key: cord-325473-hrdanbn1
authors: Ghahremanpour, Mohammad M.; Tirado-Rives, Julian; Deshmukh, Maya; Ippolito, Joseph A.; Zhang, Chun-Hui; de Vaca, Israel Cabeza; Liosi, Maria-Elena; Anderson, Karen S.; Jorgensen, William L.
title: Identification of 14 Known Drugs as Inhibitors of the Main Protease of SARS-CoV-2
date: 2020-08-28
journal: bioRxiv
DOI: 10.1101/2020.08.28.271957
sha: 
doc_id: 325473
cord_uid: hrdanbn1

A consensus virtual screening protocol has been applied to ca. 2000 approved drugs to seek inhibitors of the main protease (Mpro) of SARS-CoV-2, the virus responsible for COVID-19. 42 drugs emerged as top candidates, and after visual analyses of the predicted structures of their complexes with Mpro, 17 were chosen for evaluation in a kinetic assay for Mpro inhibition. Remarkably 14 of the compounds at 100-μM concentration were found to reduce the enzymatic activity and 5 provided IC50 values below 40 μM: manidipine (4.8 μM), boceprevir (5.4 μM), lercanidipine (16.2 μM), bedaquiline (18.7 μM), and efonidipine (38.5 μM). Structural analyses reveal a common cloverleaf pattern for the binding of the active compounds to the P1, P1’, and P2 pockets of Mpro. Further study of the most active compounds in the context of COVID-19 therapy is warranted, while all of the active compounds may provide a foundation for lead optimization to deliver valuable chemotherapeutics to combat the pandemic.

SARS-CoV-2, the cause of the COVID-19 pandemic, 1 is a coronavirus (CoV) from the Coronaviridae family. Its RNA genome is ~82% identical to that of SARS-CoV, 2 which was responsible for the Severe Acute Respiratory Syndrome (SARS) pandemic in 2003. 3 SARS-CoV-2 encodes two cysteine proteases: the chymotrypsin-like cysteine or main protease, known as 3CL pro or M pro , and the papain-like cysteine protease, PL pro . They catalyze the proteolysis of polyproteins translated from the viral genome into non-structural proteins essential for packaging the nascent virion and viral repication. 4 Therefore, inhibiting the activity of these proteases would impede the replication of the virus. M pro processes the polyprotein 1ab at multiple cleavage sites. It hydrolyzes the Gln-Ser peptide bond in the Leu-Gln-Ser-Ala-Gly recognition sequence. This cleavage site in the substrate is distinct from the peptide sequence recognized by other human cysteine proteases known to date. 5 Thus, M pro is viewed as a promising target for anti SARS-CoV-2 drug design; it has been the focus of several studies since the pandemic has emerged. 2, [4] [5] [6] [7] An X-ray crystal structure of M pro reveals that it forms a homodimer with a 2fold crystallographic symmetry axis. 2, 5 Each protomer, with a length of 306 residues, is made of three domains (I-III). Domains II and I fold into a six-stranded β-barrel that harbors the active site. 2, 4, 5 Domain III forms a cluster of five antiparallel αhelices that regulates the dimerization of the protease. A flexible loop connects domain II to domain III. The M pro active site contains a Cys-His catalytic dyad and canonical binding pockets that are denoted P1, P1, P2, P3, and P4. 2 The amino acid sequence of the active site is highly conserved among coronaviruses. 8 The catalytic dyad residues are His 41 and Cys 145 and the residues playing key roles in the binding (Figure 1 ). These residues have been found to interact with the ligands co-crystallized with M pro in different studies. 2, 4, 5 Crystallographic data also suggested that Ser 1 of one protomer interacts with Phe 140 and Glu 166 of the other as the result of dimerization. 2, 4 These interactions stabilize the P1 binding pocket, thereby, dimerization of the main protease is likely for its catalytic activity. 2, 4 Drug repurposing is an important strategy for immediate response to the COVID-19 pandemic. 9 In this approach, the main goal of computational and experimental studies has been to find existing drugs that might be effective against SARS-CoV-2. For instance, a molecular docking study suggested remdesivir as a potential therapeutic that could be used against SARS-CoV-2, 10 which was supported experimentally by an EC50 value of 23 μM in an infected-cell assay. 11 However, a clinical trial showed no statistically significant clinical benefits of remdesivir on adult patients hospitalized for severe COVID-19. 12 Nonetheless, patients who were administered remdesivir in the same trial showed a faster time to clinical improvement in comparison to the placebo-control group. 12 In another clinical trial, only patients on mechanical ventilation benefitted from remdesivir. 13 An EC50 value of 27 μM was also reported for lopinavir 11 , suggesting it may have beneficial activity against SARS-CoV-2. However, neither lopinavir nor the lopinavir/ritonavir combination has thus far shown any significant benefits against COVID-19 in clinical trials. Chloroquine, hydroxychloroquine, and favipiravir have also been explored for repurposing against COVID-19; however, clinical studies with them have been controversial. [14] [15] [16] [17] These studies reflect the urgent need for systematic drug discovery efforts for therapies effective against SARS-CoV-2.

