Virtual screening and molecular dynamics simulation study of approved drugs as a binder to the linoleic acid binding site on spike protein of SARS-CoV-2 and double mutant (E484Q and L452R)

Abstract

INTRODUCTION:

Binding of linoleic acid (LA) to the spike trimer stabilizes it in closed conformation hindering its binding to angiotensin-converting enzyme-2, thus decreasing infectivity. In the current study, we tend to repurpose Food and Drug Administration-approved drugs as binder to the LA binding pocket in wild and double mutant spike protein.

MATERIALS AND METHODS:

Approved drugs from DrugBank database (n = 2456) were prepared using Ligprep module of Schrodinger. Crystal structure of LA bound to spike trimer was retrieved (PDB: 6ZB4) and prepared using protein preparation wizard and grid was generated. A virtual screening was performed. With the help of molecular dynamics (MD) studies interaction profile of screened drugs were further evaluated. The selected hits were further evaluated for binding to the double mutant form of severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2).

RESULTS AND DISCUSSION:

Following virtual screening, a total of 26 molecules were shortlisted, which were further evaluated using 1ns MD simulation study. Four ligands showing better root mean square deviation (RMSD), RMSD to LA with interaction profile similar to LA were further evaluated using 100 ns MD simulation studies. A total of 2 hits were identified, which performed better than LA (selexipag and pralatrexate). Both these ligands were also found to bind to LA binding site of the double mutant form (E484Q and L452R); however, the binding affinity of pralatrexate was found to be better.

CONCLUSION:

We have identified 2 ligands (selexipag and pralatrexate) as possible stable binders to the LA binding site in spike trimer (wild and mutant form). Among them, pralatrexate has shown in vitro activity against SARS-CoV-2, validating our study results.

Introduction

The severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) causing coronavirus disease-19 (COVID-19) represents a global crisis.[1] There are 7 coronavirus strains that can infect humans. Among these, OC43, 229E, HKU1, and NL63 are the four endemic human coronaviruses that cause mild upper respiratory tract infections. However, SARS-CoV2, SARS-CoV, and MERS-CoV may induce severe pneumonia with acute respiratory distress syndrome and multiorgan failure, plausibly further leading to death.[2,3]

The spike protein (S1 region) of the virus facilitates binding of the virus to the host angiotensin converting enzyme-2 (ACE-2) receptor and thus facilitates the establishment of an infection in humans.[4] The receptor binding domain (RBD) of spike protein (S1) mediates direct contact with ACE-2. Following this contact, the S1/S2 polybasic cleavage site is cleaved (proteolytically) by cathepsin-L and Transmembrane protease serine 2 and the subsequent events lead to the activation of the fusion-peptide which mediates host and viral membrane fusion. After fusion of the plasma membrane, viral genome enters into the host cytosol, takes over the host cell machinery to start its own replication/translation mechanism to produce different viral structural and nonstructural protein components, which further takes part in the formation of virus particles.[5]

To summarize, S1 protein: ACE2 receptor interaction leads to the initial attachment of the coronavirus to the host and the subsequent establishment of infection. Although in all the earlier documented crystal structures of spike: ACE2 of SARS-CoV-2, the spike protein is in an open conformation, which is in favorable position for binding to ACE2.[6] A linoleic acid (LA) bound structure was documented which is less favorable to binding to ACE2. Again, supplementation of LA synergizes the inhibitory action of remdesivir.[6]

Free fatty acid bind site in spike protein

LA, a free fatty acid, forms a strong bond with the trimer spike protein's (S1) three composite binding pocket, which is close to RBD. These LA binding pockets are also found in other highly pathogenic coronavirus, i.e., SARS-CoV and MERS-CoV. After binding of LA, trimer structure of spike protein gets locked in a close conformation. The LA binding pocket is made up of hydrophobic amino acids which form a cave like structure and LA embedded into this cave. The head group of LA binds with a second S1 chain of the trimer through interaction with amino acid residues R408 and Q409. A gating helix (tyr365 and tyr369) is present at the entrance of LA binding pocket. In the LA bound form, the gating helix amino acids are displaced by about 6Å. Overall, a compaction of trimer architecture in the region formed by the three RBDs giving rise to a locked S structure.[7,8] The LA binding pocket thus represents a promising target to develop a small molecular inhibitor that is able to lock the trimer structure in closed conformation and interfere in ACE-2 interaction. Similar strategy was applied to develop anti-viral against “rhinovirus,”[7,8] and this strategy was found to be fruitful in clinical trials.[9,10]

In this study, we virtually screened approved drugs from repurposing point of view to find suitable binders which can bind and occupy the LA binding site and lock the spike protein in closed conformation.

