Unveiling mutation effects on the structural dynamics of the main protease from SARS-CoV-2 with hybrid simulation methods

Abstract

The main protease of SARS-CoV-2 (called M^pro or 3CL^pro) is essential for processing polyproteins encoded by viral RNA. Several M^pro mutations were found in SARS-CoV-2 variants, which are related to higher transmissibility, pathogenicity, and resistance to neutralization antibodies. Macromolecules adopt several favored conformations in solution depending on their structure and shape, determining their dynamics and function. In this study, we used a hybrid simulation method to generate intermediate structures along the six lowest frequency normal modes and sample the conformational space and characterize the structural dynamics and global motions of WT SARS-CoV-2 M^pro and 48 mutations, including mutations found in P.1, B.1.1.7, B.1.351, B.1.525 and B.1.429+B.1.427 variants. We tried to contribute to the elucidation of the effects of mutation in the structural dynamics of SARS-CoV-2 M^pro. A machine learning analysis was performed following the investigation regarding the influence of the K90R, P99L, P108S, and N151D mutations on the dimeric interface assembling of the SARS-CoV-2 M^pro. The parameters allowed the selection of potential structurally stable dimers, which demonstrated that some single surface aa substitutions not located at the dimeric interface (K90R, P99L, P108S, and N151D) are able to induce significant quaternary changes. Furthermore, our results demonstrated, by a Quantum Mechanics method, the influence of SARS-CoV-2 M^pro mutations on the catalytic mechanism, confirming that only one of the chains of the WT and mutant SARS-CoV-2 M^pros are prone to cleave substrates. Finally, it was also possible to identify the aa residue F140 as an important factor related to the increasing enzymatic reactivity of a significant number of SARS-CoV-2 M^pro conformations generated by the normal modes-based simulations.

Graphical abstract

1. Introduction

The 2019 coronavirus disease (COVID-19) pandemic has been the most severe health crisis in the past 100 years. As of July 20, 2022, approximately 6.365.510 cases have been reported worldwide, with more than 6.2 million deaths (WHO Coronavirus (COVID-19) Dashboard). In addition to the zoonotic agents responsible for Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS), the novel coronavirus, SARS-CoV-2, is a spherical-enveloped virus with a positive-sense single-stranded RNA (+ssRNA) which belongs to the broad Coronaviridae family and the Nidovirales order [1,2]. The SARS-CoV-2 genome comprises about 30,000 nucleotides, including open reading frames (ORFs) responsible for encoding structural and non-structural proteins. ORF1a and ORF1b are translated into two overlapping polyproteins, pp1a, and pp1ab, respectively [1]. These polyproteins are processed by the viral proteases M^pro (main protease) and PL^pro (papain-like protease) into 16 non-structural proteins (NSPs). Other ORFs undergo discontinuous transcription, and subgenomic mRNAs are translated into other non-structural, accessory, and structural molecules, such as envelope (E), spike (S), membrane (M), and nucleocapsid (N) proteins [3,4].

SARS-CoV-2 M^pro is a dimer composed of three domains (the catalytic domains I and II and the C-terminal domain III), with the catalytic dyad formed by the residues His41 and Cys145 located in a cleft between domains I (residues 10–99) and II (residues 100–182) [5]. Dimerization is essential for SARS-CoV-2 M^pro enzymatic activity since the NH₂-terminal residues (N-finger) of each monomer interact with the domain II Glu 166 residue of the other monomer, which effectively contributes to the correct arrangement of the S1 pocket of the substrate binding site. Thus, the C- and N-termini of the active dimeric protein are less flexible in comparison to their higher mobility in the free monomers [5]. This general protein architecture is highly conserved in M^pros of diverse coronaviruses [6]; therefore, despite the extensive mutagenesis of these viruses in general, these key proteins are well conserved [7] and also good targets for preventing virus replication and reducing the risk of mutation-mediated drug resistance in new viral strains. Thus, the inhibition of M^pro enzymatic activity could be an interesting strategy for exploring new therapeutic approaches to treat COVID-19. Indeed, non-structural proteins have been reported as potential targets for designing and developing antiviral agents against SARS and MERS [8].

It is well known that since 2019 SARS-CoV-2 has been presenting a continued process of evolution characterized by the accumulation of several mutations related to higher transmissibility, pathogenicity, and resistance to neutralization antibodies induced by infection or vaccination. Several M^pro mutations were found in other SARS-CoV-2 variants, including variants of concern (VOCs), variants of interest (VOIs), and formerly monitored variants, such as Alpha (B.1.1.7) [9] (first detected in the UK), Beta (B.1.351) (first detected in South Africa (Beta) [10,11], Gama (P.1) (first detected in Brazil/Japan) [12], Delta (B.1.617+) (first detected in India) [13], Epsilon (B.1.429+B.1.427) (first detected in USA/California) [14], and Eta (B.1.525) (first detected in UK/Nigeria) [15]. Notably, the genomic analysis of these variants listed in the GISAID database (https://gisaid.org) [16] highlights a set of frequent mutations, such as K90R, P108S, K236R, L220F, and R279C, in the SARS-CoV-2 M^pro. Also, Amamuddy et al. (2020) [17] identified other non-synonymous mutations in all M^pro domains, including substitutions in solvent accessible residues and some in the N-finger region. Moreover, these last authors demonstrated the collective effects of various M^pro mutations in several SARS-CoV-2 isolates using different approaches and techniques, as geometry calculations, cavity compaction analyses, molecular dynamics simulations, anisotropic network model (ANM) calculations, and coarse-grained Monte Carlo simulations, among others. Later, in November 2021, PANGO lineage B.1.1.529, the fifth SARS-CoV-2 variant of concern (VOC), was identified in Botswana and South Africa and later named Omicron by the World Health Organization (WHO). Despite its high transmissibility, even in fully vaccinated individuals and those with a booster dose [18,19], the Omicron variant seems not as pathogenic as other variants. According to Shuai et al. (2022) [20], the Omicron variant is not able to cause a significant multifocal expression of the SARS-CoV-2 nucleocapsid protein in mice as is frequently observed in wild-type (WT) virus and Delta variant infected-mice. In the same study, histological examinations did not reveal important inflammatory infiltrations in the alveoli septa of mice infected with Omicron variant compared to other SARS-CoV-2 types. Also, the authors showed that the gene expression of IP-10 and IFNγ, which are pro-inflammatory cytokines generally overexpressed by SARS-CoV-2 infection, is down-regulated by Omicron in comparison to WT SARS-CoV-2 and Delta variants. All of these results, even when considering the limitations of a mouse model, strongly indicate that the inefficient replication capacity of the Omicron variant is probably related to its attenuated lung pathology and improved mice survival compared to WT SARS-CoV-2 and variants. Otherwise, the rise of specific mutations in some SARS-CoV-2 proteins, as the proteases responsible for viral replication, could change the Omicron variant fitness and alter the infection/pathologic balance of this virus. Interestingly, a search in the literature and databases did not provide relevant information about the presence of mutations in the Omicron variant M^pro; the only sequence alteration reported up to March 2022 in this protease was the mutation P132H [21].

