A codon-optimized gene coding for full-length SARS-CoV-2 M^pro fused with the 6×His tag at the C terminus and with GST protein at the N terminus in plasmid vector pGEX6p (Zhang et al., 2020) was kindly provided by Prof. Rolf Hilgenfeld. The M ^pro gene was flanked by sequences of two protease sites for subsequent excision of native full-length M^pro: a site recognized by M^pro for auto-excision (at the N terminus of the M^pro sequence) and a PreScission™ Pro site immediately before the 6×His tag for its removal, as described in (Zhang et al., 2020). The gene of the C145A M^pro mutant was generated by PCR-mediated site-directed mutagenesis using two overlapping primers and the M^pro pGEX6p vector as a template.

The full-length M^pro protein was overproduced in E. coli BL21 (DE3) and purified as described (Zhang et al., 2020) with minor modifications. Namely, the GST-M^pro fusion protein was subjected to self-processing during E. coli expression to prepare M^pro with the intact N terminus. M^pro fusion with the GST protein was employed to improve the M^pro yield and solubility. Next, M^pro-His was purified by IMAC chromatography on TALON (Clontech) and treated with PreSсission™ Pro (M^pro/PreSсission ratio 100:1) for 48 h at 4 °C for 6×His tag removal and obtaining M^pro with the intact C terminus. Then, a mixture of M^pro with PreSсission™ Pro (which contains a GST tag and His tag) was loaded on a GST-Sepharose (Amersham Biosciences) column and a TALON (Clontech) IMAC column, connected in tandem. Pure M^pro without tags was obtained in a flow-throw fraction. Western blot analysis with anti 6×His antibodies revealed that the 6×His tag was completely removed. M^pro was concentrated up to 10 mg/mL in 50 mM Tris (pH 7.5) and stored at −80°C. C145A M^pro mutant protein was obtained in the same way with an additional step of intact N-terminus generation, because the non-active C145A M^pro form cannot process itself during E.coli expression. For this purpose, IMAC-purified GST-C145A M^proHis protein was treated by wild type M^pro (the GST-C145A M^pro-His/M^pro-ratio was 100:1), then the IMAC step was repeated to remove free M^pro. Then C145A M^proHis protein was subjected for PreSсission™ Pro cleavage and further procedures as described above.

The kinetic assays were implemented using the FRET substrate (FRET-S), Dabcyl-KTSAVLQ↓SGFRKM-E(Edans)-NH₂ (BPS Bioscience, United States), and standard covalent inhibitors GC-376, PF-00835231, boceprevir, telaprevir (Selleckchem, United States), and thimerosal (Serva). FRET-S contains a main-protease cleavage site (indicated by the arrow in the sequence above) and was utilized as the substrate in the FRET-based cleavage assay. Stock solutions of the inhibitors were prepared in DMSO (final concentration 5.0 mM).

Virtual screening was performed in July 2020 via the blind docking approach. The apo-structure of M^pro at room temperature [PDB ID 6WQF (Kneller et al., 2020)] was selected as the one representing the most physiologically relevant conditions. The protein structure was optimized in Chimera (Pettersen et al., 2004) using the Dock Prep tool: solvent molecules were deleted, while hydrogens (taking into account hydrogen bonds) and AMBER ff14SB charges (Maier et al., 2015) were added. Docking was performed in DOCK 6.9 (Allen et al., 2015). The grid box was generated to enclose the orthosteric binding site of chain A.

Docking-based virtual screening was performed on molecules previously selected from ZINC15 (Sterling and Irwin, 2015) as potential novel anticoronavirus compounds (hitlist ZINCVS/novel.-smi_id_frq) using the Generative Topographic Mapping approach (Horvath et al., 2020). Three-dimensional coordinates for the molecules were downloaded in MOL2 format from the ZINC15 website via the “Search Many” feature and used without further processing. Docked compounds were ranked by the values of grid_score for the best scored pose, and 22 of those with grid_score < −50 were designated as primary hits. They were grouped according to the molecule core and assessed for commercial availability, thereby leading to three cores available from a local supplier, Alinda (http://www.alinda.ru). The hitlist was then expanded to include compounds from the same classes with different substituents and grid_score < −45, resulting in an experimental assessment list of 10 compounds (Supplementary Table S1). Purchased compounds were used in experiments without further purification. Stock DMSO solutions were prepared with compound concentration of 5 mM.

