Potential clinical drugs as covalent inhibitors of the priming proteases of the spike protein of SARS-CoV-2

Graphical abstract


Introduction
Since its outbreak in December 2019, the COVID-19 (coronavirus disease 2019) disease from the infection of the SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) virus has caused over 20,000,000 confirmed cases and over 700,000 deaths in over 180 countries/regions as of August 12, 2020 (https://coronavirus.jhu.edu). Although most COVID-19 patients can recover from the disease, some of the severe patients might suffer from long-term health issues including irreversible lung damages [1] and fertility compromise [2]. In addition to the devastating health crisis, the ongoing COVID-19 pandemic is also inflicting heavy losses on the global economy due to city lockdowns and glitches in supply chains (https://www.imf.org/en/Topics/imf-and-covid19). Facing the escalating risk from COVID-19, the whole world is intensively working on the discovery of prevention and treatment options for SARS-CoV-2 infection [3]. According to the statistics of the Milken Institute (https://milkeninstitute.org/ covid-19-tracker), over 200 vaccines and 300 treatment options for COVID-19 are under development worldwide as of August 12, 2020. Nonetheless, the reality is that neither de novo drug discovery nor vaccine development is a task that can be accomplished in several months or several years even if the huge monetary cost is guaranteed [4]. Therefore, to save the world from the current SARS-CoV-2 crisis, a more promising option would be looking for known drugs that could be used to treat COVID-19 patients.
To develop new uses for a drug beyond its original use, or drug repurposing (also known as drug repositioning, reprofiling, redirecting, or rediscovering), can significantly cut the time and money costs in drug development [5]. For example, drug repurposing has been successfully applied on the use of sildenafil in treating erectile dysfunction and the anti-cancer uses of thalidomide [6]. Since the outbreak of COVID-19, a number of clinic drugs have been repurposed for the treatment of the infected patients, such as lopinavir/ritonavir, chloroquine, ribavirin, and arbidol [7]. However, none of these tested drugs had confirmed efficacy yet. Among them, remdesivir, an investigational drug developed to treat the Middle East respiratory syndrome coronavirus (MERS-CoV), is one of the most promising. The latest clinical study showed that remdesivir was superior to placebo in shortening the time to recovery in adults hospitalized with Covid-19 [8], but efficacious drugs for treating COVID-19 patients are still urgently needed for the current SARS-CoV-2 pandemic.
The spike (S) protein of coronaviruses facilitates viral entry into cells by binding to receptors and driving the fusion of cell membranes [9]. Prior to be functional, the S protein needs to be cleaved and activated by the cellular proteases of the host cell [9]. According to a recent study [9], both the endosomal cysteine proteases cathepsin B/L (CatB/L) and the transmembrane protease serine type 2 (TMPRSS2) can prime the S protein of SARS-CoV-2. Meanwhile, TMPRSS2 can cleave carboxypeptidase angiotensinconverting enzyme 2 (ACE2), the host cellular receptor of the S protein, to augment viral infectivity [10]. Therefore, similar with SARS-CoV [11] and MERS-CoV [12], the simultaneous inhibition of CatB/L and TMPRSS2 might be necessary for completely blocking the cellular entry of SARS-CoV-2 [9].
Historically, the drug discovery practice mainly focuses on noncovalent drugs due to potential off-target effects and toxicity issues of irreversible covalent drugs [13]. However, recent years have witnessed the resurgence of covalent drugs because many people have realized that compared to non-covalent drugs, covalent drugs might have extra advantages including: (i) better biochemical efficiency since they are more competitive than non-covalent endogenous substrates and co-factors [14]; (ii) lower patient burden and less drug resistance due to lower and less frequent dosing [14]; (iii) improved target specificity upon careful structural designs targeting specific residues [15,16]. To help the discovery of covalent drugs, we previously established a ''steric-clashes alleviating receptor (SCAR)" strategy [17] for the in silico docking and screening of covalent drugs enlightened by in silico protein design [18]. More recently, we demonstrated that our SCARdock method is also useful in drug repurposing [19,20].

