A higher flexibility at the SARS-CoV-2 main protease active site compared to SARS-CoV and its potentialities for new inhibitor virtual screening targeting multi-conformers

Abstract

The main-protease (M^pro) catalyzes a crucial step for the SARS-CoV-2 life cycle. The recent SARS-CoV-2 presents the main protease (M^CoV2_pro) with 12 mutations compared to SARS-CoV (M^CoV1_pro). Recent studies point out that these subtle differences lead to mobility variances at the active site loops with functional implications. We use metadynamics simulations and a sort of computational analysis to probe the dynamic, pharmacophoric and catalytic environment differences between the monomers of both enzymes. So, we verify how much intrinsic distinctions are preserved in the functional dimer of M^CoV2_pro, as well as its implications for ligand accessibility and optimized drug screening. We find a significantly higher accessibility to open binding conformers in the M^CoV2_pro monomer compared to M^CoV1_pro. A higher hydration propensity for the M^CoV2_pro S2 loop with the A46S substitution seems to exercise a key role. Quantum calculations suggest that the wider conformations for M^CoV2_pro are less catalytically active in the monomer. However, the statistics for contacts involving the N-finger suggest higher maintenance of this activity at the dimer. Docking analyses suggest that the ability to vary the active site width can be important to improve the access of the ligand to the active site in different ways. So, we carry out a multiconformational virtual screening with different ligand bases. The results point to the importance of taking into account the protein conformational multiplicity for new promissors anti M^CoV2_pro ligands. We hope these results will be useful in prospecting, repurposing and/or designing new anti SARS-CoV-2 drugs.

Communicated by Ramaswamy H. Sarma

Introduction

In December 2019, a cluster of pneumonia cases caused by a novel virus emerged from a local seafood and animal market of Wuhan city, the capital of Hubei province in China. The first genome sequencing analysis indicated that the genes of the novel virus share 80% nucleotide sequence identity to severe acute syndrome coronavirus (SARS-CoV). Further, RNA-based sequencing analysis showed that the virus genome was highly similar to BatCoVRaTG13, a bat coronavirus (CoV), with a sequence identity of 96.2% (Zhou et al., 2020). Therefore, this novel virus has been named as severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) by the International Committee on Taxonomy of Viruses, and pneumonia caused by it is called coronavirus disease 2019 (COVID-19). Currently, the COVID-19 has evolved to be transmitted by human-to-human and has spread rapidly around the world. People were submitted to social distancing and many countries have adopted quarantine regimes to stop the disease dissemination. Although an increasing number of different vaccines are being approved by regulatory agencies and administered worldwide, national health authorities continue to recommend the use of medical countermeasures and efforts to develop more effective drugs, given the limitations and uncertainties inherent in vaccines, especially in terms of new SARS-CoV-2 variants (Coronavirus disease (COVID-19): Vaccines). The growing set of these variants in different lineages has raised concerns about the susceptibility of SARS-CoV-2 to escape neutralizing antibodies, decreasing the effectivity of vaccines and increasing the risk of reinfection (Zhou et al., 2021). In this sense, it is urgent to screen antivirals as complementary strategies in the treatment or prophylaxis of COVID-19.

One of the best-characterized drug targets among CoVs is the viral 3 C–like protease (3CL^pro or main protease M^pro) encoded by NSP5 (Figure 1). The polyprotein pp1ab and pp1a (encoded by ORF1ab and ORF1a) are cleaved by papain–like protease (PL^pro) and the 3CL^pro, to produce NSPs such as RNA-dependent RNA polymerase and helicase. Thus, such enzymes play an essential role in viral replication and its inhibition interrupts the viral life cycle (Zumla et al., 2016). Further, M^pro has no human homologous with a similar cleavage specificity, which contributes to the idea that inhibitors for this enzyme are unlikely to be toxic (Xiong et al., 2021).

Figure 1.

Crystallographic structures showing the mains proteases: (A) M^Cov1_Pro (PDB code 2BX4) from SARS-CoV and (B) M^Cov2_Pro (PDB code 6LU7) from SARS-CoV-2. Domains I, II, III, Linker and N-Finger are shown in cartoons coloured in green, magenta, orange, yellow and blue, respectively. The active pocket is highlighted for each structure and S1₁, S1₂, S2 and S4 loops are shown in blue, green, pink and yellow sticks. Cyan sticks represent the catalytic dyad (His41 and Cys145) in both cases and the region of the mutation Ala46/Ser46 is depicted by red stars. Gray sticks represent remaining residues around 5 Å from the catalytic dyad.

M^pro is highly conserved among CoVs, sharing over 90% sequence identity, differing only 12 residues between the SARS-CoV M^pro(M^CoV1_pro) and SARS-CoV-2 M^pro(M^CoV2_pro) (Anand et al., 2002; Wang et al., 2016; Xue et al., 2008; Yang et al., 2003). This enzyme is composed of three domains (Figure 1): (i) domain I, residues 8–101; (ii) domain II, residues 102–184; and (iii) domain III, residues 201–303. Domains I and II have an antiparallel β-barrel structure, and domain III contains five α-helices grouped into an antiparallel globular cluster and linked to domain II by a loop region (residues 185–200). The substrate binding-site is in the deep cleft between domains I and II and formed by the catalytic dyad His41 and Cys145 (Jin et al., 2020).

Recent conventional molecular dynamics (MD) studies have provided new perspectives about the flexibility of the M^CoV2_pro and M^CoV1_pro binding sites, such as the assistance of the different loops on this flexibility, and its correlation with dimerization and ligand attachment (Bzówka et al., 2020; Suárez & Díaz, 2020). The flexible loop C44-P52 seems to regulate the entrance of potential inhibitors into the binding site (Bzówka et al., 2020). Significant differences were found in this same study concerning both the flexibility of this loop and the accessibility to different chemical groups to the M^CoV1_pro and M^CoV2_pro respective active sites. In the same survey, a bioinformatic analysis showed that such a loop has a potential mutability that does not affect folding. Still, it may increase the inability of putative inhibitors to access the binding pocket, which can be a challenge for drug development. Another recent virtual screening/MM-PBSA study points to the loops around the M^CoV2_pro active site as the major pharmacophoric hot spots (Mittal et al., 2020). In other MD study, initial insights have been obtained about simultaneous changes in the dynamics of the loops around the active site and the interdomain mobility, both at the dimerization and at substrate association, for M^CoV2_pro (Bzówka et al., 2020). These findings have been associated with the experimental evidence that suggests reciprocal influences of the dimerization at the substrate binding and catalysis (Hilgenfeld, 2014; Shi et al., 2004), as well as of the substrate binding at the dimerization affinity for M^CoV1pro in turn (Cheng et al., 2010).

The importance to confront the coronaviruses M_pro’s characteristic mobility and proper conformational space to optimize drug screening procedures has encouraged a crescent number of MD studies with a higher simulation time (the longer, up to now, a 100 µs simulation (Shaw)). Recently, a 15.14 µs unsupervised adaptive sampling molecular dynamics (asMD) for the dimeric M^CoV2_pro (in which these 15.14 µs have been sampled on iterations of subsequent sets of 10 ns simulations and using a bias algorithm to select the seeding for each iteration from the previously less visited configurations) has explored a substantially higher conformational space than this same longer and single one 100 µs simulation (Jaffrelot Inizan et al., 2021). This illustrates two principles: the power of adequately chosen unconventional high sampling simulation methods compared to equilibrium techniques to probe the conformational space of proteins; the fact that the proper choice of an enhanced sampling method compatible to the collective variable of interest provides a more meaningful sampling (and in a computationally cheaper way) than huge equilibrium simulations. In fact, there are significant energetic barriers between the representative conformational states, difficult to cross by equilibrium molecular dynamics (even at the micro or millisecond scales), a notoriously simpler task for ‘smartly’ chosen enhanced sampling approaches at substantially lower simulation time scales (Bernardi et al., 2015; Lazim et al., 2020; Rodriguez-Bussey et al., 2016).

The loop flexibility and the ‘breathing’ movements at the active site have been related as determinant to the substrate and antiviral accessibility for key viral proteases (Cheng et al., 2010; Tóth & Borics, 2006). In this way, the differential mobility at such loops is a crucial issue for understanding the differences in the biology, infectivity and pharmacological attributes from one viral strain to the other. For the coronaviruses’ main proteases, for which the major pharmacophoric subsites are located around such loops, these issues are still more determinant. The adequate comprehension of the intrinsic mobility of such loops, its correlation with the active site ligand accessibility and the way in which such features are modulated by single amino acid substitutions at different regions, will be crucial on drug screening procedures considering potential new M_pro variants. Conventional MD studies have pointed glimpses about the dynamics of such loops and their correlation with substrate accessibility and inter-domain motion in SARS-CoV main proteases (Bzówka et al., 2020; Cheng et al., 2010; Mittal et al., 2020). However, this mobility is represented mainly by high amplitude movements, not wholly accessible by conventional equilibrium simulations. In this sense, perturbative simulation techniques have already proved to be efficient at the characterization of the free energy landscape (FEL) related to the movements of these loops and their correlations with the whole protein dynamics (Lima et al., 2020; Mahanti et al., 2016; Tóth & Borics, 2006).

