key: cord-0840345-ujloheyi
authors: Günther, Sebastian; Reinke, Patrick Y. A.; Fernández-García, Yaiza; Lieske, Julia; Lane, Thomas J.; Ginn, Helen M.; Koua, Faisal H. M.; Ehrt, Christiane; Ewert, Wiebke; Oberthuer, Dominik; Yefanov, Oleksandr; Meier, Susanne; Lorenzen, Kristina; Krichel, Boris; Kopicki, Janine-Denise; Gelisio, Luca; Brehm, Wolfgang; Dunkel, Ilona; Seychell, Brandon; Gieseler, Henry; Norton-Baker, Brenna; Escudero-Pérez, Beatriz; Domaracky, Martin; Saouane, Sofiane; Tolstikova, Alexandra; White, Thomas A.; Hänle, Anna; Groessler, Michael; Fleckenstein, Holger; Trost, Fabian; Galchenkova, Marina; Gevorkov, Yaroslav; Li, Chufeng; Awel, Salah; Peck, Ariana; Barthelmess, Miriam; Schlünzen, Frank; Xavier, P. Lourdu; Werner, Nadine; Andaleeb, Hina; Ullah, Najeeb; Falke, Sven; Srinivasan, Vasundara; Franca, Bruno Alves; Schwinzer, Martin; Brognaro, Hévila; Rogers, Cromarte; Melo, Diogo; Zaitsev-Doyle, Jo J.; Knoska, Juraj; Peña Murillo, Gisel E.; Mashhour, Aida Rahmani; Guicking, Filip; Hennicke, Vincent; Fischer, Pontus; Hakanpää, Johanna; Meyer, Jan; Gribbon, Phil; Ellinger, Bernhard; Kuzikov, Maria; Wolf, Markus; Beccari, Andrea R.; Bourenkov, Gleb; Stetten, David von; Pompidor, Guillaume; Bento, Isabel; Panneerselvam, Saravanan; Karpics, Ivars; Schneider, Thomas R.; Garcia Alai, Maria Marta; Niebling, Stephan; Günther, Christian; Schmidt, Christina; Schubert, Robin; Han, Huijong; Boger, Juliane; Monteiro, Diana C. F.; Zhang, Linlin; Sun, Xinyuanyuan; Pletzer-Zelgert, Jonathan; Wollenhaupt, Jan; Feiler, Christian G.; Weiss, Manfred S.; Schulz, Eike-Christian; Mehrabi, Pedram; Karničar, Katarina; Usenik, Aleksandra; Loboda, Jure; Tidow, Henning; Chari, Ashwin; Hilgenfeld, Rolf; Uetrecht, Charlotte; Cox, Russell; Zaliani, Andrea; Beck, Tobias; Rarey, Matthias; Günther, Stephan; Turk, Dusan; Hinrichs, Winfried; Chapman, Henry N.; Pearson, Arwen R.; Betzel, Christian; Meents, Alke
title: Inhibition of SARS-CoV-2 main protease by allosteric drug-binding
date: 2020-11-23
journal: bioRxiv
DOI: 10.1101/2020.11.12.378422
sha: 431d953b2ce4a4244e73231dcd648e5e367039e0
doc_id: 840345
cord_uid: ujloheyi

The coronavirus disease (COVID-19) caused by SARS-CoV-2 is creating tremendous health problems and economical challenges for mankind. To date, no effective drug is available to directly treat the disease and prevent virus spreading. In a search for a drug against COVID-19, we have performed a massive X-ray crystallographic screen of two repurposing drug libraries against the SARS-CoV-2 main protease (Mpro), which is essential for the virus replication and, thus, a potent drug target. In contrast to commonly applied X-ray fragment screening experiments with molecules of low complexity, our screen tested already approved drugs and drugs in clinical trials. From the three-dimensional protein structures, we identified 37 compounds binding to Mpro. In subsequent cell-based viral reduction assays, one peptidomimetic and five non-peptidic compounds showed antiviral activity at non-toxic concentrations. We identified two allosteric binding sites representing attractive targets for drug development against SARS-CoV-2.

(11) Max Planck Institute for the Structure and Dynamics of Matter, Luruper Chaussee 149, 22761 Hamburg, Germany. (12) Deutsches Elektronen Synchrotron (DESY), Photon Science, Notkestrasse 85, 22607, 50 Hamburg, Germany. (13) directly treat the disease and prevent virus spreading. In a search for a drug against COVID-19, we have performed a massive X-ray crystallographic screen of two repurposing drug libraries against the SARS-CoV-2 main protease (M pro ), which is essential for the virus replication and, thus, a potent drug target. In contrast to commonly applied X-ray fragment 95 screening experiments with molecules of low complexity, our screen tested already approved drugs and drugs in clinical trials. From the three-dimensional protein structures, we identified 37 compounds binding to M pro . In subsequent cell-based viral reduction assays, one peptidomimetic and five non-peptidic compounds showed antiviral activity at non-toxic concentrations. We identified two allosteric binding sites representing attractive targets for 100 drug development against SARS-CoV-2.

