Introduction
The ongoing COVID-19 pandemic has resulted in over 225 million confirmed cases and claimed more than 4.62 million lives worldwide (as of September 14, 2021). The SARS-CoV-2 virus that causes COVID-19 is closely related to SARS-CoV, the coronavirus responsible for the severe acute respiratory syndrome (SARS) outbreak in 2002. Although effective vaccines are now being implemented in healthcare practice, given the magnitude of the pandemic and the high death rate from infection, there is still a need to find effective treatment strategies for infected patients.
Coronavirus infection begins with binding of the viral Spike (S) protein to a cellular receptor, which is followed by conformational changes that lead to fusion of the virus with the cellular membrane.1 The Spike protein of SARS-CoV-2 consists of subunits S1 and S2.2 In the case of SARS-CoV-2 as well as SARS-CoV, the subunit S1 binds to the angiotensin-converting enzyme 2 (ACE2) receptor,3–5 while the subunit S2 forms a six-helical bundle via the two-heptad repeat domain that supports cell membrane fusion.2 The crystal structures of complexes between the viral S protein and ACE2 suggest that the two viruses have similar but not identical modes of binding,4,6–8 which is consistent with the overall 76% identity between the SARS-CoV-2 S protein and SARS-CoV S protein (50% identity within the receptor-binding domain).5 Binding of the protein subunit S1 to the ACE2 receptor is followed by priming of S protein by extracellular TMPRSS2, a serine protease.9–11 Priming changes the conformation of S protein, allowing the viral particle to fuse with the cellular membrane and, subsequently, the nucleic acid payload to enter the host cell. Priming by extracellular proteases is required for infectivity, whereby inhibition of the proteases neutralizes the virus.3
Two basic strategies are available to prevent entry of viral DNA into the cell. The first strategy prevents binding of the S protein to the ACE2 protein using monoclonal antibodies directed at the S-protein receptor-binding domain (RBD) to inhibit attachment of the virus. The second strategy prevents binding to ACE2 using compounds that modify its glycan component12 or compounds that interfere with or block the binding site of the RBD. The second strategy can be accomplished using compounds that inhibit other viral proteins or completely new compounds that fit the residue pattern of the ACE2–S1 interface.
Techniques to accomplish these two strategies have both advantages and disadvantages for patient treatment. For instance, the use of monoclonal antibodies depends on the availability of antibodies with a required specificity. While available monoclonal antibodies against SARS-CoV are not effective against SARS-CoV-2,13,14 their neutralizing monoclonal antibodies derived from recovered patients15,16 have been used successfully.17 In addition, the use of monoclonal antibodies is now a widely-used treatment strategy for patients with mild symptoms. For patients with more severe symptoms, alternative strategies are still needed.
Targeting host enzyme activities, such as the catalytic site of ACE2 or extracellular proteases, are problematic, as these are likely to have undesirable and potentially deleterious side effects. On the other hand, affinity of the viral S protein for ACE2 seems to be unrelated to its enzymatic activity.18 Crystal structures have shown that the RBD/ACE2 interface is distinct from the catalytic site,4,6 suggesting that it can possibly interfere with the interaction between S and ACE2 without inhibiting the activity of ACE2. Herein, we report a virtual screen for compounds that will interfere with S protein binding to ACE2 at the S/ACE2 interaction site.