Thus, we decided to pursue discovery of small-molecule inhibitors of M pro .

The aim of this initial work was two-fold: to identify known drugs that may show some activity, but also to identify structurally promising, synthetically-accessible substructures suitable for subsequent lead optimization. Our expectation was that existing drugs may show activity but not at the low-nanomolar levels that are typical of effective therapies. This report provides results for the first goal. The work began by designing and executing a consensus molecular docking protocol to virtual screen were predicted using the PROPKA3 20 and the H++ severs. 21, 22 Accordingly, lysines and arginines were positively charged, aspartic and glutamic acids were negatively charged, and all histidines were neutral. All histidines were built with the proton on Nε except for His80, which was protonated at Nδ. The resulting M pro structure has a net charge of -4 e. Extensive visual inspection was carried out using UCSF Chimera. 23 Consensus Molecular Docking. Most docking programs apply methods to generate an initial set of conformations, and tautomeric and protonation states for each ligand. This is followed by application of search algorithms and scoring functions to generate and score the poses of the ligand in the binding site of a protein.

Scoring functions have been trained to reproduce a finite set of experimental ligand-binding affinities that are generally a mix of activity data converted to a free-energy scale. Therefore, the accuracy of the scores is dependent on multiple factors including the compounds that were part of the training set. To mitigate the biases, we performed four independent runs of protein-ligand docking with a library of ca. 2000 approved, oral drugs using Glide, AutoDock Vina, and two protocols with AutoDock 4.2. The results were compiled and further consideration focused on those compounds that ranked among the top 10% percent in at least 3 out of the 4 runs.

Glide. Schrödinger's Protein Prep wizard utility was used for preparing the protein. A 20-Å grid was then generated and centered on the co-crystallized ligand, which was subsequently removed. The drug library members were neutralized and/or ionized via Schrödinger's LigPrep. 24 The Epik program 25 was used for estimating the pKa values of each compound. Plausible tautomers and stereoisomers within the pH range of 7 ± 1 were generated for each compound using the OPLS3 force field. 26 These conditions resulted in a total of 16000 structures, which were then docked into M pro using Schrödinger's standard-precision (SP) Glide. 27, 28 AutoDock. The AutoDockTools (ADT) software 29 was used for creating PDBQT files from SDF and PDB files of compounds and the protein, respectively.