From a protein dynamics perspective, it would be worth investigating the structural effects of these amino acid (aa) substitutions on the stability, activation, and catalytic activity related to SARS-CoV-2 M^pro. Hence, in this study, we analyzed some molecular characteristics (solvent accessible surface areas, conformation of the catalytic dyad, and flexibility) of the regions involved in dimerization and substrate binding in wild-type SARS-CoV-2 M^pro (herein referred as WT SARS-CoV-2 M^pro) and several SARS-CoV-2 M^pro molecules presenting single mutations to study the possible influence of aa acid substitutions on the structure/activity of this viral protease. For this purpose, a set of molecular conformations was generated using a hybrid approach in silico protocol based on all-atom molecular normal mode calculations. The results indicated that specific single mutations in SARS-CoV-2 M^pro could cause important changes in structural and dynamical characteristics regarding collective movements, solvent accessible surface area (SASA) of the dimer interface, energetically accessible conformations, and arrangement of the catalytic dyad. These changes indicated that some of the mutations affect the SARS-CoV-2 M^pro structural dynamics and may potentially modify the functional properties of this protease. This finding calls attention to a possible rise of more harmful viruses in case of emergence and fixation of some of these mutations in the Omicron variant and/or other SARS-CoV-2 strains in the future. Hence, the present work aims to contribute to the identification of potential harmful mutations of the SARS-CoV-2 M^pro. Thereon, several unique surface amino acid substitutions identified in SARS-CoV-2 M^pro were analyzed in order to verify their influence on molecular stability, conformation, and reactivity, combining normal modes-based simulations, molecular dynamics (MD) simulations and machine learning techniques.

2. Material and methods

A hybrid method (VMOD, a module of the CHARMM software) that combines Normal Mode Analysis (NMA) and low-temperature Molecular Dynamics (MD) simulations (described in the next section) was employed to generate energetically accessible displaced structures along collective normal modes, using harmonic constraints applied to Cα atoms and sample the conformational space of the SARS-CoV-2 M^pro (wild-type and mutants). The displacements along the modes towards a target were achieved using a series of low-temperature MD simulations followed by energy minimizations with successively increasing restraining force constants. Moreover, NMMD the simulations were used to analyze the global motions and conformational changes for both wild-type (WT) and SARS-CoV-2 M^pro mutants [22]. The characterization of the protein structural dynamics was performed with the following in silico methods and analyses: a) flexibility assessed by root mean square fluctuation (RMSF); b) analysis of collective modes of wild-type and mutants (correspondence/correlation calculation); c) conformational sampling; d) inference of models for identification of characteristics related to the M^pro reactivity using a machine learning approach; e) catalytic dyad distance and reactivity; f) catalytic site structural analyses; and g) solvent-accessible surface area (SASA) analysis.

2.1. SARS-CoV-2 M^pro modeling and protonation

A wild-type (WT) and apo crystal SARS-CoV-2 M^pro structure with 1.91 Å resolution (PDB ID 7C2Y) was used to generate 48 mutant M^pros with one single point mutation each [17] using PyMOL software (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC). The most favorable rotamers from each mutated aa residue were selected using the PyMOL Mutagenesis tool. The protonation states of WT and mutant SARS-CoV-2 M^pros at physiological pH (pH = 7.0) were calculated in the PDB2PQR server [23]. The protonated structures were submitted to the Solution Builder module of the CHARMM-GUI web server [[24], [25], [26]] to generate the input files for the normal modes analysis [[27], [28], [29]]. Then, the protocol for these procedures was elaborated according to the following points: (1) selection of a dimeric and apo SARS-CoV-2 M^pro structure from PDB; (2) selection of SARS-CoV-2 M^pro single mutations described by Amamuddy et al. (2020) [17]; (3) modeling of 48 mutant SARS-CoV-2 M^pros with a single point mutation (Fig. 1 ); (4) energy minimization of the WT and mutant SARS-CoV-2 M^pros; (5) calculation of the first 18 lowest frequency modes with an all-atom approach (the lowest frequency normal modes represent the collective motions of the protein); (6) analysis and selection of the mutations with a significant RMSF difference compared to the WT SARS-CoV-2 M^pro (considering four important protease regions [5,17] presented in Table S1); and (7) Generation of energetically relaxed conformations along the lowest frequency normal modes by carrying out a low temperature molecular dynamics (MD) simulations and energy minimizations (first six normal modes).

Fig. 1.

SARS-CoV-2 M^pro cartoon representation with the mutated amino acid residues along the molecular tertiary structure. (A) The backbone color corresponds to the RMSFs values for the WT where blue and red represent more rigid and flexible, respectively. The positions of mutation for the mutations selected by RMSF (described in Table 1) are highlighted by spheres colored according to the characteristic of mutation: (B) conservative (no charge change) in brown (A116V, A129V, A173V, A193V, A234V, A255V, A266V, V261A, I136V, L220F, L232F, V20L, K236R, K61R, K90R, and D48E), (C) charge inversion in cyan (N151D), (D) polar to apolar without charge in red (T135I, T190I, and T45I), (E) special residue in magenta (P108S, P132L, P99L, R279C, R60C, G15D, and R105H). Figure generated using PyMOL software. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

2.2. Normal modes analysis (NMA) and conformational selection of the mutations

Initially, the protein structure was minimized in two stages. In the first stage, harmonic constraints were applied and progressively decreased (250–0 kcal mol⁻¹Å⁻²) and, for each constraint value, 500 steps of gradient conjugate (CG) were applied. After the CG minimization, the constraints were removed and 2 × 10⁵ steps of adopted basis Newton Raphson (ABNR) were used with the convergence criterion of 10⁻⁶ kcal mol⁻¹Å⁻¹ RMS to energy gradient. The normal modes were calculated using the DIMB (Iterative Mixed-Basis Diagonalization) module [29] implemented in CHARMM software [30,31] considering all the atoms of the protein and a force field corresponding to the CHARMM36 m [31].

A set of the first six normal modes describing the internal movements were analyzed to ensure that all relevant large amplitude motions were included in the analysis. The modes corresponding to global rotation and translation were discarded. Van der Waals (VDW) interactions were computed considering a distance of 5 Å, and a switching function was employed to approximate these interactions up to 9 Å. A distance-dependent dielectric constant (e = 2r _i,j) was used to treat the shielding of electrostatic interactions by the solvent. The described parameters allowed a good prediction of the intrinsic collective movements for all of the calculated structures (WT and mutant SARS-CoV-2 M^pro).

2.3. Conformational space sampling along the collective normal modes

To obtain energetically allowed displaced structures along selected normal modes, we used the VMOD module from CHARMM software [29]. The displacements along the modes were achieved using a series of low temperature MD simulations followed by energy minimizations which allowed a complete free movement of the protein, including the side chains. The structures were displaced from −1.0 to +1.0 Å (Mass Weighted Root Mean Square - MRMS values) along the selected normal modes with steps of 0.2 Å, resulting in 11 intermediate structures for each mode. The procedure was similar to that used by Batista et al. (2011) [32]. At each stage, the force constant of the harmonic restraining potential (K_d) was increased to ensure a slowly convergence towards the normal mode-based displacement. The restraining force K_d value was increased from 1,000 to 10,000 kcal mol⁻¹ Å⁻² during successive 10 ps-MD simulations. Velocities were assigned at random corresponding to a temperature of 30 K in the MD simulations. The Berendsen thermostat with a coupling constant of 0.1 ps was used in all low-temperature MD simulations, which were performed to allow a better conformational search. A final MD simulation was achieved with a K_d value of 20,000 kcal mol⁻¹ Å⁻² followed by additional 1,000 steepest descent and 1000 conjugate gradient minimization steps to reach the final displacement target along the modes. With this procedure, 66 structures were generated (11 structures per normal mode x 6 normal modes) for each molecular system considered.