Methods
Pharmacophore design and use
By analyzing the binding interface of the Spike protein with ACE2, we elucidated several possible interactions between ACE2 and this protein (PDB ID: 6VW1),19 providing a potential target for developing a pharmacophore model based on the pharmacophore centers corresponding to the S1 subunit interacting with ACE2 (Fig. 1). Using the Molecular Operating Environment (MOE) package (Chemical Computing Group ULC, Montreal, Quebec, Canada),20 we constructed a pharmacophore model of the binding interface including 16 features: 4 donors, 5 acceptors, 2 donors or/and acceptors, and 5 hydrophobic features (Fig. 2). We conducted a pharmacophore search with the whole and partial pharmacophore models using an internally created conformational database of FDA-approved drugs, containing 2,356 drugs and 600,000 conformations. The search was performed allowing for partial matches: seven and six of sixteen features. The search of seven of sixteen features (Search 1) identified 127 compounds with 64,449 conformations while the search of six of sixteen features (Search 2) identified 379 compounds with 806,486 conformations. Because Search 1 is absorbed by Search 2, we selected 152 compounds from Search 2 based on the greatest number of hydrogen-bonds and hydrophobic interactions in the best docked orientation. Then, we clustered the selected 152 compounds using the Compute/Fingerprint/Clusters application of the MOE Database Viewer, which calculates fingerprints for each molecule and a similarity matrix with Tanimoto similarity metric. From the similarity matrix, comparing to similarity threshold S, similar fingerprints are determined from which, using Tanimoto coefficient again and overlap threshold O, clusters were selected. A fingerprint GpiDAPH3 and similarity–overlap parameter SO = 45% were used to elucidate the common structure-functional features of the groups of compounds to enhance further drug development based on similarity of the selected drugs.
Docking of drug conformers using the supercomputer Comet
To dock the selected compounds, we used the crystal structure of the complex containing SARS-CoV-2 Spike protein and ACE2 (PDB ID: 6VW1). A binding interface was defined based on 14 residues of the protein’s S1 subunit with 29 contacts with ACE2. Figure 1 indicates that the S1–ACE2 interface could be divided into two interface sites: Site 1 includes ACE2 residues Gln24, Lys26, Thr27, Asp30, Lys31, His34, Glu35, Leu79, Met82, and Tyr83; and Site 2 contains residues Glu37, Asp38, Tyr41, Gln42, Gly352, Lys353, Gly354, Asp355, and Arg357.
To validate the specificity of the docked compounds, docking of random compounds was also conducted. A random number generator without repetition was used to obtain 100 random compounds and to select entries from the ZINC database that correspond to the random numbers obtained.
Conformers of each of the 152 selected compounds and 100 random compounds were generated using Compute/Conformations/Import application of MOE, which created 250 or more conformations per compound through a stochastic search. The application generated low-energy conformations of a collection of compounds and stored them in a molecular database, containing a total of 13,126 drug conformers from the selected compounds and 15,596 conformers from the randomly selected compounds.
OpenBabel [http://openbabel.org (accessed September 28, 2021)] was used to convert the pdb files into pdbqt format, which is necessary for using AutoDock Vina.21 The pdbqt files were prepared for both ligands and proteins. The binding sites of ACE2 were defined as a box that encompasses residues of the binding site. The resulting log files include poses generated by the AutoDock Vina, scores as an affinity estimation, energies of compound–protein interaction, and RMSD as geometric criteria of similarity.
AutoDock Vina (version 1.1.2) was run on the Comet supercomputer at the San Diego Supercomputer Center (SDSC) to explore docking of the 28,722 conformers (13,126 + 15,596), including both drugs and random compounds. This ultimately resulted in 258,498 total docking poses, as Autodock Vina outputs 9 poses per conformer. Details of the Comet computer used are presented in our previous publications.22,23
Final filtration of the binding poses was conducted based on the average distance of the docked compounds to the ACE2 binding residues. We calculated an upper boundary on the average distance between ligand atoms and binding residue side chain atoms. Binding poses with an average distance exceeding this upper boundary were filtered out, then poses with the lowest binding energies were selected from the remaining binding poses.
Docking workflow
Docking tasks were split up into a total of 34 separate jobs, most of which were run simultaneously (some cases were rerun with smaller splits to fit within wall clock limits on Comet) with 500–5,800 drug conformers docked in each job. The splits were made to keep the runtime between 24–48 h. During the simulations, the input dataset was extracted from aggregated zip files into local SSD space on the compute nodes. At the end of the simulation, the output files were aggregated into zip files and copied back to the home directory. This approach was successful in mitigating the IO loads on the main file system. All individual docking computations were conducted using eight cores (the parallelism is limited by the exhaustiveness parameter, which was set to 8 for the analysis), and scaling tests revealed an excellent parallel efficiency of 93.2%.