Binding of inhibitors to SARS-CoV-2 M^pro was monitored by the TSA on a QIAGEN Rotor Gene Q Real Time PCR System. An M^pro solution (5 μM) was mixed with 50 μM inhibitor (or 1% DMSO), a dye (ProteOrange Protein Gel Stain; Lumiprobe, Russia; 5X in the final volume) in the buffer solution (20 mM HEPES, pH 6.5, 120 mM NaCl, 0.4 mM EDTA, 4 mM DTT, and 20% of glycerol) to attain the final volume of 25 μL. After 30 min incubation at 30 °C, each sample was heated to 95 °C with 0.05 °C/s increment, and fluorescence was monitored in the green channel (λ_ex/λ_em = 470/510 nm). T_m was calculated as the maximum of the first derivative of the fluorescence signal in Origin 2017 software (OriginLab).

For these assays, 160 nM M^pro in reaction buffer (20 mM Tris-HCl pH 7.3, 100 mM NaCl, 1.0 mM EDTA, and 1.0 mM DTT) was incubated with or without a tested compound at various concentrations for 30 min at 30 °C. The reaction was initiated by the addition of FRET-S (0–80 μM) in reaction buffer. Substrate is cleaved by M^pro generating a product containing a free Edans group. The dequenching of fluorescence by the cleavage of the substrate catalyzed by M^pro was monitored at 460 nm with excitation at 360 nm on a Thermo Scientific Varioscan plate fluorimeter. The concentration of a fluorescent product was determined according to the calibration curve of free Edans fluorescence covering a concentration range of 0.1–20.0 μM. Initial rates of FRET-S cleavage by M^pro at each inhibitor concentration were computed from reaction kinetic curves of M^pro activity at several substrate concentrations. Kinetic constants (V_max and K _M) were derived by fitting the data to the Michaelis–Menten equation, V = V_max × [S]/(K _M + [S]). After that, k _cat was calculated according to the equation k _cat = V_max/[E].

For the preliminary screening of noncovalent M^pro inhibitors, 160 nM M^pro was incubated with 80 μM inhibitor (or 1.6% DMSO) for 30 min at 30 °C in reaction buffer, and then 16 μM FRET-S was added to initiate the reaction (final volume 20 μL). The fluorescence signal of the reaction was monitored for 2 h at λ_ex/λ_em = 355/460 nm (Thermo Scientific Fluoroskan FLash fluorimeter). The initial rate was calculated by linear regression for the first 10 min of the kinetic progress curves. Residual activities were computed by dividing the initial velocity in the presence of an inhibitor by the initial rate in its absence (DMSO control).

Stopped-flow measurements with fluorescence detection were carried out using a SX.20 stopped-flow spectrometer (Applied Photophysics Ltd., United Kingdom) equipped with a 150-W Xe arc lamp and an optical cell with 2 mm path length. The dead time of the instrument is 1.0 ms. For the analysis of enzyme–substrate interactions, the FRET-S substrate modified with the dye–quencher pair Edans/Dabcyl was utilized. The fluorescence of Edans was excited at λ_ex = 340 nm and monitored at λ_em > 435 nm as transmitted by filter GG-435 (Schott, Mainz, Germany). The binding of M^pro to PF-00835231 was monitored by means of changes in intrinsic fluorescence intensity of the inhibitor. The excitation wavelength was 300 nm, and the emission was monitored using long-pass wavelength filters at λ_em > 370 nm (Corion filter LG-370).

