A comprehensive SARS-CoV-2-human protein- protein interactome network reveals novel pathobiology and host-targeting therapies for COVID-19

23 Physical interactions between viral and host proteins are responsible for almost all aspects of the 24 viral life cycle and the host’s immune response. Studying viral-host protein-protein interactions is 25 thus crucial for identifying strategies for treatment and prevention of viral infection. Here, we use 26 high-throughput yeast two-hybrid and affinity purification followed by mass spectrometry to 27 generate a comprehensive SARS-CoV-2-human protein-protein interactome network consisting 28 of both binary and co-complex interactions. We report a total of 739 high-confidence interactions, 29 showing the highest overlap of interaction partners among published datasets as well as the 30 highest enrichment for genes differentially expressed in samples (such as upper airway and 31 bronchial epithelial cells) from patients with SARS-CoV-2 infection. Showcasing the utility of our 32 network, we describe a novel interaction between the viral accessory protein ORF3a and the host 33 zinc finger transcription factor ZNF579 to illustrate one of the first examples of a viral factor 34 mediating a direct impact on host transcription. Leveraging our interactome, we performed 35 network-based drug screens for over 2,900 FDA-approved/investigational drugs and obtained a 36 curated list of 21 drugs that had significant network proximities to SARS-CoV-2 host factors, one 37 of which, carvedilol, showed promising antiviral properties. We performed electronic health 38 record-based validation using two independent large-scale, longitudinal COVID-19 patient 39 databases and found that carvedilol usage was associated with a significantly lowered probability 40 (17%-20%, P < 0.001) of obtaining a SARS-CoV-2 positive test after adjusting various 41 confounding factors. Carvedilol additionally showed anti-viral activity against SARS-CoV-2 in a 42 human lung epithelial cell line (EC50 value of 4.1 μM), suggesting a mechanism for its beneficial 43 effect in COVID-19. Our study demonstrates the value of large-scale network systems biology 44 approaches for extracting biological insight from complex biological processes. 45 Main 46 The global coronavirus disease 2019 (COVID-19) pandemic caused by the highly transmissible 47 and pathogenic severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) remains a 48 persistent part of everyday life for much of the world. Epidemiological models predict, in line with 49 virological theory, that the pandemic is unlikely to end with the complete eradication of the virus. 50 Indeed, SARS-CoV-2 becoming endemic in pockets of the world’s population is hypothesized to 51 be a natural consequence of the virus’s widespread propagation. Recent studies are now 52 exploring the waning of immunity with time from point of natural infection or immunization, and 53 the emergence of numerous viral variants raise concern of a perennial selection for more 54 infectious or virulent mutants to again sweep through the globe. More recently, engineering of 55 SARS-CoV-2 spike (S) protein chimeras containing several mutations of concern found a variant 56 that completely evaded the immune response of all except those that were afforded protection 57 with recovery from natural infection followed by vaccination. This, along with other elusive 58 phenomena, highlight the gaps in our understanding of the interplay between this virus and its 59 host upon natural infection and immunization, and thus, much work is still to be done to elucidate 60 the pathobiology of SARS-CoV-2, especially as the maintenance of immunity against this 61 pathogen remains at utmost interest to global public health. 62 Viruses interface with host cell surfaces to gain entry, wherein they interact with 63 intracellular proteins to hijack host mechanisms that facilitate viral replication and evasion of an 64 immune response. Studying viral-host protein-protein interactions (PPIs) is therefore pivotal for 65 understanding the mechanisms by which viral infection progresses and how the host responds to 66 said infection and thus, for identifying strategies for treatment and prevention. These networks 67 of interactions are especially important as proteins rarely act in isolation and their roles should be 68 evaluated in conjunction with their neighborhood of interacting partners. Such interactomes can 69 thus reveal biological pathways and processes impacted by the viral proteome, allowing for the 70 discovery of novel drug targets that directly or indirectly affect the viral-host point of contact. 