SARS-CoV M^pro inhibitory activity of aromatic disulfide compounds: QSAR model

Abstract

The main protease (M^pro) of SARS-associated coronavirus (SARS-CoV) had caused a high rate of mortality in 2003. Current events (2019–2020) substantiate important challenges for society due to coronaviruses. Consequently, advancing models for the antiviral activity of therapeutic agents is a necessary component of the fast development of treatment for the virus. An analogy between anti-SARS agents suggested in 2017 and anti-coronavirus COVID-19 agents are quite probable. Quantitative structure-activity relationships for SARS-CoV are developed and proposed in this study. The statistical quality of these models is quite good. Mechanistic interpretation of developed models is based on the statistical and probability quality of molecular alerts extracted from SMILES. The novel, designed structures of molecules able to possess anti-SARS activities are suggested. For the final assessment of the designed molecules inhibitory potential, developed from the obtained QSAR model, molecular docking studies were applied. Results obtained from molecular docking studies were in a good correlation with the results obtained from QSAR modeling.

Graphical Abstract

Communicated by Ramaswamy H. Sarma

Introduction

The genus coronavirus appertains to the plus-strand RNA virus and currently contains about 25 species that are classified into three groups under their genetic and serological relationships (Siddell et al., 2005; Yang et al., 2005). Coronaviruses (CoVs) infect humans and different species of animals, causing severe diseases (Yang et al., 2005). For example, human coronavirus (HCoV) strains 229E (HCoV-229E), NL63 (HCoV-NL63), OC43 (HCoV-OC43), and HKU1 (HCoV-HKU1) cause a significant portion of respiratory tract infections in humans, including bronchiolitis and pneumonia. They have also been involved in otitis, asthma, diarrhea, and neurological disease (Remuzzi & Remuzzi, 2020; Siddell et al., 2005; Yang et al., 2005).

Acute respiratory syndrome is an infectious disease caused by different viruses. Symptoms can be mild to severe. Severe acute respiratory syndrome (SARS) is a highly infective respiratory disease caused by SARS coronavirus (SARS-CoV). The SARS-CoV-2 is a very dangerous virus, according to most experts, and proved to be the etiological agent of a global outbreak of a life-threatening form of pneumonia called severe acute respiratory syndrome (SARS) (Abuhammad et al., 2017; Amin et al., 2020; Jain et al., 2020; Remuzzi & Remuzzi, 2020; Worldometer's COVID-19 data, 2020). The coronavirus COVID-19 also is causing severe respiratory pathology in humans and its spread alarms health organizations around the world (Amin et al., 2020; Remuzzi & Remuzzi, 2020; Worldometer's COVID-19 data, 2020).

An attractive therapeutic target for CoVs is the main protease (M^pro) or 3-chymotrypsin-like cysteine protease (3CL^pro), as this enzyme plays a key role in polyprotein processing and is active in a dimeric form. Further, M^pro is highly conserved among various CoVs, and a mutation in M^pro is often lethal to the virus. Thus, drugs targeting the M^pro enzyme significantly reduce the risk of mutation-mediated drug resistance and display broad-spectrum antiviral activity (Goyal & Goyal, 2020).

One can expect some analogies between different coronaviruses. There are available experimental data for inhibitory activity towards SARS-CoV-2 of a set of organic compounds (Abuhammad et al., 2017).

Quantitative structure – activity relationships (QSARs) are a tool to solve problems related to medicinal chemistry when the analysis of the chemical structure becomes important (Abuhammad et al., 2017; Amin et al., 2020; Bhargava et al., 2017, 2019; Halder, 2018; Jain et al., 2020; Makhouri & Ghasemi, 2018; Wang et al., 2017). The approach can be based on the Monte Carlo technique (Abuhammad et al., 2017; Amin et al., 2020; Bhargava et al., 2017; Duhan et al., 2019; Jain et al., 2020; Kumar et al., 2020; Kumar & Kumar, 2020).

Disulfide bonds play key roles for bioactive proteins in the aspect of preferable correct folding (Wang et al., 2017). It is not a simple task to develop a method sensitive to assigned molecular features. CoMFA and CoMSIA were used to build up 3D-QSAR for novel inhibitors of SARS-CoV-2 main protease (Wang et al., 2017). The direct interface between 3D models and the chemical technique of synthesis of considered unsymmetrical aromatic disulfides is an attractive advantage of the above research work (Wang et al., 2017).

In this research in silico method - MD was used to test the inhibition effect of designed molecules, because, the calculated binding energies can correlate with inhibitory potentials (Halperin et al., 2002; Kitchen et al., 2004).

We suppose that these models can provide a bridge to develop antiviral agents for the COVID-19 when the corresponding experimental data will become available.

