After primary screening, 238 compounds dose-dependently bound to the spike-ACE2 interaction interface (Supplementary Table 1 , available online), and subsequently they were subjected to the second-round confirmation. Ten of the 238 hit compounds have been reported to have antiviral activity (Supplementary Table 1 ), and thus other hits would not be further studied here. The screened 10 antiviral compounds were tenofovir, remdesivir, atazanavir, valacyclovir, valganciclovir, abacavir, doravirine, adefovir, lopinavir, and ritonavir (Supplementary Table 1 and Supplementary Fig. 1 ). As expected, several antiviral compounds have been found in other drug screening lists, such as remdesivir, lopinavir, and ritonavir[1,9,35].
To confirm the binding activity of those antiviral compounds, we further conducted an in-depth study regarding their docking scores and modes to the active site. The detailed views of docking were shown in Fig. 4 . It is obvious that these antiviral compounds bind to the spike-ACE2 interface through hydrogen bonding and hydrophobic interactions. It is interesting to find that these hit compounds exhibit different locations. For example, remdesivir and doravirine have stronger interaction with ACE2 than spike, indicating they are biased towards to ACE2. Adefovir, tenofovir, and valacyclovir tend to be close to spike, and are expected to interact strongly with spike than with ACE2.
Figure 4. Docking conformations and interacting residues of the hit antiviral compounds in the binding pocket locating at spike-ACE2 interface.
These anti-viral compounds have some obvious structural features. For tenofovir and remdesivir, which possess the highest docking scores, they have a phosphate ester and a fused nitrogen-containing heterocyclic ring attached by an amide group (‒NH2). Furthermore, more than one cyclic side-chains are often observed. Especially for lopinavir, it has the most cyclic structures among these hit compounds. In order to evaluate the anti-viral invasion efficacy of these compounds and further perform drug modification, we selected remdesivir, tenofovir, and lopinavir for subsequent MD simulations to investigate their binding strength and binding conformations to the targeting protein.
The binding energies of the studied compounds to spike-ACE2 are displayed in Fig. 5 . All the binding energies are negative, indicating the binding possibilities of remdesivir, tenofovir, and lopinavir to the interface between spike and ACE2 (Fig. 5A ). Their binding energies increase in the sequence of tenofovir (−42.9 kcal/mol) < remdesivir (−36.4 kcal/mol) < lopinavir (−32.9 kcal/mol), demonstrating the binding strength follows the sequence of tenofovir > remdesivir > lopinavir. In order to investigate the influence of the studied compounds on the interaction between spike and ACE2, we analyzed the change in spike-ACE2 interaction energies resulted from the ligand binding. As shown in Fig. 5B , the spike-ACE2 interaction strength is decreased upon the ligand binding. This result demonstrates that remdesivir, tenofovir, and lopinavir may hinder the coupling between spike protein and ACE2.
Meanwhile, these compounds exhibited distinct binding energies to spike and ACE2 (Fig. 5C ). remdesivir has higher binding strength to ACE2 (−19.2 kcal/mol) than spike protein (−10.8 kcal/mol), and tenofovir has higher binding strength to spike (‒21.8 kcal/mol) than ACE2 (‒10.2 kcal/mol, respectively). Different from them, lopinavir has comparable binding strength to spike (−18.7 kcal/mol) and ACE2 (−14.6 kcal/mol). Among these compounds, their binding capacities to ACE2 follows the sequences of remdesivir > lopinavir > tenofovir, while was reversed totally once interacted with spike.
To understand the contributions of different energetic terms to interaction of ligands with spike protein and ACE2, Fig. 5D and E give the electrostatic and van der Waals interaction energies, respectively. Electrostatic term plays an important role in the preference of remdesivir to ACE2, and the van der Waals interaction of remdesivir with spike and ACE2 are comparable to each other. In the case of tenofovir, both the electrostatic and van der Waals contribute more to the binding to spike than ACE2. Similarly, both van der Waals interaction and electrostatic term have contributed more to the binding of lopinavir to spike.
As demonstrated above, the studied potential drugs with various structures exhibited different binding strength with spike protein and ACE2. For further understanding of the influence of structures on the ligand-receptor interaction, the hydrogen bonding and hydrophobic interactions were investigated.
The hydrogen bond occupancy (the percent of snapshots containing hydrogen bonds in the MD trajectory) is an important parameter to characterize the difficulty of hydrogen bond formation, which can reflect the binding strength between drugs and proteins. Therefore, we calculated the hydrogen bond occupancies of the studied ligand-receptor systems (Table 1 and Fig. 6 ).