In this study, we carry out computational comparison between the intrinsic conformational/dynamics space of such loops in M^CoV1_pro and M^CoV2_pro monomers, followed by the exploration of the sampled M^CoV2_pro^’s active site space on a extensive flexible virtual screening approach. For this end, we perform both well-tempered as nontempered metadynamics simulations (MetaDy) for each protein using as collective variables a pair of angles involving the four major active site loops and that have shown the higher mobility and pharmacophoric influences on crystallographic structures with different ligands (Barducci et al., 2008). In sequence, we analyse and compare the conformational, energetics, mutational and physicochemical attributes accessed by the MetaDy and by bioinformatics tools for the two proteins.

Between working with monomeric or dimeric form for the MetaDy studies, we strategically chose the first one. In this regard, we found a precedent in the use of M_pro monomers as drug targets, but with different approaches (Bolcato et al., 2020). The fact is that there is strong evidence that the active form of Mpros is dimeric, although a monomer-dimer equilibrium coexists (Xiong et al., 2021). Moreover, substrate binding to monomers can favor the dimeric state (Cheng et al., 2010). Even though only dimers are enzymatically active, some authors propose M_pro catalytic cycle models in which up (activated) and down (inactivated) states can be found at equilibrium in monomers (see figure 8 in Cheng et al., 2010 and figure 29 in Wan et al., 2020). But, only up monomers are capable of dimerization, especially those with bound substrates. Another recent study raised a ‘structure-based recognition’ hypothesis where the structural heterogeneity (more than the induced-fit theory) may be the main determinant of the greater diversity of M_pro's proteolytic activity, known to be able to recognize both N- and C-termini cleavage sites (Behnam, 2021). All in all, it seems to us that monomers have the potential to be alternative drug targets besides dimers, as they allow selection of ligands that can bind at a wider number of relevant conformational states, enabling both the virtual screening of competitive inhibitors (monomers near up state) and allosteric inhibitors (monomers near down state).

Figure 8.

Variation in the hydration enthalpy with the adoption of broader conformations for M^CoV2_pro and M^CoV1_pro. Semi-empirical QM estimated hydration enthalpy variation with the R_Av value in angstroms for the metadynamics representative conformers. The plots were carried considering the cluster centroids from Figure 3 ordered according to the crescent R_Av value and performed as window plots on a three-by-three mode. The respective values, correlations and deviation bars estimated with and without the COSMO implicit solvation are depicted for the M^CoV1_pro (blue) and M^CoV2_pro (red).

That is why we implement the two types of MetaDy aforementioned: well-tempered and nontempered. The former, its time-smoothed bias allows (at relatively short simulation times) an exploration of conformational states not so far from the starting point (the crystallographic conformation of the monomer in the dimer), keeping it closer to the up state. The latter, its free bias enables (at the same simulation time) exploration of conformational states more unrestrictedly, giving the monomer freedom to move further away from the crystallographic conformation, consequently, more chances of reaching down states. For both kinds of simulations and for both systems we have inspected the evolution of the potential of mean force (PMF) output files in order to properly calculate the free energy landscapes (FEL) only when (and immediately so) the FEL convergence was obtained (see the Theoretical Methods).

However, in order to analyse the extensibility and/or the adaptability of our findings with monomers to functional dimer, we also perform and compare the conformational statistics relevant to the binding site accessibility and activity with those obtained in freely available multiscale M^Cov2_Pro dimer simulations (Jaffrelot Inizan et al., 2021; Shaw).

We complement such MetaDy simulations with quantum mechanics (QM) methods to analyze and compare the enthalpies of solvation and electronic density descriptors of the active site of each M_Pro's main conformers, in an effort to raise hypothesis with more detailed chemical explanations about the greater flexibility of the active site found for M^CoV2_pro compared to M^CoV2_pro, as well as about the low propensity of catalytic activity for the monomers.

Then, we evaluate the implications for the differences at the M^Cov1_Pro and M^Cov2_Pro conformational space (mainly around at active site) on the druggability of a compound for which the differential affinity between these proteins is already reported. Finally, from the representative conformers of the three less energetic clusters accessed by the M^Cov2_Pro MetaDy, as well as the MD equilibrated crystallographic conformation, we carry out a multiconformer virtual screening strategy using three robust ligand databases: one for druggable compounds (DrugBank) (Wishart et al., 2018), one for commercially available compounds (ZINC15) (Irwin et al., 2012). and the other for natural compounds (SistematX) (Scotti et al., 2018). The major results point to the higher conformational promiscuity for the M_Pro from the current viruses compared to the 2009 one, the hydration changes due mainly the A46S substitution between both proteins as a crucial participant on such effect, as well as possibility of gain in recall when this robust conformational coverage could be explored on flexible virtual screening approaches. We hope these results can be useful for the rational design of new and more effective anti-SARS-CoV-2 drugs.

Theoretical methods

Pharmacophoric subsite analyses

M^pro PDB structures from SARS-CoV and SARS-CoV-2 available on 4 May 2020 were aligned and analyzed with the software Pymol. A qualitative analysis of all ligands bound to the active site of these structures was performed, focusing on protein–ligand interactions and on the diversity and chemical properties of ligands that bind to each subsite. Protein–ligand interactions in available PDBs were also analyzed with nAPOLI (Fassio et al., 2020).

Equilibrium molecular dynamics procedure

To explore the free energy landscape of the M_pro protomers from SARS-CoV and SARS-CoV-2 it was necessary to previously carry out an equilibrium MD stage. For this, the X-ray structures from the M^CoV1_pro (PDB code 2BX4) and M^CoV2_pro P(PDB code 6LU7) were used as start geometries (for this last one, previously removing the crystallographic ligand). MD simulations were performed in NAMD software version 2.12 using FF14SB AMBER force field (Maier et al., 2015; Phillips et al., 2005). Further, both monomers were inserted in a cubic water box of 20 Å containing TIP3P waters and ions to the neutrality and then, such simulations were performed according to the following parameters: (i) periodic boundary conditions, restriction of vibration in covalent bonds involving hydrogen atoms, HOH angle and OH bond distance of TIP3P water molecules (SHAKE algorithm); (iii) time steps equal to 2 fs; (iv) electrostatic interaction cutoff of 9.0 A for all steps of the simulation. Before equilibrium simulation in the NPT ensemble, the geometries were optimized by sequential stages of minimization, heating and pressurization. Then, the geometries were subjected to the equilibrium stage for 10 ns with a constant pressure and temperature of 1.0 atm and 310 K, respectively.

Metadynamics procedures and analyses

The MetaDy simulations were performed in NAMD 2.12 software using ff99SB AMBER force field at 300 K and 1 atm (Lindorff-Larsen et al., 2010; Phillips et al., 2005). The height of the Gaussians for the MetaDy was set to 0.42 kJ ·mol⁻¹, added every 2 ps with a width of 1.77 grids. All the systems used the same set of collective variables (CVs). The first CV was the angle between the center of mass of the loops (CV_ang1): 166–171 (S1₁)/17–22, 25–32 (β-sheet)/45–53 (S2), varying between −20° and 160° with an increment of 7°. The second CV was the angle between the center of mass of the loops (CV_ang2): 140–145 (S1₂)/17–22, 25–32 (β-sheet)/186–191 (S4, S5), varying between −20° and 160° with an increment (grid) of 7°. Harmonic restraints were used on the β-sheet 17–22, 25–32. For the well-tempered MetaDy, the bias temperature was 1490 K. The convergence of each MetaDy simulation (both well-tempered as nontempered) and for each system was properly checked according to the protocol described by Incerti et al. (2017), i.e. calculating the progress of the free energy landscapes (FEL) along the time and verifying about the convergence of the major minima (Supporting Information Figure S11). The equations for MetaDy projections were used similarly to Brandt et al. (2016). Similarly to recent works of our group, we used the colony method adapted by Martínez et al. to estimate the differences of conformational entropy between M^CoV1_pro and M^CoV2_pro over the CVs space (Martínez et al., 2009; Rocha & Lima, 2019). Animated gifs of both MetaDy for both M_pro can be seen in Supporting Information.

Potential of Mean Force Maps and Projections of the MetaDy at a Single Collective Variable

The resulting trajectories were visualized and analyzed with VMD software and the potential maps were made using ‘home-made’ R scripts. The MetaDy projections were made following the methodology of the similar work carried out by Brandt et al. (2016). The projections of the MetaDy simulations trajectories onto collective variables (CVs) was calculated as follows:

$$\begin{array}{c}{\mathrm{\omega }}_{{\text{CV}}_{\text{ang}1}}(x)=({\text{CV}}_{\text{ang}2})x\\ {\mathrm{\beta }}_{{\text{CV}}_{\text{ang}1}}(y)=({\text{CV}}_{\text{ang}1})y\end{array}$$ (1)

Where the projection ω for CV_ang1 is a function of a given value of CV_ang1 (x) and the minimum in CV_ang2 at this specific CV_ang1(x) value. Similarly, for CV_ang2, the projection β is a function of a given value of CV_ang2 (y) and the minimum in CV_ang1 at this specific CV_ang2(y) value.