The enzyme was placed in one of the instrument’s syringes and rapidly mixed in the reaction chamber with the substrate, inhibitor, or a substrate/inhibitor mixture from another syringe. The concentration of FRET-S in all the experiments was 2.5 μM, while concentrations of M^pro or its C145A mutant were varied from 0.1 to 3.0 μM. The reported concentrations of reactants are those in the reaction chamber after the mixing. All experiments were conducted at 25 °C in the reaction buffer.

Kinetic simulation of the time course of appearance and disappearance of various reaction intermediates was done by solving a system of differential equations in the DynaFit software (BioKin, Pullman, WA) (Kuzmic, 1996) as described before (Zakharova et al., 2009; Zakharova et al., 2017). The fast kinetic analysis combined with fluorimetry detection of conformational changes is a powerful method that may provide detailed information about mechanisms of enzyme–substrate interaction (Kuznetsov and Fedorova, 2016; Kuznetsov et al., 2017; Kladova et al., 2019; Kuznetsova et al., 2020; Kuznetsov and Fedorova, 2020). This approach is based on fluorescence intensity variation in the course of the reaction owing to sequential formation and subsequent transformation of the enzyme–substrate complex. The stopped-flow fluorescence traces were directly fitted to the fluorescence intensity at any reaction time point as the sum of background fluorescence and fluorescence intensity values of each intermediate complex that contribute to the signal.

The software performs numerical integration of a system of ordinary differential equations with subsequent nonlinear least-squares regression analysis. In the evaluated mechanisms, except for the first bimolecular step, all other reactions are first-order. In the fits, we optimized all relevant rate constants for the forward and reverse reactions as well as specific molar response factors for all intermediate complexes.

During the data processing, the kinetic information was obtained from the temporal behavior of the fluorescence intensity, not from the amplitudes of the specific signal contributions. The response factors for different states resulting from the fits were not used for determining equilibrium constants but rather provided additional information on fluorescence intensity variations in different states of the complex.

To thoroughly characterize the kinetic mechanism underlying catalytic cleavage of a peptide, pre–steady-state kinetic assays based on the Förster resonance energy transfer (FRET) effect were performed. The Dabcyl-KTSAVLQSGFRKM-E(Edans)-NH₂ peptide armed with a dye–quencher pair was used for FRET measurements (Zhu et al., 2020c; Hoffman et al., 2020). FRET analysis could reveal changes in the distance between the dye and quencher in the processes of peptide penetration into the active site and formation of specific contacts between the side chains of the substrate and the respective binding cavities—that subsequently result in the catalytic state, hydrolysis of the peptide bond, and the release the products.

The C145A substitution led to complete elimination of the catalytic activity of the protease owing to a loss of the catalytic thiol group. The interaction of C145A M^pro with the substrate can lead only to its binding and formation of a preincision complex. Indeed, as shown in Figure 3A, the FRET signal during the interaction of C145A M^pro with the substrate increased up to time point of 5 s. It is possible that the increase in the FRET signal reflects increased distance between the fluorogenic Edans residue and quenching Dabcyl residue owing to peptide stretching in the active site of the protease. An analysis of the kinetic curves suggested that the minimal kinetic mechanism of the interaction between the catalytically inactive mutant and FRET substrate involved one-step equilibrium binding (Scheme 1). The rate constants for the forward and reverse reactions are given in Table 2.

The interaction of WT M^pro with the substrate was slower and proceeded up to 100 s but caused a high amplitude increase in the FRET signal (Figure 3B). Such a growth of the FRET signal most likely reflects a release of the incised peptide products from the complex with the enzyme. Taking into account that the catalytically inactive mutant form revealed one-step binding mechanism, we assumed two-step mechanism of product formation when WT protease interacts with the FRET substrate. Indeed, the kinetic curves were satisfactorily described by Scheme 2, containing one equilibrium stage of substrate binding and one irreversible step of hydrolysis and release of the reaction products (Table 2).