Molecular dynamics of drug-protein complexes using the supercomputer Expanse
Dynamics were performed in a similar manner to that described by Tang and co-authors.24 The structures and input files were prepared with the Visual Molecular Dynamics (VMD) package25 v. 1.9.4 and CHARMM-GUI Ligand Reader and Modeler, which is available on charmm-gui.org. Nanoscale Molecular Dynamics (NAMD) software26 v. 2.14, running on the SDSC Expanse cluster, was used to simulate 100 ns of Molecular Dynamics (MD), and VMD was used to extract data from the NAMD trajectory files.
Results
Pharmacophore-based docking results
Analysis of the interface of subunit S1 with ACE2 (PDB ID: 6VW1) revealed 29 possible interactions between the two proteins, including 17 residues of ACE2 and 14 residues of the S1 subunit (see Table 1). The binding interface in Figure 1 is clearly split into two interaction sites: Site 1 with ACE2 residues Gln24, Lys26, Thr27, Asp30, Lys31, His34, Glu35, Leu79, Met82, and Tyr83; and Site 2 with ACE2 residues Glu37, Asp38, Tyr41, Gln42, Gly352, Lys353, Gly354, Asp355, and Arg357.
Table 1Residues of interface between S1 subunit of S protein and ACE2 receptor
S1 subunit
| ACE2
| Distance, Å |
---|
Residue | Atom | Residue | Atom |
---|
Tyr449 | OH [O] | Asp38 | OD1 [O] | 2.99 |
Tyr449 | OH [O] | Gln42 | NE2 [N] | 3.73 |
Tyr453 | OH [O] | His34 | NE2 [N] | 3.84 |
Leu455 | CD2 [C] | Lys31 | CD [C] | 4.88 |
Leu455 | CD2 [C] | His34 | CE1 [C] | 4.28 |
Phe456 | CZ [C] | Asp30 | CB [C] | 4.37 |
Phe456 | CE2 [C] | Lys31 | CD [C] | 4.12 |
Tyr473 | CE1 {C] | Thr27 | CG2 [C] | 4.91 |
Phe486 | CD1 [C] | Leu79 | CD2 [C] | 3.54 |
Phe486 | CD1 [C] | Met82 | CE [C] | 3.48 |
Phe486 | CE1 {C] | Tyr83 | CE2 [C] | 3.88 |
Asn487 | ND2 [N] | Gln24 | OE1 [O] | 3.03 |
Asn487 | OD1 [O] | Tyr83 | OH [O] | 2.93 |
Tyr489 | CD1 [C] | Lys31 | CG [C] | 4.02 |
Gln493 | OE1 [O] | Lys31 | NZ [N+] | 2.92 |
Gln493 | CD [C] | His34 | CB [C] | 4.55 |
Gln493 | NE2 [N] | Glu35 | OE2 [O−] | 2.93 |
Gly496 | O [O] | Lys353 | NZ [N+] | 2.98 |
Thr500 | OG1 [O] | Tyr41 | OH [O] | 2.62 |
Thr500 | O [O] | Asp355 | OD2 [O−] | 3.59 |
Thr500 | OG1 [O] | Arg357 | NH1 [N] | 3.91 |
Thr500 | CB [C] | Arg357 | CZ [C] | 4.71 |
Asn501 | ND2 [N] | Tyr41 | OH [O] | 3.29 |
Asn501 | CG [C] | Lys353 | CD [C] | 4.22 |
Gly502 | N [N] | Lys353 | O [O} | 2.82 |
Gly502 | N [N] | Gly354 | O [O} | 3.70 |
Tyr505 | CD1 [C] | Lys353 | C [C] | 3.31 |
Tyr505 | CD1 [C] | Gly354 | CA [C] | 4.30 |
Tyr505 | OH [O] | Glu374 | OE1 [O] | 3.18 |
Because our goal is to inhibit binding of the S1 subunit to ACE2, we used the S1 binding domain (Fig. 2) for pharmacophore development. The following residues from both sites were used–Site1: Tyr449, Gly496, Thr500, Asn501, Gly502, and Tyr505; and Site 2: Tyr453, Leu455, Phe456, Tyr473, Phe486, Asn487, Tyr489, and Gln493.
