Structures, dynamics, complexes, and functions: From classic computation to artificial intelligence.

Computational approaches can provide highly detailed insight into the molecular recognition processes that underlie drug binding, the assembly of protein complexes, and the regulation of biological functional processes. Classical simulation methods can bridge a wide range of length- and time-scales typically involved in such processes. Lately, automated learning and artificial intelligence methods have shown the potential to expand the reach of physics-based approaches, ushering in the possibility to model and even design complex protein architectures. The synergy between atomistic simulations and AI methods is an emerging frontier with a huge potential for advances in structural biology. Herein, we explore various examples and frameworks for these approaches, providing select instances and applications that illustrate their impact on fundamental biomolecular problems.

Decrypting Allostery in Membrane-Bound K-Ras4B Using Complementary In Silico Approaches Based on Unbiased Molecular Dynamics Simulations.

Protein functions are dynamically regulated by allostery, which enables conformational communication even between faraway residues, and expresses itself in many forms, akin to different "languages": allosteric control pathways predominating in an unperturbed protein are often unintuitively reshaped whenever biochemical perturbations arise (e.g., mutations). To accurately model allostery, unbiased molecular dynamics (MD) simulations require integration with a reliable method able to, e.g., detect incipient allosteric changes or likely perturbation pathways; this is because allostery can operate at longer time scales than those accessible by plain MD. Such methods are typically applied singularly, but we here argue their joint application─as a "multilingual" approach─could work significantly better. We successfully prove this through unbiased MD simulations (∼100 µs) of the widely studied, allosterically active oncotarget K-Ras4B, solvated and embedded in a phospholipid membrane, from which we decrypt allostery using four showcase "languages": Distance Fluctuation analysis and the Shortest Path Map capture allosteric hotspots at equilibrium; Anisotropic Thermal Diffusion and Dynamical Non-Equilibrium MD simulations assess perturbations upon, respectively, either superheating or hydrolyzing the GTP that oncogenically activates K-Ras4B. Chosen "languages" work synergistically, providing an articulate, mutually coherent, experimentally consistent picture of K-Ras4B allostery, whereby distinct traits emerge at equilibrium and upon GTP cleavage. At equilibrium, combined evidence confirms prominent allosteric communication from the membrane-embedded hypervariable region, through a hub comprising helix α5 and sheet ß5, and up to the active site, encompassing allosteric "switches" I and II (marginally), and two proposed pockets. Upon GTP cleavage, allosteric perturbations mostly accumulate on the switches and documented interfaces.

Molecular mechanisms of chaperone-directed protein folding: Insights from atomistic simulations.

Molecular chaperones, a family of proteins of which Hsp90 and Hsp70 are integral members, form an essential machinery to maintain healthy proteomes by controlling the folding and activation of a plethora of substrate client proteins. This is achieved through cycles in which Hsp90 and Hsp70, regulated by task-specific co-chaperones, process ATP and become part of a complex network that undergoes extensive compositional and conformational variations. Despite impressive advances in structural knowledge, the mechanisms that regulate the dynamics of functional assemblies, their response to nucleotides, and their relevance for client remodeling are still elusive. Here, we focus on the glucocorticoid receptor (GR):Hsp90:Hsp70:co-chaperone Hop client-loading and the GR:Hsp90:co-chaperone p23 client-maturation complexes, key assemblies in the folding cycle of glucocorticoid receptor (GR), a client strictly dependent upon Hsp90/Hsp70 for activity. Using a combination of molecular dynamics simulation approaches, we unveil with unprecedented detail the mechanisms that underpin function in these chaperone machineries. Specifically, we dissect the processes by which the nucleotide-encoded message is relayed to the client and how the distinct partners of the assemblies cooperate to (pre)organize partially folded GR during Loading and Maturation. We show how different ligand states determine distinct dynamic profiles for the functional interfaces defining the interactions in the complexes and modulate their overall flexibility to facilitate progress along the chaperone cycle. Finally, we also show that the GR regions engaged by the chaperone machinery display peculiar energetic signatures in the folded state, which enhance the probability of partial unfolding fluctuations. From these results, we propose a model where a dynamic cross-talk emerges between the chaperone dynamics states and remodeling of client-interacting regions. This factor, coupled to the highly dynamic nature of the assemblies and the conformational heterogeneity of their interactions, provides the basis for regulating the functions of distinct assemblies during the chaperoning cycle.

Gas Permeation through Mechanically Resistant Self-Standing Membranes of a Neat Amorphous Organic Cage.