71 To that effect, here, we leverage high-throughput yeast two-hybrid (Y2H) and tandem 72 mass tag affinity purification followed by mass spectrometry (TMT-AP/MS) to generate the first 73 binary and co-complex SARS-CoV-2-human protein-protein interactome network, which we 74 propose to be a more complete resource for exploration of the viral-host interactome (Fig. 1a). 75 We adopted this approach for several reasons. To date, yeast two-hybrid (Y2H) and affinity 76 purification-mass spectrometry (AP/MS) are the only two methodologies available for mapping 77 protein-protein interactome networks on a proteomic scale. While both Y2H and AP/MS alone 78 produce high-quality interactome datasets, they fundamentally capture different yet 79 complementary aspects of the full network and thus together can provide a more comprehensive 80 view of the topological and biological properties of the interactome. Moreover, labelled (e.g., 81 TMT-based) AP/MS has been shown to provide more precise, accurate, and reproducible 82 quantification of proteins compared to label-free AP/MS-based approaches, which is an important 83 criterion when trying to identify true protein interactions and generate high-quality interactome 84 networks. 85 Together, Y2H and AP/MS generated a total of 739 high confident interactions among 579 86 human proteins and 28 SARS-CoV-2 proteins. Our interactome had an unprecedented scale and 87 coverage compared with existing ones. Using our interactome, we identified important pathways 88 such as protein translation, mRNA splicing, Golgi transportation, neutrophil mediated immunity, 89 and glucose metabolism. Moreover, we prioritized host-targeting therapies by searching U.S. 90 FDA-approved/investigational drugs for their potential anti-SARS-CoV-2 effect using state-of-the91 art network proximity methods. Using two large independent COVID-19 patient databases, we 92 found that usage of one of the top candidates, carvedilol, was associated with a lowered risk 93 (17%-20%) of a positive COVID-19 test. Experimental validation shows that carvedilol inhibits 94 SARS-CoV-2 infection with an EC50 of 4.1 μM. Altogether, these results suggest that our 95 comprehensive SARS-CoV-2-human protein interactome offers great opportunities for 96 understanding the pathobiological process of SARS-CoV-2 in human and identifying host97 targeting therapies for COVID-19. 98 Results 99 A comprehensive SARS-CoV-2-human protein-protein interactome network 100 To generate a binary SARS-CoV-2-human protein-protein interactome, we systematically tested 101 all pairwise combinations of 28 SARS-CoV-2 proteins (GenBank entry MN908947) against 102 ~16,000 human proteins (hORFeome V8.1) using high-throughput Y2H screens (Fig. 1a). 103 We treated each protein as both a bait and a prey, yielding over 896,000 (28 × ~16,000 × 2) total 104 tested pair combinations. Prior to screening, all autoactivating DB ORF clones were removed 105 from further tests (see Methods). To increase experimental throughput, viral ORF AD and DB 106 clones were mated against pools of 24 human ORF DB or AD clones, respectively. Following 107 nutrition selection, AD:DB pairs were identified via PLATE-seq to generate a list of candidate 108 interactions (see Methods). Interaction candidates were then subsequently re-tested using Y2H 109 to ascertain high reproducibility. In all, we report a total of 299 high-quality binary SARS-CoV-2110 human PPIs via our high-throughput Y2H screen, 267 of which were unique to this assay (Table 111 S1). 