Small Molecule Binding to Proteins: Affinity and Binding/Unbinding Dynamics from Atomistic Simulations

Combined Physics- and Machine-Learning-Based Method to Identify Druggable Binding Sites Using SILCS-Hotspots

Identifying druggable binding sites on proteins is an important and challenging problem, particularly for cryptic, allosteric binding sites that may not be obvious from X-ray, cryo-EM, or predicted structures. The Site-Identification by Ligand Competitive Saturation (SILCS) method accounts for the flexibility of the target protein using all-atom molecular simulations that include various small molecule solutes in aqueous solution. During the simulations, the combination of protein flexibility and comprehensive sampling of the water and solute spatial distributions can identify buried binding pockets absent in experimentally determined structures. Previously, we reported a method for leveraging the information in the SILCS sampling to identify binding sites (termed Hotspots) of small mono- or bicyclic compounds, a subset of which coincide with known binding sites of drug-like molecules. Here, we build on that physics-based approach and present a ML model for ranking the Hotspots according to the likelihood they can accommodate drug-like molecules (e.g., molecular weight >200 Da). In the independent validation set, which includes various enzymes and receptors, our model recalls 67% and 89% of experimentally validated ligand binding sites in the top 10 and 20 ranked Hotspots, respectively. Furthermore, we show that the model's output Decision Function is a useful metric to predict binding sites and their potential druggability in new targets. Given the utility the SILCS method for ligand discovery and optimization, the tools presented represent an important advancement in the identification of orthosteric and allosteric binding sites and the discovery of drug-like molecules targeting those sites.

The druggable genome: Twenty years later

The concept of the druggable genome has been with us for 20 years. During this time, researchers have developed several methods and resources to help assess a target's druggability. In parallel, evidence for target-disease associations has been collated at scale by Open Targets. More recently, the Protein Data Bank in Europe (PDBe) have built a knowledge base matching per-residue annotations with available protein structure. While each resource is useful in isolation, we believe there is enormous potential in bringing all relevant data into a single knowledge graph, from gene-level to protein residue. Automation is vital for the processing and assessment of all available structures. We have developed scalable, automated workflows that provide hotspot-based druggability assessments for all available structures across large numbers of targets. Ultimately, we will run our method at a proteome scale, an ambition made more realistic by the arrival of AlphaFold 2. Bringing together annotations from the residue up to the gene level and building connections within the graph to represent pathways or protein-protein interactions will create complexity that mirrors the biological systems they represent. Such complexity is difficult for the human mind to utilise effectively, particularly at scale. We believe that graph-based AI methods will be able to expertly navigate such a knowledge graph, selecting the targets of the future.

Micellar Composition Affects Lipid Accretion Kinetics in Molecular Dynamics Simulations: Support for Lipid Network Reproduction

Mixed lipid micelles were proposed to facilitate life through their documented growth dynamics and catalytic properties. Our previous research predicted that micellar self-reproduction involves catalyzed accretion of lipid molecules by the residing lipids, leading to compositional homeostasis. Here, we employ atomistic Molecular Dynamics simulations, beginning with 54 lipid monomers, tracking an entire course of micellar accretion. This was done to examine the self-assembly of variegated lipid clusters, allowing us to measure entry and exit rates of monomeric lipids into pre-micelles with different compositions and sizes. We observe considerable rate-modifications that depend on the assembly composition and scrutinize the underlying mechanisms as well as the energy contributions. Lastly, we describe the measured potential for compositional homeostasis in our simulated mixed micelles. This affirms the basis for micellar self-reproduction, with implications for the study of the origin of life.

