key: cord-1036191-wsezhtku
authors: Olivet, Julien; Maseko, Sibusiso B.; Volkov, Alexander N.; Salehi-Ashtiani, Kourosh; Das, Kalyan; Calderwood, Michael A.; Twizere, Jean-Claude; Gorgulla, Christoph
title: A systematic approach to identify host targets and rapidly deliver broad-spectrum antivirals
date: 2022-02-28
journal: Mol Ther
DOI: 10.1016/j.ymthe.2022.02.015
sha: cdab161f83dd4c2429345b0cf734e837fee678a5
doc_id: 1036191
cord_uid: wsezhtku

nan

Early vaccine development for coronavirus disease 2019 (COVID- 19) was possible thanks to the prior knowledge that the main immunogenic protein of coronaviruses is the spike protein. Indeed, once the spike sequence of the initial severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variant was determined, vaccines could be developed quickly. In parallel, if virus-host protein-protein interactions (PPIs) are mapped for newly emerging or re-emerging pathogens, key interacting domains can be predicted, and validation of such interactions can help deliver small-molecule inhibitors in a relatively short period of time. Here, we propose a comprehensive pipeline to systematically identify broad-spectrum antivirals in a time-effective manner and complement traditional therapeutic strategies to fight viral infections. Following initial sequencing and virus-host interactome mapping, ultralarge virtual drug screening is conducted on the predicted interacting domains to quickly identify primary hits. The virtual hits and analogs are then experimentally verified using biophysical and biological assays, as well as structural biology techniques. This widely applicable pipeline should allow faster identification of broadspectrum hits and leads for antiviral drug discovery.

Viruses are obligate intracellular parasites that commonly hijack cellular pathways and host protein functions to infect cells and replicate. The basic steps of the viral life cycle include attachment, entry, replication, translation, suppression of the host immune response, assembly, maturation, and particles release. Many of these steps are orchestrated via interactions with host cellular proteins. Viruses thus manipulate host cells by perturbing key macromolecules involved in the control of cell homeostasis. Despite more than 60 years of virology research, the arsenal of antiviral drugs remains alarmingly small, with only about 100 molecules currently available in clinics and most targeting enzymes from five different pathogens: HIV-1, hepatitis C virus (HCV), HBV, herpes simplex virus (HSV), and influenza. 1 Thus, efficient therapeutics are still lacking for numerous viral diseases, and the effectiveness of existing antiviral therapies is often hampered by the occurrence of resistance to treatments. However, virus-host PPIs are, in principle, valuable targets to complement viral enzymes, as their disruption should synergize the antiviral potential of enzymatic inhibitors and minimize the occurrence of drug-resistant mutants, as previously described. 2 Viral proteins have been directly shaped through interactions with host proteomes, which has led to the development of crucial virus-host PPIs. Consequently, disruption of such PPIs by targeting the interfaces between interacting proteins provides a strong basis for the rapid development of potent antiviral drugs. Arguments to reconsider PPIs as druggable targets have been extensively reviewed elsewhere. 3, 4 In summary, the ruggedness of interaction interfaces, which often display grooves and pockets; the discovery of hot spots, i.e., amino acid residues contributing most to the binding of two protein partners; 5 the development of high-performance, quantitative interaction assays; [6] [7] [8] and the design of innovative chemical libraries, 9, 10 including macrocyclics 11 or peptide secondary structure mimetics, 12 have significantly boosted drug discovery efforts for this promising class of targets. One of the best-known examples of small molecules targeting virus-host PPIs is Maraviroc, which inhibits the interaction between the human CCR5 chemokine receptor and the HIV gp120 envelope. 13 This compound was approved by the US Food and Drug Administration (FDA) in 2007, but the underlying mechanisms allowing structure-based optimizations were only deciphered in 2013. 14 We argue that a systematic and unbiased pipeline, involving high-throughput virus-host protein interactome mapping, protein structure predictions, and ultra-large virtual screenings of millions of compounds (Figure 1 selective virus-host PPI inhibitors to complement vaccines and other therapeutic tools.