Clustering procedures and obtaining of representative conformers

The well-tempered and nontempered trajectories were merged, aligned and first divided into energetic states between 4 to 84 and 4 to 42 kJ mol⁻¹ for M^Cov1_Pro and M^Cov2_Pro, respectively, and then, subdivided into minor energy terms of 10.5 kJ mol⁻¹. A clustering hierarchical algorithm was applied to extract representative geometries from each one of the minima energy ranges. Additionally, adjustments in the epsilon term of the algorithm were carried out to improve the number of clusters. Then, the centroids of each cluster were sampled and manually filtered to reach all the conformational sets of the active site. In this procedure, when two or more centroids belonging to different energetic states were significantly closer at the two-dimensional FEL map, only the less energetic centroid was considered. At last, the filtered geometries were used as input for the next steps of this study.

CV space entropy estimation

The colony method adapted by Martínez et al. was applied to estimate the differences of conformational entropy between M^Cov1_Pro and M^Cov2_Pro at the CV_ang1 vs. CV_ang2 space (Martínez et al., 2009). Briefly, this method assumes that the conformational entropy can be recovered over the molecular dynamics simulations by clustering the microstates in the 6D phase space, which consists of coordinates in X, Y and Z vs. velocities in X, Y and Z. This 6D phase space can be approximated to a 2D phase space simply by representing the coordinates by representative coordinates (in this case, the MetaDy collective variables were used). Then, this approximated space is recursively divided into grids with decreasing steps, in which each bin can be assumed as a microstate. For each step of grid division, the number of occupied microstates is recovered and applied to a Boltzmann like approximation as:

$$\Delta S_{\text{Cov}1-2}^{\text{CVs}}=k_{\mathrm{B}}\hspace{0.17em}\text{ln}\hspace{0.17em}\left(\frac{{\omega }_{\text{Cov}1}^{\text{CVs}}}{{\omega }_{\text{Cov}2}^{\text{CVs}}}\right)$$ (2)

where k_B is the Boltzmann constant, ω^Cvs_Cov1 and ω^Cvs_Cov2 are the respective number of occupied bins for the M^Cov1_Pro and M^Cov2_Pro active sites at the CV_ang1 vs. CV_ang2 space and ΔS^CVs_Cov1-2 is the configurational entropy at this same space from M^Cov1_Pro to M^Cov2_Pro. Such ΔS^CVs_Cov1-2 as a function of the ratio of occupied bins between each system for different grid subdivisions reaches a maximum when the FEL subdivision is significant, but not excessive. At the adapted colony method, this maximum is taken as the more accurate estimation for the entropy difference between both systems. For a deeper background, we suggest the study carried by Martínez et al. (2009). Here, this adapted colony method entropy variation was recovered using an in-house script similar to previous works of the group (Rocha & Lima, 2019).

Semi-empirical quantum calculations

The enthalpies of solvation were estimated by using the PM7 semi-empirical method available on MOPAC 2016 software by having the centroid geometries of each cluster as input (see Metadynamics Procedures and Analyses section) (Stewart, 2013; Stewart Computational Chemistry). The MOZYME method and COSMO implicit solvent model were used under a dielectric constant of 78.4 (Klamt & Schüürmann, 1993; Stewart, 1996). We performed NCI calculations to explore the intermolecular interactions of water molecules around the Ala46 residue in SARS-CoV and Ser46 in SARS-CoV-2. The different conformations of both molecular structures of Mpro were used, taking into account their opening movement: closed structure (A, B), partially open (J, G) and fully open (N, O) for SARS-CoV and SARS–CoV-2, respectively.

The calculations were carried out considering a solvation shell with 2.500 water molecules around the protein structure against protein alone at the vacuum. So, the enthalpy of solvation was calculated according to

$$\Delta H_f(\text{solvation}=\Delta H_f(\text{protein}+\text{solvent})-\Delta H_f(\text{protein})$$ (3)

Electronic density quantum descriptors for the catalytic residues

Reactivity descriptors (RDs) are quantities that can be produced through mathematical manipulations of the calculated electronic structure of a molecular structure (Geerlings et al., 2003). RDs, such as Electrophilic Attack Susceptibility (EAS), Nucleophilic (NAS) and local hardness, can be used to profile the enzymatic catalysis trends (Grillo, Urquiza-Carvalho, Bachega, et al., 2020; Grillo, Urquiza-Carvalho, Chaves, et al., 2020). The same RDs were used in the present study, calculated by the PRIMoRDiA software (Grillo, Urquiza-Carvalho, Rocha, 2020), that contains these new implementations and efficiently parses the large outputs from Quantum Chemistry Packages. Thus, in this work we characterize the propensity of the proton transfer from Cys145 to His41 using the local hardness, and the nucleophilicity of the Cys145, towards to a peptide substrate that would bind with the protein, using the EAS RD.

Bioinformatics ΔΔG Estimation for M^Cov1_Pro and M^Cov2_Pro Single Residue Permutations

In order to evaluate the impact of amino acid permutations from M^Cov1_Pro to M^Cov2_Pro and the reverse, we predicted the effect of single point mutations on the stability of their MD simulation representative conformers. Usually, this effect is measured by changes in the alterations of folding energies (ΔΔG) between the wild-type and a mutant. In this article, we performed the ΔΔG estimation for M^Cov1_Pro and M^Cov2_Pro amino acid permutations using CUPSAT (Parthiban et al., 2006). We chose this software among many others because it accepts as input a generated PDB file (not deposited on Protein Data Bank) and it is more sensible to mutations, that is, it presents higher differences in the predicted ΔΔG values along the MD conformers than other methods.

We submitted representative conformers of M^Cov1_Pro and M^Cov2_Pro MD simulations to CUPSAT for predicting reverse substitutions from their enzymes at the 12 amino acid substitutions from the first to the second: 35(T/V), 46 (A/S), 65 (S/N), 86 (L/V), 88 (R/K), 94 (S/A), 134 (H/F), 180 (K/N), 202(L/V), 267 (A/S), 285 (T/A), 286 (I/L). We also performed reverse substitutions from M^Cov2_Pro and M^Cov1_Pro: 35 (V/T), 46 (S/A), 65 (N/S), 86 (V/L), 88 (K/R), 94 (A/S), 134 (F/H), 180 (N/K), 202 (S/A), 285 (A/T), 286 (L/I). All the results regarding the effect of single mutations are presented in Supporting Information Figure S1.

Protein: solvent noncovalent interaction estimations

Noncovalent interactions (NCIs) play a pivotal role in biological environments (Johnson et al., 2010; Williams et al., 2004). These interactions depend on the electronic effects of electron density and can be estimated from quantum chemistry calculations (Christensen et al., 2016; Rozengurt and Heppel, 1975). We carried out these calculations concerning the solvent interactions at position 46 for both enzymes using NCIPLOT that produces isosurfaces of topological quantities of electron density (Boto et al., 2020). This points out the type of intermolecular interactions this region of the system does, be it an atom, part of the molecule or residue.

Ligand virtual screening and docking methodologies

In order to find a selective hit for the M_pro active site, a hierarchical virtual screening cascade was performed in two sequential steps: (i) structure-based pharmacophore screening and (ii) ensemble docking. The following ligand databases were used in the virtual screening: (i) Drugbank, (ii) ZINC15 database and (iii) SistematX (Irwin et al., 2012; Scotti et al., 2018; Wishart et al., 2018).

Virtual Screening library preparation

The virtual screening libraries were selected considering three main perspectives: (i) fast clinical application; which includes FDA drug approved and investigational drugs; (ii) synthesis facilitated; here, we adopt the perspective of natural products; and (iii) long-term clinical application; includes virtual screening compounds from ZINC database, which includes compounds without biological activity described. For the ZINC database, compounds were retrieved using the following parameters: (i) Representation: 3D; (ii) Reactivity: Standard; (iii) Purchasability: In stock; (iv) pH: Reference Models; (v) Charge: Neutral. Further, the remaining compounds were subjected to a multiconformational search by using OMEGA software (Hawkins et al., 2010), which was set to find 50 low energy conformers for each compound.

Structure-based pharmacophore screening

In an attempt to increase the chance to find a selective hit into the ZINC database, which in turn has more than 1 million compounds, the binding pose of N3 inhibitor (PDB ID 6LU7) was used as a pharmacophore model. For this, we used the ROCS software (OpenEye Scientific) (Hawkins et al., 2007) to build a shape query. In the ROCS search, the conformers of the VS library were aligned with the shape query and a similarity score (Tanimoto Combo [TC]) was estimated. A high score means that the compound is similar to the shape query, thus, once the N3 is a potent inhibitor of M_pro from SARS-CoV-2, it is expected that similar compounds to its query also have a similar biological activity. The remaining compounds of the pharmacophore screening as well as the Drugbank and Sistematx compounds were subjected to molecular docking procedure.

Preparation of the receptors and ligands

The X-ray structure of the M_pro (PDB ID 6LU7) as well as the MetaDy conformers G, H and I were used as receptors for the molecular docking step. The geometries of these receptors were prepared according to the FF14SB force field by using the tLeap software, included in the Ambertools package; ligands were prepared considering the AM1-BCC atomic charges by using the Antechamber software (Maier et al., 2015).