It should be noted that the rate constants of the substrate binding were not affected by the C145A substitution. Nevertheless, the total binding constant was approximately fivefold higher in the case of the C145A variant owing to a decrease in k ₋₁. These findings suggested that the Cys145 residue influenced the stability of the enzyme–substrate complex but did not affect the rate of peptide binding.

To verify the kinetic scheme and the rate constants calculated by the global fitting procedure, we determined steady-state reaction parameters: Michaelis constant K _M and catalytic reaction rate constant k _cat (Figure 4). The Michaelis constant K _M calculated from the elementary kinetic constants via the formula K _M = (k _cat + k _–1 )/k ₁ (4.6 μM, Table 2) was three- to sixfold lower than the value obtained by the steady-state analysis (28 μM, Figure 4) or reported earlier [14 μM (Hoffman et al., 2020)]. Catalytic constants k _cat were similar between the stopped-flow and steady-state analyses, suggesting that the fast kinetic pre–steady-state approach allowed us to determine relevant characteristics of the enzymatic reaction. Fitting of the steady-state data by the Hill equation (n = 2, Supplementary Figure S1) also provides similar constant K₅₀ of 50% enzyme saturation (15.7 μM). The determined Hill coefficient indicates positive cooperativity and is in good agreement with the reported data (Lee et al., 2020; Vuong et al., 2020). However, complication of the pre–steady-state kinetic mechanism (Scheme 1) up to the two-substrate binding model does not provide a better fit of the experimental data, supporting independent action of each active site of the enzyme in the pre–steady-state conditions.

The analysis of the PF-00835231 binding kinetics in the course of its interaction with WT or C145A M^pro was performed by the stopped-flow technique with detection of intrinsic florescence intensity of PF-00835231. The association of C145A M^pro with PF-00835231 led to a two-phase increase in the fluorescence intensity up to time point 2 s (Figure 5A). The kinetic curves were satisfactorily described by Scheme 3, which contains two equilibrium stages (Table 3). It is likely that after the formation of the initial complex, there is an additional step of formation of specific interactions between inhibitor moieties and enzyme cavities resulting in full insertion of the inhibitor molecule into the active site of the protease (Figure 6). Of note, the affinity of C145A M^pro for PF-00835231 even without the covalent binding was 2.5 μM, strongly supporting perfect complementarity between the inhibitor and the active site, as first revealed by X-ray crystallography (Hoffman et al., 2020).

The process of interaction of WT M^pro with PF-00835231 is slower (up to time point 10 s) and induces a greater increase in the fluorescence intensity (Figure 5B). It should be noted that in the case of the WT enzyme, the initial phase of the fluorescence intensity increase was significantly slower as compared with C145A M^pro, indicating that the Cys145 residue must play an important role in the formation of the initial complex. As soon as the binding of the inhibitor to C145A M^pro went through two steps, the kinetic curves for the WT enzyme could be described by Scheme 4, which contains two equilibrium steps and one irreversible step (Table 3).

The value of rate constant k ₁ revealed that the formation of the initial complex is approximately 20-fold faster for C145A M^pro, eventually yielding a 10-fold difference in the equilibrium binding constant between the WT and C145A enzymes. On the other hand, rate constants of the second binding step were very similar between the two enzymes, indicating that the specific interaction of the inhibitor and enzyme is independent of the Cys145 residue. The rate constant for the formation of the covalent bond with the inhibitor was similar to the catalytic constant of peptide cleavage (Tables 2 and 3).

Overall, it can be concluded that noncovalent interaction of WT M^pro and the PF-00835231 inhibitor is not as strong as expected from the nanomolar range of the inhibition constants reported in several studies (Hoffman et al., 2020; Boras et al., 2021; de Vries et al., 2021). Nevertheless, the high efficiency of PF-00835231 is explained by the subsequent covalent modification of the enzyme; this modification significantly stabilizes the enzyme–inhibitor complex.