The synthesis and characterization of a novel film-forming organic cage and of its smaller analogue are here described. While the small cage produced single crystals suitable for X-ray diffraction studies, the large one was isolated as a dense film. Due to its remarkable film-forming properties, this latter cage could be solution processed into transparent thin-layer films and mechanically stable dense self-standing membranes of controllable thickness. Thanks to these peculiar features, the membranes were also successfully tested for gas permeation, reporting a behavior similar to that found with stiff glassy polymers such as polymers of intrinsic microporosity or polyimides. Given the growing interest in the development of molecular-based membranes, for example for separation technologies and functional coatings, the properties of this organic cage were investigated by thorough analysis of their structural, thermal, mechanical and gas transport properties, and by detailed atomistic simulations.

NLRP3 monomer functional dynamics: From the effects of allosteric binding to implications for drug design.

The protein NLRP3 and its complexes are associated with an array of inflammatory pathologies, among which neurodegenerative, autoimmune, and metabolic diseases. Targeting the NLRP3 inflammasome represents a promising strategy for easing the symptoms of pathologic neuroinflammation. When the inflammasome is activated, NLRP3 undergoes a conformational change triggering the production of pro-inflammatory cytokines IL-1ß and IL-18, as well as cell death by pyroptosis. NLRP3 nucleotide-binding and oligomerization (NACHT) domain plays a crucial role in this function by binding and hydrolysing ATP and is primarily responsible, together with conformational transitions involving the PYD domain, for the complex-assembly process. Allosteric ligands proved able to induce NLRP3 inhibition. Herein, we examine the origins of allosteric inhibition of NLRP3. Through the use of molecular dynamics (MD) simulations and advanced analysis methods, we provide molecular-level insights into how allosteric binding affects protein structure and dynamics, remodelling of the conformational ensembles populated by the protein, with key reverberations on how NLRP3 is preorganized for assembly and ultimately function. The data are used to develop a Machine Learning model to define the protein as Active or Inactive, only based on the analysis of its internal dynamics. We propose this model as a novel tool to select allosteric ligands.

Conformational Behavior of SARS-Cov-2 Spike Protein Variants: Evolutionary Jumps in Sequence Reverberate in Structural Dynamic Differences.

SARS-CoV-2 has evolved rapidly in the first 3 years of pandemic diffusion. The initial evolution of the virus appeared to proceed through big jumps in sequence changes rather than through the stepwise accumulation of point mutations on already established variants. Here, we examine whether this nonlinear mutational process reverberates in variations of the conformational dynamics of the SARS-CoV-2 Spike protein (S-protein), the first point of contact between the virus and the human host. We run extensive microsecond-scale molecular dynamics simulations of seven distinct variants of the protein in their fully glycosylated state and set out to elucidate possible links between the mutational spectrum of the S-protein and the structural dynamics of the respective variant, at global and local levels. The results reveal that mutation-dependent structural and dynamic modulations mostly consist of increased coordinated motions in variants that acquire stability and in an increased internal flexibility in variants that are less stable. Importantly, a limited number of functionally important substructures (the receptor binding domain, in particular) share the same time of movements in all variants, indicating efficient preorganization for functional regions dedicated to host interactions. Our results support a model in which the internal dynamics of the S-proteins from different strains varies in a way that reflects the observed random and non-stepwise jumps in sequence evolution, while conserving the functionally oriented traits of conformational dynamics necessary to support productive interactions with host receptors.

Integrating Protein Interaction Surface Prediction with a Fragment-Based Drug Design: Automatic Design of New Leads with Fragments on Energy Surfaces.

Protein-protein interactions (PPIs) have emerged in the past years as significant pharmacological targets in the development of new therapeutics due to their key roles in determining pathological pathways. Herein, we present fragments on energy surfaces, a simple and general design strategy that integrates the analysis of the dynamic and energetic signatures of proteins to unveil the substructures involved in PPIs, with docking, selection, and combination of drug-like fragments to generate new PPI inhibitor candidates. Specifically, structural representatives of the target protein are used as inputs for the blind physics-based prediction of potential protein interaction surfaces using the matrix of low coupling energy decomposition method. The predicted interaction surfaces are subdivided into overlapping windows that are used as templates to direct the docking and combination of fragments representative of moieties typically found in active drugs. This protocol is then applied and validated using structurally diverse, important PPI targets as test systems. We demonstrate that our approach facilitates the exploration of the molecular diversity space of potential ligands, with no requirement of prior information on the location and properties of interaction surfaces or on the structures of potential lead compounds. Importantly, the hit molecules that emerge from our ab initio design share high chemical similarity with experimentally tested active PPI inhibitors. We propose that the protocol we describe here represents a valuable means of generating initial leads against difficult targets for further development and refinement.