2.4. Inter-residues distances and reactivity analysis

According to Ramos-Guzmán et al. (2020) [33], the reaction mechanism proposed for proteolysis catalyzed by SARS-CoV-2 M^pro involves acylation and deacylation steps. In the first one, a proton transfer from C145 Sγ atom to the H41 Nε atom is followed by a nucleophilic attack of the C145 Sγ atom to a specific substrate peptide bond. Lastly, hydrolysis breaks the bond between the M^pro and the substrate in the deacylation process, enabling the enzyme to participate in a new catalytic cycle [33]. Regarding the first step of the reaction, the distance between the histidine nitrogen atom (Nε) and the cysteine sulfur (Sγ) was calculated to check its variability in the set of structures generated along the chosen six normal modes.

Furthermore, a set of measures related to the distance between important aa residues was accomplished. As summarized by Inizan e coworkers [34], some distances can be used as structural activation markers related to the oxyanion hole structural organization and are important factors for reactivity; these distances involve the residue pairs: H41/C145, H163/F140, H163/E166, and E166/H172. Some additional distances involving residues of the catalytic dyad, oxyanion hole, and substrate binding site were also computed: F140/H41, F140/C145, H163/S144, and H163/H164. All the mentioned distances are described in Table S2.

Additionally, the reactivity indices of Nε and Sγ atoms were calculated using a Quantum Mechanics (QM) approach. Reactivity indices have been successfully used to determine susceptibility to protonation and/or deprotonation processes in protein molecules and predict chemical reactions involving charge sharing or exchanges, such as covalent or ionic reactions [[35], [36], [37], [38]]. For this purpose, Mulliken electronic populations (MP) on the atoms mentioned above were calculated using the computational package Gaussian 09 via Density Functional Theory (DFT). As the protein molecule under study is too large for DFT calculations, aa residues with a maximum distance of 3 Å from any atom from catalytic residues were selected for electronic structure calculations. The obtained substructures were divided into two layers by applying the ONIOM technique using the package MolUp [39]. All atoms from catalytic residues (His41 and Cys145) were set as high layer (HL) and the atoms of the other residues were set as low layer (LL). Becke's LYP (B3LYP) exchange-correlation functional and 6-31G(p,d) basis set were employed for electronic calculation on HL, and the semi-empirical Parametric Method 6 (PM6) was employed for the LL. The Polarizable Continuum Model (PCM) was used for simulating the presence of the solvent. The combined susceptibility to Sγ deprotonation and Nε protonation (here named as CATALYSIS _SUSCEP) was determined as: CATALYSIS _SUSCEP = MP _Sγ - MP _Nε, where MP _Sγ and MP _Nε are the MP on the atoms Sγ and Nε, respectively.

2.5. Solvent-accessible surface area (SASA)

The CHARMM program was used to calculate the Solvent Accessible Surface Area (SASA) for the set of structures generated along the normal modes. We have reported the SASA values to the entire protein (SASAt), hydrophobic residues (SASAhp), dimeric interface residues (SASAdim), substrate binding site residues (SASAsb), and His41 and Cys145 residues.

2.6. Mantel test

We apply the Mantel test to check similarity (correlation) among the modes of WT with mutants of SARS-CoV-2 M^pros. Several works in the literature report that a single mutation can alter the motions present in the mutants, causing the emergence of new molecular normal modes or the loss of some motions present in the WT. This same phenomenon can be noticed in the case of the SARS-CoV-2 M^pro.

Comparisons between the movements of the WT and mutant SARS-CoV-2 M^pros were performed using the protocol described in the Mantel test. In this protocol described by Louet et al. [40], motions in 3D space were analyzed by calculating 2D matrices, exchanging the relative displacements of all pairs of Cα atoms between two structures [22,40]. Briefly, the method calculated 2D matrices reflecting the relative displacements of all pairs of Cα atoms between two structures. For each mode, these matrices were calculated from the two structures obtained at a displacement amplitude of 1.0 Å in each direction. The correlation coefficients between the two matrices were then calculated with Mantel's test [40]. With this method, it was assumed that two 2D maps sharing a correlation coefficient greater than 0.6 described highly correlated motions in cartesian space.

2.7. Data mining

Supervised machine learning (ML) and Principal Component Analysis (PCA) were performed to identify molecular features potentially associated with susceptibility to protein catalysis. For this purpose, a structured dataset was built, which was composed of a series of attributes evaluated in this study, namely: normal mode, displacement along the normal modes, the absolute value of the displacement along the normal modes, the aa sequence from the protein substructures composed only of residues at a maximum distance of 3 Å from the catalytic dyad and, finally, a series of distances between residues and atoms (Table S2 - highlighted in green). All substructures obtained by displacement of SARS-CoV-2 M^pros along the first six normal modes (in displacements ranging from −1.0 Å to +1.0 Å, in steps of 0.2 Å) were considered as instances. Finally, the values of CATALYSIS _SUSCEP were used as class. These values were categorized into quartiles (Q1, Q2, Q3, or Q4, according to an analysis of the CATALYSIS _SUSCEP distribution), where Q1 represents the lowest and Q4 the higher values. The J48 algorithm was applied to identify a set of rules that could group the instances in the above mentioned quartiles. Additionally, the InfoGainAttributeEval algorithm was used to determine the most relevant attributes for discriminating dataset classes. ML analyzes were performed using WEKA software and the PCA was performed using the package FactoMineR for the R software [41,42].

2.8. Statistical analysis

The RMSF arithmetic average was calculated for each analyzed region of the WT and mutant SARS-CoV-2 M^pros (N-finger, catalytic dyad, substrate-binding site, and dimerization site). The mutants which presented an average RMSF above the threshold of 10% for one of the chains (A or B) or both compared to the RMSF values of the control group (WT) were selected for posterior analysis.

The data obtained related to the potential energy, distance between the catalytic dyad atoms (Cys145, His41), and SASA for (i) binding site, (ii) hydriohobic residues, (iii) dimer interface residues, and (iv) total molecular surface were compared between the 48 mutants and WT of SARS-CoV-2 M^pro using ANOVA (Analysis of Variance) followed by Dunnett's test (with significance at p ≤ 0.05). All statistical analyses were performed using the ANOVA multcomp package [43] implemented in a R-Studio environment [44]. The heatmap, violing and boxplot representations were generated in Python using Pandas [45,46], NumPy [47], Matplotlib [48] and Seaborn [49] libraries. The simple correlations coefficients (Pearson and Spearman) between the numerical variables of the dataset described in section 2.7 (machine learning dataset) were also determined. Correlations with p ≤ 0.05 were considered statistically significant.

2.9. Analysis of dimer interface

The PDBe PISA (Proteins, Interfaces, Structures, and Assemblies) was used to investigate the interactions at the dimeric interfaces of the WT and mutant SARS-CoV-2 M^pros and calculate the solvation-free energy gain upon formation of the interfaces, interface areas and the number of potential hydrogen bonds and salt bridges [50]. Complementary to the PISA results, we applied the TKSA-MC [51] to estimate the contribution of the electrostatic interactions to the total free energy, especially considering the polar aa residues localized in the interface of the dimers (Arg 4, Glu 14, Glu 240 and Arg 298). The TKSA-MC calculates protein charge-charge interactions via the Tanford–Kirkwood Surface Accessibility model with the Monte Carlo method for sampling different protein protonation states. These two softwares were used to analyze the dimeric stability of four mutants which showed significant SASA and potential energy reduction (see Table 1 ).

