Ontological and bioinformatic analysis of anti-coronavirus drugs and their implication for 2 drug repurposing against COVID-19 3 4

Title: 1 Ontological and bioinformatic analysis of anti-coronavirus drugs and their implication for 2 drug repurposing against COVID-19 3 4 Yingtong Liu, Wallace K.B. Chan, Zhigang Wang, Junguk Hur, Jiangan Xie, Hong 5 Yu, Yongqun He 6 7 1. Department of Computational Medicine and Bioinformatics, University of Michigan Medical 8 School, Ann Arbor, MI 48109, USA. 9 2. Department of Pharmacology, University of Michigan Medical School, Ann Arbor, MI 10 48109, USA. 11 3. Department of Biomedical Engineering, Institute of Basic Medical Sciences and School of 12 Basic Medicine, Peking Union Medical College and Chinese Academy of Medical Sciences, 13 Beijing 100005, China. 14 4. University of North Dakota School of Medicine and Health Sciences, Grand Forks, ND 15 58202, USA. 16 5. School of Bioinformatics, Chongqing University of Posts and Telecommunications, 17 Chongqing 400065, China. 18 6. Department of Respiratory and Critical Care Medicine, Guizhou Province People’s Hospital 19 and NHC Key Laboratory of Immunological Diseases, People’s Hospital of Guizhou 20 University, Guiyang, Guizhou 550002, China. 21 7. Department of Basic Medicine, Guizhou University Medical College, Guiyang, Guizhou 22 550025, China. 23 8. Unit for Laboratory Animal Medicine and Department of Microbiology and Immunology, 24 University of Michigan Medical School, Ann Arbor, MI 48109, USA. 25 26 * Corresponding author 27 28 Email addresses: 29 YL: yingtliu@umich.edu 30 WC: wallakin@umich.edu 31 Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 29 March 2020 doi:10.20944/preprints202003.0413.v1


1.
Background 70 The COVID-19 outbreak, caused by SARS-CoV-2, has become a pandemic and is now 71 spreading worldwide. As of May 29, 2020, over 5,701,000 cases, which includes over 3572,000 72 deaths, have been reported to WHO. In addition to COVID-19, two other coronavirus-induced 73 diseases, including Severe Acute Respiratory Syndrome (SARS) (Xu, 2013) and Middle East 74 Respiratory Syndrome (MERS) (Zaki et al., 2012), had also caused huge damages previously to 75 public health. SARS emerged in China in November 2002, which lasted for 8 months and 76 resulted in 8,098 confirmed human cases in 29 countries with 774 deaths (case-fatality rate: 77 9.6%) (Control and Prevention, 2003;Xu, 2013). Approximately 10 years later in June 2012, the 78 MERS-CoV, another highly pathogenic coronavirus, was isolated in Saudi Arabia from the 79 sputum of a male patient who died from acute pneumonia and renal failure (Zaki et al., 2012). frequently used for drug studies Liu et al., 2017). ChEBI is a database and 98 ontology of over 56,000 molecular entities with a focus on small chemical compounds. ChEBI 99 ontologically classifies these compounds based on different categories such as structural and 100 saved within a single SDF file and analyzed with ChemmineR (version 3.10) package (Cao et al., 160 2008). Tanimoto coefficients were calculated between all compounds for chemical similarity using 161 1,024-bit atom pair fingerprints. Hierarchical clustering with single linkage was performed on the 162 resulting distance matrix, while the heat map was generated with the gplots package in R. 163

PCA analysis of physicochemical properties of anti-coronavirus drugs 164
SDF files corresponding to small-molecule drugs with the mechanisms, 'Inhibit viral entry ' and 165 'Inhibit viral replication', were collected from ChEBI and imported into a MOE software database. 166 All compounds were cleaned as in section 2.4. The following descriptors for drug-like compounds 167 were calculated for each compound with MOE: 1) the number of hydrogen bond acceptors, 2) the 168 number of hydrogen bond donors, 3) molecular weight, 4) octanol-water partition coefficient 169 (slogP), and 5) topological polar surface area. PCA was performed on these five physicochemical 170 properties, and the first three principal components were taken for analysis. 171

Annotation of drug targets and drug-target network 172