Phosphorylation of the Hsp90 Co-Chaperone Hop Changes its Conformational Dynamics and Biological Function.

The molecular chaperones Hsp90 and Hsp70 and their regulatory co-chaperone Hop play a key role at the crossroads of the folding pathways of numerous client proteins by forming fine-tuned multiprotein complexes. Alterations of the biomolecules involved may functionally impact the chaperone machinery: here, we integrate simulations and experiments to unveil how Hop conformational fitness and interactions can be controlled by the perturbation of just one residue. Specifically, we unveil how mechanisms mediated by Hop residue Y354 control Hop open and closed states, which affect binding of Hsp70/Hsp90. Phosphorylation or mutation of Hop-Y354 are shown to favor structural ensembles that are indeed not optimal for stable interactions with Hsp90 and Hsp70. This disfavors cellular accumulation of the stringent Hsp90 clients glucocorticoid receptor and the viral tyrosine kinase v-Src, with detrimental effects on v-Src activity. Our results show how the post-translational modification of a specific residue in Hop provides a regulation mechanism for the larger chaperone complex of which it is part. In this framework, the effects of one single alteration are amplified at the cellular level through the perturbation of protein-interaction networks.

Computational Epitope Prediction and Design for Antibody Development and Detection.

The design of optimized protein antigens is a fundamental step in the development of new vaccine candidates and in the detection of therapeutic antibodies. A fundamental prerequisite is the identification of antigenic regions that are most prone to interact with antibodies, namely, B-cell epitopes. Here, we describe an efficient structure-based computational method for epitope prediction, called MLCE. In this approach, all that is required is the 3D structure of the antigen of interest. MLCE can be applied to glycosylated proteins, facilitating the identification of immunoreactive versus immune-shielding carbohydrates.

A Fe2+-dependent self-inhibited state influences the druggability of human collagen lysyl hydroxylase (LH/PLOD) enzymes.

Multifunctional human collagen lysyl hydroxylase (LH/PLOD) enzymes catalyze post-translational hydroxylation and subsequent glycosylation of collagens, enabling their maturation and supramolecular organization in the extracellular matrix (ECM). Recently, the overexpression of LH/PLODs in the tumor microenvironment results in abnormal accumulation of these collagen post-translational modifications, which has been correlated with increased metastatic progression of a wide variety of solid tumors. These observations make LH/PLODs excellent candidates for prospective treatment of aggressive cancers. The recent years have witnessed significant research efforts to facilitate drug discovery on LH/PLODs, including molecular structure characterizations and development of reliable high-throughput enzymatic assays. Using a combination of biochemistry and in silico studies, we characterized the dual role of Fe2+ as simultaneous cofactor and inhibitor of lysyl hydroxylase activity and studied the effect of a promiscuous Fe2+ chelating agent, 2,2'-bipyridil, broadly considered a lysyl hydroxylase inhibitor. We found that at low concentrations, 2,2'-bipyridil unexpectedly enhances the LH enzymatic activity by reducing the inhibitory effect of excess Fe2+. Together, our results show a fine balance between Fe2+-dependent enzymatic activity and Fe2+-induced self-inhibited states, highlighting exquisite differences between LH/PLODs and related Fe2+, 2-oxoglutarate dioxygenases and suggesting that conventional structure-based approaches may not be suited for successful inhibitor development. These insights address outstanding questions regarding druggability of LH/PLOD lysyl hydroxylase catalytic site and provide a solid ground for upcoming drug discovery and screening campaigns.

Path to Actinorhodin: Regio- and Stereoselective Ketone Reduction by a Type II Polyketide Ketoreductase Revealed in Atomistic Detail.