Molecular docking procedures

In order to validate the docking protocol, the GOLD and Molegro Virtual Docker (MVD) softwares were used. In summary, we use the 6XHM and 6XHL crystallographic structures and the MetaDy conformers (Jones et al., 1997; Thomsen & Christensen, 2006). The structures were aligned according to the backbone and the active site was defined considering the same area of the virtual screening stage. The receptors and ligands were prepared according to the protocol presented in the previous section. For the GOLD, the standard configuration was used, with the exception, only, of the total executions of the genetic algorithm, which was modified to 30. In this validation protocol, in order to make the different docking score metrics from GOLD and MVD comparable we have created a ‘normalized relative score’ function. First, once the best scores at the MVD metric are negative, while for GOLD they are positive, we have taken the negative of the GOLD scores. Second, once our major interest in this validation was to verify the docking score variation at the MetaDy decoys concerning the redocking at the respective crystallographic structures, we have computed the variation of each score related to the proper crystallographic pose (6XHM for M^Cov2_Pro decoys and 6XHL for M^Cov1_Pro). Finally, as the significant score amplitudes for both softwares are also distinct, we have divided the score values recovered at each respective software by our data amplitude using that same program (i.e. dividing by the higher score variation considering together both M^Cov1_Pro as M^Cov2_Pro). While this normalized relative score function is shown and discussed along the validation section at the principal text, the ‘raw’ values for the respective GOLD and MVD docking outputs are also depicted at the Supporting Information Figure S13.

We also have used Molegro Virtual Docker, v. 6.0.1 (MVD) (Molexus IVS Rørth Ellevej 3, Odder, Denmark) (Thomsen & Christensen, 2006), two scoring functions Moldock and rerank were selected (Molexus IVS Rørth Ellevej 3, Odder, Denmark) and GPU screening search algorithm was performed. The following parameter settings of software package were used (Score function: MolDock and rerank scores; Ligand evaluation: Internal ES, Internal HBond, Sp2–Sp2 Torsions, all checked; Number of runs: 10 runs; Algorithm: GPU screening; Maximum Interactions: 1500; Simultaneous evaluation: 256; Max. number of poses returned: 10). The docking procedure was performed using a center grid: x = 78.5; Y = 65; z = 58, with 16 A in radius and 0.38 in resolution to cover the ligand-binding site of the M_pro structure.

Rescore

The Table 1 shows the first 10 hits of each of the three bases (Drugbank, ZINC15 SistematX) when the screening/docking is performed only on the crystallographic pose (Rigid) and when it is performed for the crystallographic and the three lower energy poses of MetaDy (Flexible).

Table 1.

Molegro’s docking scores.

DB	Hit	Rigid		Flexible
		ID	Score	ID	Min	ID	Max	ID	Average	ID	Boltzmann
Drugbank	1	DB12206	–160.29	DB14792	–199.41	DB04461	–140.98	DB14792	–157.98 ± 37.65	DB14792	–123.06
	2	DB15435	–147.03	DB04461	–187.63	DB13648	–139.29	DB04461	–157.57 ± 26.08	DB04461	–115.81
	3	DB12234	–142.19	DB12760	–161.48	DB07458	–137.00	DB13648	–144.44 ± 6.90	DB12760	–99.71
	4	DB03638	–141.22	DB14850	–160.13	DB03777	–129.61	DB07458	–142.20 ± 5.79	DB14850	–98.88
	5	DB03777	–139.90	DB06809	–158.79	DB14850	–129.25	DB14850	–141.37 ± 16.47	DB06809	–98.05
	6	DB06976	–137.83	DB07796	–158.24	DB07456	–128.45	DB11829	–140.97 ± 16.43	DB07796	–97.71
	7	DB06228	–137.57	DB11913	–153.19	DB15403	–128.13	DB03777	–139.29 ± 8.79	DB11913	–94.60
	8	DB12983	–136.97	DB15011	–152.47	DB11769	–127.96	DB07456	–136.55 ± 9.12	DB15011	–94.16
	9	DB07146	–136.83	DB11829	–152.45	DB14792	–125.86	DB11913	–136.46 ± 18.49	DB11829	–94.16
	10	DB12054	–135.30	DB13648	–152.28	DB15343	–125.02	DB11769	–134.27 ± 6.39	DB13648	–94.05
ZINC15	1	20617839	–170.76	252495534	–156.96	17026643	–140.71	17026643	–141.66 ± 1.24	252495534	–96.93
	2	72273058	–143.96	33937156	–156.19	14246524	–128.13	20617839	–139.77 ± 13.52	33937156	–96.45
	3	17026319	–143.68	15881778	–155.36	69656209	–127.14	40080363	–135.90 ± 14.67	15881778	–95.94
	4	408616086	–141.33	20617839	–153.74	20617839	–126.74	69656209	–132.57 ± 8.59	20617839	–94.94
	5	65099268	–141.14	71747809	–152.73	96099223	–125.72	96099223	–131.61 ± 7.20	71747809	–94.32
	6	11972252	–138.95	949898	–152.62	9217657	–123.82	85746622	–129.94 ± 14.59	949898	–94.26
	7	20490081	–138.93	62166167	–151.31	12419204	–121.40	3632671	–129.30 ± 22.61	62166167	–93.45
	8	17026643	–137.53	21155835	–150.27	65502553	–121.30	575443705	–128.94 ± 13.65	21155835	–92.81
	9	24992790	–137.39	9265260	–150.15	41127019	–121.14	14246524	–128.79 ± 0.72	9265260	–92.74
	10	8705677	–136.96	54318925	–150.09	14991101	–120.56	46915378	–128.21 ± 15.35	252495534	–96.93
SistematX	1	15629	–155.37	14420	–175.94	15553	–150.70	14178	–153.38 ± 11.62	14420	–108.61
	3	21181	–151.88	21418	–166.16	18176	–142.34	14543	–151.29 ± 9.56	21418	–102.59
	4	18925	–142.53	18339	–163.55	14543	–141.04	18176	–150.13 ± 11.46	18339	–100.98
	5	13895	–140.96	20761	–163.50	14094	–135.82	21181	–149.95 ± 19.49	20761	–100.95
	6	17386	–140.61	18176	–163.29	14544	–135.72	14544	–144.33 ± 10.23	18176	–100.83
	7	15117	–137.73	21181	–162.27	14545	–133.78	21047	–143.91 ± 18.07	21181	–100.26
	8	18823	–136.14	13947	–161.60	19827	–131.19	14420	–142.06 ± 29.57	13947	–99.79
	9	14543	–135.97	14083	–161.60	15629	–130.35	14083	–141.16 ± 18.06	14083	–99.78
	10	18338	–135.69	15534	–160.42	18925	–128.50	15629	–140.66 ± 9.60	15534	–99.05

Rigid vs. flexible (MetaDy) virtual screening results for M^CoV2_pro (see section Rescore for more details).

From these results, four metrics were generated:

1. the score according to the LOWEST score among the three MetaDy selected conformations (Flexible-Min.);

2. the score according to the ‘LOWEST HIGH’ score among the conformations (that is, the lowest maximum of energy - Flexible-Max.);

3. the score according to ARITHMETIC AVERAGE of the scores among the three conformations (Flexible-Average);

4. the score according to the BOLTZMANN AVERAGE among the three conformations (Flexible-Boltzmann).

In this selection, if a ligand appears as a hit in the five selection criteria (Rigid, Flexible-Min, Flexible-Max, Flexible-Average, Flexible-Boltzman) it is colored red in the matrix; if it appears in four it is colored in magenta; if it appears in three it is colored in yellow and if it appears in two it is colored in green. If it only appears in the rigid conformation, it is not colored.

Results and discussion

Pharmacophoric variance of crystallographic structures

To examine the differences between M_pro from SARS–CoV and SARS–CoV-2 from a drug discovery perspective, we first analyzed available PDB structures of ligands in complex with these enzymes (see PDB list in Supporting Information). We have also carried out this analysis to define the collective variables (CV) to be used at our MetaDy experiments.

First, we have superposed 48 M^CoV1_pro and 109 M^CoV2_pro structures complexed with ligands and profiled the intermolecular interactions at the subsites S1, S2, S4, S5 and S1’ using nAPOLI (Figure 2(A)) (Fassio et al., 2020). The only amino acid distinction in the binding site region between M^CoV1_pro and M^CoV2_pro is in position 46 at the loop 44–53, where alanine is found in SARS–CoV and serine in SARS–CoV-2. This substitution is relatively far from the binding pocket. The nAPOLI analyses revealed hydrophobic contacts among only four SARS-CoV ligands (PDB codes 3V3M, 4TWW, 4TWY, 5C5O) and Ala46. However, previous studies have suggested a substantial indirect influence of this substitution at the active site entrance dynamics (Brandt et al., 2016; Domingo et al., 2016). Indeed, analysis of the crystallographic B-factors at different M^CoV2_pro complexes points to the loop 44–53 at domain I and its neighbor, the loop 184–193 at domain II, as mobility hotspots at the enzyme active site (Figure 2(B)).