The known targets of the identified drugs were collected from DrugBank (Wishart et al., 2018). 173 For any drug without a matching DrugBank record, multiple other online resources, including 174 ChEMBL and Wikipedia, to identify any known targets. The collected drug-target interactions as 175 well as protein-protein interactions among these targets, collected from the BioGRID interaction 176 database (Oughtred et al., 2019), were used to construct a drug-target interaction network and 177 visualized using Cytoscape v3.7.2 (Smoot et al., 2011). The collected drug targets were subjected 178 to a pathway enrichment analysis using our in-house functional enrichment tool richR 179 We manually collected and identified 151 chemicals drug compounds, each of which was tested 186 at least in one cell line in vitro, and some also tested in vivo, and found effective against human 187 coronavirus infections. These 151 chemical compounds include 110 active drug compounds that 188 -8 -can be mapped to at least one of the three ontologies: ChEBI, DrON, and NDF-RT (Table 1), 15  189   drugs that do not have any record in these ontologies, and 26 biological drugs (monoclonal or  190 polyclonal antibodies) specifically targeting on coronavirus proteins (e.g., S protein) ( Table 2). 191 Note that some listed compounds (e.g., chloroquine phosphate) are the salt forms of some drugs 192 (e.g., chloroquine), and they may be both included in our lists since they have independent 193 ontology IDs and were shown in the literature reports independently. They may also have the 194 same drug bank ID, so our Table 1 includes the drug bank IDs for easy checking. 195 Our collected drugs were organized by three known mechanisms (inhibition of viral 196 entry, inhibition of viral replication, and modulation of host immune response) or unknown 197 mechanism (Table 1) Often we do not know whether a drug functions at the entry or post-entry stage, but we know 208 that the drug was able to modulate the host immune responses (Table 1). Some drugs' anti-209 coronavirus mechanisms are unknown (Table 1). Several drug therapies utilized as a 210 combination of two or more drugs. Because it is difficult to evaluate individual drug properties, 211 combination drug therapy studies were not included from our annotations. 212 Our study found that 33 drugs inhibiting viral entry to host cells, 50 drugs that inhibit 213 viral replications inside host cells, and 12 drugs modulating host immune responses to 214 coronavirus infection (Table 1). At the viral entry-level, the interaction between coronavirus and 215 host-cell receptors can be blocked in multiple ways. For example, chlorpromazine has MoA as 216 dopamine antagonist and adrenergic-alpha antagonist, and it is active against SARS-CoV and 217 MERS-CoV (Zumla et al., 2016). Chlorpromazine also modulates clathrin-coated pits at plasma 218 membrane (Chu and Ng, 2004), which inhibits viral endosomal fusion. Antagonists of adrenergic 219 receptors have been shown as potent entry inhibitors of Ebola and Marburg viruses (Cheng et al., 220 2015 Nitazoxanide blocks maturation of the viral hemagglutinin and promotes the production of 231 interferons in virus-infected cells (Rossignol, 2014). By targeting IL-6 receptor, tocilizumab has 232 been found effective in treating moderate to severe rheumatoid arthritis, cytokine storm, and 233 SARS-CoV-2 infection (Xu et al., 2020b) (Table 1). 234 We have also collected 26 anti-coronavirus antibodies, including 17 monoclonal 235 antibodies and 2 polyclonal antibodies that target on MERS-CoV, and 7 monoclonal antibodies 236 that target on SARS-CoV ( Table 2). All of these antibodies were tested for their efficacy in 237 vitro, and over half of them were also tested in vivo. The SARS-CoV-specific antibodies target 238 for S spike protein, S1 receptor-binding domain (RBD), or S2 protein. For example, S230.15 and 239 m396 were found to compete with the SARS-CoV receptor ACE2 for binding to the RBD as a 240 mechanism of their neutralizing activity (Zhu et al., 2007). The MERS-CoV-specific antibodies 241 specifically target for S spike protein, S1 RBD, or human DPP4 receptor S2. Anti-DPP4 (CD26) 242 is another therapeutic option for fighting MERS-CoV. The anti-CD26 antibodies 2F9, 1F7 and 243 YS110 target the S1-DPP4 interaction from the host side, and prevent the MERS-CoV entry into 244 cells (Rabaan et al., 2017). 245