Dynamics of the Histone Acetyltransferase Lysine-Rich Loop in the Catalytic Core of the CREB-Binding Protein

The tight control of transcriptional coactivators is a fundamental aspect of gene expression in cells. The regulation of the CREB-binding protein (CBP) and p300 coactivators, two paralog multidomain proteins, involves an autoinhibitory loop (AIL) of the histone acetyltransferase (HAT) domain. There is experimental evidence for the AIL engaging with the HAT binding site, thus interrupting the acetylation of histone tails or other proteins. Both CBP and p300 contain a domain of about 110 residues (called the bromodomain) that recognizes histone tails with one or more acetylated lysine side chains. Here, we investigate by molecular dynamics simulations whether the AIL of CBP (residues 1556-1618) acetylated at the side chain of Lys1595 can bind to the bromodomain. The structural instability and fast unbinding kinetics of the AIL from the bromodomain pocket suggest that the AIL is not a ligand of the bromodomain on the same protein chain. This is further supported by the absence of strong and persistent contacts at the binding interface. Furthermore, the simulations of unbinding show an initial fast detachment of the acetylated lysine and a slower phase necessary for complete AIL dissociation. We provide further evidence for the instability of the AIL intramolecular binding by comparison with a natural ligand, the histone peptide H3K56ac, which shows higher stability in the pocket.

Application of Site-Identification by Ligand Competitive Saturation in Computer-Aided Drug Design

Site Identification by Ligand Competitive Saturation (SILCS) is a molecular simulation approach that uses diverse small solutes in aqueous solution to obtain functional group affinity patterns of a protein or other macromolecule. This involves employing a combined Grand Canonical Monte Carlo (GCMC)-molecular dynamics (MD) method to sample the full 3D space of the protein, including deep binding pockets and interior cavities from which functional group free energy maps (FragMaps) are obtained. The information content in the maps, which include contributions from protein flexibilty and both protein and functional group desolvation contributions, can be used in many aspects of the drug discovery process. These include identification of novel ligand binding pockets, including allosteric sites, pharmacophore modeling, prediction of relative protein-ligand binding affinities for database screening and lead optimization efforts, evaluation of protein-protein interactions as well as in the formulation of biologics-based drugs including monoclonal antibodies. The present article summarizes the various tools developed in the context of the SILCS methodology and their utility in computer-aided drug design (CADD) applications, showing how the SILCS toolset can improve the drug-development process on a number of fronts with respect to both accuracy and throughput representing a new avenue of CADD applications.

Mixed-solvent molecular dynamics simulation-based discovery of a putative allosteric site on regulator of G protein signaling 4

Regulator of G protein signaling 4 (RGS4) is an intracellular protein that binds to the G_α subunit ofheterotrimeric G proteins and aids in terminating G protein coupled receptor signaling. RGS4 has been implicated in pain, schizophrenia, and the control of cardiac contractility. Inhibitors of RGS4 have been developed but bind covalently to cysteine residues on the protein. Therefore, we sought to identify alternative druggable sites on RGS4 using mixed-solvent molecular dynamics simulations, which employ low concentrations of organic probes to identify druggable hotspots on the protein. Pseudo-ligands were placed in consensus hotspots, and perturbation with normal mode analysis led to the identification and characterization of a putative allosteric site, which would be invaluable for structure-based drug design of non-covalent, small molecule inhibitors. Future studies on the mechanism of this allostery will aid in the development of novel therapeutics targeting RGS4.

Free Energies and Entropies of Binding Sites Identified by MixMD Cosolvent Simulations

In our recent efforts to map protein surfaces using mixed-solvent molecular dynamics (MixMD) (Ghanakota, P.; Carlson, H. A. Moving Beyond Active-Site Detection: MixMD Applied to Allosteric Systems. J. Phys. Chem. B 2016, 120, 8685-8695), we were able to successfully capture active sites and allosteric sites within the top-four most occupied hotspots. In this study, we describe our approach for estimating the thermodynamic profile of the binding sites identified by MixMD. First, we establish a framework for calculating free energies from MixMD simulations, and we compare our approach to alternative methods. Second, we present a means to obtain a relative ranking of the binding sites by their configurational entropy. The theoretical maximum and minimum free energy and entropy values achievable under such a framework along with the limitations of the techniques are discussed. Using this approach, the free energy and relative entropy ranking of the top-four MixMD binding sites were computed and analyzed across our allosteric protein targets: Abl Kinase, Androgen Receptor, Pdk1 Kinase, Farnesyl Pyrophosphate Synthase, Chk1 Kinase, Glucokinase, and Protein Tyrosine Phosphatase 1B.