Figure 2.

The setting of the metadynamics collective variables based on intrinsic mobility of the pharmacophore subsites. (A) Superposition of crystallographic poses of different M_Pro ligands (depicted as sticks) at the enzyme active site. The five major M_Pro pharmacophoric subsites are depicted on surface representations and with different colors (blue for the S1 subsite; red for S2; green for S4; yellow for S5 and orange for the S1₁); (B) Superposition of 52 PDB structures of the SARS–CoV-2 (M^CoV2_pro) colored by the B-factors (blue:green:red scale, where red represents higher values) and highlighting the local hot spots at the subsites S2, S4 and S5 (dashed circles). Pharmacophore side chains are depicted in sticks; (C) Structure of the SARS-CoV1 M_pro (PDB: ID 2BX4) colored by the same B-factor scale and highlighting the significant local hot spots at the subsite S1 (dashed square). The Cα atoms of the twelve positions substituted between SARS-CoV1 and 2 are depicted as vdw spheres; (D) Angular collective variables used at our MetaDy experiments: CV_ang1 (cyan) between respective geometric centers at one half of S1 (S1’ in blue), the S1 prime bound β-hairpin (orange) and S2 (blue); CV_ang2 (magenta) between geometric centers at the other half of S1 (S1’ in red), the same β-hairpin (orange) and the S4-5 region (red).

The residues with most varying behavior were Met49, Met165 and Gln189, which impact in the shape of the S2 pocket, and Asn142, located in the S1 pocket. Consequently, high plasticity is observed especially for the S2 pocket. Moreover, such residues are located on a set of four distinct loops, which in turn belongs to different pharmacophoric subsites, delimitate the same S2-1 pocket walls and encompass mobility hot spots according to the experimental B-factors (Figure 2(B,C)).

Hence, our integrated analysis of structure superposition, pharmacophoric and B-factor profiling points to two approximately orthogonal angles, involving these four highly flexible loops as adequate angular collective variables (CV angle’s) for the MetaDy experiments (Figure 2(D)). These two CV angle’s encompass an angle involving loops 44–53 (S2) and 166–173 (S1₁) (i.e. CV_ang1) and other involving loops 140–147 (S12) and 184–193 (S4–5) (i.e. CV_ang2), described at the Methods section. They entail the four loops involved at crucial ligand contacts and whose intrinsic dynamics seem to be more naturally involved in the deformation of the substrate pocket and ligand adequation.

Metadynamics simulations suggest significantly higher intrinsic flexibility for the M_Pro^Cov2 monomer active site compared to M_Pro^Cov1

Using MetaDy protocols with different attenuation degrees (i.e. well-tempered and nontempered MetaDy - wtMetaDy and ntMetaDy) we have been able to detail conformations both immediately close to the crystallographic one as highly distant from this starting configuration for each protease. For all the MetaDy procedures, It was observed the desirable increase in conformational abrangency related to the equilibrium MD with the same simulation interval (Supporting Information Figure S12). However, the two highly attenuated wtMetaDy simulations trend to sample more precisely the local minimum around the crystallographic structure (closer to the original dimmeric and up context), while the two ntMetaDy simulations sample more freely the distant minima Figure 3 and Supporting Information Figure S12), some of them, according to will be discussed below, probably monomeric typical and at the down state. Taking together the resulting free energy landscapes for each procedure, as well as their respective linearization, a more restricted conformational space was observed for M^CoV1_pro compared to M^CoV2_pro (Figures 3 and 4(A)). In fact, M^CoV1_pro is considerably more confined at a deeper minima close to the crystallographic structure. M^CoV2_pro, conversely, can easily access a wide range of minima spreading majorly at the CV_ang1 dimension, although the coverage around the CV_ang2 is also higher than for M^CoV1_pro, mostly at low CV_ang1 values (Figures 3 and 4(A)).

Figure 3.

Free energy landscape and representative cluster centroids from the SARS-CoV and SARS-CoV-2 M^pro MetaDy. Potential of mean force (PMF) maps for the SARS-CoV and SARS-CoV-2 M^pro for nontempered (above) and tempered (below) MetaDy. The zeropoint of the energetic scale is arbitrarily chosen, setting a 0 kJ · mol⁻¹ value at the local minimum closer to the crystallographic structure (structure coordinates depicted as a red star). Capital and lowercase letters are used to point the location of each representative conformers sampled for M^CoV1_pro and M^CoV2_pro, respectively. The energetic window from each macrostate is described on a ‘valueL ≤ window range < valueR’ scheme, being valueL the left term and valueR the right one.

Figure 4.

Accessibility to wider active site configurations and higher conformational promiscuity for the SARS-CoV-2 M^pro compared to the SARS-CoV one. (A) MetaDy projections at the CV_ang1 and CV_ang2 collective variables; (B) Density for the average width of the active site (R_Av) considering the energetic macrostates until 10 kcal · mol⁻¹ for the respective M^CoV1_pro and M^CoV2_pro enzymes. Dashed lines delimit the two major modes observed; (C) Variation of the PMF free energy as a function of the mean of the same R_Av value on a three-by-three scale window plot considering the representative MetaDy conformers clusterized according to Figure 3. Correlation coefficients (R²) are shown. Dashed lines delimit the same regions as in C. (D) Occupied microstates on the respective PMF maps for the M^CoV1_pro and M^CoV2_pro as estimated by the adapted colony method; (E) Conformational entropy variation at the CVang1 vs. CVang2 phase space (ΔSCVsCoV1-2) between the MCoV1pro and MCoV2pro as estimated by the adapted colony method from the plots in (D).

To estimate the consequences for ligand accessibility, we have analyzed the average active site width as the mean distance between the Cα atoms of the active site loops (R_Av) (Figure 4(B,C)). This measure has already proved to be well correlated with ligand accessibility/resistance in previous studies (Konar et al., 2019; Lima et al., 2020). First, we have measured the R_av statistics for conformations with the potential of mean force (PMF) at the 20% more favorable values from our MetaDy analyses (i.e. between −63 and +53 kJ · mol⁻¹). The data show a bimodal distribution for the M^CoV1_pro low energy conformers, one mode centered at a closer and the other one at a partially opened active site (Figure 4(B), left). M^CoV2_pro, conversely, presents just one well-defined mode at the closer conformations, but with additional open conformations, in which all of them share almost the same energetic state (Figure 4(B)). In contrast, opened states seem to be less favorable for M^CoV1_pro (Figure 4(C)), clearly resulting in entropic disadvantages due to conformational inaccessibility according to our adapted colony method estimation (Figure 4(D,E)) (Martínez et al., 2009). Altogether, these results point to the higher flexibility of M^CoV2_pro compared to M^CoV1_pro. A recent classical molecular dynamics study suggests that such increased plasticity can be an important mechanism for stabilization of the enzyme-substrate complex by Mpro (Suárez & Díaz, 2020). This results may also be in line with the structural heterogeneity of the M_pro catalytic sites found by (Behnam, 2021).

The region which comprises the β-sheet and S1, S2, S4 and S5 pockets seems to form a key flexible neighborhood important for the activity of the M^pro. Recent computational studies have shown S2 loop as a regulator to access the active site (Bzówka et al., 2020), which agrees with the high flexibility suggested by our crystals analysis and observed in our MetaDy simulations. Clearly, the exchange of A46S at the S2 loop (from M^CoV1_pro to M^CoV2_pro) can someway decrease entropic disadvantages and allows M^CoV2_pro to access more open conformations, facilitating the access to the binding site. To better explore this, we systematically sampled representative frames of each macrostate for both proteases, trying to encompass the whole range of the M^pro’s conformational multiplicity. These representative conformations were labeled alphabetically (Figure 3) and submitted to deeper flexibility comparison and quantum chemical analyses. We will return to this point later.

Finally, it is worth highlighting that M^Cov2_Pro shows increased movement correlations over the active site opening. To better understand how the internal motions of both M_Pros are affected by the mutations, we analyzed the movement correlations of the loops around the active-site (Supporting Information Tables S1 and S2) and the protein domains. Briefly, numbers between 0 and 0.5 suggest no motion dependency between two regions, whereas numbers bigger than 0.5 up to 1 imply movement correlation. In that way, an increase in the internal movements is observed in M^Cov2_Pro mainly in motions that correlate the domain III and S2 loop with the rest of the enzyme. So, the opening movement in M^Cov2_Pro is more interconnected between the active-site loops and the domain III in which S2 loop can be the regent of this choreography. Recent MD studies suggested that rotation events in domain III caused by S2 loop are important to the active site opening (Suárez & Díaz, 2020). This movement can be associated with the activity of the enzyme in a remarkable way. Experimental and computational studies suggested homodimerization as an important step for the enzymatic activation of M_Pro (Lin et al., 2008; Suárez & Díaz, 2020). In this way, a deeper understanding of the stabilization mechanisms by which M^Cov2_Pro access opened conformations without the substrate binding can be useful for rational drug design.

