Current Topics in Medicinal Chemistry - Volume 17, Issue 23, 2017

Protein-ligand binding is among the most fundamental phenomena underlying all molecular biology, and a greater ability to more accurately and robustly predict the binding free energy of a small molecule ligand for its cognate protein is expected to have vast consequences for improving the efficiency of pharmaceutical drug discovery. We briefly reviewed a number of scientific and technical advances that have enabled alchemical free energy calculations to recently emerge as a preferred approach, and critically considered proper validation and effective use of these techniques. In particular, we characterized a selection bias effect which may be important in prospective free energy calculations, and introduced a strategy to improve the accuracy of the free energy predictions.

The ability to accurately characterize the solvation properties (water locations and thermodynamics) of biomolecules is of great importance to drug discovery. While crystallography, NMR, and other experimental techniques can assist in determining the structure of water networks in proteins and protein-ligand complexes, most water molecules are not fully resolved and accurately placed. Furthermore, understanding the energetic effects of solvation and desolvation on binding requires an analysis of the thermodynamic properties of solvent involved in the interaction between ligands and proteins. WaterMap is a molecular dynamics-based computational method that uses statistical mechanics to describe the thermodynamic properties (entropy, enthalpy, and free energy) of water molecules at the surface of proteins. This method can be used to assess the solvent contributions to ligand binding affinity and to guide lead optimization. In this review, we provide a comprehensive summary of published uses of WaterMap, including applications to lead optimization, virtual screening, selectivity analysis, ligand pose prediction, and druggability assessment.

Over the last decade, a number of computational methods have been developed, which attempt to evaluate the thermodynamic properties of individual water molecules at the solute-solvent interface, in order to assess contributions to protein-ligand binding. In some cases, these tools tell us what we already know, e.g. that hydrophobic pockets prefer lipophilic substituents, and in other cases the methods only seem to add clarity when retrospectively applied. Hence we have grappled with how to utilize such approaches to understand non-intuitive results and to generate chemistry ideas that otherwise would not have been developed. Here we provide our perspective on these methods and describe how results have been interpreted and applied. We include examples from GSK and elsewhere that highlight how water methods have been (1) utilized retrospectively to explain non-intuitive structure-activity relationships and (2) applied prospectively for chemistry design. Finally, we discuss where this field of study could lead to maximal impact in drug discovery research.

Bio-molecular dynamics (MD) simulations based on graphical processing units (GPUs) were first released to the public in the early 2009 with the code ACEMD. Almost 8 years after, applications now encompass a broad range of molecular studies, while throughput improvements have opened the way to millisecond sampling timescales. Based on an extrapolation of the amount of sampling in published literature, the second timescale will be reached by the year 2022, and therefore we predict that molecular dynamics is going to become one of the main tools in drug discovery in both academia and industry. Here, we review successful applications in the drug discovery domain developed over these recent years of GPU-based MD. We also retrospectively analyse limitations that have been overcome over the years and give a perspective on challenges that remain to be addressed.

Ligand residence times and binding rates have been found to be useful quantities to consider during drug design. The underlying structural and dynamic determinants of these kinetic quantities are difficult to discern. Driven by developments in computational hardware and simulation methodologies, molecular dynamics (MD) studies on full binding and unbinding pathways have emerged recently, showing these structural and dynamic determinants in atomic detail. However, the long timescales related to drug binding and release are still prohibitive to conventional MD simulation. Here we discuss a suite of enhanced sampling methods that have been applied to the study of full ligand binding or unbinding pathways, and reveal the kinetics of drug binding and/or release. We divide these sampling methods into three families (trajectory parallelization, metadynamics, and temperature- based methods), and discuss recent applications of each, as well as their basic theoretical underpinnings including how kinetic information is extracted. We then present an outlook for how the field could evolve, and how the rich variety of sampling methods discussed here can be leveraged in the future for computationally driven drug design.

