliber approach with path reweighting methods for stochastic dynamics. When multiple solutions are found, the method selects automatically the combination that corresponds to the smallest perturbation of the entire path ensemble, as required by the maximum entropy principle. To show the validity of the approach, we illustrate the method on simple test systems undergoing rare event dynamics. Next to simple 2D potentials, we explore particle models representing molecular isomerization reactions and protein-ligand unbinding. Besides optimal interaction parameters, the methodology gives physical insights into what parts of the model are most sensitive to the kinetics. We discuss the generality and broad implications of the methodology.

Exploration and Exploitation Approaches Based on Generative Machine Learning to Identify Potent Small Molecule Inhibitors of α-Synuclein Secondary Nucleation.

The high attrition rate in drug discovery pipelines is an especially pressing issue for Parkinson's disease, for which no disease-modifying drugs have yet been approved. Numerous clinical trials targeting α-synuclein aggregation have failed, at least in part due to the challenges in identifying potent compounds in preclinical investigations. To address this problem, we present a machine learning approach that combines generative modeling and reinforcement learning to identify small molecules that perturb the kinetics of aggregation in a manner that reduces the production of oligomeric species. Training data were obtained by an assay reporting on the degree of inhibition of secondary nucleation, which is the most important mechanism of α-synuclein oligomer production. This approach resulted in the identification of small molecules with high potency against secondary nucleation.

Guide for determination of protein structural ensembles by combining cryo-EM data with metadynamics.

Metadynamics electron microscopy metaInference (MEMMI) is an integrative structural biology method that enables a rapid and accurate characterization of protein structural dynamics at the atomic level and the error in the cryo-EM experimental data, even in cases where conformations are separated by high energy barriers. It achieves this by incorporating (a) cryo-electron microscopy electron density maps with (b) metadynamic-enhanced-sampling molecular dynamics. Here, I showcase the setup and analysis protocol of MEMMI, used to discover the atomistic structural ensemble and error in the cryo-EM electron density map of the fuzzy coat of IAPP, a fibril implicated in type II diabetes.

Correlation between the binding affinity and the conformational entropy of nanobody SARS-CoV-2 spike protein complexes.

Camelid single-domain antibodies, also known as nanobodies, can be readily isolated from naïve libraries for specific targets but often bind too weakly to their targets to be immediately useful. Laboratory-based genetic engineering methods to enhance their affinity, termed maturation, can deliver useful reagents for different areas of biology and potentially medicine. Using the receptor binding domain (RBD) of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike protein and a naïve library, we generated closely related nanobodies with micromolar to nanomolar binding affinities. By analyzing the structure-activity relationship using X-ray crystallography, cryoelectron microscopy, and biophysical methods, we observed that higher conformational entropy losses in the formation of the spike protein-nanobody complex are associated with tighter binding. To investigate this, we generated structural ensembles of the different complexes from electron microscopy maps and correlated the conformational fluctuations with binding affinity. This insight guided the engineering of a nanobody with improved affinity for the spike protein.

A Structural Ensemble of a Tau-Microtubule Complex Reveals Regulatory Tau Phosphorylation and Acetylation Mechanisms.

Tau is a microtubule-associated protein that regulates the stability of microtubules. We use metainference cryoelectron microscopy, an integrative structural biology approach, to determine an ensemble of conformations representing the structure and dynamics of a tau-microtubule complex comprising the entire microtubule-binding region of tau (residues 202-395). We thus identify the ground state of the complex and a series of excited states of lower populations. A comparison of the interactions in these different states reveals positions along the tau sequence that are important to determine the overall stability of the tau-microtubule complex. This analysis leads to the identification of positions where phosphorylation and acetylation events have destabilizing effects, which we validate by using site-specific post-translationally modified tau variants obtained by chemical mutagenesis. Taken together, these results illustrate how the simultaneous determination of ground and excited states of macromolecular complexes reveals functional and regulatory mechanisms.

Determination of intermediate state structures in the opening pathway of SARS-CoV-2 spike using cryo-electron microscopy.

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the cause of COVID-19, a highly infectious disease that is severely affecting our society and welfare systems. In order to develop therapeutic interventions against this condition, one promising strategy is to target spike, the trimeric transmembrane glycoprotein that the virus uses to recognise and bind its host cells. Here we use a metainference cryo-electron microscopy approach to determine the opening pathway that brings spike from its inactive (closed) conformation to its active (open) one. The knowledge of the structures of the intermediate states of spike along this opening pathway enables us to identify a cryptic pocket that is not exposed in the open and closed states. These results underline the opportunities offered by the determination of the structures of the transient intermediate states populated during the dynamics of proteins to allow the therapeutic targeting of otherwise invisible cryptic binding sites.