In type II polyketide synthases (PKSs), which typically biosynthesize several antibiotic and antitumor compounds, the substrate is a growing polyketide chain, shuttled between individual PKS enzymes, while covalently tethered to an acyl carrier protein (ACP): this requires the ACP interacting with a series of different enzymes in succession. During biosynthesis of the antibiotic actinorhodin, produced by Streptomyces coelicolor, one such key binding event is between an ACP carrying a 16-carbon octaketide chain (actACP) and a ketoreductase (actKR). Once the octaketide is bound inside actKR, it is likely cyclized between C7 and C12 and regioselective reduction of the ketone at C9 occurs: how these elegant chemical and conformational changes are controlled is not yet known. Here, we perform protein-protein docking, protein NMR, and extensive molecular dynamics simulations to reveal a probable mode of association between actACP and actKR; we obtain and analyze a detailed model of the C7-C12-cyclized octaketide within the actKR active site; and we confirm this model through multiscale (QM/MM) reaction simulations of the key ketoreduction step. Molecular dynamics simulations show that the most thermodynamically stable cyclized octaketide isomer (7R,12R) also gives rise to the most reaction competent conformations for ketoreduction. Subsequent reaction simulations show that ketoreduction is stereoselective as well as regioselective, resulting in an S-alcohol. Our simulations further indicate several conserved residues that may be involved in selectivity of C7-12 cyclization and C9 ketoreduction. Detailed insights obtained on ACP-based substrate presentation in type II PKSs can help design ACP-ketoreductase systems with altered regio- or stereoselectivity.

Protein Allostery and Ligand Design: Computational Design Meets Experiments to Discover Novel Chemical Probes.

Herein we examine the determinants of the allosteric inhibition of the mitochondrial chaperone TRAP1 by a small molecule ligand. The knowledge generated is harnessed into the design of novel derivatives with interesting biological properties. TRAP1 is a member of the Hsp90 family of proteins, which work through sequential steps of ATP processing coupled to client-protein remodeling. Isoform selective inhibition of TRAP1 can provide novel information on the biomolecular mechanisms of molecular chaperones, as well as new insights into the development of small molecules with therapeutic potential. Our analysis of the interactions between an active first-generation allosteric ligand and TRAP1 shows how the small molecule induces long-range perturbations that influence the attainment of reactive poses in the active site. At the same time, the dynamic adaptation of the allosteric binding pocket to the presence of the first-generation compound sets the stage for the design of a set of second-generation ligands: the characterization of the formation/disappearance of pockets around the allosteric site that is used to guide optimize the ligands' fit for the allosteric site and improve inhibitory activities. The effects of the newly designed molecules are validated experimentally in vitro and in vivo. We discuss the implications of our approach as a promising strategy towards understanding the molecular determinants of allosteric regulation in chemical and molecular biology, and towards speeding up the design of allosteric small molecule modulators.

New perspectives in cancer drug development: computational advances with an eye to design.

Computational chemistry has come of age in drug discovery. Indeed, most pharmaceutical development programs rely on computer-based data and results at some point. Herein, we discuss recent applications of advanced simulation techniques to difficult challenges in drug discovery. These entail the characterization of allosteric mechanisms and the identification of allosteric sites or cryptic pockets determined by protein motions, which are not immediately evident in the experimental structure of the target; the study of ligand binding mechanisms and their kinetic profiles; and the evaluation of drug-target affinities. We analyze different approaches to tackle challenging and emerging biological targets. Finally, we discuss the possible perspectives of future application of computation in drug discovery.

SARS-CoV-2 Spike Protein Mutations and Escape from Antibodies: A Computational Model of Epitope Loss in Variants of Concern.

The SARS-CoV-2 spike (S) protein is exposed on the viral surface and is the first point of contact between the virus and the host. For these reasons it represents the prime target for Covid-19 vaccines. In recent months, variants of this protein have started to emerge. Their ability to reduce or evade recognition by S-targeting antibodies poses a threat to immunological treatments and raises concerns for their consequences on vaccine efficacy. To develop a model able to predict the potential impact of S-protein mutations on antibody binding sites, we performed unbiased multi-microsecond molecular dynamics of several glycosylated S-protein variants and applied a straightforward structure-dynamics-energy based strategy to predict potential changes in immunogenic regions on each variant. We recover known epitopes on the reference D614G sequence. By comparing our results, obtained on isolated S-proteins in solution, to recently published data on antibody binding and reactivity in new S variants, we directly show that modifications in the S-protein consistently translate into the loss of potentially immunoreactive regions. Our findings can thus be qualitatively reconnected to the experimentally characterized decreased ability of some of the Abs elicited against the dominant S-sequence to recognize variants. While based on the study of SARS-CoV-2 spike variants, our computational epitope-prediction strategy is portable and could be applied to study immunoreactivity in mutants of proteins of interest whose structures have been characterized, helping the development/selection of vaccines and antibodies able to control emerging variants.

The tumor suppressor folliculin inhibits lactate dehydrogenase A and regulates the Warburg effect.