112 To complement our binary SARS-CoV-2-human protein-protein interactome, we 113 independently expressed each of the 28 SARS-CoV-2 proteins in the human epithelial cell line 114 Caco-2 (HTB-37; ATCC) to identify viral-host co-complex interactions using TMT-AP/MS 115 proteomics (Fig. 1a). We used Caco-2 as our cell line model due to its endogenous expression 116 of TMPRSS2 required for SARS-CoV-2 S protein priming and subsequent cell entry, the line’s 117 extensive use in SARS-CoV and SARS-CoV-2 infection studies, supported by known in vivo 118 replication of SARS-CoV-2 in gastrointestinal cells, and desirable cell culture characteristics 119 including robust transfectability and rapid propagation. All Strep-, Myc-, or FLAG-tagged SARS120 CoV-2 baits and their corresponding empty vector controls were transfected in biological 121 duplicates, followed by subsequent affinity purification, TMT10 labeling, and SPS-MS3-based 122 quantification. We filtered for interactions that met stringent fold change and p-value based cutoffs 123 (see Methods). In all, we report a total of 472 high-confidence co-complex SARS-CoV-2-human 124 PPIs via AP/MS, 440 of which were unique to this assay (Table S1). Altogether, our orthogonal 125 approaches generated a network composed of 739 interactions among 28 viral and 579 host 126 proteins (Table S1). 127 We visualized the SARS-CoV-2-human protein-protein interactome through a network 128 shown in Fig. 1b. The colors of the edges between the viral proteins (represented as diamond 129 nodes) and the host proteins (represented as circle nodes) indicate the methods that detected 130 the interaction. Host proteins that interact with a single viral protein are shown in boxes connected 131 to their interacting partner. Several human proteins interact with multiple SARS-CoV-2 proteins, 132 such as ACTN4, ITGB1BP2, TRIM27, and ACTN1 (n=7, 6, 5, and 5, respectively), while the 133 majority of human proteins (469, 81%) interact with only one SARS-CoV-2 protein (Fig. S1a). 134 Among the viral proteins, N, ORF7b, and ORF9b achieved the highest network degrees (n=56, 135 56, and 54, respectively), whereas E, NSP7, and NSP1 have the lowest network degrees (n=9, 136 11, and 11, respectively) (Fig. S1b). In terms of the shared interacting partners, overall, the viral 137 proteins showed significant overlaps (Fisher’s exact test, P < 0.05) (Fig. S1c). For example, the 138 S protein overlaps significantly with 14 of the other 27 viral proteins in their host factors. Even the 139 viral protein with the lowest number of significant overlaps, NSP4, significantly overlaps with 140 NSP13 and ORF7b. These observations suggest potential functional overlap or redundancy of 141 the viral proteins. 142 Compared to the published SARS-CoV-2-human protein-protein interactome 143 networks, our interactome validated 218 (38%) human host factors previously reported. 144 Proteins including ALG5, G3BP1, CLCC1, VPS39, SIGMAR1, G3BP2, and RAP1GDS1 are 145 identified in all four interactomes (Fig. 2a). Importantly, our interactome offers 361 (62%) newly 146 discovered human host factors which in total interact with SARS-CoV-2 proteins in 493 novel 147 interactions. For S protein, which plays a key role in the entry of SARS-CoV-2 into host cells, 148 we identified 24 novel interacting partners. Among these interacting partners of S protein, we 149 found that CORO1C and STON2 express on the cell membrane, suggesting potential cell entry 150 of SARS-CoV-2 through these human proteins in addition to known mechanisms. 151 For the entire interactome, functional enrichment analysis revealed significantly 152 overrepresented biological processes (Fig. S2a), including protein translation, transcription, and 153 neutrophil-mediated immunity (highlighted with yellow background in Fig. 1b). Pathway 154 enrichment analysis show top enriched pathways including protein processing in the endoplasmic 155 reticulum, tight junction, glycolysis, ribosome, and protein export (Fig. S2b). For individual SARS156 CoV-2 proteins, many pathways and biological processes are shared in these viral proteins (Fig. 157 S3). For example, NSP12, NSP13, and NSP16 share biological processes such as regulation of 158 cellular component movement, negative regulation of cell morphogenesis involved in 159 differentiation, and negative regulation of substrate adhesion-dependent cell spreading (Fig. 160 S3a). 