Method

Data

The data on unsymmetrical aromatic disulfide compounds and their SARS-CoV M^pro inhibitory activities (IC50, µM) are available in the literature (Wang et al., 2017). This data set includes forty compounds. In our study the compounds were randomly splitting into the active training set (25%), passive training set (25%), calibration set (25%), and validation set (25%). The distribution of data into those four sets is a necessary component of building up QSAR models using the CORAL software (Toropova & Toropov, 2017). The details and tasks related to each of the above sets are described in the literature (Toropova & Toropov, 2019).

Optimal descriptor

The optimal descriptor is calculated with molecular features extracted from SMILES (Weininger, 1988):

$$\mathit{DCW}\left(T^{*},N^{*}\right)=\sum _{k=1}^{NS}CW(S_k)+\sum _{k=1}^{NS-1}CW(S{S}_k)+\sum _{k=1}^{NS-2}CW(SS{S}_k)$$ (1)

The S_k is the SMILES-atom, i.e. single symbol (e.g. C, N, O, etc.), or a group of symbols which cannot be examined apart (e.g. Cl, Br, %11, etc.); SS_k is a pair of SMILES-atoms; SSS_k is a trine of SMILES-atoms. It is to be noted the SMILES-atom can be the traditional atom (e.g. N, S, Br, etc.), but also it can represent covalent bonds (i.e. =, #, @), as well as other molecular phenomena (e.g. the number of rings, 1, 2, %11, etc.). The NS is the number of SMILES-atoms for given SMILES.

Having the correlation weight for all molecular features involved in the modelling process, one can obtain the one-variable model:

$$I{C}_{50}\left[\mathrm{\mu }M\right]=C_0+C_1\times \mathit{DCW}\left(T^{*},N^{*}\right)$$ (2)

Monte Carlo optimization

The scheme of the improved Monte Carlo optimization is described in the number of papers (Kumar & Chauhan, 2017; Manisha et al., 2019; Toropova et al., 2020). The essence of this version of the optimization procedure is in the application of the Index of ideality of correlation (IIC). Models for the inhibitory activity built up here apply the algorithm previously described (Kumar & Chauhan, 2017; Manisha et al., 2019; Toropova et al., 2020). Three models have been developed based on the three splits of data as described above.

Molecular docking

Different host receptors for cellular entry, the structural proteins (antigens), and the high mutation and recombination rates of CoVs are a significant problem in the development of wide-spectrum anti-CoV drugs. The accession number in the Protien Data Bank (PDB, http://www.rcsb.org/pdb/) for each structure of SARS M^pro is individual. The crystal structure of SARS-CoV M^pro in complex with inhibitor (PDB code 2AMD) was obtained from the PDB databank and used for molecular docking studies. All molecules that acted as ligands for docking studies were drawn using Marvin sketch (Marvin 6.1.0, 2013, ChemAxon), and MMFF94 force field implemented in Marvin sketch was used to obtain their optimal 3D geometry. In this research, Molegro Virtual Docker (MVD) was used as the main software for molecular docking studies. MVD can be successfully used in MD studies for obtaining an appropriate geometrical orientation of the ligand inside the active site of the studied enzyme. Also, MVD can identify hydrogen bonds and hydrophobic interactions between flexible studied compounds and rigid amino acids from the enzyme active site. Further, MVD can be used to calculate “scoring” functions related to relevant binding energies and these functions can be applied for the estimation of the studied compounds inhibitory effect (Thomsen & Christensen, 2006; Zivkovic et al., 2020). In this study, the used “scoring” functions were: Hbond, VdW, Steric, Pose energy, MolDock, and Rerank Score and their detailed description is given in the literature (Zivkovic et al., 2020). A published approach was used for molecular docking protocol validation (Amin et al., 2018). Maestro software version 11.6.012. was used for presenting 2D representations of interactions between the studied molecules and amino acids from the studied enzyme active site.

Results and discussion

The models for SARS-CoV M^pro inhibitory activity obtained in the three employed splits are the following:

$${\text{IC}}_{50}\left[\mu \mathrm{M}\right]=\mathrm{}-2.464\left(\pm \mathrm{}0.517\right)\mathrm{}+\mathrm{}0.1507\left(\pm \mathrm{}0.0153\right)\mathrm{}*\mathrm{}\text{DCW}\left(1,10\right)$$ (3)

$${\text{IC}}_{50}\left[\mu \mathrm{M}\right]\mathrm{}=\mathrm{}-2.807\left(\pm \mathrm{}0.463\right)\mathrm{}+\mathrm{}0.2158\left(\pm \mathrm{}0.0189\right)\mathrm{}*\mathrm{}\text{DCW}\left(1,10\right)$$ (4)

$${\text{IC}}_{50}\left[\mu \mathrm{M}\right]\mathrm{}=\mathrm{}-5.857\left(\pm \mathrm{}1.059\right)\mathrm{}+\mathrm{}0.2028\left(\pm \mathrm{}0.0265\right)\mathrm{}*\mathrm{}\text{DCW}\left(1,10\right)$$ (5)

Table 1 contains the statistical characteristics of these models. Table 1 confirms two important features of the approach: (i) the statistical quality for the validation set is high, the correlation coefficient for each split is larger than 0.90; and (ii) dispersion for all considered criteria is small.