Current Strategies and Applications for Precision Drug Design

Since Human Genome Project (HGP) revealed the heterogeneity of individuals, precision medicine that proposes the customized healthcare has become an intractable and hot research. Meanwhile, as the Precision Medicine Initiative launched, precision drug design which aims at maximizing therapeutic effects while minimizing undesired side effects for an individual patient has entered a new stage. One of the key strategies of precision drug design is target based drug design. Once a key pathogenic target is identified, rational drug design which constitutes the major part of precision drug design can be performed. Examples of rational drug design on novel druggable targets and protein-protein interaction surfaces are summarized in this review. Besides, various kinds of computational modeling and simulation approaches increasingly benefit for the drug discovery progress. Molecular dynamic simulation, drug target prediction and in silico clinical trials are discussed. Moreover, due to the powerful ability in handling high-dimensional data and complex system, deep learning has efficiently promoted the applications of artificial intelligence in drug discovery and design. In this review, deep learning methods that tailor to precision drug design are carefully discussed. When a drug molecule is discovered, the development of specific targeted drug delivery system becomes another key aspect of precision drug design. Therefore, state-of-the-art techniques of drug delivery system including antibody-drug conjugates (ADCs), and ligand-targeted conjugates are also included in this review.

Protein structure-based drug design: from docking to molecular dynamics

Recent years have witnessed rapid developments of computer-aided drug design methods, which have reached accuracy that allows their routine practical applications in drug discovery campaigns. Protein structure-based methods are useful for the prediction of binding modes of small molecules and their relative affinity. The high-throughput docking of up to 10⁶ small molecules followed by scoring based on implicit-solvent force field can robustly identify micromolar binders using a rigid protein target. Molecular dynamics with explicit solvent is a low-throughput technique for the characterization of flexible binding sites and accurate evaluation of binding pathways, kinetics, and thermodynamics. In this review we highlight recent advancements in applications of ligand docking tools and molecular dynamics simulations to ligand identification and optimization.

Quantitative Characterization of the Binding and Unbinding of Millimolar Drug Fragments with Molecular Dynamics Simulations

A quantitative characterization of the binding properties of drug fragments to a target protein is an important component of a fragment-based drug discovery program. Fragments typically have a weak binding affinity, however, making it challenging to experimentally characterize key binding properties, including binding sites, poses, and affinities. Direct simulation of the binding equilibrium by molecular dynamics (MD) simulations can provide a computational route to characterize fragment binding, but this approach is so computationally intensive that it has thus far remained relatively unexplored. Here, we perform MD simulations of sufficient length to observe several different fragments spontaneously and repeatedly bind to and unbind from the protein FKBP, allowing the binding affinities, on- and off-rates, and relative occupancies of alternative binding sites and alternative poses within each binding site to be estimated, thereby illustrating the potential of long time scale MD as a quantitative tool for fragment-based drug discovery. The data from the long time scale fragment binding simulations reported here also provide a useful benchmark for testing alternative computational methods aimed at characterizing fragment binding properties. As an example, we calculated binding affinities for the same fragments using a standard free energy perturbation approach and found that the values agreed with those obtained from the fragment binding simulations within statistical error.

Allosteric regulation of metabolism in cancer: endogenous mechanisms and considerations for drug design

Alterations in metabolic processes have been linked to various diseases, including cancer. Although gene expression can dictate long-term metabolic adaptation, many metabolic changes found in cancer are associated with altered allosteric properties of the underlying enzymes. Small molecule-protein interactions and intracellular signalling converge to orchestrate these allosteric mechanisms, which, emerging evidence suggests, constitute a promising therapeutic avenue. In this review we focus on glucose and energy metabolism to illustrate the role of allostery in cancer physiology and we discuss approaches to streamline the process of targeting aberrant allosteric pathways with small molecules.