Table 1.

The table lists the SARS-CoV-2 M^promutations presenting at least three structural characteristics with significant variations compared with WT where the data are grouped by: 1) no charge change: apolar to apolar (AP - > AP: A- > V, V- > A, I- > V, L- > F, V- > L), positive polar to positive polar (P(+) - > P(+): K- > R, R- > H), negative apolar to negative apolar (P(−) - > P(−): D- > E); 2) charge inversion: neutral to negative polar (NT - > P(−): N- > D); 3) polar to apolar without charge: polar to apolar (P - > AP: T- > I); 4) special residue: special to neutral (spe - > NT: P- > S), special to apolar (spe - > AP: P- > L), negative polar to special (P(−) - > spe: R- > C), special to negative polar (spe - > P(−): G- > D). Columns mean: Mutation: mutation and residue; Res.: type of mutation (residues); RMSF: Root mean square fluctuations of chain A and B; Energy: energy calculated of each mutation; Distance Nε: distance between Sγ atom of Cys145 and Nε atom of His41; Cys145 and His 45: catalytic dyad; Bind. site: SASA substrate binding site leads; Hidrop. SASA: SASA of hydrophobic residues; Total SASA: accessible total area; Freq.: frequency of each mutation on variants (%), by sequences deposited on GISAID (last access: 31/05/2021). Variant: variants in which the mutation is present in GISAID. The used symbols to denote significant statistics of a parameter versus WT: "*" (presence of significant statistics); "-" (decreased parameter), "+" (increased parameter), “A-" (decreased parameter on chain A), “B-" (decreased parameter on chain B), “B+" (increased parameter on chain B). In yellow are highlighted the probable stable mutation dimers.

Mutation	RMSF	Energy	Distance Nε2	CYS 145	HIS 41	Bind. site	SAS Hydrop.	Dimer interface	Total SAS	Freq.	Variants
Conservative substitutions
A116V	*	* -	* A-	* B-		* A-	* -			0.67%	B117; P1; B1429+B1427; B1525
A129V	*		* A+	* B-	* A + B-		* -	* +	* -	0.62%	B117; B1351; P1; B1617; B1429+B1427; B1525
A173V	*	* -	* A-	* A+	* A-	* A-	* +		* -	0.04%	B117
A193V	*			* A-					* -	0.78%	B117; B1351; P1; B1617; B1429+B1427
A234V	*		* A-				* +			0.29%	B117; P1; B1617; B1429+B1427
A255V	*		* A-				* +			0.15%	B117; B1351; P1; B1617; B1429+B1427
A266V		* +								0.24%	B117; B1351; P1; B1617; B1429+B1427
V261A	*	* -		* B-					* -	0.02%	B117; B1429+B1427
I136V	*	* -	* A- B+	* A+	* A-	* A-	* -	* -	* -	0%	None
L220F	*	* +	* A+	* A + B-	* A+		* -	* -	* -	0.71%	B117; B1351; P1; B1617; B1429+B1427; B1525
L232F	*		* B-		* B-			* -	* -	0.32%	B117; B1351; P1; B1617; B1429+B1427
V20L	*		* B-	* A+	* B-		* +	* +		0.04%	B117
K236R	*	* -	* A-				* -		* -	1.08%	B117; B1351; P1; B1617; B1429+B1427
K61R	*	* -				* A-	* -			0.01%	B117
K90R	*	* -	* A-					* -	* -	70.54%	B117; B1351; P1; B1617; B1429+B1427; B1525
D48E	*	* -			*B+	*B+				0.71%	B117; B1617
Electric charge inversion substitutions
N151D	*	* -	* A-		* A-		* -	* -	* -	0.01%	B117
Polarity change substitutions
T135I	*		*B+					* -		0.03%	B117; P1; B1617
T190I	*	* -			*B+	* A- B+	* +	* +	* -	0.32%	B117; B1351; P1; B1429+B1427
T45I	*	* +	* A- B+		*B+	* A- B+	* +			0.10%	B117; B1351; B1617; B1429+B1427
Special substitutions
P108S	*	* -					* -	* -	* -	5.71%	B117; B1351; P1; B1617; B1429+B1427; B1525
P132L	*	* -				* A- B+			* -	0.79%	B117; B1351; P1; B1429+B1427; B1525
P99L	*	* -	* B-	* B-	* B-			* -	* -	0.10%	B117; B1429+B1427
R279C	*	* +	* A+	* A + B-	* A+		* -	* -	* -	0.93%	B117; P1; B1617; B1429+B1427; B1525
R60C	*	* +		* B-	* B-	* A- B-	* -	* +	* -	0.27%	B117; B1351; B1429+B1427
G15D	*	* -		* A-		*B+		* -		0.01%	B117; B1429+B1427
R105H	*	* +	* B-	* B-			* -		* -	0.04%	B117; B1617; B1429+B1427

3. Results and discussion

3.1. General characteristics of the SARS-CoV-2 M^pro mutants identified in GISAID database

Several works have reported variants of SARS-CoV-2 with mutations in different regions of M^pro and their effect on its functionality [5,17]. Thereby, in this study, the collective motions of the wild-type (WT) SARS-CoV-2 M^pro and different variants with single mutations were evaluated to shed some light on the dynamic characteristics of this essential viral protease. For this purpose, it were initially calculated using an all-atom approach the collective normal modes of the WT SARS-CoV-2 M^pro and 48 mutants (grouped into set 1). Following, the structural dynamics of these molecules was analyzed based on the Cα flexibility of key protein regions, such as the N-finger, dimeric interface, substrate-binding site and H41–C145 catalytic dyad. As shown in Fig. 2 , both chains A and B from these 48 mutants present different fluctuations in comparison to the wild-type molecule considering the average Cα RMSF (root mean square deviation) of the mentioned key regions. Additionally, even when each molecule (WT and mutants) was analyzed separately, there was an evident variation of the molecular flexibility between the chains A and B, with a particular higher flexibility of the substrate-binding region and the catalytic dyad of the chains B from mutants in comparison to the WT SARS-CoV-2 M^pro correspondent chains (Figures S1, S2 and S3). This asymmetric/distinct behavior between the chains in homodimeric enzymes was also revealed by MD simulations involving the SARS-CoV-2 M^pro [17] and SARS-CoV M^pro [52] and also pointed by NMA calculations involving the HIV-1 protease [53] and DPP-IV, a related diabetes protein [54]. Thus, the analysis of set 1 indicated that some aa substitutions might potentially influence the structural/functional characteristics of SARS-CoV-2 M^pro, as also displayed previously by Amamuddy et al. (2020) [17].

Fig. 2.