Materials and Methods

For molecular docking and molecular dynamic simulation processes (using ACER Predator Helios 300 laptop), in silico experiments were carried out using LINUX Ubuntu OS 18.04.02 LTS with Schrodinger Maestro version 2019-3.

The target protein data bank structure

Our target of interest was LA bound form of spike protein crystallized in closed form. In regard to, we could find one C1 symmetry spike protein crystal structure (PDB I.D. 6ZB4). This LA bound spike protein structure of spike protein was carried forward for further computational work.

Validation of the target protein

The target selected protein structure was validated using PROCHECK tool which is provided in SAVES V5.0 server.[11] The stereochemical efficiency of selected protein was examine by PROCHECK tool and also, we analyzed the geometry of “residues by residues” or whole residues. The protein model was also evaluated using Ramachandran plot. The PROCHECK tool is helpful to differentiate good and bad quality of protein structure. While evaluation, we found PDB: 6ZB4 was found to be within acceptable quality range and was used as a model for further study.

Protein preparation process

The target protein structure (PDB i.d. 6ZB4) was imported into Maestro version 2019-3 software which contains “protein preparation wizard” tool. In this tool, firstly, we preprocessed the protein, filled the missing loops and side chains. Water molecules were deleted beyond 5.0 Å. The structure was optimized and minimized using OPLS3 force field.

Ligand's preparation process

The approved molecules library from DrugBank (n = 2456 drugs) was obtained from DrugBank database in Spatial data file format. The drug was prepared using the “ligprep module” for further docking research and molecular docking modeling. The OPLS 2003 force field was employed for the minimization method, which is useful for preparing the ligand for future investigation.

Details of linoleic acid binding and its binding pocket

In an open conformation, SARS-CoV-2 Spike protein is compatible to bind with ACE-2 receptor. However, it is not found in case of LA bound closed form. In in vitro studies, the LA bound closed form showed reduced interaction with ACE-2. LA supplementation synergizes with “remdesivir” in suppressing the replication process of SARS-CoV-2.[6] In SPR experiments LA bound closed conformation was associated with reduced level of spike protein binding in presence of LA.[6] The LA binds to a hydrophobic pocket (shaped by phenylalanine) on the B chain of S1 protein, whereas head of the LA is anchored by its interactions with the positively charged residues of the A chain (arginine [R408] and glutamine [Q409] [PDB: 6ZB4, chain A]).[6] The other residues providing additional anchoring site to this cavity are TRP436, PHE342, VAL341, PHE377, PHE374, PHE338, CYS336, CYS361, PHE392, CYS379, CYS432, and PHE515 of chain B (PDB: 6ZB4).[6] Details of binding residues are showed in Figure 1.

Figure 1.

Details of LA binding pocket: Three chains of spike protein (S1 region) form a compact trimer. LA binds to residues 408 and 409 of one chain (head region) and the other part of LA binds to the residues surrounding the hydrophobic pocket. LA: Red, A chain: Ribbon, B chain (surface). (a): LA bound to A and B chain, (b): More detailed close up view, (c): LA binding pocket at 90° of first view, (d): LA binding pocket from 180° of first view with head region residues interacting with residues 408 and 409 of the A chain, (e): LA bound to residues 408 and 409 on A chain and hydrophobic pocket can be seen in the chain B surrounding LA, (f): LA bound to residues 408 and 409 on A chain and hydrophobic pocket can be seen in the chain B surrounding LA. The residues in the hydrophobic pocket (green) can be seen surrounding LA. LA = Linoleic acid

The LA binding pocket in protein structure is a promising therapeutic target for locks the S protein in the closed conformation and also interfere with receptor interactions.[6] Again, this LA binding site and its potential to lock the structure in closed conformation makes this site an important site from drug design perspectives. The glycosylation sites are located far away from the LA binding pocket and largely native. Thus, neither glycosylation nor mutation impacts LA binding site.[6] Details of LA binding site and the interacting residues are shown in Figure 1.