161 Overall, our interactome is comprised of abundant information that can be utilized for the 162 identification of novel pathobiology and host-targeting therapies. We also developed an interactive 163 visualization tool for our interactome which can be accessed from https://github.com/ChengF164 Lab/COVID-19_PPI. 165 This SARS-CoV-2-human interactome is of high coverage and quality 166 To ensure the authenticity when applying our interactome for downstream studies, we first 167 evaluated the quality through several means. We examined three previously published SARS168 CoV-2-human protein-protein interactome networks. Importantly, all three of these 169 interactomes were generated using AP/MS-based methods alone. We found that only a few 170 interactions overlapped among these datasets (Fig. 2a and Table S2), which could be explained 171 by differences in the cell line models used (Gordon et al. and Li et al. transfected HEK-293T/17; 172 Stukalave et al. infected A549-ACE2) as well as distinct computational and/or experimental 173 methodologies implemented in their respective study. Still, we found that our interactome had the 174 highest overlap of interaction partners among published SARS-CoV-2-human protein-protein 175 interactome networks (Fig. 2a and Table S2), suggesting that our interactome had a high level of 176 coverage. 177 We next performed several comparisons using these interactomes and our own. First, we 178 examined whether these datasets contained interaction partners that coincided with genes that 179 had expression changes in response to SARS-CoV-2 infection. To this end, we performed 180 differential expression analysis for several bulk and single-cell RNA-seq datasets (see Methods). 181 For the single-cell dataset which we compared the gene expression in SARS-CoV-2 and SARS182 CoV-2 cells, we found that our interactome showed significant overlap (Fisher’s exact test, P < 183 0.01) with the differentially expressed genes (DEGs) in more cell types than that of other 184 interactomes (Fig. 2b and Table S2). Using four bulk RNA-seq datasets that contained samples 185 such as upper airway and bronchial epithelial cells, we found that our interactome had a 186 comparable number of significant overlaps to other datasets and showed the highest overall 187 Jaccard index with the bulk RNA-seq datasets (Fig. 2c and Table S2). These results suggest that 188 our interactome is highly enriched in genes differentially expressed in response to SARS-CoV-2 189 infection. 190 Gene expression patterns in disease-related tissues carry important information for 191 revealing the pathogenesis of the disease and identifying potential treatments.We therefore 192 examined the expression of the human host factors in different tissues (Fig. S4 and Table S3) 193 using the GTEx data. By normalizing the expression of each gene across different tissues (tissue 194 specificity, see Methods), we found that lung ranked the 7 out of 33 tissues in terms of the 195 number of host factors with positive tissue specificity (Fig. 2d), suggesting that lung is one of the 196 tissues where these host factors have high expression. 197 We next inspected the evolutionary factors of the SARS-CoV-2 human host factors (Fig. 198 2e-f and Table S4). Our SARS-CoV-2-human interactome showed more purifying selection 199 (quantified by lower non-synonymous versus synonymous substitution rate ratio [dN/dS ratio]), 200 as well as a lower evolutionary rate ratio, compared to the random background from the human 201 protein interactome. These bioinformatics observations further suggested high evolutionary 202 conservation of host factors of SARS-CoV-2 identified by our Y2H and TMT-AP/MS proteomics 203 platforms, consistent with previous studies. 204 Altogether, these results show the high quality of the SARS-CoV-2-human interactome 205 identified in this study and strongly encouraged us to further look into the pathobiology of COVID206 19 and potential treatment using our interactome. 