Literature simulations for the M^Cov2_Pro dimer recover a significant amount of the monomeric active site flexibility for the protomers as sampling increases

We analyze the R_av density distribution of two long multi-microsecond simulations recently published for M^CoV2_pro dimers (Jaffrelot Inizan et al., 2021; Shaw). The R_av metric (average active site width) is a measure of the active site openness, and it was taken for each chain and both together in the dimer (Figure 5). It is noticed that for both simulations the R_av density distributions are asymmetric with positive skewness, indicating a broader flexibility of the active sites. But, the unsupervised adaptive sampling simulation reveals an even more remarkable positive spread of these distributions, which reinforces its greater efficiency for the purpose of conformational explorations.

Figure 5.

Density for the average active site width (R_AV) for the respective protomers A and B (chain A and chain B), as well for both on lower and higher sampling multi-microsecond simulations for the M^Cov2_Pro dimer. (A) Conventional multi-microsecond simulation from the DESRES group (Shaw). (B) Tinker-HP unsupervised adaptive sampling multi-microsecond simulations (Jaffrelot Inizan et al., 2021). The dashed sections in different colors depict different conformational spaces representatively sampled based on the joint analysis from our monomeric MetaDy and the literature dimeric MDs.

The important thing here is the comparison of these M^CoV2_pro dimer simulations (Figure 5) with those of our MetaDy’s monomer (Figure 4(B)). We perceive equivalent profiles in the distributions for both dimers and monomers, indicating that our MetaDy simulations were able to perform similar samples from the conformational space of the active sites. In spite of this general equivalence, it is also quite evident that in the monomer the distribution expands more at high R_av values, indicating active sites even more open than in the dimers.

This is consistent with other results evidencing that, in the dimeric state, the stabilization of the N-finger close to the active site makes it more cohesive and prone to subtract-processing (Behnam, 2021; Cheng et al., 2010; Jaffrelot Inizan et al., 2021). Moreover, we show that even if the active sites of the dimers expand, the proximity of the N-finger to the active site remains almost constant or rather increases, preserving its catalytic capacity (Figure 6). Conversely, at monomeric states, the constrictions by the N-finger from the neighbor protomer (as well the dimeric environment as a whole) is no longer present, allowing wider and more diversal conformations for new virtual screening opportunities.

Figure 6.

Relative density distributions for the interchain distance between the S1 residue from the N-finger of a monomer to the F140|E166 residues at the catalytic pocket from the other at different average active site width values (R_AV) on a multi-microsecond unsupervised adaptive sampling MD for the M^Cov2_Pro dimer. Analysis carried at the freely available data from the Tinker-HP simulation described by Jaffrelot Inizan et al. (2021). From top to the bottom, the different sections with increasing average R_Av values from Figure 5 are depicted.

Open conformers present nonideal catalytic environment in M^CoV2_pro monomer according quantum calculations

The expected reaction mechanism for the catalytic dyad of M^pros is the typical for cysteine proteases: His41 receives a proton from Cys145 which, therefore, can cleave the substrate by nucleophilic attack (Blanchard et al., 2004). Thus, a residual chemical environment favorable for catalysis is expected even before substrate binding (Grillo, Urquiza-Carvalho, Chaves, et al., 2020). Hence, we estimated electronic density descriptors related to local hardness (HA), such as nucleophilic and electrophilic attack susceptibility (respectively, NAS and EAS) for the catalytic site in the representative conformers sampled by MetaDy. These same descriptors have already shown good correlations with the catalytic path in our previous study (Grillo, Urquiza-Carvalho, Bachega, et al., 2020).

The average His41 local hardness for both M^pro presents a peak at intermediary values of R_Av (between 15 and 17 Å), presenting also moderate intensity at closer conformations (being subtly higher for M^CoV1_pro at this last region) (Figure 7(C)). Uniquely for M^CoV2_pro, an increasing behavior at more open conformations (above 19 Å) is still observed for this descriptor. It is not accompanied by an equal increase on Cy145 local hardness or EAS (see below). This suggests that such increasing behavior for this unique His41 descriptor for M^CoV2_pro at broader conformations is not catalytically relevant. Interestingly, we observed a clear difference in the local hardness behavior for Cys145 between the two proteases (Figure 7(A,E)). In that case, the hardness of Cys145 seems to increase as the binding site of M^CoV1_pro opens, forming a peak at approximately the same R_Av values for this same descriptor for His41. The Cys145 local hardness for M^CoV2_pro, conversely, presents a moderate plateau at closer conformations and decreases gradually and progressively according to the increase in R_Av. The Cys145 EAS behavior for both M^pros increases approximately at the same extension according to the R_Av diminishes below 15 Å, but presents a second peak around broader conformations (about 18 Å) just for M^CoV1_pro (Figure 7(E)). This second peak, that is absent in M^CoV2_pro, presents its initial phase in superposition to the respective Cys145 and His41 peaks at M^CoV1_pro.

Figure 7.

Variation of electronic quantum chemical descriptors related to catalysis with the average active site opening for the SARS-CoV and SARS-CoV-2 main proteases. In (A), (C) and (E), are depicted the respective plots for the average values of local hardness of His41 and Cys145, as well the Cys145 electrophilic attack susceptibility (EAS) according to the increase in R_Av. Dashed lines highlight the superior deviation limit for the two ensemble modes depicted in Figure 4(B). In (B) and (D), respective images of the M^CoV1_pro and M^CoV2_pro active sites colored according to atomic local hardness are shown for representative conformers contained on each R_Av ensemble. In (F) and (G), the same is illustrated coloring according to the nucleophilic attack susceptibility (NAS) to EAS behavior. In every case, the residues His41, Cys145 and neighbors are depicted on ball and stick representations.

Taken together, these results suggest that the monomeric M^CoV1_pro can combine with a significant amount the descriptors important for proton exchange and nucleophilic attack (the two initial steps at the catalysis) in both close or mildly open conformations (Figure 7(B,D,F)). In contrast, M^CoV2_pro presents a reasonably superposition in high values both for His41 and Cys145 local hardness, as for Cy145 EAS just in closer active site modes. For this protease, the binding site opening seems to disrupt the optimum catalytic environment. In that sense, although M^CoV2_pro can easier access more opened modes (Figures 3, 4, 5 and 10), these new conformers seem to be catalytically inactive (Figure 7). As already discussed on the last topic, this effect is probably compensated at the dimer by the increase of the proximity to the N-finger from the neighbor protomer according to the active site opening at this environment (Figure 6). We will turn later to the probable importance of the accessibility to these wider conformations (even than less catalytically active at the monomer) for protein processivity, as well their potential for new approaches of drug screening and design for M^Cov2_Pro.

Figure 10.

Conformers used for ligand virtual screening and docking studies. The conformers were taken from the cluster centroids at the range between 0 and 10 kJ/mol in Figure 3. (A) M^CoV1_pro Crystallographic pose (PDB code 2BX4) and conformers C, F and H retrieved from metaDy simulations. (B) M^CoV2_pro crystallographic pose (PDB code 6LU7) and conformers G, H and I retrieved from metaDy simulations. For M^CoV1_pro only the conforme H is from well-tempered MetaDy, while for M^CoV2_pro only the conformer with this same label is from nontempered MetaDy. For all structures: RMSD (in Å) relative to crystal structure; R_av - average active site width (in Å); S1₁, S1₂, S2 and S4 loops are represented by surfaces coloured in blue, green, red and yellow, respectively. Catalytic residues His41 e Cys145 are represented in cyan sticks and residue Ala46 (M^CoV1_pro) or Ser46 (M^CoV2_pro) is represented in magenta surface. Animated gifs of both MetaDy for both M_pro can be seen in Supporting Information.

The impact of the mutations on the stability of open conformations in both M^Pros

We tried to get a glimpse of the impact on the protein structure stabilization at closer Vs wider active site conformations for each one of the 12 aa substitutions between both studied M^Pros. For this, we used bioinformatic tools to predict the protein stability changes for each point mutation that transforms M^Cov1_Pro in M^Cov2_Pro (MCov1 → MCov2) and also the opposite way (MCov2 → MCov1) (Parthiban et al., 2006). We did these calculations for each representative conformer sampled by MetaDy and plotted the results as a function of R_av (Supporting Information Figure S1).

For closer or mildly open active site conformations, it can be noticed equal incidence of stabilization both for MCov1 → MCov2 as for MCov2 → MCov1 substitutions. This is consonant with previous bioinformatics studies using just crystallographic like conformations (Bzówka et al., 2020), that shows the 12 substitutions at the M_Pro from SARS-CoV to SARS-CoV-2 do not stabilize or destabilize the native structure as a whole. According the R_av increases, however, stabilization effects for MCov1 > MCov2 and/or destabilization for MCov2 > MCov1 permutations prevail, especially considering five MCov1 > MCov2 substitutions: T35V, A46S, S65N, H134F and I286L (Supporting Information Figure S1). Between them, only I286L is located at the dimerization domain (III), being the other four located at the catalytic region (domains I and II).

K180N is the only substitution that seems to stabilize wider R_av values with the MCov2 → MCov1 transformation. The position 180 is located at the center of the hinge between the domains I and II. Once our analyses show higher correlations in M^Cov1_Pro than in M^Cov2_Pro between the R_av value and the tumbling of the domain I related to domain II (Supporting Information Tables S1 and S2), the influence of this region at this same tumbling is consonant with the role of this position on such stabilization exclusively for M^Cov1_Pro.