Grid generation

It is very crucial step for the binding of ligand to the selected site of receptor. Our area of interest is the LA binding pocket on spike protein. Hence, receptor grid was generated around LA which is anchored with residues from both A chain (anchoring head region) and B chain (hydrophobic cave forming residues). The grid was generated using “receptor grid generation” module of GLIDE, Schrodinger tool around the LA molecule which is already present in protein structure.

Virtual screening process

The Ligprep files and the receptor grid generated were used for further virtual screening. The virtual screening process was performed on GLIDE tool of maestro using a sequence of HTVS (50%)-SP (30%)-XP (10%) progression.[12] The binding affinity between different ligands and the target structure was evaluated in terms of “docking score.”

Molecular mechanics generalized born surface area

Prime molecular mechanics generalized born surface area (MM-GBSA) was used to calculate the binding free energy of docked structures (ligand-target complexes). The similar function was served by Maestro, Schrodinger, New York, USA. The docked structure complexes were submitted into Prime MM-GBSA tool and binding free energy was calculated using a Variable Surface Generalized Born (VSGB) solvent model, force field OPLS-3 and limiting residues flexibility distance to 5.0Å.[13]

Molecular dynamic simulations study

Desmond and Schrodinger were used to do molecular dynamic simulations of ligand-protein complexes. To begin, a water model was created using the “system builder” tool, and then sodium-ions and chloride-ions were introduced to neutralize. Minimization processes were done at OPLS3 force field, after which complexes were submitted for 1 ns and further 100 ns simulation at NPT and temperature 310K. Trajectory analysis was used to derive the root mean square deviation (RMSD) Ca, root mean square fluctuation (RMSF), ligand contact maps, and binding profile from the results.

Results

Protein target validation

The target protein was evaluated by PROCHECK protein assessment tool. The Ramachandran plot showed that 77.7% core of amino acid residues were in the most favored region and 22% were in the allowed region. Overall G factor was −0.03, maximum deviation was 6.1 and planar groups were 100% within limits [The data are shown in Figure 2].

Figure 2.

Details of protein validation by PROCHECK server

Virtual screening and molecular mechanics generalized born surface area result

The “Ligprep file” comprising prepared approved drugs/ligands from the “Drugbank database” and the “receptor grid” was used for virtual screening. With the virtual screening protocol (HTVS: 50%, SP: 30% and XP: 10%), a total of 26 compounds were screened and interestingly, all the compounds had better dock score than LA (−7.461 Kcal/Mol). The details of docking score MM-GBSA dG bind are shown in Table 1. The highest docking score was observed for polydatin (docking score − 12.733)

Table 1.

Result of docking score and molecular mechanics with generalised born and surface area dG bind of the screened molecules