Cosolvent-Based Molecular Dynamics for Ensemble Docking: Practical Method for Generating Druggable Protein Conformations

Protein flexibility is a major hurdle in current structure-based virtual screening (VS). In spite of the recent advances in high-performance computing, protein-ligand docking methods still demand tremendous computational cost to take into account the full degree of protein flexibility. In this context, ensemble docking has proven its utility and efficiency for VS studies, but it still needs a rational and efficient method to select and/or generate multiple protein conformations. Molecular dynamics (MD) simulations are useful to produce distinct protein conformations without abundant experimental structures. In this study, we present a novel strategy that makes use of cosolvent-based molecular dynamics (CMD) simulations for ensemble docking. By mixing small organic molecules into a solvent, CMD can stimulate dynamic protein motions and induce partial conformational changes of binding pocket residues appropriate for the binding of diverse ligands. The present method has been applied to six diverse target proteins and assessed by VS experiments using many actives and decoys of DEKOIS 2.0. The simulation results have revealed that the CMD is beneficial for ensemble docking. Utilizing cosolvent simulation allows the generation of druggable protein conformations, improving the VS performance compared with the use of a single experimental structure or ensemble docking by standard MD with pure water as the solvent.

Mixed-Probe Simulation and Probe-Derived Surface Topography Map Analysis for Ligand Binding Site Identification

Membrane proteins represent a considerable fraction of pharmaceutical drug targets. A computational technique to identify ligand binding pockets in these proteins is therefore of great importance. We recently reported such a technique called pMD-membrane that utilizes small molecule probes to detect ligand binding sites and surface hotspots on membrane proteins based on probe-based molecular dynamics simulation. The current work extends pMD-membrane to a diverse set of small organic molecular species that can be used as cosolvents during simulation of membrane proteins. We also describe a projection technique for globally quantifying probe densities on the protein surface and introduce a technique to construct surface topography maps directly from the probe-binding propensity of surface residues. The map reveals surface patterns and geometric features that aid in filtering out high probe density hotspots lacking pocketlike characteristics. We demonstrate the applicability of the extended pMD-membrane and the new analysis tool by exploring the druggability of full-length G12D, G12V, and G13D oncogenic K-Ras mutants bound to a negatively charged lipid bilayer. Using data from 30 pMD-membrane runs conducted in the presence of a 2.8 M cosolvent made up of an equal proportion of seven small organic molecules, we show that our approach robustly identifies known allosteric ligand binding sites and other reactive regions on K-Ras. Our results also show that accessibility of some pockets is modulated by differential membrane interactions.

Simulation of Ligand Binding to Membrane Proteins

Membrane proteins are involved in a large variety of functions. Most of these protein functions are regulated by ligand binding with diverse modes of action: agonists, partial agonists, antagonists, and allosteric modulators, potentiators and inhibitors. From the pharmacological point of view, membrane proteins are one if not the major target for drug development. However, experimental structure determination of membrane proteins in complex or in free form still represents a great challenge. Molecular dynamics (MD) simulations commonly reach the microsecond scale on membrane systems. This numerical tool is mature enough to predict and add molecular details on the different ligand-binding modes. In the present chapter, I will present the different steps to design, simulate, and analyze a MD simulation system containing a protein embedded in a membrane and surrounded by water and ligand. As an illustration, the simulation of the ligand-gated ion channel γ-aminobutyric acid type A receptor (GABA_AR) surrounded by one of its allosteric potentiators, bromoform, will be presented and discussed.