The Role of A46S Mutation on the Attenuation of the Energetic Disadvantages for Binding Site Opening in M^CoV2_pro

Suárez and Díaz already observed greater exposure of the hydrophobic gorge of M^CoV2_pro active site because of the interdomain opening movements (Bzówka et al., 2020). Moreover, the mutation that occurs around the binding site (A46S) is located exactly in the S2 loop that presents increased mobility in M^CoV2_pro. This loop also shows increased movement correlations over the active site opening, suggesting a key role for binding site exposure motions (Supporting Information Tables S1 and S2). Additionally, A46S changing (a hydrophobic to hydrophilic substitution) seems to significantly impact on the open conformations stability in both M^pros (Supporting Information Figure S1(B)).

Thus, hydration can have a key effect to explain the differences in the conformational set of both M^pros. To check this hypothesis, we used semi-empirical quantum chemistry methods based on Neglect of Diatomic Differential Overlap (NDDO) formalism, which are a good option for predicting heat of formation for biological systems (Stewart, 2004). We estimated the variation of hydration enthalpy for each representative conformer sampled by MetaDy simulations (Figure 8) and we observed no enthalpic variation over the binding site opening for M^CoV1_pro, but significant decrease on this same enthalpy with this opening in M^CoV2_pro. The Ala/Ser exchange at S2 loop possibly improves the chances of hydrophilic contacts with the solvent and contributes to decreased hydration penalties. Noncovalent interaction (NCI) calculations that can verify the influence of the different binding site modes on the hydration of the A46S position seem to corroborate this interpretation (Figure 9). Basically, the polar S46 in M^Cov2_Pro allows favorable solvent interactions at the wider conformations, while the hydrophobic A46 in M^Cov1_Pro seems to promote unfavorable clathrates at the same condition (see higher discussions on the SI text) (Homans, 2007; Makhatadze & Privalov, 1996; Maurer & Oostenbrink, 2019).

Figure 9.

Hydration effects on the residue 46 of M^CoV1_pro and M^CoV2_pro. For each plot, on the left, is presented the 3D representation of noncovalent interactions, while on the right is shown the promolecular density. (A) Noncovalent interactions between A46 and water molecules of three conformers (A, J and N) from M^CoV1_pro; (B) Noncovalent interactions between S46 and water molecules of three conformers (B, G and O) from M^CoV2_pro.

Our results agree with recent MD studies which pointed to the importance of Ser46 (Bzówka et al., 2020; Suárez & Díaz, 2020). Therefore, Ser46 seems to work like a weak lock that releases the S2 loop by breaking H–bonds with the enzyme and interacting with environmental water molecules, at the same time that it elegantly decreases the solvation penalties when opening the binding site. Then, we were curious about the druggability and enzymatic activity impact of these opened conformations in both M^pros, but, especially in monomers of M^CoV2_pro (see docking analyses below).

Relative advantages for the M^CoV2_pro active site conformational promiscuity concerning ligand accessibility and adaptability

The evolutionary reasons by which the enzyme monomer can adopt a wide range of stable open conformations for the M^CoV2_pro, compared to the older M^CoV1_pro is not clear yet, although this may have to do with the greater heterogeneity of the M_Pro proteolytic recognition process (Behnam, 2021). In order to better understand how the greater conformational variety of the active site of these proteases can orchestrate their differences on ligand accessibility and/or catalytic activities, we redocked the PF-00835231 ligand into the crystallographic structures of M^CoV1_pro (PDB code 6XHL) and M^CoV2_pro (PDB code 6XHM), as well as into their respective most representative and energetically favorable MetaDyn's conformers (G,H,I) and (C,F,H) (Figures 10 and 11).

Figure 11.

Analysis of the active site accessibility for the covalent inhibitor PF-00835231 on M^Cov2_Pro and M^Cov1_Pro in different conformations by virtual docking procedures. At the top, plots relating redocking score statistics (best three poses) with the active site dimensions are depicted for the ligand PF-00835231 (Boras et al., 2020; Hoffman et al., 2020), considering both crystallographic structures as the MetaDy decoys for M^Cov1_Pro and M^Cov2_Pro. The different score metrics from the GOLD and Molegro virtual docker (MVD) softwares, are normalized for comparison according to the docking methodology. (A) Variation of the docking normalized score according to the increase in the active site R_av value; (B) Variation according to the active site RMSD considering the crystal conformation. The respective values for the crystallographic redocking (Crystal), the less favorable scores (dashed circle) and the scores for wider conformations (solid circle) are highlighted. (C, D), redocking poses for the crystallographic conformations for M^CoV2_pro and M^CoV1_pro, respectively. The atomic colors on the surface highlight the region of Ser46 and Ala46 on M^CoV2_pro and M^CoV1_pro, respectively. At the bottom, 3D plot depicts the redocking of PF-00835231 into superimposed conformer structures of M^CoV2_pro (E) and M^CoV1_pro (F). The black asterisk and the little black arrow indicate a region poorly accessed in M^CoV1_pro compared to M^CoV2_pro and Ala46, respectively. (See Figure 10).

PF-00835231 is a potent ketone-based covalent inhibitor for both M^CoV1_pro and M^CoV2_pro with IC50 about 0.004 and 0.00027 µM, respectively (Boras et al., 2020; Hoffman et al., 2020). It is interesting (at the same time a conundrum) so high variation in affinity (i.e. one magnitude order) of the proteases for this same ligand since the only difference bordering all the active site regions is the A46S substitution at the loop C44-P52, a region that does not contact directly the ligand. In fact, no notorious differential contact with the ligand can be glimpsed at the respective 6XHL and 6XHM PDB structures. In this way, we have inferred that the differences in dynamics and conformational subspaces for the active site of both enzymes (differences for which our analyses point to a great participation of the A46S substitution) can play a pivotal role in this phenomena and carried out the differential docking analysis described below.

At the top of Figure 11, we see plots of a comparative redocking score metric for both M^pros conformers considering two different softwares and algorithms: GOLD and the Molegro Virtual Docker (MVD) (Jones et al., 1997; Thomsen & Christensen, 2006). It can be noticed, for both proteins, than the crystallographic conformation (with active site R_av measure around 14 Å) presents the more favorable scores, as well the lower RMSD considering the ligand heavy atoms at the respective crystal structures (close to 0 Å) for both softwares (Figure 11(A–D) and Supporting Information Figure S13). Although the variant MetaDy major decoys have, all of them, less favorable docking scores related to the crystal structures and considering both programs, it can be noticed two apparent advantageous features for M^CoV2_pro compared to M^CoV1_pro. First the ‘worst of all’ scores are always found between the mildly open or most closed M^CoV1_pro conformers (Figure 11(A,B), dashed circles). Second, it can be noted that with the wider M^CoV2_pro conformations (here represented by the M^CoV2_pro MetaDy decoy H), this protein gain a new option of ligand access with similar affinity than the average for the remained conformations (Figure 11(A,B), solid circles). In other words, the apparent higher conformational entropy gained for the M^CoV2_pro active site compared to M^CoV1_pro and depicted at Figure 4(D,E) seems to be reflected also on a gain in entropy related to paths of ligand access. This can be better glimpsed if we compare the structural visualization of all the more favorable docking poses (here, considering just the GOLD poses to avoid visual pollution) for all the conformations in both proteins (Figure 11(E,F)). It can be noted that the higher diversity of wider poses in M^CoV2_pro allows a higher exploration of the active site in all its extension for the ligand access, while for the closer M^CoV1_pro conformations, some regions are less accessed (asterisk in Figure 11(F)). In fact, a previous QM/MM study has already shown for other similar covalent inhibitor (the aldehyde derivative GC373) than the primary and noncovalent access to the M^CoV2_pro active site occurs at a wider and noncatalytic conformation (Ramos-Guzmán et al., 2021). Following this initial binding, the active site better fit to the ligand and the first catalytic steps happen. In this way, this higher facility to adopt wider and ligand accessible conformations could confer to the M^CoV2_pro enzyme advantages related both to substrate affinity as to catalysis mechanism compared to M^CoV1_pro. If the ligand were a peptide substrate, for instance, we could conceive that it would have easier access by a higher amount of paths to this first noncovalent event in M^CoV2_pro than in M^CoV1_pro, increasing the chances of proteolysis. It is expected that this apparently entropic facilitation of access to the M^CoV2_pro active site could reflect both at a diminishing of the catalytic free energy (and hence on a higher catalytic efficiency and the consequent facilitation to the first Cys145 nucleotide attack that could justify the higher affinity to the PF-00835231 covalent inhibitor), as (and maybe fortunately) on a higher multiplicity of accessible conformational microstates to be explored in drug screening. In this sense, we look for an exploration of such accessibility on a multiconformational ligand screening strategy on the last part of this study.