Drug code	Zinc ID	Drug name	Docking score	MMGBSA
LA		LA	−7.540	−60.127
DB11263	ZINC000004098633	Polydatin	−12.733	−89.087
DB01046	ZINC000004654889	Lubiprostone	−12.319	−62.819
DB00654	ZINC000012468792	Latanoprost	−12.197	−94.34
DB09038	ZINC000036520252	Empagliflozin	−12.023	−91
DB04908	ZINC000052716421	Flibanserin	−11.921	−73.988
DB00398	ZINC000001493878	Sorafenib	−11.81	−75.047
DB00650	ZINC000009212428	Leucovorin	−11.804	−84.8
DB08907	ZINC000043207238	Canagliflozin	−11.754	−44.998
DB01216	ZINC000003782599	Finasteride	−11.614	−85.248
DB00661	ZINC000003871832	Verapamil	−11.514	−62.565
DB11362	ZINC000003990451	Selexipag	−11.463	−83.185
DB06210	ZINC000011679756	Eltrombopag	−11.422	−76.433
DB01132	ZINC000000968327	Pioglitazone	−11.31	−89.849
DB08877	ZINC000043207851	Ruxolitinib	−11.253	−56.6
DB00850	ZINC000019228902	Perphenazine	−11.228	−81.496
DB00315	ZINC000000015515	Zolmitriptan	−11.101	−55.309
DB00642	ZINC000001540998	Pemetrexed	−11.091	−56.462
DB09076	ZINC000034608502	Umeclidinium	−11.082	−66.81
DB01026	ZINC000000643138	Ketoconazole	−11.048	−99.285
DB00310	ZINC000000020253	Chlorthalidone	−10.984	−61.073
DB11256	ZINC000002005305	Levemifolic acid	−10.848	−70.118
DB06684	ZINC000001542113	Vilazodone	−10.778	−82.411
DB06813	ZINC000011616925	Pralatrexate	−10.758	−44.729
DB01101	ZINC000003806413	Capecitabine	−10.68	−61.296
DB09080	ZINC000034636383	Olodaterol	−10.643	−81.454
DB00308	ZINC000008214402	Ibutilide	−10.558	−82.799

MMGBSA=Molecular mechanics with generalised born and surface area solvation, LA=Linoleic acid

Molecular dynamics studies

The S protein stays in trimeric conformation and binding of LA to the LA binding pocket and its subsequent interactions results in a locked trimeric conformation of S protein, which results in a locked trimeric spike protein conformation, which is less favorable to ACE-2 binding. LA thus acts as a connector between two spike protein chains interacting via a hydrophobic pocket to one spike chain and interacts with residues 408 and 409 forming hydrogen bond with the other spike protein chain and thus stabilizing the locked Spike trimer.[6] Hence, to discover successful molecule, which should have the potential to lock the spike protein in a locked trimeric conformation, we can expect better locking when the molecule interacts with both the chains [Head-group residues: chain A: R408, Q409) and hydrophobic pocket].

For a detailed evaluation of the binding profile of the ligands selected on the basis of HTVS algorithm, we conducted molecular dynamics (MD) studies of all the molecules screened by virtual screening with 1ns MD simulation. Compounds performing equal or better than LA in the 1ns MD simulation were further evaluated in 100 ns MD simulation studies.

Molecular dynamics simulation studies of linoleic acid

The details of binding profile of LA (control/standard in case of our study) in MD studies are shown in Figure 3a. Regarding the binding of LA to the spike protein trimer, we can see that atoms 20 and 21 took main part in the binding of LA to both chain A and B of the spike protein and thus bind them together. The oxygen at atomic position 19 of LA interacted with both chain A residues (A: ARG408) and chain B residues (B: PHE374). Similarly, the oxygen with atomic position 20 also interacted with both chain A residues (A: LYS417, A: GLN409) and chain B residues (B: SER373, B: ALA372). Rest of the atoms of LA resided within a hydrophobic pocket.

Figure 3.

(a) Detailed interactions of Linoleic acid to its binding pocket in molecular dynamics studies. (b) Evaluation of effect of mutation and deletion of B: PHE 374 on the hydrophobic pocket formation binding profile of LA. LA = Linoleic acid

Physiological importance of major interacting residues of the LA binding pocket

Four major residues took part in the interaction between LA and spike A and B chain, these are A: LYS417, A: GLN409, A: ARG408, B: SER373 and B: PHE374 residue and they interacted for more than 50% of the simulation time. The importance of the interactions of A chain residues 408 and 409 as anchor residues.[6] and locking the headgroup of LA. Is already highlighted by previous authors. A: LYS417 residue takes part in direct interaction with ACE 2.[4,14] Hence, LA binding to this residue may take part critical role in inhibition of binding to ACE 2 theoretically. Again, among our study drugs, none of the ligands interacted with B: SER373 residue. B: PHE374 lines the gate of the gating helix (at the entrance of binding pocket). However, its relative importance is still unknown.

To evaluate the importance of B: PHE374, we conducted a series of mutation/deletion experiments followed by LA binding to evaluate the impact of those changes on binding pocket formation and LA binding profile [Data shown in Figure 3b].