RMSF Boxplot of important regions of the SARS-CoV-2 M^pro reported in Table S1. (A) Chain A and (B) chain B. N-finger in yellow, dimeric interface residues in red, catalytic dyad in green and substrate-binding site in purple. Black circles represent the mean of mutant SARS-CoV-2 M^pros, and the blue circles represent the same measure for WT SARS-CoV-2 M^pro. The residues R4 and C145 are not in ascending order in the N-finger and catalytic dyad regions, respectively, to clarify that they are present in more than one region. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

These initial results allowed the selection of 35 SARS-CoV-2 M^pro mutants with a high degree of flexibility (grouped into set 2). Although the molecular flexibility is not per se an indication of function diversity, some additional criteria ⎼ potential energy (Fig. S4), total solvent-accessible surface area (SASAt) (Fig. S5), solvent-accessible hydrophobic surface area (SASAhp) (Fig. S6A), solvent-accessible surface area of the dimeric interface (SASAdim) (Fig. S6B), SASA of His41 and Cys 145 (Figs. S7A and S7B), solvent-accessible surface area of the substrate-binding site (SASAsb) (Fig. S8), and catalytic dyad distance (Fig. S9) ⎼ were evaluated in this second set of structures to identify aa substitutions more prone to induce structural/functional modifications. The analysis of the calculated structures from set 2 underscored 26 mutants with a significant statistical variation in at least three of the mentioned criteria in comparison to WT SARS-CoV-2 M^pro. Then, the structures with these 26 surface single mutations were grouped in a third set (set 3) and sorted according to the following types of aa residue substitution: i) conservative substitution (when a given aa residue is replaced by a different one with similar biochemical properties); ii) electric charge or polarity modification; and iii) special cases, which are related to the inclusion/exclusion of aa residues with unique properties (cysteine, glycine, histidine, or proline) (Table 1).

Most of the mutations from set 3 (present in 16 molecules) involved conservative replacement; in this group, besides significant Cα RMSF alterations, ten SARS-CoV-2 mutants M^pros also showed relevant modifications of potential energy. More specifically, eight of these 16 mutants (D48E, K61R, K90R, I136V, A173V, L220F, K236R, and V261A) presented a negative variation of the potential energy and only two (L220F and A266V) showed a positive variation of the potential energy concerning the WT SARS-CoV-2 M^pro. The aa residue substitution D48E was the only one that did not present a significant negative variation in SASAdim, SASAsb, SASAhp, and SASAt. Taking into account the total potential energy and SASAdim variation and at least one of the parameters related to total surface exposition (SASAhp and SASAt) is possible to point out five SARS-CoV-2 mutants M^pros with statistically significant differences in comparison to the WT SARS-CoV-2 M^pro (K90R, I136V, A173V, L220F, and L232F). Particularly, the mutation K90R is present in the B.1.1.7 (first detected in the UK), B.1.351 (first detected in South Africa), P.1 (first detected in Brazil/Japan), B1.617 (first detected in India), B.1.429+B.1.427 (first detected in USA/California), and B.1.525 (first detected in UK/Nigeria) variants and also markedly prevalent in the SARS-CoV-2 M^pro sequences deposited in the GISAID database (around 70% of frequency). Another relatively prevalent mutation identified in the GISAID database was the K236R, which presents 1.08% of frequency and could be found in the B.1.1.7, B.1.351, P.1, B1.617, and B.1.429+B.1.427 variants. In this case, there is no significant difference in the SASAdim in comparison to the WT molecule; however, it was verified SASAt and SASAhp alterations. Curiously, the mutation I136V is the only one that presents a significant variation in the SASAdim, SASAt and SASAhp parameters, but it was not found in the variants from the GISAID database.

The other SARS-CoV-2 M^pro aa residue substitutions from set 3 (11) were grouped according to the electric charge/polarity and inclusion/exclusion of cysteine, glycine, histidine, or proline. Only one substitution lead to the introduction of a negative-charged aa residue (N151D); this mutation was only found in the variant B.1.1.7, being associated with a significant variation of the potential energy, SASAdim, SASAt and SASAhp in comparison to the WT SARS-CoV-2 M^pro. Another three SARS-CoV-2 M^pro mutations presented polar to nonpolar aa residue substitutions (T45I, T135I, and T190I). In this case, the mutation T190I was found in four variants (B.1.1.7, B.1.351, P.1, and B.1429+B.1427). Remarkably, these mutations introduced important modifications related to the potential energy, SASAdim, SASAsb, SASAhp, and SASAt in relation to WT SARS-CoV-2 M^pro. Finally, it was identified in the GISAID database seven aa residue substitutions related to the inclusion/exclusion of cysteine, glycine, histidine, or proline residues (G15D, R60C, P99L, R105H, P108S, P132L, and R279C). Here, the mutation P108S was identified in six variants (B.1.1.7, B.1.351, P.1, B1.617, B.1.429+B.1.427, and B.1525) and it is also the most prevalent among the seven mutations of this subgroup, with 5.71% of frequency in the GISAID database. In addition, the mutation P108S promotes significant alterations of the SASAdim, SASAsb, and SASAt parameters, as well as the mutations R60C and R279C. However, the mutations R60C and R279C are associated with a potential energy increase compared to the WT SARS-CoV-2 M^pro, whereas P108S decreases the potential energy of the calculated structures.

3.2. Modifications of the SARS-CoV-2 M^pro dimeric interface induced by mutations at protein surface

The M^pro protomers from different coronaviruses consist of three well-characterized domains: two with a chymotrypsin fold and a third extra helix domain, which is essential for the dimerization of this viral protease [[55], [56], [57], [58], [59]]. Also, since the mid 2000s, a series of scientific works confirmed that the dimerization process is fundamental to the full catalytic activity of the M^pros from coronavirus [5]. Up to now, the solved crystallographic structures of M^pro molecules also indicate they are structurally very similar to each other. Hence, it is reasonable to suppose that the combination of low levels of potential energy, lesser SASA values and larger dimeric interfaces could favor the protein stabilization and affect the activity of SARS-CoV-2 M^pro. About this issue, Hu et al. (2009) [60] showed that two specific and neighboring mutations at the SARS-CoV-1 M^pro dimeric interface lead this molecule to assume distinct oligomeric conformations. In fact, this last work concluded that certain key aa residues control the SARS-CoV-1 M^pro dimerization, and it also suggested the dimeric stability of this protease is heavily dependent on the extent and integrity of the contacts between the monomers. In general, the calculated mutant structures from set 3 showed an expected tendency of SASAt to decrease associated with a negative variation of the potential energy. Although the SARS-CoV-2 M^pro mutants do not present aa substitutions at their dimeric interfaces, some of the calculated structures showed wider contact areas in relation to the calculated wild-type molecules. Specifically, four surface aa residue substitutions from set 3 (K90R, P99L, P108S, and N151D) (Table 1) allowed reach a combination of lower energy, smaller SASAt, and larger dimeric interfaces) in comparison to other structures. Thus, these SARS-CoV-2 M^pro mutations were carefully examined to assess their structural differences to the WT SARS-CoV-2 M^pro, particularly in relation to the dimeric interface and the influence of the K90R, P99L, P108S, and N151D mutations on the conformational space accessed by the calculated structures.

Moreover, PDBePISA web server [61,62] indicated that some K90R and N151D mutant structures from in silico simulations present a higher dimeric interface contribution to the total solvation free energy (ΔⁱG) (not including the effect of satisfied hydrogen bonds and salt bridges across the interface) in relation to the WT molecule and the other mutants (Fig. 3 A). This fact indicates that the K90R and N151D SARS-CoV-2 M^pro mutants seem able to assume oligomeric conformations kept by a dimeric interface more hydrophobic than the WT and mutant structures. On the other hand, as shown in Fig. 3A, there is a noticeable concentration of WT structures that reach ΔⁱG values around −11.0 kcal/mol, whereas the most mutant structures showed higher ΔⁱG values. Hence, the K90R, P99L, P108S, and N151D mutations can also promote the formation of dimeric interfaces with lower hydrophobic contribution to the ΔiG compared to the WT SARS-CoV-2 M^pro. Indeed, Fig. 3C indicates the K90R, P99L, P108S, and N151D mutations can be potentially linked to the organization of a higher number of dimeric interface hydrogen bonds, contributing thus to the formation of more specific and stable contacts between the SARS-CoV-2 M^pro protomers. An example regarding this difference between the WT and the calculated mutant structures is the influence of the mutation P99L. In this case, it was possible to obtain a P99L dimer presenting approximately ten additional hydrogen bonds at the dimeric interface compared to the calculated WT structure with the highest number of hydrogen bonds at this same region.