The pharmacophore developed using MOE is based on both interaction sites and consists of 16 centers: 5 acceptors, 4 donors, 2 donor or/and acceptor, and 5 hydrophobics. The pharmacophore search was conducted using an FDA-approved drug database with 2,356 drugs using full and partial pharmacophores. The search with the partial pharmacophore having 6 of 16 features resulted in 379 compounds with 806,486 conformations. After analysis of the binding properties, 152 compounds were selected.
Then, we clustered the selected 152 compounds as described in the Methods section. We identified four clusters (A, B, C, and D) in our pharmacophore search of the FDA-approved drug database containing more than ten compounds (18, 18, 14, and 11 correspondingly); three clusters (E, F, and G) containing eight, six, and five compounds correspondingly; three clusters (H, I, and J) with four compounds; three clusters (K, L, and M) with three compounds; and eight two-compound clusters (N-R and T-V) and 35 not clustered single compounds (S). Compounds in clusters A-R and T-V are listed in Table 2, and the 35 single-cluster compounds are provided in Table 3.27
Table 2Drug-candidates clustered by fingerprint similarity–overlap alignment
Cluster |
---|
A | B | C | D | E | F | G |
Angiotensin II | Acarbose | Calcifediola | Cefmenoxime | Daunorubicin | Abemaciclib | Alatrofloxacina |
Anidulafunginb | Amikacin | Calcitriol | Cefonicid | Diosmin | Afatinib | Daclatasvirb |
Bleomycin | Deslanosidea | DMPC*a | Cefoperazone | Doxorubicin | Brigatinib | Ombitasvirb |
Carfilzomib | Dibekacina | Epoprostenola | Ceforanide | Epirubicin | Gilteritinib | Pibrentasvira |
Caspofungin | Framycetina | Fingolimod | Cefotetana | FAD** | Imatinib | Pralatrexate |
Desmopressin | Gentamicin | Ibutilide | Cefotiamb | Hesperidin | Nilotinib | |
Etelcalcetidea | Hyaluronana | Linoleic acid | Cefpiramide | Mithramycina | | |
Goserelin | Kanamycin | Lutein | Ceftibuten | Rutin | H | I |
Lanreotidea | Lactulose | Montelukast | Ceftobiprolea | | Flavin mononucleotidea | Betaxolol |
Lopinavir | Micronomicina | Paricalcitola | Ceftolozanea | | Regadenoson | Bisoprolol |
Nafarelin | Netilmicin | Pravastatin | Latamoxef | | Riboflavin | Esmolol |
Pentagastrina | Paromomycin | Retapamulin | | | Ticagrelor | Levobetaxolol |
Rifapentine | Pentosan polysulfatea | Thonzoniuma | | | | |
Ritonavir | Plazomicina | Vitamin E succinate | J | K | L | M |
Saquinavira | Ribostamycin | | Lymecyclinea | Ledipasvir | Arformoterol | Dapagliflozin |
Tacrolimus | Steviolbioside | N | Methacycline | Lusutrombopaga | Indacaterol | Empagliflozin |
Viomycina | Streptomycin | Glimepiride | Rolitetracyclinea | Velpatasvir | Protokylola | Ertugliflozin |
Xifaxana | Tobramycin | Glyburide | Tigecycline | | |
O | P | Q | R | T | U | V |
Doripenem | Valacyclovira | Irinotecan | Reserpininea | Mitoxantrone | Bosutinib | Florbetaben (18F)a |
Ertapenem | Valganciclovir | Simeprevir | Vilazodone | Pixantrone | Neratinib | Florbetapir (18F)a |
Table 3Compounds that are single in a cluster
Singles (S) |
---|
Novobiocin | Hydrocortamatea | Dasatinibb | Ponatinib | Bisoctrizolea | DPTA*a |
Calceina | Chlorohexidine | Oxiglutationea | Aliskiren | Curcumin | Betrixaban |