Infection of host cells by SARS-CoV-2 is critically governed by the complex interplay of several molecular factors of both the host and the virus(1, 2). Coronaviruses are RNA-viruses 105 with a genome of approximately 30,000 nucleotides. The viral open-reading frames, essential for replication of the virus, are expressed as two overlapping large polyproteins, which must be separated into functional subunits for replication and transcription activity(1). This proteolytic cleavage, which is vital for viral reproduction, is primarily accomplished by the main protease (M pro ), also known as 3C-like protease 3CL pro or nsp5. M pro cleaves the viral 110 polyprotein pp1ab at eleven distinct sites. The core cleavage motif is Leu-Gln↓(Ser/Ala/Gly)(1). M pro possesses a chymotrypsin-like fold appended with a C-terminal helical domain, and harbors a catalytic dyad comprised of Cys145 and His41(1). The active site is located in a cleft between the two N-terminal domains of the three-domain structure of the monomer, while the C-terminal helical domain is involved in regulation and dimerization 115 of the enzyme, with a dissociation constant of ~2.5 µM(1). Due to its central and vital involvement in virus replication, M pro is recognized as a prime target for antiviral drug discovery and compound screening activities aiming to identify and optimize drugs which can tackle coronavirus infections(3). Indeed, a number of recent publications confirm the potential of targeting M pro for inhibition of virus replication(1, 2). 120 A rational approach to the identification of new drugs is structure-based drug design (4, 5) .

The first step is target selection followed by biochemical and biophysical characterization and its structure determination. This knowledge forms the basis for subsequent in silico screening of up to millions of potential drug molecules, leading to the identification of potentially binding compounds. The most promising candidates are then subjected to screening in vitro 125 for biological activity. Lead structures are derived from common structural features of these biologically active compounds. Further chemical modifications of lead structures can then create a drug candidate that can be tested in animal models and, finally, clinical trials.

In order to speed up this process and find drug candidates against SARS-CoV-2, we performed a massive X-ray crystallographic screen of the virus' main protease against two 130 repurposing libraries containing in total 5953 unique compounds from the "Fraunhofer IME Repurposing Collection" (6) and the "Safe-in-man" library from Dompé Farmaceutici S.p.A.

Analysis of the derived electron-density maps revealed 37 structures with bound compounds.

Further validation by native mass spectrometry and viral reduction assays led to the identification of six of those compounds showing significant in vitro antiviral activity against 135 SARS-CoV-2, including inhibitors binding at allosteric sites.

In contrast to crystallographic fragment-screening experiments that use small molecules of low molecular weight typically below 200 Da, the repurposing libraries are chemically more complex and contain compounds twice the molecular weight (Fig. 1A) and thus likely to bind more specifically and with higher affinity(7). Due to the higher molecular weights, we performed co-crystallization experiments instead of compound soaking into native crystals (8) . Crystals were grown at a physiological pH-value of 7.5. X-ray data collection was performed at beamlines P11, P13 and P14 at the PETRA III storage ring at DESY. In total, datasets from 6288 crystals were collected over a period of four weeks. From the 5953 unique compounds in our screen, we obtained 3089 high-quality 145 diffraction datasets to a resolution better than 2.5 Å. Datasets from 1152 compounds were suitable for subsequent automated structure refinement followed by cluster analysis (9) and pan dataset density analysis (PanDDA) (10) . In total, 43 compounds were found that bound to M pro . Seven of these compounds had maleate as a counterion and in these structures maleate was found in the active site but not the compounds themselves, resulting in 37 unique 150 binders. A summary of these, together with additional experimental information, is provided in tables S1 and S2. The binding mode could be unambiguously determined for 29 molecules.

The majority of hits were found in the active site of the enzyme. Six of 16 active-site binders covalently bind as thioethers to Cys145, one compound binds covalently as a thiohemiacetal to Cys145, one is coordinated through a zinc ion and eight bind non-covalently. The 155 remaining 13 compounds bind outside the active site at various locations (Fig. 1B) .

Out of the 43 hits from our X-ray screen, 39 compounds were tested for their antiviral activity against SARS-CoV-2 in cell assays. Ten compounds reduced viral RNA replication by at least two orders of magnitude in Vero E6 cells (Fig. S1 ). Further evaluation to determine the effective concentrations that reduced not only vRNA but also SARS-CoV-2 160 infectious particles by 50% (EC 50 ) (Fig. 2) showed that six compounds exhibit either selectivity indexes (SI = CC 50 / EC 50 ) greater than five or hundredfold viral reduction with no cytotoxicity in the tested concentration range (table S3) . 6 In the following we focus on a more detailed description of the most relevant compounds.

The compounds are grouped according to their different binding sites. All other compounds 165 are described in more detail in the supplementary text.

Tolperisone, HEAT and isofloxythepin bind covalently to the active site. Tolperisone is antivirally active (EC 50 = 19.17 µM) and shows no cytotoxicity at 100 µM (Fig. 2) , whereas HEAT and isofloxythepin show activity but unfavorable cytotoxicity. For all three compounds only breakdown products are observed in the active site. Tolperisone and HEAT 170 are β-ketoamines, but we only observe the part of the drug containing the activating ketone, while the remaining part with the amine group is missing in the electron-density maps. The breakdown product of the parent drug is observed to bind as Michael-acceptor to the thiol of Cys145. Similarly, the aromatic ring system of both tolperisone (Fig. 3A) and HEAT ( Fig.   3B ) protrudes into the S1 pocket and forms van der Waals contacts with the backbone of 175 Phe140 and Leu141 and the side chain of Glu166. In addition, the keto group accepts a hydrogen bond from the imidazole side chain of His163. Tolperisone and HEAT bind exclusively in the (S)-configuration. Interestingly, for HEAT, this binding mode was confirmed independently by mass spectrometry (Fig. S2 and table S3). A similar observation has been reported for binding of β-ketoamines to type-1 methionine aminopeptidases, where 180 the parent compound decomposes into an amine and an α,β-unsaturated ketone which subsequently binds to the thiol of the catalytic cysteine(11). This is a typical situation for a pro-drug (12) . Tolperisone is in use as a skeletal muscle relaxant (13) . Isofloxythepin binds similarly as a fragment to Cys145 (Fig. 3C ).