We can see that in the B: PHE374 deleted protein: LA complex, RMSD was higher and the interaction profile was quite different compared to the wild form of protein. In the Phe374 deleted protein, no significant hydrogen bonding (>30% interaction fraction) was seen with the B chain, highlighting the importance of the residue in connecting the A and B chain by binding to both.

As in the interaction between LA and PHE374, COOH of LA interacted with NH2 + of Phe374, we mutated this residue with a negatively charged residue to evaluate the changes in LA binding in case of Phe374 and thus to further elucidate the importance of interaction with this residue. When Phe374 was mutated to Glu374 (Glutamic acid), compared to the wild type, RMSD and ligand RMSF was quite higher in the mutated form highlighting the importance of the interaction of the Phe374 residue. It is to be noted that despite mutation the formation of the hydrophobic pocket was seen, however, in case of deletion, the hydrophobic pocket formation was partially hampered. Thus, Phe374 represents an important residue in the formation of the hydrophobic pocket in the B chain [Data shown in Figure 3b].

Molecular dynamics simulation studies (1ns) of all the HTVS screened compounds

The details result of MD studies all the HTVS screened compounds at 1ns (velocity randomized every 500 Pico second) are shown in Table 2. When we compared the RMSD and RMSF result of LA bound protein (RMSD 2.05 ± 0.26 and RMSF 0.9 ± 0.4) with all 26 ligand-protein complexes, we found 20 ligands were shown better or similar RMSD and RMSF result compared to LA at 1ns (nano second) molecular dynamic simulation (polydatin, latanoprost, empagliflozin, sorafenib, leucovorin, canagliflozin, finasteride, verapamil, selexipag, eltrombopag, pioglitazone, ruxolitinib, perphenazine, ketoconazole, vilazodone, pralatrexate, capecitabine, and ibutilide). However, among all these 20 ligands, the interacting residues (A: LYS417, GLN409, ARG408, B: PHE374) were similar to LA only in case of 4 ligands (leucovorin, selexipag, pemetrexed, and pralatrexate) [Table 3a]. These four ligands were further subjected to further studied using 100 ns MD simulation studies [Figure 4].

Table 2.

Details of result from molecular binding studies (1 ns, randomized velocity every 500 ps)

Drug name	RMSD (A)	CΑ RMSF (A)	Interacting residues
LA	2.05±0.26	0.9±0.4	A: LYS417, GLN409, ARG408 B: PHE374, SER373, ALA372
Polydatin	1.92±0.17	0.89±0.42	A: GLN414, ARG408, GLU406, ARG403, ASP405, GLN409 B: TYR369, PHE374, LEU387, ALA363, LEU368, ALA372
Lubiprostone	1.97±0.22	0.91±0.47	A: ARG408, LYS417 B: TYR365
Latanoprost	1.90±0.11	0.87±0.39	A: 408 B: ALA363, PHE515
Empagliflozin	1.96±0.186	0.90±0.379	A: LYS417, ARG408 B: PHE377
Flibanserin	1.99±0.179	0.92±0.453	A: ASP405, GLU406, ARG408 B: PHE374, TYR369
Sorafenib	2.04±0.146	0.89±0.40	A: ARG408, LYS417 B: TYR369, LEU368
Leucovorin	1.91±0.11	0.87±0.398	A: LYS417, GLN409, ARG408 B: PHE374, PHE377, PHE392
Canagliflozin	2.05±0.18	0.884±0.42	A: ARG408, GLN409, ASP405 B: PHE374, PHE377
Finasteride	1.90±0.16	1.89±0.17	A: ARG408 B: TYR369
Verapamil	1.97±0.166	0.89±0.41	A: ARG408 B: ASP364, PHE338, PHE342, TYR365, LEU368, PHE374
Selexipag	2.05±0.187	0.89±0.46	A: LYS417, GLN409, ARG408 B: PHE374, PHE342
Eltrombopag	1.92±0.12	0.88±0.43	A: LYS417, GLN409, ARG408 B: PHE377
Pioglitazone	1.97±0.17	0.87±0.41	A: ARG408, ASP405
Ruxolitinib	1.91±0.19	0.89±0.38	A: GLN414, THR415, ARG408 B: PHE377, TYR369, TYR365, PHE374
Perphenazine	2.01±0.15	0.87±0.38	A: ARG408, THR415 B: ASN388, TYR365, PHE338
Zolmitriptan	2.24±0.16	1.07±0.54	LYS417, GLN414 B: PRO384, PHE377
Pemetrexed	1.95±0.18	0.88±0.41	A: ASP405, GLN409, LYS417, ARG408 B: ALA372, ALA363, LEU387, PHE374
Umeclidinium	1.97±0.20	0.94±0.43	B: PHE377, TYR365, PHE392, PHE342
Ketoconazole	1.91±0.173	0.86±0.39	A: LYS417, B: PHE377, TYR365
Chlorthalidone	2.18±0.19	1.06±0.59	A: ASP420, ARG408, LYS417, 409 B: TYR369, PHE374
Levemifolic acid	1.97±0.212	0.918±0.42	A: ASP405, ARG408, LYS417 B: PHE374, LEU368, ALA372, PHE392, TYR369
Vilazodone	1.96±0.15	0.85±0.39	A: GLU406, GLN409 B: ALA372, TYR369, PHE392
Pralatrexate	1.91±0.16	0.90±0.40	A: GLU406, GLN409, ARG408, LYS417 B: PHE374, ALA372, PHE377
Capecitabine	1.98±0.136	0.87±0.38	A: GLN414, ARG408, THR415
Olodaterol	2.094±0.19	0.880±0.40	A: ASP405, LYS417, GLU406, THR415, ARG408 B: PHE377, PHE374
Ibutilide	2±0.164	0.89±0.45	B: ALA372, TYR369, PHE392 A: LYS417, GLN409