A method of incorporating rate constants as kinetic constraints in molecular dynamics simulations.

From the point of view of statistical mechanics, a full characterization of a molecular system requires an accurate determination of its possible states, their populations, and the respective interconversion rates. Toward this goal, well-established methods increase the accuracy of molecular dynamics simulations by incorporating experimental information about states using structural restraints and about populations using thermodynamics restraints. However, it is still unclear how to include experimental knowledge about interconversion rates. Here, we introduce a method of imposing known rate constants as constraints in molecular dynamics simulations, which is based on a combination of the maximum-entropy and maximum-caliber principles. Starting from an existing ensemble of trajectories, obtained from either molecular dynamics or enhanced trajectory sampling, this method provides a minimally perturbed path distribution consistent with the kinetic constraints, as well as modified free energy and committor landscapes. We illustrate the application of the method to a series of model systems, including all-atom molecular simulations of protein folding. Our results show that by combining experimental rate constants and molecular dynamics simulations, this approach enables the determination of transition states, reaction mechanisms, and free energies. We anticipate that this method will extend the applicability of molecular simulations to kinetic studies in structural biology and that it will assist the development of force fields to reproduce kinetic and thermodynamic observables. Furthermore, this approach is generally applicable to a wide range of systems in biology, physics, chemistry, and material science.

Approximating free energy and committor landscapes in standard transition path sampling using virtual interface exchange.

Transition path sampling is a powerful technique for investigating rare transitions, especially when the mechanism is unknown and one does not have access to the reaction coordinate. Straightforward application of transition path sampling does not directly provide the free energy landscape nor the kinetics. This drawback has motivated the development of path sampling extensions able to simultaneously access both kinetics and thermodynamics, such as transition interface sampling, and the reweighted path ensemble. However, performing transition interface sampling is more involved than standard two-state transition path sampling and still requires (some) insight into the reaction to define interfaces. While packages that can efficiently compute path ensembles for transition interface sampling are now available, it would be useful to directly compute the free energy from a single standard transition path sampling simulation. To achieve this, we present here an approximate method, denoted virtual interface exchange transition path sampling, that makes use of the rejected pathways in a form of waste recycling. The method yields an approximate reweighted path ensemble that allows an immediate view of the free energy landscape from a standard single transition path sampling simulation, as well as enables a committor analysis.

Enhanced Sampling of Protein Conformational Transitions via Dynamically Optimized Collective Variables.

Protein conformational transitions often involve many slow degrees of freedom. Their knowledge would give distinctive advantages because it provides chemical and mechanistic insight and accelerates the convergence of enhanced sampling techniques that rely on collective variables. In this study, we implemented a recently developed variational approach to conformational dynamics metadynamics to the conformational transition of the moderate size protein, L99A T4 Lysozyme. To find the slow modes of the system, we combined data coming from NMR experiments as well as from short MD simulations. A Metadynamics simulation based on these information reveals the presence of two intermediate states, at an affordable computational cost.

Accelerating the Calculation of Protein-Ligand Binding Free Energy and Residence Times Using Dynamically Optimized Collective Variables.

Elucidation of the ligand/protein binding interaction is of paramount relevance in pharmacology to increase the success rate of drug design. To this end, a number of computational methods have been proposed; however all of them suffer from limitations since the ligand binding/unbinding transitions to the molecular target involve many slow degrees of freedom that hamper a full characterization of the binding process. Being able to express this transition in simple and general slow degrees of freedom would give a distinctive advantage, since it would require minimal knowledge of the system under study, while in turn it would elucidate its physics and accelerate the convergence speed of enhanced sampling methods relying on collective variables. In this study we pursuit this goal by combining for the first time variational approach to conformational dynamics with funnel metadynamics. In so doing, we predict for the benzamidine/trypsin system the ligand binding mode, and we accurately compute the absolute protein-ligand binding free energy and unbinding rate at unprecedented low computational cost. Finally, our simulation protocol reveals the energetics and structural details of the ligand binding mechanism and shows that water and binding pocket solvation/desolvation are the dominant slow degrees of freedom.

Folding a small protein using harmonic linear discriminant analysis.