Triglycidyl isocyanurate shows antiviral activity and adopts a covalent and non-covalent 185 binding mode to the M pro active site. In both modes, the compound's central ring sits on top of the catalytic dyad (His41, Cys145) and its three epoxypropyl substituents reach into subsites S1', S1 and S2. The non-covalent binding mode is stabilized by hydrogen bonds to the main chain of Gly143 and Gly166, and to the side chain of His163. In the covalently bound form, one oxirane ring is opened by nucleophilic attack of Cys145 forming a thioether 190 (Fig. 3D) . The use of epoxides as warheads for inhibition of M pro offers another avenue for covalent inhibitors, whereas epoxysuccinyl warheads have been extensively used in biochemistry, cell biology and later in clinical studies (14) . Triglycidyl isocyanurate (teroxirone, Henkel's agent) has been tested as antitumor agent (15) .

Calpeptin shows the highest antiviral activity in the screen, with an EC 50 value in the lower 195 µM range. It binds covalently via its aldehyde group to Cys145, forming a thiohemiacetal.

This peptidomimetic inhibitor occupies substrate pockets S1 to S3, highly similar to inhibitor 17) , calpain inhibitors (18) and other peptidomimetic inhibitors such as N3 (2) and the α-ketoamide 13b(1). The peptidomimetic backbone forms hydrogen bonds to the main chain of His164 and Glu166, whereas the norleucine side chain is in van der Waals 200 contacts with the backbone of Phe140, Leu141 and Asn142 (Fig. 3E ). Calpeptin has known activity against SARS-CoV-2(16). The structure is highly similar to leupeptin, which served as positive control in our screen (Fig. S3B ). In silico docking experiments verified the peptidomimetic compound Calpeptin as a likely M pro binding molecule (table S4) .

MUT056399 is an active-site binding compound without a covalent bond to Cys145 and 205 reduced viral replication. The diphenyl ether core of MUT056399 blocks access to the catalytic site consisting of Cys145 and His41. The terminal carboxamide group occupies pocket S1 and forms hydrogen bonds to the side chain of His163 and the backbone of Phe140 (Fig. 3F) . The other part of the molecule reaches deep into pocket S2, which is enlarged by a shift of the side chain of Met49 out of the substrate binding pocket. MUT056399 was 210 developed as an antibacterial agent against multidrug-resistant Staphylococcus aureus strains (19) .

In general, the enzymatic activity of M pro relies on the architecture of the active site, which critically depends on the dimerization of the enzyme and the correct orientation of the subdomains to each other. In addition to the active site, as the most obvious target for drug 215 development, we discovered two allosteric binding sites of M pro which have previously not been reported. (21), developed to bind to a cysteine in the active site of a tyrosine kinase. But from its observed binding position it is impossible for it to reach into the active site and no evidence for covalent binding to Cys145 is found in the 235 electron-density maps. Pelitinib is an irreversible epidermal growth factor receptor inhibitor and developed as an anticancer agent (22) .

Ifenprodil, RS-102895 and PD-168568 all exhibit an elongated structure, consisting of two aromatic ring systems separated by a linker containing a piperidine or piperazine ring ( Fig   4C) . All three compounds have a distance of at least 12 Å between the terminal aromatic 240 rings. Thus, this binding mode is unlikely to be identified through fragment screening. The hydrophobic pocket in the helical domain is covered by the side chain of Gln256. In our complex structures, this side chain adopts a different conformation. One of the terminal aromatic ring systems is inserted into the hydrophobic groove in the dimerization domain.

The linker moiety stretches across the native dimer interface and the second aromatic ring is 245 positioned close to Asn142, adjacent to the active site loop where residues 141-144 contribute to the pocket S1. In particular, in the case of RS-102895, two hydrogen bonds are formed to the side and main chains of Asn142. In contrast to ifenprodil, RS-102895 and PD-168568 do not exhibit selective antiviral activity (SI<5). All three compounds are GPCR antagonists.

Ifenprodil antagonizes N-methyl-D-aspartate receptors (23) Arg298Ala causes a reorientation of the dimerization domain relative to catalytic domain, leading to changes in the oxyanion hole and destabilization of the S1 pocket by the Nterminus. AT7519 was evaluated for treatment of human cancers (28) and shows weak antiviral activity but a poor selectivity index against SARS-CoV-2 (Fig. 2) . 270 Our X-ray screen revealed six compounds with previously unreported antiviral activity against SARS-CoV-2. Two of them, calpeptin and pelitinib, show strong antiviral activity combined with low cytotoxicity and are suitable for preclinical evaluation. The remaining compounds are valuable lead structures for further drug development. A general advantage of using drug-repurposing libraries for such a screening is the proven bioactivity of the 275 compounds and key properties such as cell-permeability are usually known(29).

The most active compound, calpeptin binds in the active site in the same way as other members of the large class of peptide-based inhibitors that bind as thiohemi-acetals or -ketals to M pro (30) . However, in addition to this peptidomimetic inhibitor, we discovered several non-peptidic inhibitors. Those compounds binding to the active site of M pro contained new 280 Michael acceptors based on β-ketoamines (tolperisone and HEAT). These lead to the formation of thioethers and have not previously been described as prodrugs for viral proteases. We also identified a non-covalent binder, MUT056399, blocking the active site.

