In total, we collected DNA sequences and protein sequences for 15 HCoVs, including three most recent 2019-nCoV/SARS-CoV-2 genomes, from the NCBI GenBank database (28 January 2020, Supplementary Table S1). Whole-genome alignment and protein sequence identity calculation were performed by Multiple Sequence Alignment in EMBL-EBI database (https://www.ebi.ac.uk/) with default parameters. The neighbor joining (NJ) tree was computed from the pairwise phylogenetic distance matrix using MEGA X with 1000 bootstrap replicates. The protein alignment and phylogenetic tree of HCoVs were constructed by MEGA X.

We collected HCoV–host protein interactions from various literatures based on our sizeable efforts. The HCoV-associated host proteins of several HCoVs, including SARS-CoV, MERS-CoV, IBV, MHV, HCoV- 229E, and HCoV-NL63 were pooled. These proteins were either the direct targets of HCoV proteins or were involved in critical pathways of HCoV infection identified by multiple experimental sources, including highthroughput yeast-two-hybrid (Y2H) systems, viral protein pull-down assay, and in vitro co-immunoprecipitation and RNA knock down experiment. In total, the virus–host interaction network included 6 HCoVs with 119 host proteins.

Next, we performed Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) enrichment analyses to evaluate the biological relevance and functional pathways of the HCoV-associated proteins. All functional analyses were performed using Enrichr.

Here, we collected drug–target interaction information from the DrugBank database (v4.3), Therapeutic Target Database (TTD), PharmGKB database, ChEMBL (v20), BindingDB, and IUPHAR/BPS Guide to PHARMACOLOGY. The chemical structure of each drug with SMILES format was extracted from Drug- Bank. Here, drug–target interactions meeting the following three criteria were used: (i) binding affinities, including Ki, Kd, IC50, or EC50 each ≤10 μM; (ii) the target was marked as “reviewed” in the UniProt database; and (iii) the human target was represented by a unique Uni- Prot accession number. The details for building the experimentally validated drug–target network are provided in our recent studies.

To build a comprehensive list of human PPIs, we assembled data from a total of 18 bioinformatics and systems biology databases with five types of experimental evidence: (i) binary PPIs tested by high-throughput yeasttwo- hybrid (Y2H) systems; (ii) binary, physical PPIs from protein 3D structures; (iii) kinase-substrate interactions by literature-derived low-throughput or high-throughput experiments; (iv) signaling network by literature-derived low-throughput experiments; and (v) literature-curated PPIs identified by affinity purification followed by mass spectrometry (AP-MS), Y2H, or by literature-derived low-throughput experiments. All inferred data, including evolutionary analysis, gene expression data, and metabolic associations, were excluded. The genes were mapped to their Entrez ID based on the NCBI database as well as their official gene symbols based on GeneCards (https://www.genecards.org/). In total, the resulting human protein–protein interactome used in this study includes. 351,444 unique PPIs (edges or links) connecting 17,706 proteins (nodes), representing a 50% increase in the number of the PPIs we have used previously. Detailed descriptions for building the human protein–protein interactome are provided in our previous studies.

To date, seven pathogenic HCoVs (Fig. 2a, b) have been found: (i) 2019-nCoV/SARS-CoV-2, SARS-CoV, MERS-CoV, HCoV-OC43, and HCoV-HKU1 are β genera, and (ii) HCoV-NL63 and HCoV-229E are α genera. We performed the phylogenetic analyses using the whole genome sequence data from 15 HCoVs to inspect the evolutionary relationship of 2019-nCoV/SARS-CoV-2 with other HCoVs. We found that the whole genomes of 2019-nCoV/SARS-CoV-2 had ~99.99% nucleotide sequence identity across three diagnosed patients (Supplementary Table S1). The 2019-nCoV/SARS-CoV- 2 shares the highest nucleotide sequence identity (79.70%) with SARS-CoV among the six other known pathogenic HCoVs, revealing conserved evolutionary relationship between 2019-nCoV/SARS-CoV-2 and SARS-CoV (Fig. 2a). HCoVs have five major protein regions for virus structure assembly and viral replications, including replicase complex (ORF1ab), spike (S), envelope (E), membrane (M), and nucleocapsid (N) proteins (Fig. 2b). The ORF1ab gene encodes the non-structural proteins (nsp) of viral RNA synthesis complex through proteolytic processing.