Pantethine | Mezlocillin | Panobinostat | Edoxabanb | Naldemedinea | Siponimod |
Bemotrizinola | Cromoglicic acid | Peramivira | Lenvatinibb | Tezacaftor | Pegvaliasea |
Indinavir | Balsalazide | Dabigatran etexilate | Nintedanib | Glasdegib | Netarsudila |
Cerivastatina | Ketoconazole | Raltegravir | Sonidegib | Sultamicillin | |
To validate the search results and select the best binding drugs, we docked the conformers from the set of 152 drugs selected from the pharmacophore-based search and from a set of 100 random compounds to the binding site of the ACE2 receptor (Protein Data Bank entry, 6VW1).
Multiple conformations molecular docking with Comet results
For more accurate docking, Conformational Import (a special conformer-generating application) was performed on the selected compounds from MOE, resulting in 13,126 conformations. Similarly, MOE Conformational Import was conducted for the 100 random compounds selected from the public ZINC database, resulting in 15,596 conformations.
The 28,722 total conformations were fed into Autodock Vina to find docking poses; 9 docking poses were found per conformation creating a total of 258,498 docking poses. The local SSD and job bundling approach used for Autodock Vina docking runs is described in the Docking Workflow subsection in the Methods section. The interaction energies of the selected drugs with each of interaction sites of ACE2 are shown in Table 4. Thirteen compounds with highest affinities to both sites are listed in Table 5.
Table 4Top 40 docked compounds sorted by their energies of interaction with S1 subunit of COVID-19 Spike protein binding sites on ACE2 receptor
SITE1 docking | DFE* | Cluster | SITE2 docking | DFE* | Cluster |
---|
Abemaciclib | −9.9 | F | Ledipasvir | −9.1 | K |
Flavin adenine dinucleotide | −9.9 | E | Raltegravir | −9.1 | S |
Nilotinibb | −9.9 | F | Angiotensin II | −9.0 | A |
Ponatinib | −9.9 | S | Ertapenem | −8.8 | O |
Saquinavira | −9.9 | A | Flavin adenine dinucleotide | −8.6 | E |
Siponimod | −9.9 | S | Velpatasvirb | −8.4 | K |
Vilazodoneb | −9.9 | R | Deslanosidea | −8.3 | B |
Alatrofloxacin | −9.8 | G | Arformoterol | −8.1 | L |
Diosmin | −9.8 | E | Indacaterol | −8.1 | L |
Irinotecan | −9.8 | Q | Pibrentasvira | −8.1 | G |
Naldemedinea | −9.8 | S | Sonidegib | −8.1 | S |
Sonidegib | −9.8 | S | Siponimod | −8.0 | S |
Sultamicillin | −9.8 | S | Desmopressin | −7.9 | A |
Glyburide | −9.6 | N | Irinotecan | −7.9 | Q |
Deslanosidea | −9.5 | B | Afatinib | −7.8 | F |
Hesperidin | −9.5 | E | Diosmin | −7.8 | E |
Brigatinib | −9.4 | F | Lanreotidea | −7.8 | A |
Chlorohexidine | −9.4 | S | Mithramycina | −7.8 | E |
Cromoglicic acid | −9.4 | S | Rifapentine | −7.8 | A |
Novobiocin | −9.4 | S | Vilazodoneb | −7.8 | R |
Table 5Top 13 docked compounds with highest affinities of interaction energies of COVID-19 Spike protein–ACE2 receptor ranked by average DFE* for both binding Sites 1 and 2
Thirteen top best compounds | Site 1 DFE* | Site 2 DFE* | Cluster |
---|
Flavin adenine dinucleotide | −9.9 | −8.6 | E |
Siponimod | −9.9 | −8.6 | S |
Sonidegib | −9.8 | −8.1 | S |
Deslanoside | −9.5 | −8.3 | B |
Irinotecan | −9.8 | −7.9 | Q |
Raltegravir | −8.6 | −9.1 | S |
Vilazodeneb | −9.9 | −7.8 | R |