Besides this common orthosteric inhibition, we discovered compounds that inhibit the enzyme through binding at two previously unreported allosteric sites of M pro .

The first allosteric site (dimerization domain) is in direct vicinity of the S1 pocket of the adjacent monomer within the native dimer. The potential for antiviral inhibition through this site is demonstrated by ifenprodil and pelitinib. A comparison of coronavirus M pro sequences shows that the compound binding residues of this allosteric site are conserved (Fig. S4) .

Consequently, potential drugs targeting this allosteric binding site can be assumed to be 290 robust against mutational variations and might also be effective against other coronaviruses.

The potential of the second allosteric site, connecting the dimerization and catalytic domain, as a druggable target is demonstrated by the observed weak antiviral activity of AT7519.

Recently, the potential of allosteric inhibition of M pro through modulation of its dimerization has been demonstrated by mass spectrometry (31) .

Protein production and purification 300 The protein was overexpressed in E. coli and purified for subsequent crystallization according to previously published protocols and plasmid constructs(1). Lysis was carried out in 20 mM HEPES buffer supplemented with 150 mM NaCl using ultrasound for cell disruption. After separation of the cell fragments and the dissolved protein, a subsequent nickel NTA column was used to extract the M pro -histidine-tag fusion. The cleavage of the 305 histidine tag was achieved by a 3C protease during an overnight dialysis step. The histidine tag and the 3C protease were removed using a nickel NTA column, and as a final step a gel filtration was performed with an S200 Superdex column.

Co-crystallization with the compounds was achieved mixing 0.23 μL of protein solution (6.25 310 mg/mL) in 20 mM HEPES buffer (pH 7.8) containing 1 mM DTT/TCEP (respectively), 1 mM EDTA, and 150 mM NaCl with 0.22 μL of reservoir solution consisting of 100 mM MIB, pH 7.5, containing 25% w/w PEG 1500 and 5% (v/v) DMSO, and 0.05 μL of a microseed crystal suspension using an Oryx4 pipetting robot (Douglas Instruments). This growth solution was equilibrated by sitting drop vapor diffusion against 40 μL reservoir solution. 315 Prior to crystallization 125 nL droplets of 10 mM compound solutions from the two libraries in DMSO were applied to the wells of SwissCI 96-well plates (2-well or 3-well low profile, respectively) and subsequently dried in vacuum. Taking the crystallization drop volume into account this resulted in a final compound concentration of 2.5 mM and a molar ratio of ~13.6 of compound to protein. To obtain well-diffracting crystals in a reproducible way micro-320 seeding was applied for crystal growth(32). Crystals appeared within a few hours and reached their final size (~200×100×10 µm 3 ) after 2 -3 days. Crystals were manually harvested and flash-frozen in liquid nitrogen for subsequent X-ray diffraction data collection. We aimed at harvesting two crystals per crystallization condition as a compromise between through-put and increasing the probability to collect data from well diffracting crystals. 325 Data collection Data collection was performed at beamlines P11, P13 and P14 at the PETRA III storage ring at DESY in Hamburg within a period of four weeks. Exclusive use of DESY beamline P11 was generously granted by the DESY directorate for the project.

Data processing and structure refinement 330 An automatic data processing and structure refinement pipeline "xia2pipe" as written specifically to support this project. Raw diffraction images from the PETRA III beamlines were processed using three crystallographic integration software packages: XDS(33), autoPROC(34) followed by staraniso (35) , and DIALS via xia2 (36, 37) . Diffraction data quality indicators for all datasets and the 43 hits are summarized in Fig. S5 . In total, 7857 335 unique crystals were harvested and frozen, of which 7258 were studied by X-ray diffraction at PETRA-III. Of these, 5934 produced diffraction data consistent with a protein lattice and were labeled as "successful" experiments. In some cases, multiple datasets were collected on a single crystal, so in total 8304 diffraction experiments were conducted with 6831 successful protein diffraction datasets obtained. As processed by DIALS, these 6831 datasets had an 340 average resolution of 2.12 Å (criterion: CC1/2 > 0.5), CC1/2 of 0.97, and Wilson B of 27.8 Å 2 (Fig. S5) . Crystallographic data of all structures submitted to the PDB are summarized in table S2.

For clustering and hit identification, all datasets were integrated and merged to a resolution of 1.7 Å. In order to reduce the influence of noise for lower resolution datasets, the following 345 processing was applied to standardize the Wilson plot for each dataset: the datasets were split into equally sized bins, each covering 1000 reflections, and a linear fit was applied to the logarithm of the average intensities in each shell. The residual between the data and the Wilson fit was calculated, considering sequentially one additional bin from low to high resolution until the residual exceeded 10%, if applicable. The intensities in all higher 350 resolution bins beyond this point were scaled to fit the calculated Wilson B factor.

The results of each dataset were then automatically refined using Phenix (38) . Refinement began by choosing one of two manually refined starting models (differing in their unit cell, table S2), selecting the starting model with the closest unit cell parameters, then proceeding in four steps: (a) rigid body and ADP refinement, (b) simulated annealing, ADP, and 355 reciprocal space refinement, (c) real-space refinement, and (d) a final round of reciprocal space refinement as well as TLS refinement, with each residue pre-set as a TLS group. This procedure was hand-tuned on 5 test datasets; the procedure and parameters were manually adjusted to minimize Rfree until deemed satisfactory for the continuation of the project. All processing and refinement results were logged in a database, which enabled comparison 360 between methods and improvement over time. All code and parameters needed to reproduce this pipeline are available online (39) .