Table 1.

The statistical characteristics of the developed model for SARS-CoV M^pro inhibitory activities for three random splits of data.

Split	Set	R2	Q2	CCC	IIC*	RMSE	MAE
1	Active training set	0.6464	0.4450	0.7852		0.956	0.795
	Passive training set	0.7586	0.5802	0.8205		0.886	0.610
	Calibration set	0.9115	0.8512	0.9026	0.9539	0.470	0.371
	Validation set	0.9566	0.9320	0.9216		0.542	0.438
2	Active training set	0.7439	0.5840	0.8531		0.893	0.762
	Passive training set	0.7654	0.6003	0.7366		1.07	0.714
	Calibration set	0.9174	0.8799	0.9358	0.9577	0.413	0.336
	Validation set	0.9442	0.9154	0.9479		0.454	0.309
3	Active training set	0.6983	0.4234	0.8223		0.824	0.650
	Passive training set	0.8445	0.7742	0.5553		1.54	1.33
	Calibration set	0.9124	0.8829	0.9081	0.9548	0.345	0.267
	Validation set	0.9175	0.8916	0.9249		0.489	0.403

^*IIC = index of ideality of correlation (Toropova & Toropov, 2019); RMSE = root mean squared error; MAE = mean absolute error.

The statistical quality of the model based on 3D representation of the molecular structure suggested in the literature (Wang et al., 2017) is defined as R²=0.916, Q²=0.681. Three compounds were removed as influential outliers (Wang et al., 2017). Thus, the comparison of the above model with the predictive potential of models suggested here confirms that the described approach gives models with quite good predictive potential. Several runs of the Monte Carlo optimization with a different distribution of data into the training and validation sets allow obtaining the statistical and mechanistic interpretation of the model (Table 2). It should be noted that promoters of increase for IC50[µM] have stable prevalence, whereas promoters of decrease are relatively rare ones.

Table 2.

Promoters of increase and decrease of the inhibitory activity of SARS‐CoV M^pro (IC₅₀, μM).

No.	SMILES attribute	CWs Probe 1	CWs Probe 2	CWs Probe 3	N1	N2	N3
	Promoters of IC50 increase
1	1………..	2.63153	1.15486	3.78669	10	10	10
2	2………..	2.55878	1.21207	1.64933	10	10	10
3	S………..	0.48795	0.06021	2.20339	10	10	10
4	S…S…….	2.02730	0.56998	2.44765	10	10	10
5	c………..	0.05582	0.34232	0.34661	10	10	10
6	c…1…….	0.07471	0.40638	1.18716	10	10	10
7	c…S…S…	2.96490	1.14621	2.40403	10	10	10
8	c…c…….	0.04212	0.58132	0.22205	10	10	10
9	n………..	0.30264	0.00072	0.28500	10	10	10
10	n…c…….	1.35271	0.05507	0.12055	10	10	10
11	C…(…….	0.43671	0.17893	0.17255	9	8	6
12	C………..	1.10523	0.50393	0.47038	9	8	7
13	O…(…….	0.16441	0.57525	0.88699	9	7	7
14	O…=…….	0.13313	0.43790	0.28864	9	7	6
15	n…c…(…	0.10720	0.32142	0.33964	8	6	4
	Promoters of IC50 decrease
1	N………..	−0.02294	−0.17040	−0.48815	6	3	4
2	3…n…(…	−0.15624	−0.05464	−0.16157	3	2	0
3	n…3…n…	−0.25960	−0.10767	−0.08532	3	2	0
4	n…(…O…	−2.54052	−0.12802	−2.47903	2	0	0
5	1…n…(…	−2.13926	−1.32809	−2.11675	1	3	1
6	s…(…1…	−0.05921	−0.10068	−0.01523	1	1	2

N1, N2, and N3 are frequencies of molecular features in the active training set, passive training set, and calibration set, respectively; CWs are the correlation weight.

The described approach indicates that the molecular features related to nitrogen atoms hint on how to select promising molecular structures (Table 3). In other words, the analysis of various structures based on the suggested CORAL model is transparent and convenient for practical applying.

Table 3.