Ertapenem | −8.9 | −8.8 | O |
Diosmin | −9.8 | −7.8 | E |
Ponatinib | −9.9 | −7.6 | S |
Nilotinib | −9.9 | −7.5 | F |
Alatrofloxacin | −9.8 | −7.5 | G |
Ledipasvir | −8.0 | −9.1 | K |
The Venn diagram created with Venny server28 shows that 27 of the top 39 compounds dock to both interaction sites (Fig. 3).
After analyzing the list of common compounds, one can see that ten best compounds with a docking energy less than −7.0 kcal/mol are bound to both sites with higher affinity, half of which belong to cluster A.
The summary of binding energies is shown in Table 6, which includes minimal (min), maximal (max), median (med), and mean (mean) values as well as lower q1 and upper q3 quartiles. Table 6 also indicates that all binding energy values for selected drugs are better for both sites. The values for both sites are almost identical. Figure 4 illustrates the box plots of docking free energies for the selected drugs and random compounds, which correspond to the first five values of Table 6. In the plots, whiskers represent the minimal and maximal values (q0 and q4), the box outlines the lower and upper quartiles (q1 and q3), and the line with × inside indicates the median (q2). Outliers are plotted as individual points. It is clear from Table 6 and Figure 4 that free binding energies are better for selected drugs than random compounds.
Table 6Summary of binding energies (kcal/mol) for selected drugs and random FDA-approved compounds
| min | q1 | med | q3 | max | mean |
---|
SITE 1 | | | | | | |
Selected | −9.9 | −8.7 | −8.2 | −7.5 | −5.9 | −8.16 |
Random | −9.8 | −8.1 | −7.4 | −6.4 | −3.4 | −7.17 |
SITE 2 | | | | | | |
Selected | −9.1 | −7.4 | −6.80 | −6.4 | −4.9 | −6.89 |
Random | −8.7 | −7.0 | −6.4 | −5.5 | −3.4 | −6.29 |
Note that the values of docking energy were only used to prioritize compounds for further experimental testing. In general, all of the pharmacophore-selected drugs could be valuable inhibitors.
Molecular dynamics results
Given the results from Autodock Vina, the top five drugs with the best binding energies were selected for validation with MD. Figure 5 shows the positions of the top binding drugs after 100-ns MD, which are within the initial binding sites of these drugs. Figure 6 displays the simulation results by tracking the distance between the geometric centers of each drug and an arbitrary residue in each binding site over time. The plots indicate the stability of the drug complexes with ACE2, in which most drugs have distances that remain mostly stable.
Discussion
Based on the crystal structure of the SARS-CoV-2 chimeric receptor binding domain (PDB ID: 6VW1),19 we created a pharmacophore model of the interface of this protein and ACE2 to screen a conformational database of FDA-approved drugs. Of the 379 identified drugs, 152 were selected and clusterized to determine the most promising candidates. Then, the selected drugs were used for multi-conformational docking to the interface region of ACE2. The drug list selected includes drugs tested for the treatment of viral infections. In particular, the following drugs that were found using the DrugVirus.info database29 have been reported in various studies for the treatment of several viruses (Fig. 7).