Non-polar hydrogen atoms were removed and Gasteiger-Marsili charges were assigned for both the protein and the ligands using ADT. The AutoGrid 4. simulations. 31 The protonated M pro dimer, with a net charge of -8 e, was represented by the OPLS-AA/M force field. 32 TIP4P water was used as the solvent. 33 Sodium counterions were added to neutralize the net charge of each system. The selected ligand candidates were represented by the OPLS/CM1A force field, 34 as assigned by the BOSS software 35 (version 4.9) and the LigParGen Python code. 36 The parameters were converted to GROMACS format using LigParGen. 36 For neutral ligands, the CM1A partial atomic charges were scaled by a factor of 1.14. 34 Each M pro -ligand complex was put at the center of a triclinic simulation box with 10-Å padding. An energy minimization was then performed until the steepest descent algorithm converged to a maximum force smaller than 2.4 kcal mol -1 Å -1 . A cutoff radius of 12 Å was used to explicitly calculate non-bonded interactions. Long-range electrostatic interactions were treated using the Particle Mesh Ewald (PME) algorithm. 37 The PME was used with an interpolation order of 4, a Fourier spacing of 1.2 Å, and a relative tolerance of 10 -6 . The van der Waals forces were smoothly switched to zero between 10 and 12 Å. Analytical corrections to the long-range effect of dispersion interactions were applied to both energy and pressure. All covalent bonds to hydrogen atoms were constrained at their equilibrium lengths using the LINCS algorithm 38 with the order of 12 in the expansion of the constraint-coupling matrix. Each system was subsequently simulated for 1 ns in the canonical ensemble (NVT) in order for the solvent to relax and the temperature of the system to equilibrate. Initial velocities were sampled from a Maxwell-Boltzmann distribution at 310 K. The V-rescale thermostat with a stochastic term 39 was used for keeping the temperature at 310 K. The stochastic term ensured that the sampled ensemble was canonical. 39 The coupling constant of the thermostat was set to 2.0 ps. Table 1 , and the structures of some of the ones that turned out to be most interesting are shown in Figure 2 . The primary indications include bacterial and viral infections, hypertension, psychosis, inflammation, and cancer. Their mechanisms of action are also broad ranging from kinase and protease inhibitors to dopamine receptors agonists/antagonists, and calcium channel blockers. It is not surprising that peptidic protease inhibitors are well-represented in view of the peptide substrate and prior discovery of peptidic inhibitors for M pro and its SARS-CoV relative. 7, 42, 43 In almost all cases the predicted poses for the 42 compounds from the different docking programs agreed well. The poses from Glide were then subjected to extensive visual scrutiny to check for unsatisfied hydrogen-bonding sites, electrostatic mismatches, and unlikely conformation of the ligand. About half of the compounds were ruled out for further study due to the occurrence of such liabilities and the presence of multiple ester groups (e.g., methoserpidine and nicomol) or overall size and complexity (e.g., bromocriptine and benzquercin). A repeated motif was apparent with high-scoring ligands having a cloverleaf pattern with occupancy of the P1, P1', and P2 pockets, as illustrated in Figure 3 for the complex of azelastine.

Other common elements are an edge-to-face aryl-aryl interaction with His41 and placement of a positively-charged group in the P1 pocket in proximity to Glu166, e.g., the methylazepanium group of azelastine, the protonated trialkylamino group of bedaquiline, and protonated piperazine of periciazine. However, Glu166 forms a saltbridge with the terminal ammonium group Ser1B (Figure 1 ). The electrostatic balance seems unclear in this region, so our final selections included a mix of neutral and positively-charged groups for the P1 site.

The analysis of the high-scoring 42 compounds also considered structural variety and potential synthesis of analogs. In the end, we settled on 17 compounds, which are highlighted in Table 1 , for purchase and assaying. Sixteen were commercially available, mostly from Sigma-Aldrich. The seventeenth, cinnoxicam, was not available, but it was readily prepared in a one-step synthesis from the commercially-available ester components. It may be noted that three calcium channel blockers, efonidipine, lercanidipine, and manidipine were purchased ( Figure 2 ). This was not done owing to the characteristic dihydropyridine substructure, since this end of the molecule protrudes out of the P1' site in the docked poses. It was for the variety in the left-sides of the molecules in Figure 2 , which form the cloverleaf that binds in the P1, P1', and P2 pockets, as illustrated in Figure 4 for manidipine. The steric fit in this region appears good, though the only potential hydrogen bond is between the nitro group and the catalytic Cys145. Remarkably, fourteen of the drugs at 100 μM decreased M pro activity (100 nM), as shown in Figure 5 and Table 2 . Five drugs decreased M pro activity to below 40%. The top five hits from the kinetic assay were manidipine, boceprevir, efonidipine, lercanidipine, and bedaquiline. Dose-response curves were obtained to determine IC50

values, when possible, as shown in Figure 6 for the five most potent inhibitors, with the raw data as a function of time and concentration given in Figure S1 . The computed structures for the complexes of boceprevir and bedaquiline are illustrated in Figure 7 . For boceprevir, the dimethylcyclopropyl subunit is predicted to μM. Further study of these compounds in the context of COVID-19 therapy is warranted, while all of the active compounds reported here may provide a foundation for lead optimization to deliver valuable chemotherapeutics to combat the pandemic.

The Supporting Information is available free of charge on the ACS Publications website. An Excel file with the names and docking scores for the full drug library, a 

The authors have no competing interests.

Gratitude is expressed for support to the U. S. National Institutes of Health (GM32136) and to the Yale University School of Medicine for a CoReCT Pilot Grant.