Hitfinding: cluster4x and PanDDA Analysis

The resulting model structure Cα positions were then ingested into cluster4x (40) , which briefly (a) computes a correlation coefficient between each structure over the position of all 365 C α atoms, (b) performs PCA the resulting correlation matrix, (c) presents 3 chosen principal components to a human, who then manually annotates clusters. Clusters were ordered chronologically and separated into groups of 1500 and subsequently clustered into groups of approximately 60-120 datasets based on a combination of reciprocal and Cα-atom differences using cluster4x. In an earlier version of the software, structure factor amplitudes were used 370 for clustering instead of refined Cα positions, and both methods were applied for hitfinding.

The resulting clusters were then analyzed via PanDDA(10) using default parameters. The resulting PanDDA analyses were manually inspected for hits which were recorded.

Identified hits were further refined by alternating rounds of refinement using refmac (41), 375 phenix.refine (38) or MAIN(42) , interspersed with manual model building in COOT (43) .

To enable a preselection of potentially promising compounds to support the experimental Xray screening effort and to get an idea about the most promising compounds, we pursued a virtual screening workflow consisting of the selection of a representative ensemble of binding 380 site conformations, non-covalent molecular docking and rescoring. We performed this study with 5,575 compounds of the Fraunhofer IME Repurposing Collection. UNICON (44) was applied to prepare the library compounds. To consider binding site flexibility, we used multiple receptor structures. We applied SIENA (45) to extract five representative binding site conformations for the active site of M pro . We chose the structures with the PDB IDs 5RFH, 385 5RFO, 6W63, 6Y2G and 6YB7 The SIENA-derived aligned structures were used and the proteins were preprocessed using Protoss(46) to determine protonation states, tautomeric forms, and hydrogen orientations. The binding site was defined based on the active site ligand of the structure with the PDB ID 6Y2G (ligand ID O6K). A 12.5 Å radius of all ligand atoms was chosen as binding site definition. The new docking and scoring method JAMDA was 390 applied with default settings for the five selected binding sites (47) . Subsequently, HYDE (48) was used for a rescoring of all predicted poses of the library compounds. The 200 highest ranked compounds of all 5,575 compounds according to the HYDE score were extracted. For 70 of these compounds, well-diffracting crystals were obtained in the X-ray screening.

Intriguingly, only calpeptin, a known cysteine protease inhibitor, could be co-crystallized and 395 was found on rank 3 (table S5) with a micropipette puller (P-1000, Sutter instruments) using a squared box filament (2.5 × 2.5 mm 2 , Sutter Instruments) in a two-step program. Subsequently capillaries were gold-coated using a sputter coater (CCU-010, safematic) with 5.0 × 10-2 mbar, 30.0 mA, 100 s, 3 410 runs to vacuum limit 3.0 × 10-2 mbar argon. Native MS was performed using an electrospray quadrupole time-of-flight (ESI-Q-TOF) instrument (Q-TOF2, Micromass/Waters, MS Vision) modified for higher masses (49) . Samples were ionized in positive ion mode with voltages of 1300 V applied at the capillary and of 130 V at the cone. The pressure in the source region was kept at 10 mbar throughout all native MS experiments. For desolvation and 415 dissociation, the pressure in the collision cell was adjusted to 1.5 × 10 -2 mbar argon. Nativelike spectra were obtained at an accelerating voltage of 30 V. To calibrate raw data, CsI (25 mg/ml) spectra were acquired. Calibration and data analysis were carried out with MassLynx 4.1 (Waters) software. In order to determine each inhibitor binding to M pro , peak intensities of zero, one or two bound ligands were analyzed from three independently recorded mass 420 spectra at 30 V acceleration voltage. Results are shown in table S4.

Compounds. All compounds were diluted to a 50 mM concentration in 100% DMSO and stored at -80°C. 

In the following, we discuss those compounds that did not show significant antiviral activity 455 but for which we could determine the binding pose based on the crystal structures.

Isofloxythepin binds as breakdown product (Fig. 3C) . Here, the piperazine group is not found in the crystal structure but the dibenzothiepine moiety is observed in the active site, 460 bound as a thioether to Cys145. The tricyclic system stretches from the S1 across to the S1' pocket. According to the electron-density maps, two orientations of the molecule are possible, with either the fluorine or the isopropyl group placed inside the S1 pocket.

Degradation of the drug with piperazine as the leaving group has been previously reported (50) and was confirmed by mass spectrometry (Fig. S2) . Isofloxythepin is an 465 antagonist of dopamine receptors D1 and D222 and has been tested as a neuroleptic in phase II clinical trials.

Leupeptin is a well-known cysteine protease inhibitor and was therefore included in our screening effort as a positive control (51) . Structurally, it is highly similar to calpeptin. Indeed this peptidomimetic inhibitor also forms a thiohemiacetal and occupies the substrate pocket, 470 much like calpeptin ( Fig. S3B and 3E) . The binding mode is identical to the recently released room-temperature structure of M pro with leupeptin (PDB-ID 6XCH).