The nsp12 is a viral RNA-dependent RNA polymerase, together with co-factors nsp7 and nsp8 possessing high polymerase activity. From the protein 3D structure view of SARS-CoV nsp12, it contains a larger N-terminal extension (which binds to nsp7 and nsp8) and polymerase domain (Fig. 2c). The spike is a transmembrane glycoprotein that plays a pivotal role in mediating viral infection through binding the host receptor. Figure 2d shows the 3D structure of the spike protein bound with the host receptor angiotensin converting enznyme2 (ACE2) in SARS-CoV (PDB ID: 6ACK). A recent study showed that 2019-nCoV/SARS-CoV-2 is able to utilize ACE2 as an entry receptor in ACE2-expressing cells, suggesting potential drug targets for therapeutic development. Furthermore, cryo-EM structure of the spike and biophysical assays reveal that the 2019-nCoV/SARS-CoV- 2 spike binds ACE2 with higher affinity than SARS-CoV. In addition, the nucleocapsid is also an important subunit for packaging the viral genome through protein oligomerization, and the single nucleocapsid structure is shown in Fig. 2e.

Protein sequence alignment analyses indicated that the 2019-nCoV/SARS-CoV-2 was most evolutionarily conserved with SARS-CoV (Supplementary Table S2). Specifically, the envelope and nucleocapsid proteins of 2019-nCoV/SARS-CoV-2 are two evolutionarily con- served regions, with sequence identities of 96% and 89.6%, respectively, compared to SARS-CoV (Supplementary Table S2). However, the spike protein exhibited the lowest sequence conservation (sequence identity of 77%) between 2019-nCoV/SARS-CoV-2 and SARS-CoV. Meanwhile, the spike protein of 2019-nCoV/SARS-CoV-2 only has 31.9% sequence identity compared to MERS- CoV.

To depict the HCoV-host interactome network, we assembled the CoV-associated host proteins from four known HCoVs (SARS-CoV, MERS-CoV, HCoV-229E, and HCoV-NL63), one mouse MHV, and one avian IBV (N protein) (Supplementary Table S3). In total, we obtained 119 host proteins associated with CoVs with various experimental evidence. Specifically, these host proteins are either the direct targets of HCoV proteins or are involved in crucial pathways of HCoV infection. The HCoV–host interactome network is shown in Fig. 3a. We identified several hub proteins including JUN, XPO1, NPM1, and HNRNPA1, with the highest number of connections within the 119 proteins. KEGG pathway enrichment analysis revealed multiple significant biological pathways (adjusted P value < 0.05), including measles, RNA transport, NF-kappa B signaling, Epstein-Barr virus infection, and influenza (Fig. 3b). Gene ontology (GO) biological process enrichment analysis further confirmed multiple viral infection-related processes (adjusted P value < 0.001), including viral life cycle, modulation by virus of host morphology or physiology, viral process, positive regulation of viral life cycle, transport of virus, and virion attachment to host cell (Fig. 3c). We then mapped the known drug–target network (see Materials and methods) into the HCoV-host interactome to search for druggable, cellular targets. We found that 47 human proteins (39%, blue nodes in Fig. 3a) can be targeted by at least one approved drug or experimental drug under clinical trials. For example, GSK3B, DPP4, SMAD3, PARP1, and IKBKB are the most targetable proteins. The high druggability of HCoV–host interactome motivates us to develop a drug repurposing strategy by specifically targeting cellular proteins associated with HCoVs for potential treatment of 2019-nCoV/SARS-CoV-2.

The basis for the proposed network-based drug repurposing methodologies rests on the notions that the proteins that associate with and functionally govern viral infection are localized in the corresponding subnetwork (Fig. 1a) within the comprehensive human interactome network. For a drug with multiple targets to be effective against an HCoV, its target proteins should be within or in the immediate vicinity of the corresponding subnetwork in the human protein–protein interactome (Fig. 1), as we demonstrated in multiple diseases using this network-based strategy. We used a state-of-theart network proximity measure to quantify the relationship between HCoV-specific subnetwork (Fig. 3a) and drug targets in the human interactome. We constructed a drug–target network by assembling target information for more than 2000 FDA-approved or experimental drugs (see Materials and methods). To improve the quality and completeness of the human protein interactome network, we integrated PPIs with five types of experimental data: (1) binary PPIs from 3D protein structures; (2) binary PPIs from unbiased high-throughput yeast-two-hybrid assays; (3) experimentally identified kinase-substrate interactions; (4) signaling networks derived from experimental data; and (5) literature-derived PPIs with various experimental evidence (see Materials and methods). We used a Z-score (Z) measure and permutation test to reduce the study bias in network proximity analyses (including hub nodes in the human interactome network by literature-derived PPI data bias) as described in our recent studies. In total, we computationally identified 135 drugs that were associated (Z < −1.5 and P < 0.05, permutation test) with the HCoV-host interactome (Fig. 4a, Supplementary Tables S4 and 5). To validate bias of the pooled cellular proteins from six CoVs, we further calculated the network proximities of all the drugs for four CoVs with a large number of know host proteins, including SARS-CoV, MERS-CoV, IBV, and MHV, separately. We found that the Z-scores showed consistency among the pooled 119 HCoV-associated proteins and other four individual CoVs (Fig. 4b). The Pearson correlation coefficients of the proximities of all the drugs for the pooled HCoV are 0.926 vs. SARS-CoV (P < 0.001, t distribution), 0.503 vs. MERSCoV (P > 0.001), 0.694 vs. IBV (P < 0.001), and 0.829 vs. MHV (P < 0.001). These network proximity analyses offer putative repurposable candidates for potential prevention and treatment of HCoVs.