Driving Structure-Based Drug Discovery through Cosolvent Molecular Dynamics

Identifying binding hotspots on protein surfaces is of prime interest in structure-based drug discovery, either to assess the tractability of pursuing a protein target or to drive improved potency of lead compounds. Computational approaches to detect such regions have traditionally relied on energy minimization of probe molecules onto static protein conformations in the absence of the natural aqueous environment. Advances in high performance computing now allow us to assess hotspots using molecular dynamics (MD) simulations. MD simulations integrate protein flexibility and the complicated role of water, thereby providing a more realistic assessment of the complex kinetics and thermodynamics at play. In this review, we describe the evolution of various cosolvent-based MD techniques and highlight a myriad of potential applications for such technologies in computational drug development.

Protein Structural Memory Influences Ligand Binding Mode(s) and Unbinding Rates

The binding of small molecules (e.g., natural ligands, metabolites, and drugs) to proteins governs most biochemical pathways and physiological processes. Here, we use molecular dynamics to investigate the unbinding of dimethyl sulfoxide (DMSO) from two distinct states of a small rotamase enzyme, the FK506-binding protein (FKBP). These states correspond to the FKBP protein relaxed with and without DMSO in the active site. Since the time scale of ligand unbinding (2-20 ns) is faster than protein relaxation (100 ns), a novel methodology is introduced to relax the protein without having to introduce an artificial constraint. The simulation results show that the unbinding time is an order of magnitude longer for dissociation from the DMSO-bound state (holo-relaxed). That is, the actual rate of unbinding depends on the state of the protein, with the protein having a long-lived memory. The rate thus depends on the concentration of the ligand as the apo and holo states reflect low and high concentrations of DMSO, respectively. Moreover, there are multiple binding modes in the apo-relaxed state, while a single binding mode dominates the holo-relaxed state in which DMSO acts as hydrogen bond acceptor from the backbone NH of Ile56, as in the crystal structure of the DMSO/FKBP complex. The solvent relaxes very fast (∼1 ns) close to the NH of Ile56 and with the same time scale of the protein far away from the active site. These results have implications for high-throughput docking, which makes use of a rigid structure of the protein target.

Computational allosteric ligand binding site identification on Ras proteins

A number of computational techniques have been proposed to expedite the process of allosteric ligand binding site identification in inherently flexible and hence challenging drug targets. Some of these techniques have been instrumental in the discovery of allosteric ligand binding sites on Ras proteins, a group of elusive anticancer drug targets. This review provides an overview of these techniques and their application to Ras proteins. A summary of molecular docking and binding site identification is provided first, followed by a more detailed discussion of two specific techniques for binding site identification in ensembles of Ras conformations generated by molecular simulations.

pMD-Membrane: A Method for Ligand Binding Site Identification in Membrane-Bound Proteins

Probe-based or mixed solvent molecular dynamics simulation is a useful approach for the identification and characterization of druggable sites in drug targets. However, thus far the method has been applied only to soluble proteins. A major reason for this is the potential effect of the probe molecules on membrane structure. We have developed a technique to overcome this limitation that entails modification of force field parameters to reduce a few pairwise non-bonded interactions between selected atoms of the probe molecules and bilayer lipids. We used the resulting technique, termed pMD-membrane, to identify allosteric ligand binding sites on the G12D and G13D oncogenic mutants of the K-Ras protein bound to a negatively charged lipid bilayer. In addition, we show that differences in probe occupancy can be used to quantify changes in the accessibility of druggable sites due to conformational changes induced by membrane binding or mutation.

Peptide Binding to a PDZ Domain by Electrostatic Steering via Nonnative Salt Bridges

We have captured the binding of a peptide to a PDZ domain by unbiased molecular dynamics simulations. Analysis of the trajectories reveals on-pathway encounter complex formation, which is driven by electrostatic interactions between negatively charged carboxylate groups in the peptide and positively charged side chains surrounding the binding site. In contrast, the final stereospecific complex, which matches the crystal structure, features completely different interactions, namely the burial of the hydrophobic side chain of the peptide C-terminal residue and backbone hydrogen bonds. The simulations show that nonnative salt bridges stabilize kinetically the encounter complex during binding. Unbinding follows the inverse sequence of events with the same nonnative salt bridges in the encounter complex. Thus, in contrast to protein folding, which is driven by native interactions, the binding of charged peptides can be steered by nonnative interactions, which might be a general mechanism, e.g., in the recognition of histone tails by bromodomains.

