Molecular docking and dynamics simulation of FDA approved drugs with the main protease from 2019 novel coronavirus.

Design and development of an effective drug to combat the 2019 novel coronavirus remains a challenge. Therefore, it is of interest to study the binding features of 1615 FDA approved drugs with the recently known 2019-nCoV main protease structure having high sequence homology with that from SARS-CoV. We document the binding features of top 10 drugs with the target protein. We further report that Conivaptan and Azelastine are mainly involved in hydrophobic interactions with active site residues. Both drugs can maintain close proximity to the binding pocket of main protease during simulation. However, these data need further in vitro and in vivo evaluation to repurpose these two drugs against 2019-nCoV.

©Biomedical Informatics (2020) Background: Coronaviruses (CoVs) are positive-sense and single-stranded RNA viruses that can cause a number of respiratory diseases in mammals [1,2]. In the last two decades, two beta coronaviruses were responsible for the epidemic outbreak of atypical pneumonia cases. The first outbreak was caused by severe acute respiratory syndrome coronavirus (SARS-CoV) while the later epidemic was related to Middle East respiratory syndrome coronavirus (MERS-CoV) [3,4]. The global mortality rate for SARS-CoV was 9.6% and 34.4% as documented for MERS-CoV, death was the outcome of progressive acute respiratory distress syndrome or multiple organs failure [5,6]. Respiratory diseases caused by CoVs is believed to be the result of zoonotic transmission from specific animals [7]. Genomic sequencing studies suggest that the potential zoonotic reservoir of SARS-CoV is bats while camels are thought to be a probable vector or reservoir for MERS-CoV [8,9]. In December 2019, multiple cases of unknown viral pneumonia had been reported in Wuhan, China. By that time, most of the patients were living or working near a local seafood market [10]. Genomic sequencing of patients' specimens collected from lower respiratory tract had revealed the implication of unprecedented type of coronavirus, which was later named 2019 novel coronavirus (2019-nCoV) [11]. Analysis of next generation sequencing data has showed that 2019-nCoV is similar to SARS-CoV and MERS-CoV with a genomic identity of 79% and 50% respectively [12]. According to WHO situation report of February 27, 2020, the current epidemic outbreak of pneumonia caused by 2019-nCoV has a global distribution that affects 47 countries. The confirmed cases of coronavirus disease 2019 (COVID-19) are 82,294 and most of them are living in China, the reported number of deaths is 2,804 [13]. Pneumonia cases of 2019-nCoV had clinical features very similar to those reported with SARS-CoV and MERS-CoV. COVID-19 patients were presented with fever, dry cough, dyspnea and bilateral ground-glass opacity and consolidation of chest as seen in CT images. Unlike SARS-CoV and MERS-CoV, patients with 2019-nCoV had rarely showed signs of enteric disease like diarrhea. Also, few COVID-19 patients had presented with upper respiratory tract symptoms like sore throat and rhinorrhea [11,14,15]. The transmission mode of 2019-nCoV may be similar to SARS-CoV and MERS-CoV through airborne droplets and contact with infected persons [13]. The mean incubation period is estimated to be 5 days with 95% confidence interval range of 4-7 days [16]. Like SARS-CoV, 2019-nCoV may use angiotensin-converting enzyme 2 (ACE2) as a potential receptor for host cell infectivity [17]. Currently, no antiviral therapy or vaccine is available against coronavirus infection and attempts in this trend are accelerated to combat current epidemic outbreak [7,11]. On February 6, 2020, worldwide protein data bank has established COVID-19 coronavirus resources to facilitate target based drug design efforts against current global threat [18]. As listed in COVID-19 coronavirus resources, a crystallization team in Shanghai Technical University was able to resolve the structure of main protease for 2019-nCoV [19]. The three-dimensional representation for the monomer of 2019-nCoV main protease as deciphered from 6LU7 crystal can be seen in Figure 1. The main protease (M pro ), also less commonly known as 3CL protease, is believed to be essential for coronaviruses replication cycle through posttranslational processing of RNA replicase machinery [21]. The M pro is usually present as a homodimer, many coronaviruses share a significant homology regarding three-dimensional structure and amino acids sequence of this proteolytic enzyme. Therefore, M pro represents a conserved molecular target to design a broad spectrum anti-CoV drug [22]. It is of interest to study the binding features of 1615 FDA approved drugs with the recently known 2019-nCoV main protease structure having high sequence homology with that from SARS-CoV.

Methodology: Superimposition and alignment analysis:
The 6LU7 crystal for 2019-nCoV main protease was superimposed and then aligned with a reference crystal of SARS-CoV main protease with code 2AMQ [23]. Both MatchMaker and Match (Align) tools embedded in UCSF chimera had been utilized to superimpose and align chain A only for these two crystals [24]. MatchMaker tool can superimpose two proteins by using residues types and/ or secondary structure information. On the other hand, Match (Align) tool can align the superimposed proteins depending on residues spatial proximities.

Binding pockets prediction:
Although the binding site is well characterized for N3 inhibitor within many M pro crystals [22]. We have applied DoGSiteScorer online tool to predict and describe potential binding pockets within the recently released 6LU7 crystal [25]. DoGSiteScorer tool can detect potential binding pockets within a specific protein and then rank these pockets according to their size, surface area and druggability score.