A two-step virtual screening procedure was carried out to prioritize commercially available small-molecule compounds as potential M^pro inhibitors. At the first step, an antiviral chemical space (Nikitina et al., 2019) analysis was performed using the Generative Topographic Mapping approach (Horvath et al., 2020). Of 800 million ZINC compounds, 574 were predicted as hits potentially possessing an anticoronaviral activity. Items of this hitlist were then docked into the active-site cavity of room temperature M^pro crystal structure 6WQF (Kneller et al., 2020), and the best compounds were selected according to the scoring function (Figure 7A, Supplementary Table S1). Ten compounds with grid scores less than −50 were designated as primary hits and subjected to the initial screening of M^pro inhibition (Figure 7B). Three compounds—IBS-E0680092, IBS-E0183442, and IBS-E0474913—showed more than 50% enzyme inhibition, whereas IBS-E0530026 manifested a less prominent inhibitory activity.

These four compounds were chosen for a more detailed comparison with covalent inhibitors of M^pro. First of all, the protein stabilization by inhibitor binding was compared using thermal shift assay (TSA). TSA allows a direct comparison of protein binding efficiency between small-molecule compounds. A substantial change of protein melting temperature (T_m) may be an indicator of tight binding (Figure 8; Table 4). The binding of covalent inhibitors GC-376, PF-00835231, and boceprevir caused changes of protein T_m, pointing to strong stabilization of the enzyme molecule in the covalent complex with these compounds. On the other hand, the interaction of telaprevir and all noncovalent inhibitors did not result in T_m changes, suggesting that the binding of these compounds with the enzyme is less efficient.

Steady-state analysis of the inhibition efficacy of all the tested compounds revealed (Table 4) that the thermal shift is directly related to inhibition constant K _i. Indeed, the decrease of ΔT_m correlated with an increase of the inhibition constant. Moreover, the thermal shift was negligible when K _i approached ∼10–15 μM. Furthermore, the elimination of the M^pro covalent binding with PF-00835231 by means of the C145A mutation also strongly decreased the change in T_m (to ∼2.2°C) as compared to the covalently bound WT adduct (∼14.0°C), consistently with the pre-steady-state data, which yielded an inhibition constant of 2.5 μM (Table 3).

The efficiency of enzymatic hydrolysis of the substrate by M^pro in the presence of a standard inhibitor (boceprevir, telaprevir, GC-376, PF-00835231, or thimerosal; Figure 9A) or a new inhibitor from the virtual screening (Figure 9B) was determined by comparison of the kinetics of the substrate cleavage. The protease was mixed with an inhibitor and kept on ice for 5 min to obtain an enzyme–inhibitor complex. After that, the substrate hydrolysis was initiated by stopped-flow fast mixing of this complex with FRET-S, and the signal was monitored. To estimate the remaining enzymatic activity, the initial slope of the FRET signal increase was calculated (Figure 9C). A comparison of the obtained data allowed us to conclude that both GC-376 and PF-00835231 were the most potent covalent inhibitors of M^pro among the tested ketone-based and organometallic compounds. Boceprevir showed an intermediate level of activity, whereas telaprevir and thimerosal were on the lowest activity tier, in line with the specificity of the design approach and binding-site complementarity for each of these compounds. Indeed, GC-376 and PF-00835231 were specifically designed as coronavirus M^pro inhibitors; boceprevir is less complementary to the binding site but still has a similar molecular size. On the other hand, telaprevir is much larger and engages in fewer specific interactions although it is still able to form a covalent bond with Cys145. Thimerosal, on the contrary, is substantially smaller and does not form specific interactions in the binding site; these features reduce the inhibitory mechanism of thimerosal to pure Cys145 blockage.

An assessment of relative efficacy of M^pro inhibition by the noncovalent binders revealed that two of them, IBS-E0183442 and IBS-E0680092, significantly inhibit the enzymatic activity, albeit less potently than the covalent inhibitors. This observation supports possible further inhibitor optimization via introduction of additional chemical groups to improve specific interactions as well as indicates that such a pre–steady-state analysis may be a good platform for rapid low-cost screening of small molecule compounds to reveal their inhibitory potential.