Drugs verified against SARS-CoV-2 infections in vitro or in vivo 246
Several drugs, including remdesivir, chloroquine phosphate, favipiravir, and tocilizumab 247 (Table 1), have been experimentally or clinically evaluated and found effective at various levels 248 against the SARS-CoV-2 infections in vitro or in vivo and have potential in treating COVID-19 249 (Table 1 and Table 2). 250 As a drug used to successfully treat the first case of COVID-19 patient in the USA 251 (Holshue et al., 2020), remdesivir has become an highly promising drug for treating  Remdesivir is a nucleoside analog which inhibits viral proliferation (Wang et  A recent study that a triple combination of Interferon beta-1b, Lopinavir-Ritonavir, and 300 Ribavirin in the treatment of mild to moderate COVID-19 patients was safe and superior to 301 lopinavir-ritonavir alone in alleviating symptoms and shortening the duration of viral shedding 302 (Hung et al., 2020). The adverse events for the early triple antiviral therapy included self-limited 303 nausea and diarrhea, which had no difference with the double lopinavir-ritonavir therapy. 304

Anti-coronavirus drugs are primarily inhibitors or antagonists, and many having the 305 anticancer role based on ontology analysis 306
These manually annotated 110 chemical drugs were mapped to and analyzed by three ontologies: 307 ChEBI, NDF-RT, and DrON. Among them, 99 have ChEBI IDs, 70 have NDF-RT IDs, and 60 308 have DrON IDs. We extracted these terms out from their ontologies using the tool Ontofox 309 (Xiang et al., 2010) and built relatively independent subsets of these ontologies. By analyzing 310 these subsets, we were able to identify more scientific insights. 311 ChEBI can be ontologically and systematically represent various anti-coronavirus drugs 312 and how such representation can be used for advanced analysis. After extracting these drugs and 313 their associations using Ontofox (Xiang et al., 2010), the hierarchical structure among different 314 anti-coronavirus drugs were clearly displayed . For example, we can find that chloroquine, 315 chlorpromazine, dasatinib, and three other drugs all belong to chlorobenzenes ( Figure 1A). 316 Chloroquine has many roles including the antimalarial role. Since ontology is computer-317 understandable, we can query the ontology using different approaches including SPARQL 318 (Group, 2013) and Description Logic (DL) query (Pan et al., 2019). As shown in Figure 1B were inhibitors of enzymes such as kinase or protease. The antagonists were specific for G 328 Protein-coupled receptors. There was only one agonist drug, methylprednisolone, a 329 glucocorticoid hormone receptor agonist that is used to suppress the immune system and 330 decrease inflammation. 331 Using a DL-query, 29 chemical entities were identified to have the antineoplastic role 332 ( Figure 2B). These 29 chemical entities are used as active ingredients compounds of 23 drugs. 333 To support viral program, viruses often hijack infected cells to have enhanced synthesis of 334 nucleic acids, proteins, and lipids and maintain boosted energy metabolism. The cell patterns are 335 also seen in cancer cells. Therefore, it is likely to use anticancer drugs to inhibit SARS-CoV-2 336 viral replication (Ciliberto et al., 2020). The 23 antineoplastic or anticancer drugs identified in 337 our study ( Figure 2B) have potential to be used for COVID-19 treatment. 338 Our identified drugs were also found to have many other roles, including various 339 antimicrobial (e.g., antimalarial, antiviral, and antibacterial) roles. Figure 1B provides an 340 example of the drugs having the antimalarial role. 341

Anticoronaviral drug patterns based on ontology classification and heatmap analyses 342