Structure-based virtual screening:
We have used FDA approved drugs library as a resource for potential hits in our virtual screening. The FDA approved drugs library was downloaded from ZINC 15 database on February 13, 2020 [26]. By that time, the downloaded library did contain 1615 FDA approved drugs. These drugs were uploaded as 3D conformations into MCULE online drug discovery platform [27]. We have employed MCULE platform to screen these FDA approved drugs against 2019-nCoV main protease crystals. AutoDock Vina tool embedded in MCULE platform was applied to carry out an accelerated screening [28]. M pro crystal with code 6LU7 had been downloaded from COVID-19 coronavirus resources of the worldwide protein data bank [18,19]. Before screening, we have removed both water molecules and the bounded ligand (N3 inhibitor) from 6LU7 crystal by using UCSF chimera version 1.13.1 on local machine [20]. MCULE platform had automatically added both Gasteiger charges and polar hydrogen atoms to the uploaded crystal of M pro by using online AutoDock tools [29]. For screening protocol, we have used a binding site area of (22 * 22 * 22) Angstrom with coordinates of (X: -12, Y: 12.5, Z: 67). Otherwise, we have used the default parameters and options to screen these FDA approved drugs against the processed 6LU7 crystal. The screening output had listed and ranked these drugs according to their minimum binding energy to M pro crystal.

Molecular docking:
We have picked up the top ten hits from output of structure-based virtual screening process. The screening of these top hits against M pro crystal was repeated to confirm MCULE platform results. Here, we used UCSF chimera version 1.13.1 on local machine to run the screening process [20]. UCSF chimera software provides an easily accessible interface to process both ligands and targets, the software can easily add polar hydrogen atoms and Gasteiger charges and also ignore non-standard amino acid residues. The UCSF chimera also provide a flexible way to customize AutoDock     . The protocol for MD simulation did include an optimization for the hydrogen bonding network to enhance stability of the solute, and a pKa anticipation to fine-tune the protonation states of amino acid residues at pH value of 7.4 [35]. Sodium chloride ions were added with a concentration of 0.9%, with an excess of either Na or Cl to neutralize the complex. Following steepest descent and simulated annealing minimizations to remove possible clashes, the simulation was allowed to run for 7 nanoseconds using AMBER14 force field . After evaluation of the solute root-mean-square deviation (RMSD) as a function of simulation time, the first 7 nanoseconds were considered an equilibration time and excluded from any further analysis.

Results and Discussion:
By using 2AMQ crystal for SARS-CoV M pro as a reference, we were able to successfully align and superimpose 6LU7 crystal for 2019-nCoV main protease. According to (Figure 2), both 6LU7 and 2AMQ crystals are well superimposed. The only significant overlap deviation can be noticed in the C-terminus region for these two crystals. For simplicity of illustration, only chain A was considered for alignment and superimposition. These two superimposed crystals were then aligned according to residues spatial proximities. (Figure 3) reports sequence alignment output for 2019-nCoV M pro and SARS-CoV M pro . These two crystals are very well aligned with identity percentage of 93.79. These results fall in favor of previous studies indicating that the main protease enzyme is highly conserved in many coronavirus members, therefore this proteolytic enzyme may be used in structure-based screening studies to design a broad spectrum anti-CoV drug [22,23]. Then by employing DoGSiteScorer grid-based tool [25], we were able to detect several potential binding pockets within chain A of 2019-nCoV main protease crystal. Here, we have reported only the first three pockets in (Table 1), these binding pockets are ranked according to their size, surface area and druggability score. The first pocket is the preferred binding site for N3 inhibitor as reported in many crystallization studies of CoVs M pro [19, 22,23]. The location of pocket one was used as coordinates for our grid box setup and docking analysis. These predicted binding pockets within 6LU7 crystal are also well illustrated in (Figure 4). The chemical and clinical characteristics for the top ten hits with the least binding energy (docking score) are listed in (Table 2). These FDA approved drugs were ranked according to their minimum binding energy to 2019-nCoV main protease. According to this table, four of these drugs are anticancer agents. Due to safety concerns, they have been neglected from further analysis in this study. We also ignored Perampanel and Loxapine from our consideration due to their significant central nervous system effect. We have chosen both Conivaptan and Azelastine for further molecular docking and dynamics analysis due to their relative safety [41] and thereby they may be repurposed. Molecular dynamics (MD) analysis is computationally demanding process; therefore we have focused only on Conivaptan and Azelastine to save our limited computational power. The chemical structures for these ten drugs are shown in (Figure 5). Molecular docking images show the predominant involvement of different hydrophobic interactions between ligand (Conivaptan or Azelastine) and amino acid residues of 2019-nCoV main protease. The only exception is the formation of a hydrogen bond between glutamine 189 residue of M pro and Conivaptan. The two-dimensional image for docking analysis of Conivaptan and Azelastine against M pro crystal is shown in (Figure 6). It is evident that Conivaptan has more interactions with M pro active site than does the Azelastine. The ligand-protein complex with least binding energy pose was saved as PDB file for MD simulation. According to potential energy plot in (Figure 7), both Conivaptan-M pro complex and Azelastine-M pro complex were stable during equilibrium phase. Based on MD analysis report, Conivaptan has more interactions with M pro active site residues than does Azelastine during simulation period. Most of these interaction bonds are hydrophobic. These results can be seen in (Figure 8), which fall in agreement with findings of docking studies. Ligand proximity to active site residues of the target may indicate stronger binding. In this regard, both Conivaptan and Azelastine were able to maintain low ligand movement root-meansquare deviation (RMSD) that didn't exceed 3.25 Angstrom as seen in (Figure 9). Superposing the receptor on its reference structure generated this plot. Finally, the analysis of molecular dynamics has shown that Conivaptan may undergo more conformational changes as compared to Azelastine during simulation time. (Figure 10) shows ligand conformation root-mean-square deviation (RMSD) by superposing the ligand on its reference structure.

Conclusion:
We report the optimal binding features of Conivaptan and Azelastine with the main protease protein target from 2019-nCoV using molecular docking and simulation studies for further consideration.