207 This SARS-CoV-2-human interactome reveals novel pathobiology of COVID-19 208 ORF3a is a SARS-CoV-2 accessory protein that has been reported to induce apoptosis in 293T 209 cells and to suppress the innate immune response via unclear molecular mechanisms. Our 210 interactome revealed that ORF3a physically interacts with ZNF579, a previously uncharacterized 211 human protein likely to be a transcription factor. We were able to validate this interaction using 212 co-immunoprecipitation (co-IP) western blotting (Fig. 3a). Furthermore, we found that the level of 213 ZNF579 protein is decreased after overexpression of ORF3a in 293T cells (Fig. 3b). As a result, 214 we hypothesized that the presence of ORF3a in cells might trigger changes in the transcriptional 215 state of human genes that are normally regulated by ZNF579. Using ENCODE ChIP-seq data, 216 we identified that ZNF579 is bound strongly to the promoter of HSPA6 (Fig. 3c). We found that 217 overexpression of ORF3a in 293T cells causes massive induction of HSPA6 using qPCR (Fig. 218 3d). These results indicate that the multifunctional SARS-CoV-2 accessory protein ORF3a can 219 induce expression of HSPA6, presumably by disrupting ZNF579, which is likely to normally exert 220 repressive activity at the HSPA6 promoter. This represents an additional previously unknown 221 activity of this multifunctional viral accessory protein. 222 The oligosaccharyltransferase (OST) complex catalyzes the N-glycosylation of nascent 223 polypeptides in the endoplasmic reticulum. Glycoproteins are critical for normal cell-cell 224 interactions, RNA replication and pathogenesis. Interestingly, OST inhibitor has been shown 225 to have activity against Dengue virus, Zika virus, West Nile virus, yellow fever viruses, and 226 HSV1 by affecting the viral replication. The OST complex was also found to be crucial for 227 innate immune responses triggered by lipopolysaccharide. Notably, the OST complex subunits 228 STT3A/B, RPN1/2, and DDOST were all shown to be present in our Y2H and AP/MS 229 interactome datasets, which we further validated using co-IP (Fig. 3e). Additionally, we also found 230 Sec61 (Fig. 3f-g), which is a major component of the ER translocon that facilitates the entry of 231 nascent polypeptides into the ER lumen for protein processing. Evidence suggests that Sec61 232 may participate in the replication and transcription of several viruses like Ebola virus, Influenza 233 virus, HIV, and Dengue virus. Thus, we hypothesize that OST and Sec61 may also participate 234 in SARS-COV-2 replication and/or the host immune response, offering potential targets for host235 targeting therapy development. 236 The SARS-CoV-2 nucleocapsid (N) protein binds to the viral RNA genome and is 237 multifunctional in viral RNA transcription, replication, and genome condensation. N protein is 238 conserved and stable with ~90% amino acid homology to the SARS-COV N protein. From our 239 dataset, we confirmed known interactions, including the stress granule core protein G3BP1/2 also 240 found in three other interactome datasets. In addition to these known interactions, we identified a 241 novel interaction between histone H1.4 and N protein. To validate this histone H1.4 and N protein 242 interaction, we overexpressed both N protein and histone H1.4 to perform co-IP, confirming their 243 interaction (Fig. 3h). Histone H1, also known as linker histone, mainly functions in chromatin 244 condensation and transcriptional repression. Accumulating evidence suggests that linker 245 histone is essential in the pathogenesis of several diseases, particularly for viral infection. There 246 is also evidence that Histone H1 could regulate IFN and inhibit influenza replication, in addition 247 to playing a role in the regulation of viral gene expression. Thus, we hypothesize that this novel 248 viral-host interaction could also be involved in mediation viral replication and/or gene expression. 249 Discovery of interactome-based host-targeting therapies for COVID-19 250 In our previous studies, we demonstrated that by using the SARS-CoV-2 host factors and the 251 human protein-protein interactome, we could prioritize existing drugs for their anti-SARS-CoV-2 252 potential. Using our newly discovered SARS-CoV-2-human protein-protein interactome 253 network, we performed network-based drug screening for more than 2,900 FDA254 approved/investigational drugs (Fig. 4a). We obtained a list of 189 drugs with significantly small 255 network proximities to the SARS-CoV-2 host factors, among which 27 had clinical trials for SARS256 CoV-2 (Table S5). To refine this list, we obtained the antiviral profiles of the top 189 drugs from 257 NCATS (https://opendata.ncats.nih.gov/covid19/assays, National Center for Advancing 258 Translational Sciences) and evaluated each drug for their desired antiviral properties (see 259 Methods). From this, we obtained a curated list of 21 drugs with significant network proximities 260 to the SARS-CoV-2 host factors as well as desired anti-SARS-CoV-2 activities in at least two 261 NCATS assays (Fig. 4b, Fig. S5, Table S6). 