Number Acceptor Donor H Donor Distance (Å) Angle (°) Occupancy (%) Between remdesivir and spike-ACE2a 1 OE@Glu17@ACE2 H29@remdesivir N6@remdesivir 2.90 159.3 71.3 2 OG@Ser480@spike H30@ remdesivir N6@remdesivir 3.08 159.5 28.2 Between tenofovir and spike-ACE2b 1 OE1@Gln479@spike H29@tenoforvir N3@tenoforvir 2.97 161.6 49.1 2 O@Ser480@spike H29@tenoforvir N3@tenoforvir 2.92 154.8 34.2 3 ND1@Hie16@ACE2 H30@tenoforvir N3@tenoforvir 3.13 158.7 29.3 4 N2@tenoforvir H@Ser480@spike N@Ser480@spike 3.08 160.5 83.2 5 O9@tenoforvir H@Phe476@spike N@Phe476@spike 3.06 160.9 31.4 Between lopinavir and spike-ACE2c 1 OE@Glu17@ACE2 H33@lopinavir N4@lopinavir 2.98 156.9 87.9 2 O1@lopinavir H@Ser480@spike N@Ser480@spike 2.94 156.2 96.8 3 O4@lopinavir HZ@Lys13@ACE2 NZ@Lys13@ACE2 2.90 155.9 71.3 a,b, and cAtom indexes were shown in Supplementary Fig. 5 , 6, and 7 respectively, available online.
Table 1. Hydrogen bonds between the studied antiviral compounds and spike-ACE2 with occupancy time more than 20%
Figure 6. Hydrogen bonding and hydrophobic interaction diagrams of studied antiviral compounds with spike and ACE2.
There are two hydrogen bonds between remdesivir and spike-ACE2, both of which are involved in the hydrogen atom of amide group bonding to the triazine ring in remdesivir (H29@remdesivir and H30@remdesivir, Table 1 , Fig. 6A , and Supplementary Fig. 5 , available online). Its hydrogen bond formed with side-chain carboxyl anion (‒COO−) in Glu17 of ACE2 (OE@Glu17@ACE2) has hydrogen bond length (donor-acceptor) of 2.90 Å and bond angle (donor‒H···acceptor) of 159.3°. The other hydrogen bond is formed between amide hydrogen in remdesivir with the hydroxyl oxygen atom in Ser480 of spike protein. The hydrogen bond length and angle are 3.08 Å and 159.5°, respectively. The higher occupancy of hydrogen bond between remdesivir with ACE2 (71.3%) than that with spike (28.2%) indicates that remdesivir has a strong tendency to form hydrogen bonding interaction with ACE2 than spike protein.
There are much more hydrogen bonding interactions of tenofovir with spike than ACE2. Five hydrogen bonds are formed with Gln479, Ser480, and Phe476, while only one hydrogen bond forms with Hie16 of ACE2 (Table 1 ). The amide hydrogen atoms (H29 and H30) in the purine ring of tenofovir can form hydrogen bonds not only with Gln479 and Ser480 of spike protein, but also with Hie16 of ACE2 (Fig. 6C and Supplementary Fig. 6 , available online). The long chains of phosphate ester increase the flexibility of tenofovir, resulting in the competition of the hydrogen bonding with spike and ACE2. This can be seen from the fluctuated and lower occupancies (smaller than 50%). In comparison, the hydrogen bonds lying at rigid purine ring (N2@tenofovir···H@Ser480 of spike) are much more stable with a high occupancy of 83.2%.
In the case of hydrogen bonds between lopinavir and spike-ACE2, the hydrogen bonding interaction between groups in the six-membered heterocyclic ring and some residues of ACE2 are found to be dominating (Table 1 , Fig. 6E and Supplementary Fig. 7 , available online). For instance, lopinavir uses amide hydrogen in six-membered heterocyclic ring to form hydrogen bond with side-chain carboxyl anion (‒COO‒) in Glu17 of ACE2. The occupancy is 87.9% with an average bonding length of 2.98 Å and an angle of 156.9°. It also forms a hydrogen bond between oxygen in the six-membered heterocyclic ring and Lys13 of ACE2 with an occupancy of 71.3%. The average length and angle of this hydrogen bond is 2.90 Å and 155.9°, respectively. In addition, the hydrogen bonding of ligand formed with hydroxyl hydrogen atom in Ser480 of spike is also found with the highest occupancy of 96.8% among these three hydrogen bonds.
Comparing the hydrogen bonding acceptors and donors in Table 1 , one can find that both Glu17 of ACE2 and Ser480 of spike protein are key residues to form hydrogen bonding with the studied ligands. For the ligands, the amide group attaching to the rigid fused nitrogen-containing heterocyclic ring plays an essential role in the hydrogen bonds with receptors. In addition, we also noticed that a rigid ligand skeleton is critical to support a stable hydrogen bonding.