Aerobic glycolysis in cancer cells, also known as the 'Warburg effect', is driven by hyperactivity of lactate dehydrogenase A (LDHA). LDHA is thought to be a substrate-regulated enzyme, but it is unclear whether a dedicated intracellular protein also regulates its activity. Here, we identify the human tumor suppressor folliculin (FLCN) as a binding partner and uncompetitive inhibitor of LDHA. A flexible loop within the amino terminus of FLCN controls movement of the LDHA active-site loop, tightly regulating its enzyme activity and, consequently, metabolic homeostasis in normal cells. Cancer cells that experience the Warburg effect show FLCN dissociation from LDHA. Treatment of these cells with a decapeptide derived from the FLCN loop region causes cell death. Our data suggest that the glycolytic shift of cancer cells is the result of FLCN inactivation or dissociation from LDHA. Together, FLCN-mediated inhibition of LDHA provides a new paradigm for the regulation of glycolysis.

The molecular chaperone TRAP1 in cancer: From the basics of biology to pharmacological targeting.

TRAP1, the mitochondrial component of the Hsp90 family of molecular chaperones, displays important bioenergetic and proteostatic functions. In tumor cells, TRAP1 contributes to shape metabolism, dynamically tuning it with the changing environmental conditions, and to shield from noxious insults. Hence, TRAP1 activity has profound effects on the capability of neoplastic cells to evolve towards more malignant phenotypes. Here, we discuss our knowledge on the biochemical functions of TRAP1 in the context of a growing tumor mass, and we analyze the possibility of targeting its chaperone functions for developing novel anti-neoplastic approaches.

Targeting the mitochondrial chaperone TRAP1: strategies and therapeutic perspectives.

TRAP1, the mitochondrial isoform of heat shock protein (Hsp)90 chaperones, is a key regulator of metabolism and organelle homeostasis in diverse pathological states. While selective TRAP1 targeting is an attractive goal, classical active-site-directed strategies have proved difficult, due to high active site conservation among Hsp90 paralogs. Here, we discuss advances in developing TRAP1-directed strategies, from lead modification with mitochondria delivery groups to the computational discovery of allosteric sites and ligands. Specifically, we address the unique opportunities that targeting TRAP1 opens up in tackling fundamental questions on its biology and in unveiling new therapeutic approaches. Finally, we show how crucial to this endeavor is our ability to predict the activities of TRAP1-selective allosteric ligands and to optimize target engagement to avoid side effects.

Exploiting Folding and Degradation Machineries To Target Undruggable Proteins: What Can a Computational Approach Tell Us?

Advances in genomics and proteomics have unveiled an ever-growing number of key proteins and provided mechanistic insights into the genesis of pathologies. This wealth of data showed that changes in expression levels of specific proteins, mutations, and post-translational modifications can result in (often subtle) perturbations of functional protein-protein interaction networks, which ultimately determine disease phenotypes. Although many such validated pathogenic proteins have emerged as ideal drug targets, there are also several that escape traditional pharmacological regulation; these proteins have thus been labeled "undruggable". The challenges posed by undruggable targets call for new sorts of molecular intervention. One fascinating solution is to perturb a pathogenic protein's expression levels, rather than blocking its activities. In this Concept paper, we shall discuss chemical interventions aimed at recruiting undruggable proteins to the ubiquitin proteasome system, or aimed at disrupting protein-protein interactions in the chaperone-mediated cellular folding machinery: both kinds of intervention lead to a decrease in the amount of active pathogenic protein expressed. Specifically, we shall discuss the role of computational strategies in understanding the molecular determinants characterizing the function of synthetic molecules typically designed for either type of intervention. Finally, we shall provide our perspectives and views on the current limitations and possibilities to expand the scope of rational approaches to the design of chemical regulators of protein levels.

The Answer Lies in the Energy: How Simple Atomistic Molecular Dynamics Simulations May Hold the Key to Epitope Prediction on the Fully Glycosylated SARS-CoV-2 Spike Protein.

SARS-CoV-2 is a health threat with dire socioeconomical consequences. As the crucial mediator of infection, the viral glycosylated spike protein (S) has attracted the most attention and is at the center of efforts to develop therapeutics and diagnostics. Herein, we use an original decomposition approach to identify energetically uncoupled substructures as antibody binding sites on the fully glycosylated S. Crucially, all that is required are unbiased MD simulations; no prior knowledge of binding properties or ad hoc parameter combinations is needed. Our results are validated by experimentally confirmed structures of S in complex with anti- or nanobodies. We identify poorly coupled subdomains that are poised to host (several) epitopes and potentially involved in large functional conformational transitions. Moreover, we detect two distinct behaviors for glycans: those with stronger energetic coupling are structurally relevant and protect underlying peptidic epitopes, and those with weaker coupling could themselves be prone to antibody recognition.