262 Overall, these top drugs fall into several major categories, including anti-infective 263 (amodiaquine, azithromycin, tetracycline, adefovir dipivoxil, tipranavir), anti-inflammatory 264 (apremilast, mefenamic acid, balsalazide, fenoprofen), antihypertensive (carvedilol, 265 hydrochlorothiazide, nilvadipine), and antineoplastic (toremifene, decitabine, venetoclax). Among 266 these drugs, apremilast, toremifene, decitabine, amodiaquine, and azithromycin are currently 267 being or have been tested in clinical trials for SARS-CoV-2. These top 21 drugs offer candidate 268 treatments for SARS-CoV-2 infections across diverse mechanism-of-actions identified from our 269 human-SARS-CoV-2 interactome. For example, balsalazide, toremifene, tetracycline, venetoclax, 270 tipranavir, and brimonidine may inhibit viral replication by inhibiting papain-like protease 3CL (Fig. 271 S5 and Table S6). Other drugs, such as carvedilol and hydrochlorothiazide, may directly inhibit 272 viral entry by disrupting the Spike-ACE2 PPI (Fig. S5 and Table S6). 273 In our previous efforts using existing SARS-CoV-2 interactomes and literature evidence, 274 we identified toremifene, a selective estrogen receptor modulator, as a one of the top candidates 275 for SARS-CoV-2 treatment. Previous studies show that toremifene blocks various viral 276 infections efficiently, including SARS-CoV-2 (IC50 = 3.58 μM), SARS-CoV-1 (EC50 = 11.97 μM), 277 MERS-CoV (EC50 = 12.9 μM), and Ebola virus (IC50 = ~1 μM). Indeed, NCATS data show that 278 toremifene is active in four assays, including Spike-ACE2 protein-protein interaction (AC50 = 11.92 279 μM), SARS-CoV pseudotyped particle entry (AC50 = 15.85 μM), MERS-CoV pseudotyped particle 280 entry (AC50 = 31.62 μM), and 3CL enzymatic activity (AC50 = 5.01 μM) (Fig. S5). Mechanistically, 281 a previous study that toremifene may inhibit SARS-CoV-2 cell entry by blocking the S and NSP14 282 proteins. These comprehensive validations show potential implications of SARS-CoV-2 283 interactome-predicted drug candidates (i.e., toremifene) offer promising drug candidates to be 284 tested further in COVID-19 patients. 285 Population-based and experimental validation of interactome-predicted drugs 286 We further selected candidate drugs for patient-level data validation and experimental validation 287 using subject matter expertise based on a combination of factors: (1) strength of the interactome 288 network-based prediction associations (a stronger network proximity score in Table S5); (2) 289 novelty of predicted drugs; (3) availability of sufficient patient data for meaningful evaluation 290 (exclusion of infrequently used medications); and (4) ideal pharmacokinetics properties in lung of 291 interactome-predicted drugs. Applying these criteria resulted in 2 top candidate drugs, carvedilol 292 (Z = -2.195, FDR = 0.03) and hydrochlorothiazide (Z = -2.428, FDR = 0.005), which are originally 293 approved for treatment of hypertension. 294 To identify the drug-outcome relationships of these drugs, we used a state-of-the-art active 295 user-design approach based on large-scale electronic health record (EHR) data. Using the 296 Northwestern Medicine Enterprise Data Warehouse (NMEDW) COVID-19 dataset (481,526 total 297 patients, 66,541 COVID-19 positive cases, Table 1), we found that both carvedilol (odds ratio 298 [OR] = 0.8, 95% confidence interval [CI] 0.68-0.94, P = 0.008) and hydrochlorothiazide (OR = 299 0.62, 95% CI 0.47-0.82, P < 0.001) were associated with a significantly lowered risk of positive 300 COVID-19 test after confounding adjustment (age, sex, race, and comorbidities) using a 301 propensity score (PS) matching approach (Fig. 4c-d and Table S7). The effect of carvedilol 302 was consistent for different race and sex subgroups (Fig. 4c and Table S7). To validate these 303 observations, we used a second EHR database as an external validation set (168,712 total 304 individuals, 83,340 SARS-CoV-2 positive cases, Table S8). We found that carvedilol had a 305 sufficient number of usage cases for the drug-outcome evaluation. By comparing individuals with 306 and without carvedilol usages (PS-matched by age, sex, race, and/or comorbidities), we found 307 that carvedilol usage was associated with a 17% (OR = 0.83, 95% CI 0.78-0.88, P < 0.001) 308 significantly lowered risk of COVID-19 positive test (Fig. 4e). This effect was also consistent when 309 we examined subgroups from the registry in terms of race and sex (Fig. 4e). 