In this section, we introduce our discovery of different anticoronaviral drug patterns based on 343 ontology-based hierarchical classification, and two different types of heatmap analyses ( Figure  344 3-5). have high similarities and were mostly organic halide salt. As ChEBI annotates drugs from both 365 molecular structure and functional radical aspects at a higher level than chemical structure, this 366 similarity clusters may infer potential drug design. 367 the drugs and mechanism of action; glycyrrhizic acid was shown to modulate immune response 375 and inhibit viral replication, while everolimus only inhibits viral replication (Table 1). It should be 376 noted that the aforementioned glycosides will very likely not be useful as anti-coronaviral drugs 377 due to their high toxicity even in low doses (Cock, 2015). Though not grouped into the cluster, 378 azithromycin was also found to be chemically similar to the aforementioned compounds ( Figure  379 5, asterisk). Interestingly, azithromycin was recently reported to be effective against SARS-CoV-380 2 (Gautret et al., 2020), suggesting that macrolides could be a potential class of chemical 381 compounds that could be utilized against coronaviruses. Though they are traditionally used in the 382 clinical setting against retroviruses (Pulido et al., 2008), the antiretrovirals, ritonavir, lopinavir, 383 nelfinavir, and SG85 also were observed to have a mixed mechanisms of action ( Figure 5, green 384 bar); the former three act as inhibitors of viral entry, while SG85 inhibits viral replication (Table  385 1). Additionally, the structurally-similar anti-coronavirus drugs, toremifene, tamoxifen, and 386 triparanol ( Figure 5, purple bar), all operate as inhibitors of viral entry, as annotated in this study 387 (Table 1). Apart from this function, they have been shown conventionally to inhibit cholesterol 388 biosynthesis (Jordan et al., 2014). However, bazedoxifene was also in the aforementioned cluster, 389 though it has currently an unknown mechanism (Table 1); judging by its chemical similarity to the 390 others, it is likely it could also function as an inhibitor of viral entry. 391

Different physicochemical profiles of drugs inhibiting viral entry and inhibiting viral 392
replication 393 Five common physicochemical properties associated with druggability were examined for anti-394 coronaviral drugs responsible for: 1) inhibiting viral entry and 2) inhibiting viral replication; there 395 were 25 and 44 small-molecule compounds, respectively. Utilizing PCA analysis on these 396 compounds, no clear separation of mechanism of action was observed among the 69 examined 397 drugs ( Figure 6A). Interestingly, chloroquine was annotated to work as an inhibitor of both viral 398 entry and replication and was also observed to be located in a dense area of PCA space of mixed 399 efficacy ( Figure 6A, arrow), suggesting that it is possible that many of the drugs may possess a 400 dual mechanism of action. Overall, the compounds associated with inhibiting viral replication 401 possessed a greater number of hydrogen bond acceptors and donors and higher topological polar 402 surface area than those that inhibit viral entry after removal of five compounds (valinomycin, 403 telavancin, oritavancin, dalbavancin, and cyclosporine) greater than 1,000 Da ( Figure 6B); this 404 removal was performed because these compounds were significantly larger in size than the others, 405 which would have allowed for a more fair comparison between groups. Consequently, compounds 406 that inhibit viral replication also had higher topological polar surface areas. However, octanol-407 water partition coefficients did not seem to be different between the two groups. Our study found many drugs having similar roles or grouping to chloroquine and 500 hydroxychloroquine. Chloroquine and hydroxychloroquine are both antimalarial drugs. Our 501 collection includes three other antimalarial drugs, i.e., mefloquine, conessine, and lycorine 502 ( Figure 1B). These three antimalarial drugs may also be effective against COVOD-19. 