Maleate was observed covalently bound in seven structures during hit finding. In all cases maleate served as the counter ion of the applied compound. In these crystal structures the maleate, rather than the applied compound, forms a thioether with the thiol of Cys145, 475 modifying it to succinyl-cysteine. The thiol of Cys145 undergoes a Michael-type nucleophilic attack on the C2 of maleate. A similar adduct has been described for maleate isomerase (52) as an intermediate structure in the isomerization reaction. The covalent adduct is further stabilized by hydrogen bonds to the backbone amide of Gly143 and Cys145 to the carboxylate group (C1) of succinate. The terminal carboxylate (C4) is positioned by 480 hydrogen bonds to the side chain of Asn142 and a water-bridged hydrogen bond to the side chain of His163 (Fig. S3A ).

is covalently linked to Cys145 through nucleophilic substitution of the bromine, leading to thioether formation (Fig. S3C) . The other bromine-alkane chain occupies the S1 pocket while the nitro-imidazole stretches into pocket S2. The substitution of 485 chlorine or hydroxyl for bromines in TH-302 has been demonstrated in cell culture (53) . Our mass spectrometry analysis suggested the loss of a bromine atom (Fig. S2C) .

Triglycidyl isocyanurate shows antiviral activity and adopts a covalent and non-covalent two binding modes to the M pro active site, one covalent and one non-covalent. In both modes, the compound's central ring sits on top of the catalytic dyad (His41, Cys145) and its three 490 epoxypropyl substituents reach into subsites S1', S1 and S2. The non-covalent binding mode is stabilized by hydrogen bonds to the main chain of Gly143 and Gly166, and to the side chain of His163. In the covalently bound form, one oxirane ring is opened by nucleophilic attack of Cys145 forming a thioether (Fig. 3D) . The use of epoxides as warheads for inhibition of M pro offers another avenue for covalent inhibitors, whereas epoxysuccinyl 495 warheads have been extensively used in biochemistry, cell biology and later in clinical studies23. Triglycidyl isocyanurate (teroxirone, Henkel's agent) has been tested as antitumor agent24.

Zinc pyrithione was already demonstrated to have inhibitory activity against SARS-CoV-1 M pro (54) .The pyrithione chelates the Zn 2+ ion which coordinates the thiolate and imidazole of 500 the catalytic dyad residues Cys145 and His41 (Fig. S3D) . The remaining part of the ionophore protrudes out of the active site. This tetrahedral binding mode of zinc has previously been described for other zinc-coordinating compounds in complex with HCoV-229E M pro (55) . Interestingly, antiviral effects against a range of corona-and noncoronaviruses have already been ascribed to zinc pyrithione, although its effect had been 505 attributed to inhibition of RNA-dependent polymerase (56) . Zinc pyrithione exhibits both antifungal and antimicrobial properties and is known in treatment of seborrheic dermatitis.

Adrafinil mainly binds mainly through van der Waals interactions to M pro . In particular, its 510 two phenyl rings are inserted into pockets S1' and S2 (Fig. S3E) . A hydrogen bond is formed between the backbone amide of Cys145 and the hydroxylamine group. The side chain of Met49 is wedged between the two phenyl rings.

Fusidic acid interacts with M pro mainly through hydrophobic interactions, especially through the alkene chain within pocket S2 and the tetracyclic moiety packing against Ser46 (Fig.   515 S3F). Moreover, the carboxylate group forms indirect hydrogen bonds, mediated via two water molecules, to the main chain of Thr26, Gly143 and Cys145. In addition, the same carboxylate group forms a hydrogen bond to an imidazole molecule from the crystallization conditions. This imidazole occupies pocket S1' and mediates hydrogen bonds to the backbone of His41 and Cys44. These indirect interactions offer opportunities for optimization 520 of compounds binding to M pro . Fusidic acid is a well-known bacteriostatic compound, with a steroid core structure.

LSN-2463359 binds mainly to M pro by interaction of the pyridine ring with the S1 pocket ( Fig. S3G ). Besides van der Waals interactions with the β-turn Phe140-Ser144, contributing to the pocket, it also forms a hydrogen bond to the side chain of His163. 525 SEN1269 binds only to the active site of one protomer in the native dimer. This causes a break in the crystallographic symmetry, leading to a different crystallographic space group (table S2) . The central pyrazine ring forms a hydrogen bond to Gln189 (Fig. S3H) . The terminal dimethylaniline moiety sits deep in pocket S2 which is enlarged by an outwards movement of the short α-helix Ser46-Leu50 by 1.7 Å (Ser46 Cα-atom) compared to the 530 native structure. This includes a complete reorientation of the side chain of Met49 which now points outside of the S2 pocket. Additionally, the C-terminus of a crystallographic neighboring M pro protomer is trapped between SEN1269 and part of the S1 pocket, including a hydrogen bond between Asn142 and the backbone amide of Phe305 and Gln306 of the Cterminus. 535 Tretazicar binds at the active site entrance at pocket S3/S4 (Fig. S3I) . The amide group forms hydrogen bonds to the backbone carbonyl of Glu166, the adjacent nitro group forms hydrogen bonds to the side chain of Gln192 and the backbone amide of Thr190.

UNC2327 binds to active site of M pro by stacking its benzothiadiazole ring against the loop Glu166-Pro168 that forms the shallow pocket S3 (Fig. S3J ). This is stabilized by a hydrogen 540 bond between the benzothiadiazole and the main chain carbonyl of Glu166. The piperine ring and adjacent carbonyl are inserted into pocket S1' and interact with Thr25 and His41.