With recent advances in DNA sequencing, it has become possible to map and analyze viral genomes within hours or days. New or re-emerging viruses can thus be classified quickly and assigned to a family following phylogenetic studies. Using genomic information, open reading frames (ORFs) can also be rapidly identified, synthesized, and cloned into many expression systems to test viral proteins in the desired experimental conditions ( Figure 1A ). In particular, ORFs can be cloned into entry vectors to be rapidly shuttled into various destination plasmids using the highthroughput Gateway cloning technology. 15, 16 This strategy has allowed systematic expression of proteins and mapping of PPIs in different assays and multiple environments. 7, 8 There are two main complementary approaches to identify protein interactions at the proteome scale: affinity purification followed by mass spectrometry (AP-MS) and binary interaction assays, such as yeast two-hybrid (Y2H). 15, 16 AP-MS has been widely used to confidently identify the composition of distinct cellular complexes 17 and understand how viral proteins can disrupt them. 18 However, as recently described in Trepte et al., 8 it remains challenging to differentiate direct interactions (i.e., proteins sharing a physical interface) from indirect associations (i.e., proteins that do not directly interact) within the different complexes identified by AP-MS. Structural biology techniques can provide such information and generate high-resolution 3D models of directly interacting proteins, but their modest throughput currently limits their applications for proteome-wide efforts. In this context, several complementary, quantitative binary interaction assays, such as Y2H, NanoLuc two-hybrid (N2H), or luminescence-based two-hybrid (LuTHy), can be applied to rapidly identify direct PPI targets. [6] [7] [8] To systematically map direct virus-host PPIs with binary interaction assays, two strategies can be applied in parallel. First, all possible pairwise combinations between the viral proteins and the 20,000 human reference protein-coding genes can be mapped and validated, similar to what was done for SARS-CoV-2. 16, 19 Second, AP-MS from different cell lines expressing tagged viral proteins can be performed and binary interaction assays applied between the relevant viral proteins and the members of the identified disrupted complexes. This will output a list of high-quality, direct virus-host PPI targets, from which the interacting domains can be predicted in silico using motif-domain and/ or domain-domain complementarity approaches, as described in Maseko et al. 20 Indeed, many viral proteins contain short linear motifs (SLiM), which are predicted to bind known complementary host protein 

Virtual drug discovery Opinion domains. Alternatively, structure-based protein-protein docking can be carried out using experimental structures available in the Protein Data Bank or predicted structures in the AlphaFold Protein Structure Database. 21 The best candidates for potentially interacting domains can then be systematically tested and the corresponding interactions confirmed experimentally using several in vitro (e.g., fluorescence polarization, isothermal titration calorimetry [ITC], etc.) and/or in cellulo binary interaction assays (e.g., N2H, LuTHy, Y2H, etc.). 7 Finally, the validated viral and host protein domains can be used to conduct ultra-large virtual drug screenings and potentially identify small-molecule inhibitors of key virus-host PPIs.

Ultra-large virtual screening using structurebased molecular dockings has several key advantages over experimental high-throughput approaches, as highlighted by recent successes for protein structures with resolutions above 3Å. [22] [23] [24] [25] First, it can predict potent compounds with dissociation constants (K D ) often lying in the lower nanomolar range. This is a direct consequence of sampling a vast chemical space, which allows the identification of exceptionally well-fitting molecules. This aspect leads to the second advantage: the ability to find sufficiently strong binders to targets traditionally considered challenging, such as PPI interfaces and allosteric sites. The third major advantage is that a large-scale screen improves the true hit rate, i.e., the number of active compounds divided by the number of experimentally tested compounds, since the error robustness increases when scaling up. 26 Furthermore, ultra-large virtual screenings can be carried out in days and thus save a substantial amount of time and costs compared with traditional high-throughput screens.

Published in 2020, VirtualFlow is the first freely accessible, ultra-large virtual screening platform able to routinely sample billions of commercially available compounds. 26 VirtualFlow also provides ultralarge ligand libraries in a ready-to-dock format. These include 1.48 billion compounds from the ZINC library and 1.4 billion molecules from the REAL library of Enamine. 26 Platforms such as VirtualFlow can thus be used to rapidly find small-molecule binders to the sites identified in systematic protein interactome mapping studies ( Figure 1B) . Furthermore, Virtual-Flow can perform multi-stage screenings, beneficial for the proteins exhibiting high flexibility at the target site. In such cases, a conventional docking to a rigid protein is carried out first, followed by rescoring of the top hits with a flexible protein model. The starting protein structures can be obtained from the Protein Data Bank (PDB) or by using structure prediction methods, such as AlphaFold. 21 When needed, the molecular dynamic simulations can be performed during the protein structure preparation step prior to docking. Once the virtual screening step is completed, the highest ranked hits can be tested in the lab. If an experimentally confirmed compound is promising but requires further optimization, a virtual analog library can be created and screened and the best virtual analogs then re-tested in the same setting. Alternatively, classical medicinal chemistry and structure-activity relationship (SAR) studies can be implemented to further optimize the compounds, which is particularly suitable at the later stages of the optimization procedure.