In addition to the hydrogen bonding, the ligand-receptor hydrophobic interactions were also investigated (Fig. 6 and Supplementary Fig. 8‒10 , available online). The results showed that there were hydrophobic interactions of methyl group of remdesivir with the alkyl side-chain of Lys13 of ACE2, as shown by the distance (4.81 Å) between hydrophobic centers in Supplementary Fig. 8 . The interaction between isobutyl group at the end of the remdesivir and benzene ring of Phe54 is stronger with a smaller distance of 4.58 Å (Supplementary Fig. 8 ). In addition, the π-π stacking interaction between the triazine ring of remdesivir and benzene ring of Phe476 of spike protein was also found. The average value of distance between the mass centers of two rings along the simulation time is about 4.72 Å (Supplementary Fig. 8 ).
In the case of lopinavir, three kinds of hydrophobic interactions with spike protein and ACE2 were detected. T-shaped π-π interactions was formed between 2,6-dimethylphenyl group of lopinavir and 4-hydroxylphenyl side-chain of Tyr435 of spike protein. The phenyl group of lopinavir also formed a π-π stacking with phenmethyl group of Phe476 of spike protein. The distances between the stacked fragments were 4.45 and 4.67 Å, respectively (Fig. 6F and Supplementary Fig. 10 ). Furthermore, alkyl-π interaction involving methyl of 2,6-dimethylphenyl group of lopinavir and imidazole ring of Hie16 of ACE2 is weaker, with a longer distance of 5.22 Å.
In comparison with remdesivir and lopinavir, the hydrophobic interaction of tenofovir with its receptor is weaker. Only T-shaped π-π interactions were found between its purine ring and aromatic rings of Tyr491 of spike protein as well as Hie16 of ACE2. As exhibited in Supplementary Fig. 9 , the distance between the stacked fragments were 4.21 and 4.44 Å, respectively.
In order to further improve the therapeutic efficacy, structural modification of remdesivir was conducted. Since the spike-ACE2 interaction involves a large interface (Supplementary Fig. 2B ), it is expected that increasing the area occupied by remdesivir on ACE2 will help improve its ability to resist viral invasion. Hence the compounds with binding abilities to ACE2 were considered. Because of the side-effect of ACE2 antagonists such as lowering blood pressure, diminazene, which is an agonist of ACE2 and has myocardial protective effect, was employed to investigate its binding capacity to the spike-ACE2 interface. As shown in Fig. 7 , the binding capacity (with docking score of 8.0891) of diminazene is slightly lower than that of remdesivir (with docking score of 8.9948 in Supplementary Fig. 1 ). Compound JL-01 was obtained through structural modification by Nitrogen-benzyl substituted methyl benzoate. Its binding capacity was slightly increased to 8.3138 (Fig. 7 ). Subsequently, JL-01 was employed to modify remdesivir through an ester bond, which ended producing an N-benzyl substituted diamidine derivative (JL-02 in Fig. 7 ). This modification of remdesivir showed a significant improvement in the binding capacity to 10.5261. Considering that remdesivir is the phosphate prodrug of GS-441524, we also linked GS-441524 to JL-01 in the way of ester bond. The binding ability of the obtained compound (JL-03, Fig. 7 ) was not significantly improved.
Identification of therapeutic drugs against COVID-19 through computational investigation on drug repurposing and structural modification
- Received Date: 2020-04-01
- Accepted Date: 2020-06-30
- Rev Recd Date: 2020-06-22
- Available Online: 2020-08-31
- Publish Date: 2020-11-27
Abstract: Global prevalence of coronavirus disease 2019 (COVID-19) calls for an urgent development of anti-viral regime. Compared with the development of new drugs, drug repurposing can significantly reduce the cost, time, and safety risks. Given the fact that coronavirus harnesses spike protein to invade host cells through angiotensin-converting enzyme 2 (ACE2), hence we see if any previous anti-virtual compounds can block spike-ACE2 interaction and inhibit the virus entry. The results of molecular docking and molecular dynamic simulations revealed that remdesivir exhibits better than expected anti-viral invasion potential against COVID-19 among the three types of compounds including remdesivir, tenofovir and lopinavir. In addition, a positive correlation between the surface area occupied by remdesivir and anti-viral invasion potential was also found. As such, the structure of remdesivir was modified by linking an N-benzyl substituted diamidine derivative to its hydroxyl group through an ester bond. It was found that this compound has a higher anti-viral invasion potential and greater specificity.
|Citation:||Yangfang Yun, Hengyi Song, Yin Ji, Da Huo, Feng Han, Fei Li, Nan Jiang. Identification of therapeutic drugs against COVID-19 through computational investigation on drug repurposing and structural modification[J]. The Journal of Biomedical Research, 2020, 34(6): 458-469. doi: 10.7555/JBR.34.20200044|