Open science discovery of potent noncovalent SARS-CoV-2 main protease inhibitors

Abstract

We report the results of the COVID Moonshot, a fully open-science, crowdsourced, and structure-enabled drug discovery campaign targeting the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) main protease. We discovered a noncovalent, nonpeptidic inhibitor scaffold with lead-like properties that is differentiated from current main protease inhibitors. Our approach leveraged crowdsourcing, machine learning, exascale molecular simulations, and high-throughput structural biology and chemistry. We generated a detailed map of the structural plasticity of the SARS-CoV-2 main protease, extensive structure-activity relationships for multiple chemotypes, and a wealth of biochemical activity data. All compound designs (>18,000 designs), crystallographic data (>490 ligand-bound x-ray structures), assay data (>10,000 measurements), and synthesized molecules (>2400 compounds) for this campaign were shared rapidly and openly, creating a rich, open, and intellectual property-free knowledge base for future anticoronavirus drug discovery.

Fig. 1.. Crowdsourcing rapidly identified chemotype scaffolds by merging fragment hits.

(A) A Diamond/XChem fragment screen that initiated this SARS-CoV-2 Mpro inhibitor discovery campaign generated 71 hits that completely cover the Mpro active site, with a variety of chemotypes engaging each pocket; 1638 x-ray datasets were collected, and 96 solved structures for hits were publicly posted (20). The peptidomimetic N3 ligand is shown on the left for comparison to indicate natural substrate engagement in the binding site, defining the peptide sidechain numbering scheme used throughout this work. The nucleophilic Cys¹⁴⁵ reacts with the scissile peptide bond between P1 and P1’; His⁴¹-Cys¹⁴⁵ form a catalytic dyad whose coupled charge states shuttle between zwitterionic and neutral states (90). (B) On 18 March 2020, the COVID Moonshot set up a crowdsourcing website to empower scientists across the globe to contribute molecule designs. The number of designs actioned for synthesis each quarter (except for the 2020 Q2, which is shown per-month in brackets) is shown, subdivided by the region of the submitter of the design idea. The total number of unique submitters that contributed actioned designs for that quarter is shown on top of the bars. (C) Many submissions, such as TRY-UNI-714a760b-6, exploited spatially overlapping fragment hits to design potent leads that are synthetically facile. (D) Experimental biochemical potency of designs broken down by submission group. Multiple submissions in 2020 from the community were more potent than the best designs from the core team, as seen for the top three chloroacetamide structures (left) and noncovalent structures (right). (E) Distribution of synthetic accessibility scores (SAScores) for designs contributed by the core team and the community. The concern that community submissions may be of poor quality is not supported by the fact that these were as synthetically accessible as those designed by the core team (median: community, 0.17; core, 0.13). Half of the outliers (SAScore = 1) were primarily natural products, which are hard to achieve through organic chemistry.

Fig. 2.. Strategies to support rapid optimization cycles.

(A) Machine learning forecasts experimental synthesis time (left) and returns efficient routes that leverage more than 10 million in-stock advanced intermediates (right). Our algorithm predicts the probability of each step being successful and predicts synthetic accessibility by taking the product of the probabilities along the whole route. We analyzed all compounds made in COVID Moonshot from 1 May 2020 to 1 July 2021 (n = 898). The right panel exemplifies the experimental execution of the predicted routes, demonstrating the ability of the algorithm to build on functionalized intermediates to shorten synthesis. (B) Applying alchemical free-energy calculations at scale enables us to estimate the potency of compounds. Retrospective assessment of our automated free-energy calculation workflow on early compounds in the 3-aminopyridine series in the first month of the COVID Moonshot campaign suggested that free-energy calculations could provide good predictive utility, which inspired confidence for large-scale deployment during this campaign. Here, the absolute free energy of binding (ΔG) is shown in the rightmost panel by adding a constant offset to the computed relative free-energy differences. (C) Alchemical free-energy predictions for all submissions elaborating on the depicted scaffold for three representative batches of prospective free-energy calculations plotted as calculated (converted using Cheng-Prusoff equation) versus experimental pIC₅₀. Simulations were run using Mpro in dimer form, with neutral catalytic dyad and no restraints. Each batch (numbered 1 to 3 from left to right) is annotated with its scaffold, and top-scoring candidates are shown on the right-hand side (numbered 1 to 3 from top to bottom)–for these, the structure names are shown together with their predicted and experimental pIC₅₀ (“Pred” and “Exp,” respectively). Statistical performance with 95% confidence intervals for each batch is shown as a table in each scatterplot. (D) Two examples of nanomole-scale HTC campaigns used to optimize the potency of intermediate binders, centering on the Chan-Lam reaction (fig. S7) and amide couplings (fig. S8). Direct biochemical screening of crude reactions identified candidates that were resynthesized and in both cases were able to improve the potency of the parent compound. Soaking of crude reaction mixtures of the most potent biochemical hits into Mpro crystals provided complex structures with the identified hits (Chan-Lam PDBs: 7GJJ/7GJZ, resolution: 1.75Å/1.65Å; Amide coupling PDBs: 7GNL/7GNQ, resolution: 1.68Å/1.53Å). In both cases, new interactions were discovered, explaining the improved activity. Although for the Chan-Lam reaction campaign, the extended compounds occupied the intended P4, for the amide-coupling vector, all compounds extended into the P3/5 pockets.