We created a conformational set of 100 random compounds from the ZINC database and performed multi-conformational docking of these compounds to the interface region of ACE2. After comparing these results with the docking results for the selected drugs, the range of significance in the values of binding energies was found to be moderately better for the selected drugs than the randomly selected drugs. Note that the pharmacophore-based search of the conformational databases by itself identified possible valuable therapeutic compounds, and the calculated docking energies may be useful for prioritizing these candidates for further investigation.
We further note that eight of the drugs selected by the pharmacophore-based database screening have already been tested in various experimental and clinical settings.
In cluster A (Table 2), ritonavir and lopinavir were shown to be effective in treatment of COVID-19 patients. Compared to treatment with pneumonia-associated adjuvant drugs alone, the combination therapy of lopinavir/ritonavir and adjuvant drugs has a more evident therapeutic effect in restoring normal health of the patients with no toxic side effects.30 Saquinavir demonstrated antiviral activity, as quantified by Taqman RT-PCR. Drug-induced effects on cells were monitored by quantifying LDH release and ATP levels.31
In cluster C (Table 2), three drugs have been tested by other scientists. In an ELISA assay, calcitriol was found to inhibit the receptor binding domain of SARS-CoV-2 S1 interaction with ACE2 at both low and high concentrations.32 Khan and colleagues reported montelukast to be associated with a reduction in clinical deterioration for COVID-19 confirmed patients.33 Jonsson and co-authors demonstrated that the duration of COVID-19 symptoms can be shortened by early initiation of nebulized isomerized linoleic acid in outpatient treatment.34 A recent pilot study indicates that calcitriol may decrease hospitalization and increase oxygenation.35
In cluster E (Table 2), the molecule similar to rutin–quercetin–was tested by Pan and colleagies.36 Interaction of quercetin with ACE2 was confirmed by a surface plasmon resonance assay, which showed that rutin (quercetin) interacts strongly with the main protease 3CLpro.36
In cluster G (Table 2), daclatasvir was shown to prolong the life expectancy of COVID-19 patients.37
Our approach uses a well-understood methodology to identify a set of existing, approved drugs that have therapeutic potential in treating individuals infected with SARS-CoV-2. The methodology identified a group of drugs that have been shown to improve the treatment outcomes of SARS-CoV-2 and related viral infections, compounds that are currently under active investigation, as well as a group of related drugs that have not yet been investigated. The calculated binding energies for the related drugs provide a basis on which to prioritize further experimental and clinical investigations on the potential impact of these compounds in treating COVID-19.
Relation to experimental results
Recently, Tsegay and colleagues reported the results of binding experiments with a similar set of pre-approved pharmaceutical compounds.27 This work provides an opportunity to evaluate the screening method reported in our presented work.
True Positives: Of the 39 compounds examined in our final screen, 3 drugs, namely nilotinib, velpatasvir, and vilazodone, were found to significantly inhibit Spike protein binding, with EC50 values of 4.2, 15, and 70 µM, respectively.27 Binding inhibition by the tyrosine kinase inhibitor nilotinib was also confirmed by Chtita and colleagues,38 owing to its strong binding to the ACE protease. Recent reports show that nilotinib inhibits viral replication in cell culture, with a low EC50 of 1.88 µM,39 and exhibits potential as a therapeutic agent. On the other hand, initial clinical trials indicate that vilazodone is ineffective for the treatment of COVID-19 in humans.40
False Positives: Of the top 39 candidates identified through our second screen, 26 were also examined by Tsegay and co-authors.27 Of the total 46 drugs that significantly inhibit ACE protease/Spike protein binding in vitro (EC50 < 10−4 M), only the three noted above (nilotinib, velpatasvir, and vilazodone) showed significant inhibition of Spike protein binding to the ACE2 protease. Further experiments are needed to evaluate the status of the remaining seven candidates.