RMSD=Root-mean-square deviation, CΑ RMSF=Calculatig root mean square fluctuation, LA=Linoleic acid

Table 3a.

Detail of selected 5 ligand complex from 1 ns molecular dynamics simulation study on the basis of physiological interacting residues (A: LYS417, GLN409, ARG408, B: PHE374)

Drug code	Drug name	Docking score	Interacting residues
	LA (normal)	−7.540	A: LYS417, GLN409, ARG408 B: PHE374
DB00650	Leucovorin	−11.804	A: LYS417, GLN409, ARG408 B: PHE374
DB11362	Selexipag	−11.463	A: LYS417, GLN409, ARG408 B: PHE374
DB00642	Pemetrexed	−11.091	A: LYS417, ARG408, GLN409 B: PHE374, phe377
DB00310	Chlorthalidone	−10.984	A: ARG408, GLN409, LYS417, THR415 B: PHE377, PHE374
DB06813	Pralatrexate	−10.758	A: LYS417, GLN409, ARG408 B: PHE392, PHE338, PHE377, PHE374

LA=Linoleic acid

Figure 4.

(a) Detail result of RMSD of selected 5 top ligands. (b) Detail result of RMSF of selected 5 ligands. RMSD = Root mean square deviation, RMSF = Root mean square fluctuation

Molecular dynamics simulation studies of four selected compounds (100 ns)

In 100 ns MD simulation studies, we can see that compared to LA, although the interacting amino acid residues [Table 3b] and RMSF [Figure 4b] of all the four agents were similar. However, much stable interactions could be seen in case of both selexipag and pralatrexate in terms of RMSD [Figure 4a]. The details of interaction of all the selected two ligands (selexipag and pralatrexate) are shown in Figure 5a and b.

Table 3b.

Detail result of the root-mean-square deviation and root mean square fluctuation of selected 5 complexes after 100 ns molecular dynamic simulation study

Compound name	RMSD	RMSF
LA	4.9±1.5Å	1.5±2.7Å
Leucovorin	5.74±6.1Å	4.0±6.5Å
Selexipag	3.0±0.3Å	1.3±0.9Å
Pemetrexed	4.3±1.3Å	1.4±2.5Å
Pralatrexate	3.1±0.4Å	1.4±1.0Å

RMSD=Root-mean-square deviation, RMSF=Root mean square fluctuation, LA=Linoleic acid

Figure 5.