Fig. 3.. Analysis of 367 complex crystal structures reveals hotspots for ligand engagement and a variety of ways to engage substrate recognition subpockets.

(A) The five substrate recognition subpockets exhibit distinct preferences for intermolecular interactions. The figure highlights the locations of different types of interactions, with the shading indicating the frequency. The bottom row tallies the number of times that each interaction was seen in our structures for different residues. The interaction map was generated using PLIPify (Materials and methods) and summarizes the interactions witnessed across 367 complexes from the perspective of the protein, distinguishing between backbone (bb) and sidechain (sc) interactions (which might be more vulnerable to point mutations). (B) Representative examples of protein-ligand interactions engaging the P1’, P1, P2, and P3-5 subpockets. Hydrogen bonds and π-stacking interactions are depicted as yellow and cyan dashed lines, respectively. The rows above each set of complexes tally the number of times that each interaction was seen with the specific residues within the subpockets. See data S4 for Protein Data Bank (PDB) IDs and crystallography statistics. Single-letter abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

Fig. 4.. Structural plasticity of the binding subpockets.

The subpockets have different degrees of plasticity, which is also dependent on the chemical series (fig. S2). The corners of the figure show the distribution of sidechain root mean square deviation (RMSD) deviations from the structure of MAT-POS-e194df51-1 (middle panel; PDB: 7GAW). The boxes exemplify ligands that significantly deform the pockets.

Fig. 5.. Iterative medicinal chemistry furnished an orally bioavailable inhibitor.

(A) Summary of key medicinal chemistry milestones in developing the initial crowdsourced lead compound into a potent antiviral. X-ray structures for each milestone compound are available via Fragalysis, and each compound can be obtained from Enamine using the corresponding catalog numbers. Retrospective alchemical free-energy calculation predictions for each transformation (ΔΔG_FEP) are shown for each step between milestones, along with the corresponding experimental free-energy difference (ΔΔG_exp) derived from biochemical activities. As positive control, under our assay condition, nirmatrelvir has an IC₅₀ of 2.6 nM. (B) Antiviral activity of MAT-POS-e194df51-1 cellular antiviral assays, with an EC₅₀ of 64 nM in A549-ACE2-TMPRSS2 cells assessing CPE (black; plotted as 100 – percent viability) and 126 nM in HeLa-ACE2 assays (blue; plotted as percentage of infected cells). Both assays were performed with P-gp inhibitors. (C) Histogram comparing antiviral efficacy of all COVID Moonshot compounds measured to date in an A549-ACE2-TMPRSS2 CPE cellular antiviral assay. (D) Detailed cellular antiviral assessment of key compounds composing the synthetic strategy (A) across different cell lines and assay techniques, with and without p-gp inhibitors, demonstrating efficacy of MAT-POS-e194df51-1 in various setups and laboratories. (E) MAT-POS-e194df51-1 shows good cross-reactivity against known circulating variants of SARS-CoV-2 in antiviral cellular assays in a CPE assay in HeLa-ACE2 cells. (F) PK profile of MAT-POS-e194df51-1 in rats with a 2 mg/kg intravenous and 10 mg/kg oral dosing with good oral availability. (G) ADME characteristics of MAT-POS-e194df51-1 demonstrate a favorable safety profile, indicating translational potential of the lead series.