As soon as interacting virus-host protein domains are experimentally validated (Figure 1A) , they can be expressed and purified to conduct follow-up binding assays and structural biology studies. For example, ITC, surface plasmon resonance (SPR), and/or nuclear magnetic resonance (NMR) spectroscopy can be employed to validate the virtual screening results and confirm binding of the small molecules to their targets. 26 Since viruses often acquire resistance by developing mutations in response to inhibitors, in vitro resistance selection antiviral assays can also be conducted to validate the binding sites of the compounds. In parallel, inhibition of the corresponding virus-host PPIs by the candidate small molecules can be evaluated using the aforementioned experimental techniques ( Figure 1A ). Ana-logs of a hit compound can be screened to identify more potent molecules, i.e., those with higher binding affinities for their targets ( Figure 1B ). In addition to NMR and X-ray crystallography, single-particle cryoelectron microscopy (cryo-EM) can also be applied to solve the 3D structures of the desired PPI complexes.

In parallel to those biophysical assays, primary cell-based experiments can be conducted to evaluate whether the candidate small molecules can cross cellular membranes while avoiding cytotoxicity. Comparative proteomic, transcriptomic, and metabolomic analyses can then be implemented to model potential perturbed cellular functions and anticipate off-target and side effects in vivo. For example, genome-scale metabolic network models could be used to identify targets that alter cellular metabolism in a disease state. 27, 28 Currently, Recon3D, 29 a reconstruction of the human metabolic network with over 3,200 ORFs, 13,500 metabolic reactions, and 12,800 protein structures, serves as a comprehensive computational resource to predict proteins that should contract metabolic perturbations following viral infections or disruptions by other agents, including small molecules. These in silico evaluations can further reduce the time and costs of wet-lab experimentations. Follow-up studies can then be performed to confirm that the small molecules indeed impact the viral life cycle and/or enhance the resistance of the target cells to infection at low micromolar to low nanomolar concentrations.

Based on these results, the best PPI inhibitors can then be used to conduct pharmacokinetic (PK) and pharmacodynamic (PD) assays and initiate pre-clinical and clinical studies to rapidly deliver novel antivirals ( Figure 1C ). This pipeline presents two obvious challenges: (1) complicated structural optimization of the PPI inhibitors at early discovery stages and (2) potential side effects resulting from blocking other functions of the targeted host proteins. These might potentially be overcome by early determination of 3D structures of the PPI targets and by testing molecules in relevant animal models, respectively. Importantly, our methodology should allow 

C.G. is the cofounder of two companies, Virtual Discovery, Inc. and Quantum Therapeutics, Inc., which are engaged in drug discovery activities, in part using VirtualFlow.

Principles of virology

Small-molecule inhibitors of the LEDGF/p75 binding site of integrase block HIV replication and modulate integrase multimerization

Modulators of protein-protein interactions

Recent advances in the development of protein-protein interactions modulators: mechanisms and clinical trials

A hot spot of binding energy in a hormone-receptor interface

LuTHy: a double-readout bioluminescence-based two-hybrid technology for quantitative mapping of proteinprotein interactions in mammalian cells

Maximizing binary interactome mapping with a minimal number of assays

A quantitative mapping approach to identify direct interactions within complexomes

Protein-protein interaction inhibition (2P2I)-oriented chemical library accelerates hit discovery

The iPPI-DB initiative: a community-centered database of protein-protein interaction modulators

Macrocycles as protein-protein interaction inhibitors

Rational design of peptide-based inhibitors disrupting proteinprotein interactions

Maraviroc (UK-427,857), a potent, orally bioavailable, and selective smallmolecule inhibitor of chemokine receptor CCR5 with broad-spectrum anti-human immunodeficiency virus type 1 activity

Structure of the CCR5 chemokine receptor-HIV entry inhibitor maraviroc complex

A reference map of the human binary protein interactome

The ORFeome collaboration: a genome-scale human ORF-clone resource

Dual proteome-scale networks reveal cell-specific remodeling of the human interactome

A SARS-CoV-2 protein interaction map reveals targets for drug repurposing

A map of binary SARS-CoV-2 protein interactions implicates host immune regulation and ubiquitination

Interactome and structural basis for targeting the human T-cell leukemia virus Tax oncoprotein

Highly accurate protein structure prediction for the human proteome

Structural ensemble-based docking simulation and biophysical studies discovered new inhibitors of Hsp90 N-terminal domain

Discovery of potent disheveled/dvl inhibitors using virtual screening optimized with NMR-based docking performance index

Structurebased characterization of novel TRPV5 inhibitors

Structure-based identification and characterization of inhibitors of the epilepsy-associated K Na 1.1 (KCNT1) potassium channel

An open-source drug discovery platform enables ultra-large virtual screens

Model-based identification of drug targets that revert disrupted metabolism and its application to ageing

MoVE identifies metabolic valves to switch between phenotypic states

Recon3D enables a three-dimensional view of gene variation in human metabolism