False negatives: Of the 152 compounds that showed promise in our initial virtual screen, 7 exhibited inhibition of Spike protein binding in vitro (EC50 < 75 µM),27 which include: anidulafungin; cefotiam hexetil hydrochloride; daclatasvir, ombitasvir; dasatinib; edoxaban; and lenvatinib mesylate. However, these compounds were not among the top 39 in our secondary docking screening, indicating that the pharmacophore search alone is an effective tool for selection of potential drug candidates. The results presented here provide an opportunity to further tune our screening protocol.
Future directions
After further experimental testing of the suggested compounds for repurposing drugs, we plan to improve the pharmacophore model based on the residues and molecules that participate in the binding of compounds to both sides of ACE2. Subsequently, molecular docking would be conducted.
Conclusions
A set of FDA-approved drug compounds with the best parameters for interacting with the spike protein of SARS-CoV-2 is presented. The compounds were identified considering consistently several steps. Pharmacophore was developed on the basis of the ACE2 residues that participate in the interface with the SARS-CoV-2 spike protein. Pharmacophore-based docking, and unrestrained molecular docking of the compounds to the ACE2 were conducted. The stability of the selected compounds’ binding to ACE2 was confirmed by 100-ns molecular-dynamics simulation of the bound protein–drug complexes. We suggest that the selected drugs might bind to the interface of the ACE2/spike protein of SARS-CoV-2 and prevent spike binding to ACE2. We suggest further testing of the selected compounds for treatment of COVID-19 in future works.
Abbreviations
- 3D:
three-dimensional
- ACE2:
angiotensin-converting enzyme 2
- CMV:
cytomegalovirus
- CoV:
coronavirus
- COVID-19:
coronavirus disease 2019
- DB:
database
- DFE:
docking free energy
- DMPC:
dimyristoylphosphatidylcholine
- FAD:
flavin adenine dinucleotide
- GpiDAPH3:
Graph three-point pharmacophore (fingerprint)
- HCV:
hepatitis C virus
- HIV:
human immunodeficiency virus
- HMPV:
human metapneumovirus
- IO:
input/output (file system)
- MD:
molecular dynamics
- MERS:
Middle East respiratory syndrome
- MOE:
Molecular Operating Environment
- NAMD:
Nanoscale Molecular Dynamics
- RBD:
receptor-binding domain
- RVFV:
Rift Valley fever virus
- S:
Spike (protein)
- S1:
Spike protein 1
- S2:
Spike protein 2
- SARS:
severe acute respiratory syndrome
- SDSC:
San Diego Supercomputer Center
- SSD:
solid-state drive
- VMD:
Visual Molecular Dynamics
- ZIKV:
Zika virus
- ZINC:
“ZINC is not commercial” (database)
Declarations
Acknowledgement
We would like to thank the people of the San Diego Supercomputer Center for their friendly support.
Data sharing statement
No additional data are available.
Funding
The SDSC Comet supercomputer is supported by the NSF grant: ACI #1341698 Gateways to Discovery: Cyberinfrastructure for the Long Tail of Science. The SDSC Expanse Cluster by the NSF grant: #1928224 Computing without Boundaries: Cyberinfrastructure for the Long Tail of Science.
Conflict of interest
Prof. Igor F. Tsigelny has been an associate editor of Journal of Exploratory Research in Pharmacology since November 2021. Prof. Igor F. Tsigelny is a President of BiAna, and Prof. Valentina L. Kouznetsova is the CEO of BiAna. The authors have no other conflicts of interest related to this publication.
Authors’ contributions
IFT and VLK introduced the basic idea of the project. VLK conducted pharmacophore models development, databases searches, pharmacophore-based clustering and multiconformational alignment. MAM, IFT, and VLK conducted interpretation of the results. AZ and MT conducted computational docking on the Comet supercomputer. JPG conducted molecular dynamics simulations on Expanse cluster. IFT, VLK, MAM, and AZ wrote the article. VLK, MAM, and AZ contributed equally to this work.