Preparation of the screening library
The structure files of the screening compounds were downloaded as mol2 files from the ZINC15 database (http://zinc15.docking.org). The 3D conformations were protonated at physiological pH, and biologically relevant tautomers were generated for each molecule as described in ZINC15 [25]. The ''in-trials" catalog (2019-04-22 version) was downloaded, which contained 5811 approved or investigational (clinically tested but not approved) drugs worldwide. MGLTools (version 1.5.6) was used to generate the PDBQT files from the mol2 files for docking.

Preparation of protein structures
The 3D structures of the indicated proteins were downloaded from the RCSB database (http://www.rcsb.org/). The homology model of TMPRSS2 was obtained from the SWISS-MODEL repository (https://swissmodel.expasy.org/repository/uniprot/O15393), which was built from the homologous protein TMPRSS1 (PDB ID: Table 1 The drugs repurposed as potential covalent inhibitors of the indicated target proteins using SCARdock. 5CE1). As mentioned previously [17], the protein structures were relaxed in Rosetta 3 [26] to eliminate possible structural conflicts. MGLTools (version 1.5.6) was used to generate the corresponding PDBQT files for docking.

Preparation of existing inhibitors
The small molecule inhibitors were extracted from the complex structures mentioned above. The structures were visually checked, and incorrect bonds/atoms were manually corrected in IQmol (version 2.14.0). MGLTools (version 1.5.6) was used to generate the corresponding PDBQT files for docking.

SCARdock screening and filtering
To prepare the SCAR proteins [17] used in SCARdock, Cys29 of CatB, Cys25 of CatL, and Ser441 of TMPRSS2 were computationally mutated to Gly to eliminate the sidechain clashes. The small molecules were docked into the corresponding pockets of the proteins with AutoDock Vina (version 1.1.2) [27]. The docking process did not consider the flexibility of the protein. The space coordinates of the S atom (for Cys) or the O atom (for Ser) in the wild-type protein were used for calculating the atom distances of the bonding atoms in the warhead groups. The distance cutoff between the bonding atoms and the S/O atom in the protein was set to 1.8 Å, indicating that the conformation with a distance above 1.8 Å is not accepted. Since the results (Table 1) had distances between 1.2 and 1.8 Å, no score punishment for steric conflict was applied for the cases in this study. For each ligand, top 10 poses were used for evaluation. The score cutoff was set as À7.5 for CatB, À7.0 for CatL, and À6.0 for TMPRSS2.

The overall screening results
Among the 5811 approved or investigational clinic drugs, we obtained 75 containing potential reactive groups (warheads) targeting cysteine and 9 containing potential warheads targeting serine (Fig. 1A). Based on previous studies [28,29], Cys29 of CatB and Cys25 of CatL are nucleophilic and can covalently bind to electrophilic ligands (Fig. 1B & 1C). Following the SCARdock protocol [17], these reactive residues were computationally mutated to glycine to generate the SCAR proteins for docking. As of TMPRSS2, the homology model showed that its Ser441 is comparable to the residue Ser353 of TMPRSS1 (Hepsin) (Fig. 1D & 1E). Since previous studies showed that TMPRSS1-Ser353 covalently bound to substrates [30], TMPRSS2-Ser441 was mutated to glycine for SCARdock.
From the RCSB database, we obtained four X-ray structures of human CatB and seventeen X-ray structures of human CatL (Supplementary Table 1). To identify the most suitable structures for SCARdock, we docked all the known inhibitors in these complex structures to each SCAR proteins of the respective PDB structure. After evaluating the recaptured X-ray poses of the docked inhibitors (Supplementary Table 1), the protein structure from 1CSB (PDB ID) was chosen for CatB, and the protein structure from 5MAE (PDB ID) was chosen for CatL. The human TMPRSS2 structure was obtained from the SWISS-MODEL repository since there were no X-ray structures available. These structures were then used for SCARdock screening, and after distance and score filtering, we identified five potential covalent inhibitors for CatB, three for CatL, and four for TMPRSS2 (Table 1). Overall, three cysteine covalent warheads and two serine covalent warheads were observed in the identified drugs (Fig. 1F).