Examples of proposed modifications for structure #38 together with variations of model values of SARS-CoV M^pro inhibitory activity.

Structures and SMILES	Model IC50[µM]	Comment
Basis Clc2ccc(SSc1ncccn1)cc2 Experiment IC50[µM] = 0.684	0.921
Improvement Clc2cnc(SSc1ncccn1)cc2	0.850	Fragment [N] is added
Improvement Fc2ccc(SSc1ncccn1)cc2	0.909	Fragment [F] is added
Improvement Fc2cnc(SSc1ncccn1)cc2	0.838	Fragments [N] and [F] are added

Figure 1 represents experimental and calculated with Equations (3)–(5) the inhibitory activity of SARS‐CoV M^pro. One can see (Figure 1), the split (distribution of available data into the training set, and the validation set) have a significant influence on the prediction for a given structure. However, as noted above, in all the described experiments, the statistical quality of models is quite good.

Figure 1.

The experimental and calculated SARS-CoV M^pro inhibitory activities for three random splits.

For the assessment of designed molecules, inhibitory potential MD studies were used. Binding energies of the studies compound with the active site of SARS‐CoV M^pro are presented in Table 4 and the best-calculated poses for all studied compounds inside the SARS‐CoV M^pro were shown in Figure 2.

Table 4.

Score values (kcal/mol) for all computer-aided designed compounds.

Molecule	Pose Energy	HBond	Steric	VdW	MolDock Score	Rerank Score
A	−95.3978	0	−108.245	−31.6369	−93.9852	−81.9477
A1	−96.2774	−2.5	−109.457	−21.6198	−94.3305	−78.3661
A2	−96.357	0	−108.868	−30.6175	−94.9384	−82.1304
A3	−97.1479	−2.5	−107.498	−33.5078	−93.8624	−85.0888

Figure 2.

Binding energies of the studied compound with the active site of SARS-CoV M^pro.

“Scoring” functions can be associated with the inhibitory potential of the studied compound and according to obtained results for MolDock score function molecule with the highest potential is A2, while results for ReRank score function indicate that molecule with the highest potential is A3. According to results for both above stated “scoring” functions molecule with the lowest inhibitory potential is A. The highest energy interactions related to Van der Waals interactions were determined for molecule A3, while molecule A1 had the highest energy-related to steric interactions. The highest pose energy was calculated for molecule A3. All obtained results for “scoring” functions are in good agreement with results obtained from QSAR modeling. MD studies were applied to identify all interactions between the designed molecules and amino acids from SARS‐CoV M^pro, including hydrogen bonds, hydrophobic, and hydrophilic interactions, and they are presented in Figure 3. Most significant components of the interaction are histidine, valine, and glutamine (Figure 3).

Figure 3.

MD studies are applied to identify all interactions between the designed molecules and amino acids from SARS-CoV M^pro, including hydrogen bonds, hydrophobic, and hydrophilic interactions.

Supplementary materials section contains experimental and calculated SARS-CoV M^pro inhibitory activities for three random splits.

Conclusions

The described approach provides a quite good model for the inhibitory activity of SARS‐CoV M^pro by 50% (IC50, μM). The model is accompanied by the mechanistic interpretation that can help to compare the potentials of different molecular structures as possible antiviral agents. This facilitates the exploration of efficient drug candidates. The CORAL software is freely available on the Internet (www.insilico.eu/coral) and provides a capable tool for QSAR studies. Molecular docking studies were applied to calculate the energy and determine interactions between the designed compounds and amino acids inside the SARS‐CoV M^pro. In presented research calculated “scoring” functions for designed molecules were used as estimators for their inhibitory potential and obtained results were in good correlation with the results obtained from QSAR modeling.

Author contributions

The authors contributed equally to this work.

Glossary

Abbreviations

CoV

Coronaviruses

COVID-19

(CO: Corona, VI: Virus, D: Disease, 19: 2019)

CoMFA

Comparative molecular field analysis

CoMSIA

comparative molecular similarity indices analysis

DCW

optimal descriptor of correlation weights

HCoV

human coronavirus

MD

molecular docking

M^pro

main protease

MVD

Molegro Virtual Docker

PDB

Protein Data Bank

QSAR

quantitative structure activity relationships

RNA

ribonucleic acid

SMILES

simplified molecular input-line entry system

SARS

Severe acute respiratory syndrome

Funding Statement

A.A.T and A.P.T. are grateful for the contribution of the project LIFE-VERMEER contract (LIFE16 ENV/IT/000167) for the support. A.M.V. would like to thank the Ministry of Education and Science, the Republic of Serbia, under Project Number 172044. J.L. and DL would like to thank the NSF-CREST program for the support (grant HRD #154774).

Disclosure statement

The authors confirm they have no conflict of interest.