To further validate the 135 repurposable drugs against HCoVs, we first performed gene set enrichment analysis (GSEA) using transcriptome data of MERS-CoV and SARS-CoV infected host cells (see Methods). These transcriptome data were used as gene signatures for HCoVs. Additionally, we downloaded the gene expression data of drug-treated human cell lines from the Connectivity Map (CMAP) database to obtain drug–gene signatures. We calculated a GSEA score (see Methods) for each drug and used this score as an indication of bioinformatics validation of the 135 drugs. Specifically, an enrichment score (ES) was calculated for each HCoV data set, and ES > 0 and P < 0.05 (permutation test) was used as cut-off for a significant association of gene signatures between a drug and a specific HCoV data set. The GSEA score, ranging from 0 to 3, is the number of data sets that met these criteria for a specific drug. Mesalazine (an approved drug for inflammatory bowel disease), sirolimus (an approved immunosuppressive drug), and equilin (an approved agonist of the estrogen receptor for menopausal symptoms) achieved the highest GSEA scores of 3, followed by paroxetine and melatonin with GSEA scores of 2. We next selected 16 high-confidence repurposable drugs (Fig. 5a and Table 1) against HCoVs using subject matter expertise based on a combination of factors: (i) strength of the network-predicted associations (a smaller network proximity score in Supplementary Table S4); (ii) validation by GSEA analyses; (iii) literature-reported antiviral evidence, and (iv) fewer clinically reported side effects. Specifically, we showcased several selected repurposable drugs with literature-reported antiviral evidence as below.

An overexpression of estrogen receptor has been shown to play a crucial role in inhibiting viral replication. Selective estrogen receptor modulators (SERMs) have been reported to play a broader role in inhibiting viral replication through the non-classical pathways associated with estrogen receptor. SERMs interfere at the post viral entry step and affect the triggering of fusion, as the SERMs’ antiviral activity still can be observed in the absence of detectable estrogen receptor expression. Toremifene (Z = –3.23, Fig. 5a), the first generation of nonsteroidal SERM, exhibits potential effects in blocking various viral infections, including MERS-CoV, SARSCoV, and Ebola virus in established cell lines. Compared to the classical ESR1-related antiviral pathway, toremifene prevents fusion between the viral and endosomal membrane by interacting with and destabilizing the virus membrane glycoprotein, and eventually inhibiting viral replication. As shown in Fig. 5b, toremifene potentially affects several key host proteins associated with HCoV, such as RPL19, HNRNPA1, NPM1, EIF3I, EIF3F, and EIF3E. Equilin (Z = –2.52 and GSEA score = 3), an estrogenic steroid produced by horses, also has been proven to have moderate activity in inhibiting the entry of Zaire Ebola virus glycoprotein and human immunodeficiency virus (ZEBOV-GP/HIV). Altogether, network-predicted SERMs (such as toremifene and equilin) offer candidate repurposable drugs for 2019- nCoV/SARS-CoV-2.

Angiotensin receptor blockers (ARBs) have been reported to associate with viral infection, including HCoVs. Irbesartan (Z = -5.98), a typical ARB, was approved by the FDA for treatment of hypertension and diabetic nephropathy. Here, network proximity analysis shows a significant association between irbesartan’s targets and HCoV-associated host proteins in the human interactome. As shown in Fig. 5c, irbesartan targets SLC10A1, encoding the sodium/bile acid cotransporter (NTCP) protein that has been identified as a functional preS1-specific receptor for the hepatitis B virus (HBV) and the hepatitis delta virus (HDV). Irbesartan can inhibit NTCP, thus inhibiting viral entry. SLC10A1 interacts with C11orf74, a potential transcriptional repressor that interacts with nsp-10 of SARS-CoV. There are several other ARBs (such as eletriptan, frovatriptan, and zolmitriptan) in which their targets are potentially associated with HCoV-associated host proteins in the human interactome.