Potential covalent inhibitors targeting CatB
The five potential covalent drugs identified for CatB are trapoxin B, neratinib (HKI-272), HKI-357, domatinostat (4SC-202) and (Z)dacomitinib ( Fig. 2 & Table 1). Trapoxin B contains an epoxide warhead. For this warhead, the nucleophilic attack might occur on the ring carbon next to the carbonyl carbon, and then a covalent bond can form between the sulfur atom of Cys29 and the bonding carbon of the oxirane moiety, accompanied by the ring opening and the formation of a hydroxyl group (Fig. 1F & Fig. 3A). Neratinib (HKI-272) and HKI-357 contain a nitrile warhead. For this warhead, a covalent thioimidate bond might form at the electrophilic nitrile carbon after the attack of the cysteine sulfur atom (Fig. 3B & 3C). Domatinostat (4SC-202) and (Z)-dacomitinib are amide-based ligands. A covalent bond might form between the b-carbon and the cysteine sulfur atom (Fig. 3D & 3E). In addition to these possible covalent bonds, the non-covalent scaffolds of these drug can also form suitable hydrogen bonds with CatB except (Z)dacomitinib (Fig. 4A-4F).

Potential covalent inhibitors targeting TMPRSS2
Four compounds, i.e. lodoxamide, aceneuramic acid, (S)boceprevir and (R)-boceprevir, were identified as potential TMPRSS2 covalent inhibitors (Fig. 5A). Lodoxamide, aceneuramic acid and aleplasinin possess an a-ketoacid group, whereas (S)and (R)-boceprevir have an a-ketoamide group. Both of these groups could be used as covalent warheads targeting serine (Fig. 1F). As shown in Fig. 5B, the warheads in these inhibitors are positioned well for the covalent bonding between the bonding atoms and the hydroxyl group of TMPRSS-Ser441. In addition, multiple hydrogen bonds were also observed between the inhibitors and the protein (Fig. 5C).