Many processes of scientific importance are characterized by time scales that extend far beyond the reach of standard simulation techniques. To circumvent this impediment, a plethora of enhanced sampling methods has been developed. One important class of such methods relies on the application of a bias that is a function of a set of collective variables specially designed for the problem under consideration. The design of good collective variables can be challenging and thereby constitutes the main bottle neck in the application of these methods. To address this problem, recently we have introduced Harmonic Linear Discriminant Analysis, a method to systematically construct collective variables as linear combinations of a set of descriptors. The method uses input information that can be gathered in short unbiased molecular dynamics simulations in which the system is trapped in the metastable states. Here, to scale up our examination of the method's efficiency, we applied it to the folding of chignolin in water. Interestingly, already before any biased simulations were run, the constructed one-dimensional collective variable revealed much of the physics that underlies the folding process. In addition, using it in metadynamics, we were able to run simulations in which the system goes from the folded state to the unfolded one and back, where to get fully converged results, we combined metadynamics with parallel tempering. Finally, we examined how the collective variable performs when different sets of descriptors are used in its construction.

Water structure and dynamics in the hydration layer of a type III anti-freeze protein.

We report on a molecular dynamics study on the relation between the structure and the orientational (and hydrogen bond) dynamics of hydration water around the ocean pout AFP III anti-freeze protein. We find evidence for an increasing tetrahedral structure from the area opposite to the ice binding site (IBS) towards the protein IBS, with the strongest signal of tetrahedral structure around the THR-18 residue of the IBS. The tetrahedral structural parameter mostly positively correlates with increased reorientation decay times. Interestingly, for several key (polar) residues that are not part of the IBS but are in its vicinity, we observe a decrease of the reorientation time with increasing tetrahedral structure. A similar anti-correlation is observed for the hydrogen-bonded water molecules. These effects are enhanced at a lower temperature. We interpret these results in terms of the structure-making and structure-breaking residues. Moreover, we investigate the tetrahedral structure and dynamics of waters at a partially dehydrated IBS, and for the protein adsorbed at the air-water interface. We find that the mutation changes the preferred protein orientation upon adsorption at an air-water interface. These results are in agreement with the water-air Vibration Sum Frequency Generation spectroscopic experiments showing a strongly reduced tetrahedral signal upon mutation at the IBS.

Stability and growth mechanism of self-assembling putative antifreeze cyclic peptides.

Cyclic peptides (CPs) that self-assemble in nanotubes can be candidates for use as antifreeze proteins. Based on the cyclic peptide sequence cyclo-[(l-LYS-d-ALA-l-LEU-d-ALA)2], which can stack into nanotubes, we propose a putative antifreeze cyclic peptide (AFCP) sequence, cyclo-[(l-LYS-d-ALA)2-(l-THR-d-ALA)2], containing THR-ALA-THR ice binding motifs. Using molecular dynamics simulations we investigate the stability of these cyclic peptides and their growth mechanism. Both nanotube sequences get more stable as a function of size. The relative stability of the AFCP sequence CPNT increases at sizes greater than a dimer by forming intermolecular THR side chain H-bonds. We find that, like the naturally occurring AF protein from spruce budworm (Choristoneura fumiferana), the THR distances of the AFCP's ice binding motif match the ice prism plane O-O distances, thus making the AFCP a suitable AF candidate. In addition, we investigated the nanotube growth process, i.e. the association/dissociation of a single CP to an existing AFCP nanotube, by Transition Path Sampling. We found a general dock-lock mechanism, in which a single CP first docks loosely before locking into place. Moreover, we identified several qualitatively different mechanisms for association, involving different metastable intermediates, including a state in which the peptide was misfolded inside the hydrophobic core of the tube. Finally, we find evidence for a mechanism involving non-specific association followed by 1D diffusion. Under most conditions, this will be the dominant pathway. The results yield insights into the mechanisms of peptide assembly, and might lead to an improved design of self-assembling antifreeze proteins.

A one-way shooting algorithm for transition path sampling of asymmetric barriers.

We present a novel transition path sampling shooting algorithm for the efficient sampling of complex (biomolecular) activated processes with asymmetric free energy barriers. The method employs a fictitious potential that biases the shooting point toward the transition state. The method is similar in spirit to the aimless shooting technique by Peters and Trout [J. Chem. Phys. 125, 054108 (2006)], but is targeted for use with the one-way shooting approach, which has been shown to be more effective than two-way shooting algorithms in systems dominated by diffusive dynamics. We illustrate the method on a 2D Langevin toy model, the association of two peptides and the initial step in dissociation of a ß-lactoglobulin dimer. In all cases we show a significant increase in efficiency.