The M pro plasmid was kindly provided by the Hilgenfeld lab. 5 

A New Coronavirus Associated with Human Respiratory Disease in China

Structure of Mpro from SARS-CoV-2 and Discovery of Its Inhibitors

Identification of a Novel Coronavirus in Patients with Severe Acute Respiratory Syndrome

Structural Basis for the Inhibition of SARS-CoV-2 Main Protease by Antineoplastic Drug Carmofur

Crystal Structure of SARS-CoV-2 Main Protease Provides a Basis for Design of Improved α-Ketoamide Inhibitors

Prediction of Novel Inhibitors of the Main Protease (M-pro) of SARS-CoV-2 through Consensus Docking and Drug Reposition

GC-376, and Calpain Inhibitors II, XII Inhibit SARS-CoV-2 Viral Replication by Targeting the Viral Main Protease

Learning from the Past: Possible Urgent Prevention and Treatment Options for Severe Acute Respiratory Infections Caused by 2019-nCoV

Drug Repurposing: Progress, Challenges and Recommendations

Repurposing against COVID-19

Lopinavir, Emetine, and Homoharringtonine Inhibit SARS-CoV-2 Replication in Vitro

Favipiravir and Other Drugs for the Treatment of the New Coronavirus

Chloroquine and Hydroxychloroquine in the Treatment of COVID-19 with or without Diabetes: A Systematic Search and a Narrative Review with a Special Reference to India and Other Developing Countries

In Vitro Antiviral Activity and Projection of Optimized Dosing Design of Hydroxychloroquine for the Treatment of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2)

Observational Study of Hydroxychloroquine in Hospitalized Patients with COVID-19

PanDDA analysis of COVID-19 main protease against the DSI-poised Fragment Library

Asparagine and Glutamine: Using Hydrogen Atom Contacts in the Choice of Side-Chain Amide Orientation

PROPKA3: Consistent Treatment of Internal and Surface Residues in Empirical p K a Predictions

H++: A Server for Estimating PKas and Adding Missing Hydrogens to Macromolecules

H++ 3.0: Automating PK Prediction and the Preparation of Biomolecular Structures for Atomistic Molecular Modeling and Simulations

A Visualization System for Exploratory Research and Analysis

Towards the Comprehensive, Rapid, and Accurate Prediction of the Favorable Tautomeric States of Drug-like Molecules in Aqueous Solution

Epik: A Software Program for pKa Prediction and Protonation State Generation for Drug-like Molecules

OPLS3: A Force Field Providing Broad Coverage of Drug-like Small Molecules and Proteins

A New Approach for Rapid, Accurate Docking and Scoring. 1. Method and Assessment of Docking Accuracy

Glide: A New Approach for Rapid, Accurate Docking and Scoring. 2. Enrichment Factors in Database Screening

AutoDock4 and AutoDockTools4: Automated Docking with Selective Receptor Flexibility

AutoDock Vina: Improving the Speed and Accuracy of Docking with a New Scoring Function, Efficient Optimization, and Multithreading

5: A High-Throughput and Highly Parallel Open Source Molecular Simulation Toolkit

Improved Peptide and Protein Torsional Energetics with the OPLS-AA Force Field

Comparison of Simple Potential Functions for Simulating Liquid Water

Potential Energy Functions for Atomic-Level Simulations of Water and Organic and Biomolecular Systems

Molecular Modeling of Organic and Biomolecular Systems Using BOSS and MCPRO

LigParGen Web Server: An Automatic OPLS-AA Parameter Generator for Organic Ligands

A Smooth Particle Mesh Ewald Method

LINCS: A Linear Constraint Solver for Molecular Simulations

Canonical Sampling through Velocity Rescaling

Molecular Dynamics with Coupling to an External Bath

Polymorphic Transitions in Single Crystals: A New Molecular Dynamics Method

Structure-based Design of Antiviral Drug Candidates Targeting the SARS-CoV-2 Main Protease

An Overview of Severe Acute Respiratory Syndrome -Coronavirus (SARS-CoV) 3CL Protease Inhibitors: Peptidomimetics and Small Molecule Chemotherapy

Calcium Channel Blockers: A Possible Potential Therapeutic Strategy for the Treatment of Alzheimer's Dementia Patients with SARS-CoV-2 Infection

Targeting the SARS-CoV-2 Main Protease to Repurpose Drugs for

Single-Molecule Pulling Simulations Can Discern Active from Inactive Inhibitors

Investigating Drug-Target Association and Dissociation Mechanisms Using Metadynamics-Based Algorithms

Estimation of Drug-Target Residence Times by τ-Random Accelerated Molecular Dynamics Simulations

Dynamic Undocking and the Quasi-bound State as Tools for Drug Discovery