Fig. 3.

Dimeric interface analysis of WT and mutant SARS-CoV-2 M^pros using PISA web server. (A) ΔⁱG indicates the solvation free energy gain upon the formation of the interface (kcal/mol). The mean and standard deviation values of solvation free energy (in kcal/mol) are the following: WT (−10.3 ± 1.1); K90R (−9.7 ± 1.4); P99L (−7.2 ± 1.2); P108S (−8.8 ± 1.3); and N151D (−10.3 ± 1.3). (B) Interface area (Ų). (C) Number of potential hydrogen bonds. (D) Number of potential salt bridges.

As well, according to the TKSA-MC server [51], the electrostatic contribution to the total free energy (ΔGqq) of four essential interface aa residues (R4, E14, E290, and R298) was clearly distinct for WT and mutant SARS-CoV-2 M^pros. These aa residues are essential for homodimerization due to the following contributions: R4 (chain A) and E290 (chain B) form an inter-protomer salt bridge; E14 (chains A and B) contributes to maintaining the H-bond net between the N-finger helices from both protomers, and R298 makes an intra-protomer interaction with M6 in chain B which helps to properly insert the chain B N-finger into the chain A and also interacts with a serine establishing an important interface H-bond (R298 (chain B)/S123 (chain A)) [60]. As shown in Fig. S10 (Supplementary Material), the ΔGqq values related to the mutant chains A are lower in comparison to their counterparts chains of the calculated WT structures. Otherwise, this difference is not observed for the chains B from the calculated WT and the K90R, P99L, P108S, and N151D mutants. This finding corroborates the previous results (calculated by the PDBePISA web server), which also indicated the prevalence of polar dimeric interface interactions in several mutant structures. Indeed, Ding et al. (2005) [63] also highlighted that hydrophobic contacts and electrostatic interactions play major roles for SARS-CoV M^pro dimerization based on affinity capillary electrophoresis experiments. Then, based on the preceding findings, it is feasible to suggest that the surface aa residue substitutions K90R, P99L, P108S, and N151D were able to induce a higher density of structures with different and larger dimeric interfaces (or distinct oligomers), showing a higher number of polar contacts. Then, these results imply that the K90R, P99L, P108S, and N151D mutations may lead to more stable SARS-CoV-2 M^pro conformational structures.

Curiously, most of the data literature point that oligomeric alterations of the SARS-CoV-1 M^pro are caused by substitutions of aa residues close to the dimeric interface and/or those directly involved in the formation of the dimeric contacts. However, the results shown here indicated that the quaternary conformations reached by the calculated WT SARS-CoV-2 M^pro and the K90R, P99L, P108S, and N151D mutants are probably due to specific surface aa residues since there was no alteration in the dimeric interface composition of these molecules. This is another interesting fact, mainly considering the notable sequence and structural similarity between the SARS-CoV-1 and SARS-CoV-2 M^pro molecules. Indeed, it is possible to suppose that the surface substitutions K90R, P99L, P108S, and N151D may are able to alter the dimeric interface and also modulate the enzymatic reactivity and activity of the SARS-CoV-2 M^pro.

3.3. Influence of the mutations on SARS-CoV-2 M^pro reactivity

A machine learning analysis was performed following the investigation regarding the influence of the K90R, P99L, P108S, and N151D mutations on the functionality of the SARS-CoV-2 M^pro. Thereby, the combined susceptibility to Sγ-C145 deprotonation and Nε-His41 protonation (named henceforth as CATALYSIS _SUSCEP) was first evaluated. The variable CATALYSIS _SUSCEP (in arbitrary units) can be understood as the susceptibility for the catalytic mechanism to occur in each protein structural conformation. As shown in Fig. 4 A, the chains A of the WT and mutant SARS-CoV-2 M^pros present higher values of CATALYSIS _SUSCEP than their counterparts (chains B). In part, this result is according to MD simulations performed by Chen et al., 2006 [52], which indicated that only one of the monomers forming the active dimer presents catalytic activity at a given time. However, based on our data, it is not possible to conclude that the chains B are catalytically inactive; instead, they generally display catalytic dyads with lower susceptibilities to nucleophilic group formation.

Fig. 4.

Analysis of the influence of the K90R, P99L, P108S, and N151D mutations on the functionality of the SARS-CoV-2 M^pro performed by machine learning. (A) Combined susceptibility to Sγ-C145 deprotonation and Nε-His41 protonation (CATALYSIS_SUSCEP) of chains A and B of WT and mutant SARS-CoV-2 M^pros. (B) The inferred decision tree (AUC = 0.90) with structures which do present or not present the aa residue F140 in the SUBSTRUCTURE_3Å. (C) New decision tree. (D) Distances between F140 and C145 (D_140-145 values) associated with the normal modes calculated for the SARS-CoV-2 M^pro structures of WT and mutants. The numbers in parentheses indicate the number of correctly/incorrectly classified instances.

Furthermore, as it follows and according to the CATALYSIS _SUSCEP values, it was possible to find other interesting issues related to the influence of the K90R, P99L, P108S, and N151D mutations on the enzymatic reactivity of the SARS-CoV-2 M^pros. Firstly, the machine learning analysis (Information Gain Ranking Filter) revealed the following attributes concerning to CATALYSIS _SUSCEP discrimination: (1) the aa residues at a maximum distance of 3 Å from any atoms of the catalytic dyad (named henceforth as SUBSTRUCTURE _3Å) (InfoGain score = 0.76), and (2) the distance between the atoms Sγ-C145 and Nε2-H41 (InfoGain score = 0.44). To better explore these data, the CATALYSIS _SUSCEP values from WT and the mutant chains A were ranked and discretized in quartiles (Q1 to Q4, in ascending order of enzymatic reactivity according to the CATALYSIS _SUSCEP values) and then defined as classes for subsequent machine learning analyses (Fig. 4A). The inferred decision tree (AUC = 0.90) revealed that the main parameter for class discrimination is the attribute (1); this result agrees with those obtained using the Information Gain algorithm. Additionally, the more populated decision tree branches indicate that the structures which enclose the aa residue F140 in the SUBSTRUCTURE _3Å also belong to the quartile with the higher CATALYSIS _SUSCEP values (Q4). On the other hand, the structures which do not present F140 in the SUBSTRUCTURE _3Å show lower CATALYSIS _SUSCEP values (in this case, these structures were allocated in the first two quartiles Q1 and Q2) (Fig. 4B). Since shorter distances between the aa residues C145 and F140 (named henceforth as D _145-140) are also associated with higher CATALYSIS _SUSCEP, these parameters were also identified as correlated variables (ρ = ₋0.20 and p-value = 0.0004). It is very known that the SARS-CoV-1 M^pro substrate-binding site (located in a crevice dividing the domains I and II) encloses the following substructures: a subsite denominated S1 (formed by the residues F140, H163, M165, E166, and H172), which gives to this enzymes a strict specificity related to a glutamine residue at the substrate P1 position (QP1); an oxyanion hole (formed by the main chain amides of the residues G143, S144 and C145); and an oxyanion loop (formed by the protease segment S139 to L141) [57,60,64]. The main QP1/substrate-binding site interactions are hydrogen bonds involving (i) the QP1, H163, and E166 side chains, and (ii) the QP1 and oxyanion hole residues' main chains [64]. As already described, the oxyanion hole is crucial to stabilize the tetrahedral intermediate derived from the substrate during the catalytic reaction [60]. Thus, the role of the residue F140 is indispensable for proteolysis, since the insertion of its large aromatic ring into the S1 subsite helps to keep this structural element in an open and functional conformation [60]. Indeed, the correct structural arrangement of the whole SARS-CoV-1 M^pro substrate-binding site and its corresponding region in SARS-CoV-2 M^pro comes from a stacking interaction between the side chains of H163 and F140 and the formation of a hydrogen bond network comprising other residues from both monomers [65]. The stacking also keeps H163 uncharged in different pH conditions [64], avoiding the “incorrect” trapping of this histidine between two neighboring tyrosine residues (Y126 and Y161) and the consequent structural alteration of S1 subsite and substrate-binding site [60]. These features strongly support that F140 is the most important stabilizing element for the active conformation of the S1 subsite and substrate-binding site. Further, as demonstrated by several works, the correct conformation of the substrate-binding site is a sine qua non condition for dimerization/activation of the SARS-CoV-1 and SARS-CoV-2 M^pros [5,17,60,63,64,66,67]. The same dataset was also evaluated using PCA, and the results were congruent with the ML analyses. PCA results are presented in the supplementary material (Fig. S11).