Aurothioglucose 545 In the crystal structure of the aurothioglucose complex, the strong nucleophile Cys145 becomes oxidized to a sulfinic acid. The initial reaction is the disproportionation of Aurothioglucose into Au(0) and a disulfide dimer of thioglucose. This is followed by a cascade of redox reactions of thioglucose, its disulfide and sulfenic acid. A disulfide linkage to thioglucose is only observed at Cys156 on the surface of M pro (Fig. S3K ). Here the 550 thioglucose moiety is located between Lys100 and Lys102.

Glutathione isopropyl ester binds to the surface-exposed Cys156 via a disulfide linkage (Fig. S3L) . Additionally, the ester forms a hydrogen bond to the backbone amide of Tyr101, while the amine of the other arm of the molecule is interacting with the side chain amine of Lys102. 555 Surface pockets AR-42 binds with its phenyl ring to a small hydrophobic pocket in the dimerization domain formed by residues Gly275, Met276, Leu286 and Leu287 (Fig. S3M) . Additionally, the central amide forms a hydrogen bond to the backbone carbonyl of Leu272. 560 AZD6482 binds to a pocket on the back of the catalytic domain, away from the native dimer interface (Fig. S3N) . The nitrobenzene ring is inserted in a pocket formed by His80, Lys88, Leu89 and Lys90. The central aromatic system and morpholine ring lie flat on the surface of M pro . Furthermore, Asn63 forms a hydrogen bond to the keto-group in the pyrimidine ring.

Climbazole binds in a shallow surface pocket, wedged between two crystallographic 565 symmetry-related molecules (Fig. S3O ). Only van der Waals interactions are observed. One monomer contributes with residues Phe103, Val104, Arg105 and Glu178 to this binding site, while the other monomer contributes Asn228, Asn231, Leu232, Met235 and Pro241.

Clonidine also sits in between two crystallographic, symmetry-related molecules and binds through van der Waals interactions (Fig. S3P ). Here one protomer mainly forms the binding 570 site, by contributing Asp33, Aps34 and Ala94. The other protomer contributes Lys236, Tyr237 and Asn238. The amine ring of clonidine forms a loose ring stacking interaction to Tyr237, while a hydrogen bond between the backbone carbonyl of Lys236 and the ring connecting amine of clonidine is formed. The side chain of Lys236 is flipped to the side to make room for the chlorine containing ring system.

Ipidacrine is in contact with two different M pro protomers (Fig. S3Q) . The tricylic ring system is packed against a surface loop, including residues Pro96 and Lys97 as well as Lys12. It also interacts with the end of an α-helix including residues Gln273, Asn274 and Gly275.

Tegafur binds to a in a shallow surface pocket generated by residues Asp33, Pro99, Lys100 580 and Tyr101. The main interaction is through π-stacking of the aromatic ring of Tyr223. The side chain of Lys100 flips away and generates space for the compound (Fig. S3R ).

Tofogliflozin binds to the same hydrophobic pocket as pelitinib, ifenprodil, RS-102895, and 585 PD-168568 but no antiviral activity was observed at 100 µM, the highest concentration tested. In contrast to the previous four compounds, it does not reach across to the opposing protomer in the native dimer. Its main interaction with M pro is via its isobenzofuran moiety that occupies the hydrophobic pocket (Fig. S3S ). were determined by RT-qPCR, immunofocus assays, and the CCK-8 method, respectively. EC 50 for the viral titers reduction are shown. Values were calculated from three independent replicates in one experiment. Individual data points represent mean ± SD. Additional information is provided in table S1. interactions with M pro , detailed compound information, biochemical and cell-based antiviral reduction data. Summary of X-ray crystallographic data processing and refinement statistics. Table S3 .

In vitro antiviral activity, cytotoxicity and selectivity of selected compounds against SARS-CoV-2. EC 50 -half-maximal effective concentration; CC 50 -half-maximal cytotoxic concentration; n.a.-not available. Viral titers, vRNA yield and cell viability were determined by RT-qPCR, immunofocus assays, and the CCK-8 method, respectively. Values were 955 calculated from three independent replicates in one experiment. 

Crystal structure of SARS-CoV-2 main protease provides a basis for design of improved α-ketoamide inhibitors

Structure of M pro from COVID-19 virus and discovery of its inhibitors

From SARS to MERS: crystallographic studies on coronaviral proteases enable antiviral drug design

The Process of Structure-Based Drug Design

A Structure-Based Drug Discovery Paradigm

Identification of inhibitors of SARS-CoV-2 in-vitro cellular toxicity in human (Caco-2) cells using a large scale drug repurposing collection

Molecular Complexity and Its Impact on the Probability of Finding Leads for Drug Discovery

Soaking suggests "alternative facts": Only co-615 crystallization discloses major ligand-induced interface rearrangements of a homodimeric tRNA-binding protein indicating a novel mode-of-inhibition

Pre-clustering data sets using cluster4x improves the signal-to-noise ratio of high-throughput crystallography drug-screening analysis

A multi-crystal method for extracting obscured crystallographic states from conventionally uninterpretable electron density

MRSA Infections: From Classical Treatment to Suicide Drugs

Tolperisone: A Typical Representative of a Class of Centrally Acting Muscle Relaxants with Less Sedative Side Effects

Cysteine 635 cathepsins: From structure, function and regulation to new frontiers

Kenis, Phase I clinical trial with alpha 1,3,5-triglycidyl-s-triazinetrione (NSC-296934)