Mechanism and Kinetics of Acetyl-Lysine Binding to Bromodomains

Bromodomains are four-helix bundle proteins that specifically recognize acetylation of lysine side chains on histones. The available X-ray structures of bromodomain/histone tail complexes show that the conserved Asn residue in the loop between helices B and C is involved in a hydrogen bond with the acetyl-lysine side chain. Here we analyze the spontaneous binding of acetyl-lysine to the bromodomain TAF1(2) by the first molecular dynamics simulations of histone mark binding to an epigenetic reader protein. Multiple events of reversible association sampled along the unbiased simulations allow us to determine the pathway and kinetics of binding. The simulations show that acetyl-lysine has two major binding modes in TAF1(2) one of which corresponds to the available crystal structures and is stabilized by a hydrogen bond to the conserved Asn side chain. The other major binding mode is more buried than in the crystal structures and is stabilized by two hydrogen bonds with conserved residues of the loop between helices Z and A. In the more buried binding conformation, three of the six structured water molecules at the bottom of the binding pocket are displaced by the acetyl-lysine side chain. The kinetic analysis shows that the two binding modes interconvert on a faster time scale with respect to the association/dissociation process. The atomic-level description of the binding pathway and binding modes is useful for the design of small molecule modulators of histone binding to bromodomains.

Structured water molecules in the binding site of bromodomains can be displaced by cosolvent

Bromodomains are α-helical bundles of approximately 110 residues that recognize acetylated lysine side chains mainly on histone tails. Bromodomains are known to play an important role in cancer and inflammation, and as such, significant efforts are being made to identify small-molecule inhibitors of these epigenetic reader proteins. Here, explicit solvent molecular dynamics (MD) simulations of two bromodomains (BAZ2B and CREBBP) are used to analyze the water molecules that seem to be conserved at the bottom of the acetyl-lysine binding site in most crystal structures of bromodomains. The MD runs suggest that the occupancy of the structured water molecules is influenced by conformational transitions of the loop that connects helices Z and A. Additional simulations in the presence of 50 molecules of cosolvent (i.e., 440 mM of dimethylsulfoxide, methanol, or ethanol) indicate that some of the structured water molecules can be displaced transiently. The residence time in the acetyl-lysine binding site is calculated to be about 1 ns, 2-5 ns, and 10-30 ns for methanol, ethanol, and dimethylsulfoxide, respectively, while the affinity of the three cosolvents for BAZ2B and CREBBP is in the range of 50-500 mM. The results described have implications for ligand design, suggesting that only structured water molecules that do not exchange with cosolvent should be maintained in crystal structures used for docking campaigns, and that hydroxy substituents should be incorporated in the ligand so as to map the structured water molecules replaced by (m)ethanol.

Free Energy Guided Sampling

A free energy-guided sampling (FEGS) method is proposed for accelerating exploration of conformational space in unbiased molecular dynamics. Using the cut-based free energy profile and Markov state models, FEGS speeds up sampling of the canonical ensemble by iteratively restarting multiple short simulations in parallel from regions of the free energy surface visited rarely. This exploration stage is followed by a refinement stage in which multiple independent runs are initiated from Boltzmann distributed conformations. Notably, FEGS does not require either collective variables or reaction coordinates and can control the kinetic distance from the starting conformation. We applied FEGS to the alanine dipeptide, which has a human-comprehensible two-dimensional free energy landscape, and a three-stranded antiparallel β-sheet peptide of 20 residues whose folding/unfolding process is governed by a delicate interplay of enthalpy and entropy. For these two systems, FEGS speeds up the exploration of conformational space by 1 to 2 orders of magnitude with respect to conventional sampling and preserves the basins and barriers on the free energy profile.