Interestingly, the detailed structural analysis of the WT and mutant structures also exposed some subtle traits possibly associated with F140 and the SARS-CoV-2 M^pro enzymatic mechanism. As depicted in our results, some SARS-CoV-2 M^pro displacements allow F140 to access the SUBSTRUCTURE _3Å, thereby generating different conformations which are associated with higher CATALYSIS _SUSCEP. A careful analysis of the SUBSTRUCTURE _3Ås associated with higher CATALYSIS _SUSCEP values shows a slight movement of the Nε2-H163 atom towards the O-carboxyl-H164 atom, shortening in approximately 0.2 Å the distance between these two atoms (4.3 Å is the distance between the Nε2-H163 and O–H164 O-carboxyl-H164 atoms when F140 is close whereas this measure increases to 4.5 Å when the F140 phenyl ring is not in the neighborhood). Notably, this small distance alteration was able to modify the electronic distribution on the O-carboxyl-H164 atom and increment the CATALYSIS _SUSCEP since the C145 sulfhydryl group is placed at 1.9 Å from the H164 carboxyl group. Actually, there is a negative Pearson correlation between the Nε2-H163/O-carboxyl-H164 atomic distance and the CATALYSIS _SUSCEP (ρ = ₋0.27 and p-value <0.0001), revealing thus an association between these parameters. Another interesting point related to the F140 is that it also interacts with the oxyanion hole residues S144 and G143. Remarkably, Firouzi et al. (2022) [68] highlighted that the residue S144 interacts with ligands in 58.2% of the protein/ligand complexes available in the Protein Data Bank and also indicated this residue as the most rigid of the substrate-binding site, presenting a similar configuration in practically all of the analyzed SARS-CoV-2 M^pro X-ray crystal structures with a resolution less than 3 Å. Hence, it seems clear that the lack of interaction of some key residues with the residue F140 may probably affect the substrate-binding site, oxyanion hole stability, and the SARS-CoV-2 M^pro enzymatic reactivity.

In light of these findings, the following question to be investigated was the particular influence of the K90R, P99L, P108S, and N151D mutations on the CATALYSIS _SUSCEP values. In this context, when the CATALYSIS _SUSCEP values were evaluated without considering the dataset attributes (1) and (2), a new decision tree (Fig. 4C) was generated with the following findings: (i) the N151D mutation (located at the domain II) is associated with a lower CATALYSIS _SUSCEP (Q1); (ii) the WT and P99L (located at the domain I) were allocated in branches with lower/intermediate CATALYSIS _SUSCEP; (iii) the mutation K90R (located at the domain I) was related to intermediate/high CATALYSIS _SUSCEP; and (iv) the mutation P108S (located at the domain II) is found in structures with the high CATALYSIS _SUSCEP (Q4). Remarkably, this machine learning analysis also suggests that broader displacements lead to lower enzymatic reactivity, as confirmed by the Pearson correlation test calculated with the CATALYSIS _SUSCEP values and the normal modes-based displacements applied to the SARS-CoV-2 M^pro structure (ρ = ₋0.19, p-value = 0.048). Rather, there are different distances between F140 and C145 (D _140-145 values) associated with the normal modes calculated for the SARS-CoV-2 M^pro structures. Although there is a lack of direct correspondence between the modes 1 to 6 of the WT and K90R, P99L, P108S, and N151D mutant structures, it is possible to identify a distinct D _140-145 pattern between the calculated structures, with a particular difference in relation to the K90R, P108S mutants, and WT SARS-CoV-2 M^pro structures. Regarding K90R mutant structures, D _140-145 values related to the normal modes 1 to 6 varied between 5 and 7 Å, whereas the corresponding distances for the WT structures showed values below 5 Å in approximately 50% of the structures (Fig. 4D). However, as mentioned before, many K90R mutant structures were allocated in Q3 and Q4 quartiles, i.e., they presented intermediate and high CATALYSIS _SUSCEP. On the other hand, most of P108S substructures were allocated in the quartile Q4 (high CATALYSIS _SUSCEP values). In this case, the number of P108S mutant SARS-CoV-2 M^pro structures with D _140-145 values between 3 and 4 Å is significantly higher compared to the wild-type population (Fig. 4D). Also, the displacements along the normal modes 1, 2, 3, and 6 enclose about 50% of the P108S structures that present D _140-145 around 3 Å. Thus, this result suggests that the mutation P108S could favor an ensemble of SARS-CoV-2 M^pro structures with higher enzymatic reactivity induced by a closer distance of the residues F140/C145. It is feasible, therefore, that the reactivity of more than half of the P108S mutant structures could be related, as described before, to the structural arrangement of the substrate-binding site induced by the proximity of F140 to the catalytic dyad. Thereby, the mutation P108S could favor an ensemble of more enzymatically reactive SARS-CoV-2 M^pro due to the closer distance of the residues F140/C145. Indeed, Amamuddy et al. (2020) [17] pointed out that the catalytic dyad is more stable than other residues from the substrate-binding site; consequently, a closer proximity of F140 may likely lead to an even higher structural stabilization of the pair H41/C145. On the other hand, the higher D140-145 values found in some K90R mutant structures (compared to the WT structures) may indicate a border line (in this case, a F140/C145 distance range of 5–7 Å) between high and intermediate CATALYSIS _SUSCEP. The K90R mutant intrinsic molecular movements could probably reach more unrestrained substrate-binding and catalytic sites and, seemingly, a higher number of conformations associated with lower enzymatic reactivity. Here, it is important to mention that Amamuddy et al. (2020) [17] demonstrated that the domain II aa residues S139, F140, and Q189 of the substrate-binding site could show substantial conformational fluctuations during coarse grained (CG) simulations, while other aa residues (G143, S144, H164, H163, E166, P168, and C145) presented only moderate structural alterations. The analysis of the K90R mutant structures gathered after the simulations showed that these molecules present longer distances between some important aa residues from the substrate-binding and catalytic sites in relation to the WT and the other mutants (ex.: pairs F140/H163, E166/H172, H41/F140, and F140/C145).