Detail interaction structure of both selected ligands, Pralatrexate bound to the LA binding pocket in spike protein. Represent selexipag bound to the bound to the LA binding pocket in spike protein B chain. bound to the LA binding pocket in spike protein. The green surface structure and helix represent the A chain and gray surface structure represent the B chain. Pralatrexate is represented in purple and selexipag in Red. LA: Linoleic acid

Further, we also analyzed the individual chain A and chain B RMSD and RMSF result of the selected two ligands (selexipag and pralatrexate). Both ligands were compared with LA [Table 4]. Both ligands fit into the binding pocket site and have shown interaction with A chain and B chain which is.

Table 4.

Detail of the root-mean-square deviation and root mean square fluctuation result of selected 2 ligand complex with chain A and chain B

Compounds	A chain interaction RMSD			B chain interaction RMSD
	RMSD	RMSF	Interaction	RMSD	RMSF	Interaction
DB06813 (pralatrexate)	3.3±0.4Å	1.5±1.0Å	LYS417, GLN409, ARG408, ARG403	2.9±0.3Å	1.3±0.8Å	PHE374, ASN270, ALA372
DB11362 (selexipag)	2.9±0.2Å	1.2±0.9Å	LYS417, GLN409, ARG408, ASP405	2.9±0.4Å	1.3±0.9Å	PHE374, LEU368, PRO384, ASN388

RMSD=Root-mean-square deviation, RMSF=Root mean square fluctuation

Effect of drug on double mutant (E484Q and L452R) coronavirus disease-19

Phylogenetic analysis indicated that the predominant clade in circulation was an unique newly found lineage B.1.617, which had mutations at residue locations 452 and 484 with enhanced infectivity and was observed in globally circulating lineages during COVID-19 case spikes.[15] The formation of such variations via accumulation of convergent mutations must be researched further for their public health implications.[15]

The structural study of RBD mutations L452R and E484Q found in the furin cleavage region of spike protein may result in greater ACE 2 binding and a faster rate of S1-S2 cleavage, resulting in improved transmissibility.[15]

In our study, we modeled the mutant protein structure from wild protein (PDB: 6zb4) using Schrodinger. After preparation of the mutated protein, the mutant and the wild proteins were aligned using Schrodinger software tool. We found the alignment score was 0.038 and RMSD difference was 0.049 Angstrom [Figure 6a].

Figure 6.

(A) The Alignment of structure of wild and mutated protein of spike SARS-Cov-2. (a) the whole trimer spike protein structure alignment with wild (green) protein and mutant at E484Q and L452R protein (Red). (b) Is represent the aligned image with surface structure (here green is A chain and gray is B chain (B) Detailed of results of molecular dynamics studies with the mutated protein with linoleic acids and two selected hits (Selexipag and pralatrexate). SARS-Cov-2 = Severe acute respiratory syndrome coronavirus-2

After the alignment study, we intended to evaluate if there is any effect of these mutations which hinders the binding of LA and our hits Selexipag and Pralatrexate. Protein was prepared after incorporating double mutation (mutant E484Q and L452R) and was used for molecular docking with selected ligands (Pralatrexate, Selexipag). We compared the docking score between mutant and wild protein structure, found that there is a difference in docking result [Table 5].

Table 5.

Comparative result from molecular docking results of the wild type and the double mutant form with linoleic acid and selected hits (selexipag and pralatrexate)

Ligand name	Docking score
	Variant reported in PDB: 6ZB4 (unmutated form)	Double mutated variant
LA	−7.540	7.152
Selexipag	−11.463	−12.189
Pralatrexate	−10.758	−9.504

LA=Linoleic acid, PDB=Protein data bank

Molecular dynamic simulation study of mutant protein-ligand complexes

After docking of both selected ligands with the mutated protein, we preformed molecular dynamic simulation studies of the same using “Desmond” module of Schrodinger. Compared to LA, RMSD and RMSF of selexipag was higher, however, pralatrexate showed a consistent lower RMSD and RMSF value when compared to LA indicating better binding affinity of the same. Table 6 and Figure 6b illustrate the data.