Cellular drug targets exist within networked function-generating systems whose constituent molecular species undergo dynamic interdependent non-equilibrium state transitions in response to specific perturbations (i.e.. inputs). Cellular phenotypic behaviors are manifested through the integrated behaviors of such networks. However, in vitro data are frequently measured and/or interpreted with empirical equilibrium or steady state models (e.g. Hill, Michaelis-Menten, Briggs-Haldane) relevant to isolated target populations. We propose that cells act as analog computers, “solving” sets of coupled “molecular differential equations” (i.e. represented by populations of interacting species)via “integration” of the dynamic state probability distributions among those populations. Disconnects between biochemical and functional/phenotypic assays (cellular/in vivo) may arise with targetcontaining systems that operate far from equilibrium, and/or when coupled contributions (including target-cognate partner binding and drug pharmacokinetics) are neglected in the analysis of biochemical results. The transformation of drug discovery from a trial-and-error endeavor to one based on reliable design criteria depends on improved understanding of the dynamic mechanisms powering cellular function/dysfunction at the systems level. Here, we address the general mechanisms of molecular and cellular function and pharmacological modulation thereof. We outline a first principles theory on the mechanisms by which free energy is stored and transduced into biological function, and by which biological function is modulated by drug-target binding. We propose that cellular function depends on dynamic counter-balanced molecular systems necessitated by the exponential behavior of molecular state transitions under non-equilibrium conditions, including positive versus negative mass action kinetics and solute-induced perturbations to the hydrogen bonds of solvating water versus kT.

The development and application of quantum mechanics (QM) methodologies in computer-aided drug design have flourished in the last 10 years. Despite the natural advantage of QM methods to predict binding affinities with a higher level of theory than those methods based on molecular mechanics (MM), there are only a few examples where diverse sets of protein-ligand targets have been evaluated simultaneously. In this work, we review recent advances in QM docking and scoring for those cases in which a systematic analysis has been performed. In addition, we introduce and validate a simplified QM/MM expression to compute protein-ligand binding energies. Overall, QMbased scoring functions are generally better to predict ligand affinities than those based on classical mechanics. However, the agreement between experimental activities and calculated binding energies is highly dependent on the specific chemical series considered. The advantage of more accurate QM methods is evident in cases where charge transfer and polarization effects are important, for example when metals are involved in the binding process or when dispersion forces play a significant role as in the case of hydrophobic or stacking interactions.

The rapid delayed rectifier current IKr is one of the major K+ currents involved in repolarization of the human cardiac action potential. Various inherited or drug-induced forms of the long QT syndrome (LQTS) in humans are linked to functional and structural modifications in the IKr conducting channels. IKr is carried by the potassium channel Kv11.1 encoded by the gene KCNH2 (commonly referred to as human ether-a-go-go-related gene or hERG) [1, 2]. The first necessary step for predicting emergent drug effects on the heart is determining and modeling the binding thermodynamics and kinetics of primary and major off-target drug interactions with subcellular targets. The bulk of drugs that target hERG channels are known to have complex interactions at the atomic scale. Accordingly, one of the goals for this review is to provide comprehensive guide in the universe of computational models aiming to refine our understanding of structure-function relations in Kv11.1 and its isoforms. The special emphasis is placed on the mapping of drug binding sites and tentative mechanisms of channel inhibition and activation by drugs. An overview over recent structural models and mapping of binding sites for blockers and activators of IKr current along with the discussion on agreements and discrepancies among different models is presented. There is an apparent reciprocity or feedback loop between drug binding and action potential of the cardiac myocytes. Thus one has to connect drug binding to a particular receptor so that its functional consequences impact on the action potential duration. The natural pathway is to develop multi-scale models that connect between receptor and cellular scales. The potential for such multi-scale model development is discussed through the lens of common gating models. Accordingly, the second part of this review covers an ongoing development of the kinetic models of gating transitions and cardiac ion currents carried by hERG channels with and without drug bound.

Protein-protein interactions play key roles in all biological processes, motivating numerous campaigns to seek small-molecule disruptors of therapeutically relevant interactions. Two decades ago, the prospect of developing small-molecule inhibitors was thought to be perhaps impossible due to the potentially undruggable nature of the protein surfaces involved; this viewpoint was reinforced by the limited successes provided from traditional high-throughput screens. To date, however, refinement of new experimental approaches has led to a multitude of inhibitors against many different targets. Having thus established the feasibility of attaining success in this valuable and diverse target space, attention now turns to incorporating computational techniques that might assist during various stages of drug design and optimization. Here we review cases in which computational approaches – virtual screening, docking, and ligand optimization – have contributed to discovery of new inhibitors of protein-protein interactions. We conclude by providing an outlook into the upcoming challenges and recent advances likely to shape this field moving forward.