GC-376, and calpain inhibitors II, XII inhibit SARS-CoV-2 viral replication by targeting the viral main protease

Feline coronavirus drug inhibits the main protease of SARS-CoV-2 and blocks virus replication

Structure and inhibition of the SARS-CoV-2 main protease reveals strategy for developing dual inhibitors against Mpro and cathepsin L

The MUT056399 655 Inhibitor of FabI Is a New Antistaphylococcal Compound

pH-dependent Conformational Flexibility of the SARS-CoV Main Proteinase (Mpro) Dimer: Molecular 660

Dynamics Simulations and Multiple X-ray Structure Analyses

Anilinoquinoline-3-carbonitriles. The Design of an Orally Active, Irreversible Inhibitor of the Tyrosine Kinase Activity of the Epidermal Growth Factor Receptor (EGFR) and the Human Epidermal Growth Factor Receptor-2 (HER-2)

Phase I Study of EKB-569, an Irreversible Inhibitor of the Epidermal Growth Factor Receptor, in Patients With Advanced Solid Tumors

Antagonist properties of the stereoisomers of ifenprodil at NR1A/NR2A 675 and NR1A/NR2B subtypes of the NMDA receptor expressed in Xenopus oocytes

Identification of the Binding Site for a Novel Class of CCR2b Chemokine Receptor 680

Isoindolinone enantiomers having affinity for the dopamine D4 receptor

Ifenprodil and Flavopiridol Identified by Genomewide RNA Interference Screening as Effective Drugs To Ameliorate Murine Acute Lung Injury after Influenza A H5N1 Virus Infection. mSystems

Mechanism for Controlling the Dimer-Monomer Switch and Coupling Dimerization to Catalysis of the Severe Acute Respiratory Syndrome Coronavirus 3C-Like Protease

Identification of N-(4-piperidinyl)-4-(2,6-dichlorobenzoylamino)-1H-pyrazole-3-carboxamide (AT7519), a novel cyclin dependent kinase inhibitor using fragment-based X-ray crystallography and structure based drug design

Drug repurposing: progress, challenges and recommendations

The development of Coronavirus 3C-Like protease (3CLpro) inhibitors from 2010 to 2020

Allosteric Inhibition of 710 the SARS-CoV-2 Main Protease: Insights from Mass Spectrometry Based Assays

Pre-clustering data sets using cluster4x improves the signal-to-noise ratio of high-throughput crystallography drug-screening analysis

F2X-Universal and F2X-Entry: Structurally Diverse Compound Libraries for Crystallographic Fragment Screening

A poised fragment library 720 enables rapid synthetic expansion yielding the first reported inhibitors of PHIP(2), an atypical bromodomain

Microseed matrix screening for optimization in protein crystallization: what have we learned?

Data processing and analysis with the autoPROC toolbox

Scaling diffraction data in the DIALS software package: algorithms and new approaches for multi-crystal scaling

xia2: an expert system for macromolecular crystallography data reduction

Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix

REFMAC5 for the refinement of 745 macromolecular crystal structures

MAIN software for density averaging, model building, structure refinement and validation

UNICON: A Powerful and Easy-to-Use Compound Library Converter

SIENA: Efficient Compilation of Selective Protein Binding Site Ensembles

Protoss: a holistic approach to predict tautomers and protonation states in protein-ligand complexes

A Consistent Scheme for Gradient-Based Optimization of Protein-Ligand Poses

A consistent description of HYdrogen bond and DEhydration energies in protein-ligand complexes: methods behind the HYDE scoring function

Improving the Performance of a Quadrupole Time-of-Flight Instrument for Macromolecular Mass Spectrometry

Determination of isofloxythepin in biological fluids by gas chromatography-mass spectrometry

A Covalent Succinylcysteine-like Intermediate in the Enzyme-Catalyzed Transformation of Maleate to Fumarate by Maleate Isomerase

Cellular pharmacology of evofosfamide (TH-302): A critical re-780 evaluation of its bystander effects

Evaluation of metal-conjugated compounds as inhibitors of 3CL protease of SARS-CoV

Structural Basis of Inhibition Specificities of 3C and 3C-like Proteases by Zinc-coordinating and Peptidomimetic Compounds

Zn2+ Inhibits Coronavirus and Arterivirus RNA Polymerase Activity In Vitro 790 and Zinc Ionophores Block the Replication of These Viruses in Cell Culture

Forschungsgemeinschaft (DFG) -EXC 2056 -project ID 390715994, the Federal Ministry of Education and Research (BMBF) via project 05K19GU4

STOP CORONA" ), the Joachim-Herz-Stiftung Hamburg via the project Infecto-Physics

DT is 815 supported by Slovenian Research Agency (ARRS; research program P1-0048, Infrastructural program IO-0048 -both awarded to D.T.). B.S. was supported by an Exploration Grant from the Boehringer Ingelheim Foundation. R.C. is supported by DFG grants INST 187/621-1 and INST 187/686-1. Author contributions: SeG

Code used in this analysis has been previously published(40). The code for forcing adherence to the Wilson distribution is included in the same repository under a GPLv3 license

Correspondence and requests for data and materials should be addressed to sebastian.guenther@desy.de or alke

Ligands binding non-covalently to the active site: E, adrafinil. F, fusidic acid. G, LSN-2463359

SEN1269 (C-terminus of neighboring M pro protomer shown as pink stick model). I, tretazicar. J, UNC2327. Covalent binders to Cys156: K, aurothioglucose. L, glutathione isopropylester. Other surface pockets: M