Inflammatory pathways play essential roles in viral infections. As a biogenic amine, melatonin (N-acetyl- 5-methoxytryptamine) (Z = –1.72 and GSEA score = 2) plays a key role in various biological processes, and offers a potential strategy in the management of viral infections. Viral infections are often associated with immune-inflammatory injury, in which the level of oxidative stress increases significantly and leaves negative effects on the function of multiple organs. The antioxidant effect of melatonin makes it a putative candidate drug to relieve patients’ clinical symptoms in antiviral treatment, even though melatonin cannot eradicate or even curb the viral replication or transcription. In addition, the application of melatonin may prolong patients’ survival time, which may provide a chance for patients’ immune systems to recover and eventually eradicate the virus. As shown in Fig. 5e, melatonin indirectly targets several HCoV cellular targets, including ACE2, BCL2L1, JUN, and IKBKB. Eplerenone (Z = –1.59), an aldosterone receptor antagonist, is reported to have a similar anti-inflammatory effect as melatonin. By inhibiting mast-cell-derived proteinases and suppressing fibrosis, eplerenone can improve survival of mice infected with encephalomyocarditis virus. In summary, our network proximity analyses offer multiple candidates repurposable drugs that target diverse cellular pathways for potential prevention and treatment of 2019-nCoV/SARS-CoV-2. However, further preclinical experiments and clinical trials are required to verify the clinical benefits of these network-predicted candidates before clinical use.

Drug combinations, offering increased therapeutic efficacy and reduced toxicity, play an important role in treating various viral infections. However, our ability to identify and validate effective combinations is limited by a combinatorial explosion, driven by both the large number of drug pairs and dosage combinations. In our recent study, we proposed a novel network-based methodology to identify clinically efficacious drug combinations. Relying on approved drug combinations for hypertension and cancer, we found that a drug combination was therapeutically effective only if it was captured by the “Complementary Exposure” pattern: the targets of the drugs both hit the disease module, but target separate neighborhoods (Fig. 6a). Here we sought to identify drug combinations that may provide a synergistic effect in potentially treating 2019-nCoV/SARS-CoV-2 with well-defined mechanism-of-action by network analysis. For the 16 potential repurposable drugs (Fig. 5a, Table 1), we showcased three network-predicted candidate drug combinations for 2019-nCoV/SARS-CoV-2. All predicted possible combinations can be found in Supplementary.

Toremifene is among the approved first-generation nonsteroidal SERMs for the treatment of metastatic breast cancer. SERMs (including toremifene) inhibited Ebola virus infection by interacting with and destabilizing the Ebola virus glycoprotein. In vitro assays have demonstrated that toremifene inhibited growth of MERSCoV and SARA-CoV (Table. 1). Emodin, an anthraquinone derivative extracted from the roots of rheum tanguticum, has been reported to have various anti-virus effects. Specifically, emdoin inhibited SARSCoV- associated 3a protein, and blocked an interaction between the SARS-CoV spike protein and ACE2 (ref. 71). Altogether, network analyses and published experimental data suggested that combining toremifene and emdoin offered a potential therapeutic approach for 2019-nCoV/ SARS-CoV-2 (Fig. 6c).

As shown in Figure.5a, targets of both mercaptopurine and melatonin showed strong network proximity with HCoVassociated host proteins in the human interactome network. Recent in vitro and in vivo studies identified mercaptopurine as a selective inhibitor of both SARS-CoV and MERS-CoV by targeting papain-like protease. Melatonin was reported in potential antiviral infection via its anti-inflammatory and antioxidant effects. Melatonin indirectly regulates ACE2 expression, a key entry receptor involved in viral infection of HCoVs, including 2019-nCoV/SARS-CoV-2 (ref. 33). Specifically, melatonin was reported to inhibit calmodulin and calmodulin interacts with ACE2 by inhibiting shedding of its ectodomain, a key infectious process of SARS-CoV. JUN, also known as c-Jun, is a key host protein involving inHCoV infectious bronchitis virus. As shown in Fig. 6d, mercaptopurine and melatonin may synergistically block c-Jun signaling by targeting multiple cellular targets. In summary, combination of mercaptopurine and melatonin may offer a potential combination therapy for 2019-nCoV/SARS-CoV-2 by synergistically targeting papainlike protease, ACE2, c-Jun signaling, and anti-inflammatory pathways (Fig. 6d). However, further experimental observations on ACE2 pathways by melatonin in 2019-nCoV/SARS-CoV-2 are highly warranted.