However, literature data indicate that the P108S and K90R mutations diminish or do not change the enzymatic activity of the SARS-CoV-2 M^pros. Abe et al. (2019) [69] identified four non-synonymous mutations in the SARS-CoV-2 sub-lineage B.1.1.284 inversely correlated with COVID-19 severity, including the P108S mutation found in the SARS-CoV-2 M^pro. These authors identified patients infected with the P108S mutant sub-lineage B.1.1.284 who presented, in general, a comparatively milder clinical course in comparison to patients infected with non-mutant virus from the same sub-lineage. Further, the same authors, based on a hydrogen/deuterium exchange mass spectrometry (HDX-MS) assay, revealed that the specificity constant (Kcat/Km) of the P108S mutant SARS-CoV-2 M^pro was 58% lower than that of the P108 SARS-CoV-2 M^pro. Also, Ullrich et al. (2022) [70] studied, among others, the K90R mutant in different lineages of SARS-CoV-2 (C.37 Lambda, B.1.1.318, B.1.2, B.1.351 Beta, B.1.1.529 Omicron, P.2 Zeta) and showed that this mutation presents a catalytic activity equivalent to the WT SARS-CoV-2 M^pro. Similarly, the catalytic efficiency of the K90R SARS-CoV-2 M^pro is comparable to the wild-type enzyme, according to Greasley et al. (2022) [71]. Thus, how is it possible to solve the paradox arising from the fact that there is a clear catalytic reactivity difference between different conformations of P108S and K90R mutant SARS-CoV-2 M^pros, as shown above?

At first sight, this puzzle could be solved by taking into account the work performed by Kuroda and Tsumoto (2021) [72], which revealed an interesting particularity of the emergent P108S mutation identified by Abe et al. (2019) [68]. After analyzing 5.0-μs MD trajectories of the P108S mutant SARS-CoV-2 M^pro in comparison to the WT protein, Kuroda and Tsumoto (2021) [72] suggested that their work explains, in structural basis, the significant increase of the K _m value and little K _cat decrease related to the P108S mutant SARS-CoV-2 M^pro (Abe et al., 2019) [69]. According to these authors, the predominant effects of SARS-CoV-2 M^pro K _m come from the substrate-binding site obstruction caused by a flexibility restraint of a loop segment (aa residues C44 to N53) hindering the access to the catalytic dyad. Thus, even though the P108S mutations may to become SARS-CoV-2 M^pro more prone to reach reactive conformations, this theoretical advantage of the mutant protease is practically neutralized by the stereochemical hindrance related to the mentioned rigidified loop bordering the substrate-binding site. A similar mechanism could also justify the lack of a higher enzymatic activity of the K90R mutant SARS-CoV-2 M^pro at least for the conformations classified as more reactive, albeit in this case, our simulations were not able to indicate a higher rigidity of the loop comprising the aa residues C44 to N53. Nevertheless, the evidence shown in this work clearly highlights that it is not possible to rule out the possibility that the combination of certain, although yet unknown, aa substitutions could, at the same time, increase the SARS-CoV-2 M^pro reactivity and promote an easier path of substrates to the catalytic dyad. Such straightforward interaction of the SARS-CoV-2 M^pro with their target molecules may be then a reason for concern regarding the emergence of novel viral variants potentially more pathogenic.

Hence, based on the presented findings, it is possible to propose two SARS-CoV-2 M^pro structural characteristics that could be observed and considered for the detection and surveillance of new SARS-CoV-2 VOCs. First, the reactivity of the catalytic dyad, since it is feasible, as shown in this work, that different viruses can carry main proteases with distinct enzymatic activities. A clue for the identification of more reactive enzymes could be the inspection of F140. According to our results, an approximate F140/catalytic dyad distance around or below 3 Å is a potential trigger for generating SARS-CoV-2 M^pro conformations with higher reactivities. Presumably, there are also other structural mechanisms related to the increasing of the enzymatic reactivity, since approximately 50% of the conformations generated during the simulations that were classified into the quartiles Q4 did not present F140 in the SUBSTRUCTURE_3Å. The lack of significant correlations in these cases is probably linked to the distance-based metrics used for the analyses, therefore, new atomic and aa residues network connections should be explored to completely clarify the reasons for the increase of enzymatic reactivity. Following, it would be useful to analyze a second structural characteristic related to the degree of flexibility/rigidity of loops and aa residues around the entrance of the catalytic site, as suggested by Kuroda and Tsumoto (2021) [72].

4. Conclusions

This work compared the wild-type and single-point mutant SARS-CoV-2 M^pros using a normal modes-based protocol to analyze simple parameters such as Cα flexibility, potential energy, hydrophobic and electrostatic contributions to free energy, solvent accessible surface areas, and catalytic dyad distance. These parameters allowed the selection of potential structurally stable dimers, which demonstrated that some single surface aa substitutions not located at the dimeric interface (K90R, P99L, P108S, and N151D) are able to induce significant quaternary changes. This is a novel finding, since previous works that showed similar dimeric modifications based their conclusions exclusively on the influence of aa residues directly involved in the formation of dimeric interfaces or placed at other key regions of the SARS-CoV-2 M^pro as, for example, the substrate-binding region and the catalytic site. Based on this finding, a machine learning analysis was performed to identify possible functional alterations coming from the influence of the surface single-point mutations K90R, P99L, P108S, and N151D.

For the first time, it was demonstrated, by using a QM method, the influence of SARS-CoV-2 M^pro mutations on the catalytic mechanism. This finding confirms that only one of chains of the WT and mutant SARS-CoV-2 M^pros are prone to cleave substrates. Additionally, the AI analysis also indicated that the aa residues at a maximum distance of 3 Å from any atoms of the catalytic dyad and the distance between the atoms Sγ-C145 and Nε2-H41 are significant parameters associated with the SARS-CoV-2 M^pro reactivity considering the mutations K90R, P99L, P108S, and N151D. At this point, it was also possible to identify the aa residue F140 as an important factor related to the increasing enzymatic reactivity of a significant number of SARS-CoV-2 M^pro conformations generated by the normal modes-based simulations. This fact, associated with the previous suggestion of the influence of a rigidified loop segment on the SARS-CoV-2 M^pro activity, addresses the possibility of the emergence of different, although yet unknown, mutation combinations that could lead to more pathogenic viruses. Hence, we propose two SARS-CoV-2 M^pro structural characteristics that may be important for surveillance of new SARS-CoV-2 VOCs. The first one is the reactivity of the catalytic dyad; in this case, a hint to identify potential molecules with high reactivity could be the inspection of F140, although this is almost certainly not the unique structural trait that defines the increase of enzymatic reactivity. Finally, in our opinion, it is also essential to evaluate the degree of flexibility/rigidity of loops and aa residues around the entrance of the catalytic site, since a clear and direct path to the catalytic site is essential for SARS-CoV-2 M^pro activity as well highlighted in the literature.

Funding

This work was supported by Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) - Brazil: Grant Numbers: 423717/2021-9 and 164052/2020-8; and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) - Brazil, Fundação de Amparo à Pesquisa (Fapesp) [Finance Code 2021/14-7] - São Paulo, Brazil.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Ana ligia Scott reports financial support was provided by National Council for Scientific and Technological Development.

Appendix A. Supplementary data

The following is the Supplementary data to this article.

Data availability

Data will be made available on request.