Discussion and conclusions
In his TED talk five years ago, Bill Gates warned that ''if anything kills over 10 million people in the next few decades, it's most likely to be a highly infectious virus rather than a war." The ongoing COVID-19 pandemic caused by the SARS-CoV-2 virus is undoubtedly reminding us his warning was not a hoax. The high infection rate and mortality ratio of COVID-19 are unexpected [31], and the SARS-CoV-2 virus has made 2020 a difficult year for a lot of people in the world. Although this virus would unlikely kill over 10 million people, it is still posing an unprecedent threat to both the health and the economy of the whole world due to the shortage of effective prevention and treatment therapeutics so far. In the regular research pipeline, the development of a new drug will cost over 10 years, but drug repurposing might significantly save the time [5], which might be the best chance for combating the current COVID-19 pandemic in time [32].
During the infection of coronaviruses, the spike (S) protein mediates host recognition and binding. However, the S protein needs to be cleaved and primed by the host cell before the virus can enter and hijack the host cell [9]. CatB, CatL, and TMPRSS2 of the human cell can prime the S protein of SARS-CoV-2 [9]. Therefore, as the other similar proteases [33], targeting these priming proteases might be an effective choice to disrupt the infection of this virus. Based on our recent work [17,19,20], we adopted the SCARdock protocol to repurpose clinic drugs as potential inhibitors of these priming proteases in this study. We identified several clinic drugs that might be useful as the covalent inhibitors of CatB, CatL, and TMPRSS2.
Neratinib (HKI-272), HKI-357 and dacomitinib are identified as potential covalent inhibitors for both CatB and CatL. All of these three drugs have an acrylamide group that is suitable for binding cysteine covalently, but neratinib (HKI-272) and HKI-357 also have an additional nitrile group, another potential covalent warhead targeting cysteine (Fig. 1F). More interestingly, all these drugs are covalent pan-HER (human epidermal growth factor receptor) kinase inhibitors targeting the nucleophilic cysteine in the ATP binding site of EGFR and/or HER2 [34]. This fact indicated that these three molecules are electrophilic, and it is highly possible that they can form covalent bonds with the indicated cysteine residues of CatB and CatL if the non-covalent binding affinity is high enough. Nontheless, we want to note that in our docking results, the nitrile groups, instead of the acrylamide groups, of neratinib (HKI-272) and HKI-357 were close to the cysteine residues of CatB/CatL in the top 10 poses (Fig. 3B, 3C, 3F & 3G). However, there were docked poses of these two drugs with suitably positioned acrylamide groups in less optimal poses (Supplementary Fig. 1). In addition, domatinostat (4SC-202) and trapoxin B were identified as potential CatB covalent inhibitors. Interestingly, both domatinostat (4SC-202) and trapoxins B are the inhibitors of histone deacetylases (HDACs), and they have been used for the treatment of advanced hematological malignancies [35,36]. Therefore, neratinib (HKI-272), HKI-357 and dacomitinib might be the most attractive drugs worth experimental validation for CatB/CatL among all of these hits.
As of the identified hits of TMPRSS2, the most attractive drug might be boceprevir, which is a first-generation inhibitor of hepatitis C virus non-structural protease 3 (HCV NS3) [37]. Based on the docking score and the SCAR enriching score (Table 1), (S)boceprevir might be better than (R)-boceprevir. Since boceprevir has anti-virus activity, it will be worth a try for SARS-CoV-2 infection. In addition, lodoxamide is a mast cell stabilizing compound with anti-inflammatory activity [38], aceneuramic acid is used for the treatment of hereditary inclusion body myopathy [39], and aleplasinin was developed to treat Alzheimer disease [40]. A note is that the screening of TMPRSS2 inhibitors was based on a homology model of human TMPRSS2, which might make the result not as reliable as the CatB/L cases. However, it would be worthwhile to test if these three drugs will have anti-SARS-CoV-2 activities.
A theoretical method to evaluate if a ligand could bind to two different proteins is to compare the similarity of the binding pockets of the target proteins [41]. Presumably, the protein-ligand interactions between a same ligand and different proteins should be similar, although the extent to which should vary on a caseby-case basis [41]. Therefore, to validate the binding potential of our computational hits with the target proteins, the ligand-protein interactions were generated (Fig. 6). For meaningful comparison, only the hits with experimental complex structures available were included. As shown in Fig. 6, compared to experimental complex structures, the docked ligands had similar hydrophobic interactions and hydrogen bonds with the target proteins. This analysis also showed that (S)-boceprevir formed more interactions with TMPRSS2 than (R)-boceprevir did, which, in agreement with their docking scores and SCAR enriching scores (Table 1), supported the conclusion that (S)-boceprevir might be better than (R)boceprevir for inhibiting TMPRSS2.
Taken together, using our SCARdock protocol, we identified nine drugs that might be repurposed as the covalent inhibitors of the priming proteases of the S protein of SARS-CoV-2. Among these nine drugs, neratinib (HKI-272), HKI-357, dacomitinib, and boceprevir might be highly potential with moderate side effects (Supplementary Table 2). We hope our work will provide the scientific community additional options for tackling the unprecedented COVID-19 threat.

Author contributions
S.L. conceived the idea and did the computational work. Z.Y.W., S.L., and Q.Z. did the computational work. S.L., Q.Z.L., and Z.Y.W. analyzed and interpreted the data. Q.Z.L., Z.Y.W., and S.L. wrote the manuscript. All authors reviewed and approved the submitted manuscript.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.