Table 6.

Molecular dynamic simulation result of selected hits (selexipag and pralatrexate) and linoleic acid with double mutant form (double mutant E484Q and L452R)

Drug name	RMSD	RMSF	Interactions
LA	3.19±0.52	1.44±1.01	A: GLN409, ARG408, LYS417, THR415 B:
Selexipag (DB011362)	4.16±0.97Å	1.53±2.04Å	A: GLN409, ARG408, LYS417, THR415, ASP405 B: PHE374, TYR365
Pralatrexate (db006813)	3.08±0.39	1.33±0.93	A: GLN409, ARG408, LYS417, ARG403, THR415, GLU406 B: PHE374, SER371, LEU368, ALA372, ala363

LA=Linoleic acid, RMSD=Root-mean-square deviation, RMSF=Root mean square fluctuation

Discussion

In this study, we have virtually screened Food and Drug Administration-approved drugs from re-purposing point of view to find suitable binders which can bind and occupy the LA binding site and lock the spike protein in closed conformation. The LA bonding pocket is therefore a suitable target for the development of a small molecule inhibitor capable of locking the trimer structure in closed conformation and interfering with ACE-2 interaction. Similar strategy was applied to develop anti-viral against “rhinovirus,”[7,8] and this strategy was found to be fruitful in clinical trials.[9,10] Virtual screening yielded a total of 26 substances, all of which had a higher dock score than LA (−7.461 Kcal/Mol). At 1ns molecular dynamic simulation, we discovered that 20 ligands had better or similar RMSD and RMSF results than LA. However, only four ligands (leucovorin, selexipag, pemetrexed, and pralatrexate) of the 20 ligands had interaction residues (A: LYS417, GLN409, ARG408, B: PHE374) that were similar to LA. These four ligands were then further investigated utilizing 100 ns MD simulation, the RMSD and RMSF result were found favorable in selexipag (RMSD: 3.0 ± 0.3Å, RMSF: 1.3 ± 0.9Å) and pralatrexate (RMSD: 3.1 ± 0.4Å, RMSF: 1.4 ± 1.0Å). another side the structural study of RBD mutations L452R and E484Q found in the furin cleavage region of spike protein may result in greater ACE2 binding and a faster rate of S1-S2 cleavage, resulting in improved transmissibility.[15] In our study, we also modeled the mutant protein structure from wild protein (PDB: 6ZB4) using Schrodinger. We compared the docking score between mutant and wild protein structure. We performed molecular dynamic simulation of mutant protein ligand complexes. Result indicated better affinity of Pralatrexate ligand with mutant protein (double mutant E484Q and L452R).

We reviewed the in vitro evidence and safety of efficacy of the selected hits (selexipag and pralatrexate). There is a no data available for selexipag. In the in vitro study, it is confirmed that the pralatrexate (EC50 = 0.008 μM) and Azithromycin (EC50 = 9.453 μM) able to effectively inhibit replication process of SARS-CoV-2.[16] The in vitro potency of pralatrexate was far higher than azithromycin. The efficacy of pralatrexate is found to be even better than remdesivir in vitro.[17]

Conclusion

We identified two drugs (pralatrexate and selexipag) as potential binders to the LA binding pocket on SARS-CoV-2 spike protein (S1). In silico experiments, pralatrexate were able to bind to both A and B chain of wild and mutant protein similar to LA. In in vitro studies, pralatrexate is already reported to have inhibitory action on SARS-CoV-2 highlighting robustness of our study methodology. We do not have any in vitro or in vivo data regarding selexipag. Selexipag needs to be evaluated against COVID-19 in different platforms.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.

Graphical Figure 1

The spike protein can exist in two form Open (A) and Closed (B). Linoleic acid binding stabilizes the spike trimer in closed conformation by binding to two chains of spike protein through cationic interactions to oner chain and hydrophobic interactions to the other chain. The closed conformation of spike trimer shows less binding to the ACE2 (B). In this study, we have screened the FDA approved drugs as possible binder to the linoleic acid binding site which bind in a similar manner to the linoleic acid binding site and stabilizes the trimer in closed conformation (C)