Exploration of the Significant Active Site Flexibility for M^Cov2_Pro on a MultiConformational Strategy for anti-SARS–CoV-2 Drug Development

Whatever the exact reasons for these M^pro flexibilities we are evidencing here, the realisation of more open conformations sheds light on new possibilities for drug design targeting M^CoV2_pro, especially with monomers. Leastwise two possible ways can be considered for the development of ligands targeting the protease monomers: (i) the planning of allosteric or mixed inhibitors that are able to avoid entropic penalties by favorable binding modes at any open state of the enzyme; (ii) the design of enthalpically directed orthosteric inhibitors that are able to overcome entropic penalties by strong contacts with the crystallographic conformation. Besides that, we investigate a third way for drug development targeting M^CoV2_pro monomers, which regards the overstabilization of its open conformations. The presence of an allosteric or mixed inhibitor able to displace the conformational equilibrium to these open conformations should decrease the proteolytic activity of the enzyme. Beyond the active site, the dimerization pocket seems to be a potential binding spot for this kind of ligand, as already suggested by Suárez and Díaz and by our analyses (Suárez & Díaz, 2020). In our analyses, this can be inferred by the high correlation between the dimerization domain tumbling and the active site loops movements, specially for M^Cov2_Pro (Supporting Information Table S2), as well the influence of substitutions at the position 286 at the domain III on the stabilization of wider conformations on the active site (Supporting Information Figure S1(I)). For the case of a ligand that would prevent the enzyme from adopting the crystallographic mode by binding on the active site, this molecule should have higher affinity for opening states.

In order to test this hypothesis, as well the potentiality of the M^Cov2_Pro higher abrangency on the conformational space for drug discover, we perform a multiconformational target ligand virtual screening and docking with the aim of targeting M^CoV2_pro monomers from three bases: DrugBank, Zinc15 and SistematX (Irwin et al., 2012; Scotti et al., 2018; Wishart et al., 2018). The first two are canonical bases of ligands. The latter is a base of natural products and secondary metabolites hosted at the Federal University of Paraiba, Brazil. We use Molegro Virtual Docker (MVD) for the preparation of complexes, cavity prediction and docking (see Table 1) (Thomsen & Christensen, 2006).

In Figure 12(A–D) we see the docking of ligand x21181 (one of our best consensus scores) into four conformers: MD crystal (i.e. the last snapshot from the equilibrium MD simulation from the crystallographic structure), wtMetaDy G and I and ntMetaDy H (see also Figure 10). The blue and yellow spots on the molecular surface indicate the presence of the catalytic dyad (His41 and Cys145, respectively). These conformers (MD crystal, G, I, H) are in ascending order of RMSD, having the crystallographic structure as a reference. Hence, they represent targets in increasing order of deviation from the protomers of the dimer (the crystal form). It is interesting to see how this ligand was able to assume different poses at all four targets. In the first three, it is positioned more directly on the catalytic site core, as a candidate for competitive inhibitor. But, the last one (H conformer) occupied a newly formed pocket (or subpocket) just above His41 (Figure 12(D)), and now it also becomes a candidate for allosteric inhibitor, given that it may stabilize an inactive state and even interfere with dimerization. In fact, the conformer H differs from all other conformers and crystal structures, with a big conformational change that affects mostly the regions of the S2, S4 and S5 subsites, drastically changing the shape of this region, and possibly allowing larger moieties to occupy these pockets. The docking score value for this ligand in H conformer is equivalent (or rather higher) than in other conformers. It is worth noting that such H conformer becomes bicavitary, with His41 in the middle of the two cavities (Figure 12(E)).

Figure 12.

Multi-conformer flexible docking’s hits explore the M^Cov2_Pro active site space in different ways. For all, the blue and yellow patches on the molecular surface indicate the presence of the catalytic dyad (His41 and Cys145, respectively). In (A), a representative conformer of MD at equilibrium from the crystallographic structure (MD - crystal). In (B, C), the representative conformers G and I from a well-tempered MetaDy, respectively. In (D), the representative conformer H from a nontempered MetaDy. Conformers G, I and H are in increasing order of RMSD in relation to the crystallographic structure (see Figure 10). Also, from (A) to (D) we show the docking of the same ligand x21181 into MD crystal, G, I and H conformes, with respective scores in blue. In (E), the H conformer highlighting its bicavitary site. Note that the dyad residues are separated and this opens a cavity in between. From (F) to (H), superimposed docking of several ligands selected from DrugBank, ZINC15 and Sistemamatx, respectively. In (I), an example of ligand (x18176) that binds both cavities of the H conformer. (See Figure 10).

In this sense, it is significant to pay more attention to the docking into H like conformers. It is possible to verify that the H conformer was the most open of all between the low energy conformations, which implied even in the separation of the two residues that integrate the catalytic dyad (colored spot on surface), making the monomer certainly inactive. But, this separation induces the formation of an unexpected cavity between them. Thus, a ligand that could bind strongly in this state (even enzymatically inactive), could shift the equilibrium from this nonfunctional state, acting as a kind of allosteric inhibitor. Curiously, the overlap of the various ligands with better docking scores from Drugbank, ZINC and SistematX shows that there are preferences in the exploration for one of these two cavities around His45 (being more evident in the results from Drugbank) (Figure 12(F–H)). However, it is possible to find ligands that sought to cover both cavities, like x18176 (Figure 12(I)). Even if this conformation is not relevant for direct contacts with eventual ligands, the easier access to open conformations by M^CoV2_pro should be considered due to entropic penalties associated with stabilization of closed conformations (Figure 4(D,E)).

In this work, we gather evidence that the facility for M^CoV2_pro to access more open conformations, even if not initially catalytic, could confer processivity advantages to this enzyme. In fact, the openness of active-site is well correlated with the S2 loop movements, which is quite distant from the dimerization interface. So, it is reasonable to suggest that this set of open conformations could persist some way, even if residual, after dimerization. If this is true, an energetically facilitated transition between closed and open conformations can particularly favor the cycle involving substrate entrance, adaptation to the catalytic pose, catalysis, product liberation and diffusional searching for a new substrate, as well the inactive monomer/active dimer cycle. Indeed, similar accessibility to the ‘open and close’ movements at the binding site loops has already been shown functionally advantageous in previous studies in our group (Costa et al., 2019; Lima et al., 2020). In particular for viral M^pro, whose facility to bind on different points at the polyprotein sequence in a timely manner is crucial for the viral replication, such enhanced processivity could be determinant. If true, why not explore such ‘breathing’ movements in the search of innovative inhibitors for these viral proteases? Animated gifs illustrating such movements of both MetaDy for both M_pro can be seen in Supporting Information.

It is essential to point out that the docking tests performed here were only a proof of concept in terms of indicating the potentiality of the intrinsic conformational variety of M^CoV2_pro monomers as alternative druggable targets, rather than an in-depth study of drug discovery. Our present work (already quite extensive) was not designed to go deep into such study. But, we intend to do it in a second article (in progress) that will also include in vitro validations.

Conclusion

Through a combination of perturbative molecular dynamics, pharmacophoric analyses and quantum mechanics calculations, we brought light to motion, druggability and activity differences between M^pro monomers from both SARS-CoV and SARS-CoV-2. A new set of highly open conformations emerges from M^CoV2_pro which seems to be facilitated by solvation effects. Also, NCI calculations suggested that the hydrophobic to hydrophilic A46S substitution from M^CoV1_pro to M^CoV2_pro at the more flexible loop S2 participates significantly on such solvation improvement. Comparison with dimer statistics, suggests some extensibility of this same active site flexibility at the dimer. Our quantum calculations revealed, also, a disturbed chemical environment of the active site for highly open states of M^CoV2_pro monomers, indicating loss of proteolytic capacity at this state.

The reasons why these inactive and flexible conformers would be more remarkable in the M^CoV2_pro monomer than M^CoV1_pro are still under debate, and this work is part of it. Whatever, the rising of these open conformations affect directly on the pharmacophoric exposition of the M^CoV2_pro active site, reflecting on possible new drug development strategies. Our preliminary docking studies in different conformers were able to show promising possibilities in the virtual screening of new ligands, whether competitive inhibitors for monomers closer to the up state (near to the protomer in the dimer) or allosteric inhibitors for those closer to the down state, with more open but inactive catalytic sites. Even such nonfunctional monomers should still be considered as drug targets, since a ligand that binds strongly in these down states could stabilize it and even displace the monomer-dimer balance and shut off viral replication.

We believe that our comparative simulations on M^pro monomers of SARS-CoV and SARS-CoV-2 have generated intriguing results and raised hypotheses that should be considered by the scientific community in the search for new antiviral drugs against COVID19.

Disclosure statement

No potential conflict of interest was reported by the authors.

Author’s contributions

L.H.F.L. and G.B.R. designed research; R.E.O.R., E.J.F.C., P.H.C., L.S.C.C., I.B.G., L.E.G.C., F.C.G., C. H.S., M.T.S., A.D.C., K.S.M., A.V.W., R.S.F., G.B.R. and L.H.F.L. performed research; R.E.O.R., E.J.F.C., I.B.G., L.E.G.C., R.S.F., G.B.R., C.H.S. and L.H.F.L. analyzed data; and R.E.O.R., E.J.F.C., G.B.R., C.H.S. and L.H.F.L. wrote the paper.