503 Meanwhile, our study found that chloroquine, chlorpromazine and dasatinib are all 504 chlorobenzenes ( Figure 1A). Chlorpromazine and dasatinib were also found as hubs from our 505 drug-target interaction network ( Figure 6). Given that chloroquine and hydroxychloroquine are 506 likely effective anti-COVID-19 drugs (Wang et al., 2020a), it is worth testing the role of 507 chlorpromazine and dasatinib as effective anti-COVID-19 drug as well. 508 For better therapeutic effect, it is common to use a combination therapy with two or more 509 synergistically acting drugs. The combined usage of two or more drugs may achieve the three 510 main biological processes, including prohibit viral entry, viral replication and induce host 511 immune response (Table 1). Those combinational choices that target these different aspects of 512 the viral lifecycle would be preferred choices. Our identification of the 110 drugs provides 513 choices of drugs for targeting these three areas (Table 1 and (Table 1). Anisomycin was introduced earlier in the 538 article. Teicoplanin is an glycopeptide antibiotic used to prevent or treat serious infections 539 caused by Gram-positive bacteria such as methicillin-resistant Staphylococcus aureus and 540 Enterococcus faecalis. Teicoplanin inhibits bacterial cell wall synthesis. Teicoplanin was also 541 found to potently prevents the entry of Ebola envelope pseudotyped viruses into the cytoplasm, 542 and was able to block the entry of MERS-CoV and SARS-CoV envelope pseudotyped viruses as 543 well (Zhou et al., 2016). Valinomycin is a K + ionophore that can change the membrane potential 544 by conducting ions directly through membrane. It can treat B. gibsoni, a species of bacterial 545 infection, by changing bacteria's cellular cation concentration in vitro (Yamasaki et al., 2009). In 546 a study, TZMbl cells were treated with valinomycin before HIV infection, and valinomycin at 547 10 nM concentration impeded HIV entry to TZMbl cells by nearly 50% (Dubey et al., 2019). The 548 study also showed that valinomycin inhibits HIV entry by increasing the membrane 549 depolarization, which indicates membrane polarization is crucial for HIV entry. Therefore, 550 valinomycin may use the same mechanism against virus and bacteria. However, valinomycin is 551 classified as an extremely hazardous substance in the United States. 552 It is likely that the antibacterial role of the antibiotics may play a critical role in COVID-553 19 treatment. Microbiomes exist in many parts of human body such as the gut and lung (He et 554 al., 2020). The lung is now known to hold a large amount of bacteria (Fabbrizzi et al., 2019). 555 Bacterial microbiomes play an important role in human metabolism and immune response to 556 various pathogens. However, in a COVID-19 patient, the normally healthy bacteria inside human 557 lung and other parts of human body may become harmful and cause infections. Therefore, the 558 usage of an antibiotics may support the treatment by killing these previously good but now bad 559 bacteria in the infected lung. Such a hypothesis may still need experimental verification. 560 A recent study demonstrates that a network-based quantification and analysis of the host-561 coronavirus interactions and drug targets would help identification of candidate repurposable 562 drugs (Zhou et al., 2020). Their network-based drug repurposing method is aligned with our 563 ontology-based and bioinformatics-based study. Our drug-target network provides a backbone 564 for further network analysis and our ontology approaches provide more logically defined 565 relations between drugs and other types of entities (such as MoA, roles, and diseases). Our 566 ontology-based bioinformatics strategy can enhance the network standardization and computer 567 interpretation in a logical, interoperable, and consistent way, leading to improved prediction of 568 drugs for COVID-19. 569 As an important part of the network study of interactions among host, host cells, 570 coronavirus, and drugs, we will need to systematically understand the SARS-CoV-2-human 571 protein-protein interactions (PPIs). Recently a bioRxiv paper introduced a SARS-CoV-2-human 572 PPI map that reveals drug targets and potential drug-repurposing (Gordon et al., 2020). This Foundry library ontology (http://obofoundry.org/ontology/cido.html). The drug information 600 introduced in this article will be included in the CIDO. CIDO will also represent the fundamental 601 host-coronavirus molecular and cellular interaction networks and how the drugs can interact with 602 such interaction networks, and such computer-interpretable mechanism presentation can be used 603 to support different applications. We also welcome more participation from the community to 604 support its deep development and applications.