310 We found that carvedilol not only showed favorable results in the EHR-based validation, 311 but also has a promising antiviral profile from NCATS, showing high potencies for multiple desired 312 activities (Fig. 4f). The NCATS profile of carvedilol is comparable to that of remdesivir, whose 313 profile was deemed highly desirable (Fig. 4f). We then investigated carvedilol’s anti-SARS-CoV314 2 activity experimentally. We treated A549-ACE2 cells with 0.3-20 μM of carvedilol for 2 hours 315 followed by infection with SARS-CoV-2 at a multiplicity of infection (MOI) of 0.5 and incubation for 316 2 days. Cells were subsequently fixed and immunostained to detect for S protein, which was used 317 as a marker for infection. We found that carvedilol inhibited SARS-CoV-2 infection with an EC50 318 value of 4.1 μM (Fig. 4g), mechanistically supporting our SARS-CoV-2-human interactome-based 319 prediction and EHR-based findings. Lastly, we conducted drug-target network analysis of 320 carvedilol’s targets and SARS-CoV-2 host factors (Fig. S6 and Table S9). We found that 321 carvedilol could potentially affect the SARS-CoV-2 host factors through PPIs with its targets (Fig. 322 S6). 323 Discussion 324 In this study, we leveraged high-throughput Y2H and TMT-AP/MS to generate the first binary and 325 co-complex SARS-CoV-2-human protein-protein interactome network that overcomes the 326 shortcomings of using AP/MS as the sole approach for PPI identification. This interactome 327 validated 218 previously published SARS-CoV-2 host factors, and more importantly, revealed 361 328 novel ones. By comparing with previous interactomes, this interactome is the most 329 comprehensive in terms of overlaps among the interactomes and differentially expressed genes 330 captured by bulk and single-cell RNA-seq of SARS-CoV-2 infection. These host factors identified 331 in this interactome, particularly those altered in response to SARS-CoV-2 infection, present an 332 invaluable opportunity for understanding the disease pathobiology of COVID-19 and prioritizing 333 potential drug targets for treatment development. 334 Among the novel interacting partners for S protein, we identified several human proteins 335 which may play important roles in SARS-CoV-2 infection. CORO1C and STON2 are expressed 336 on the cell membrane. CORO1C is highly expressed in lung (Table S3). STON2 is ubiquitously 337 expressed and involved in endocytic machinery. It is possible that SARS-CoV-2 can enter host 338 cells through binding of S protein not only to ACE2, NRP1 and BSG, but also other 339 (unknown) factors such as CORO1C and STON2. We also noticed two proteins, EPPK1 and 340 SPECC1L, that both express on the cell junctions. It has been suggested that SARS-CoV-2 341 could spread through cell-to-cell transmission. These cell junction proteins that can be targeted 342 by SARS-CoV-2 S protein may facilitate its cell-to-cell transmission. 343 We noticed an overall high functional overlap of the SARS-CoV-2 proteins in terms of both 344 shared interacting partners (Fig. S1c) and shared enriched pathways and biological processes 345 (Fig. S3). This observation shows a high redundancy of SARS-CoV-2 proteins, which further 346 suggests that SARS-CoV-2 therapies that targets only specific viral protein or its host factors may 347 not be sufficient. This observation also justified the advantage and necessity of using network348 based approach of drug discovery for SARS-CoV-2. As shown by the drug repurposing results, 349 the top drugs can potentially affect multiple SARS-CoV-2 host factors (Fig. 4b). 350 We identified a previously uncharacterized human transcription factor, ZNF579, that 351 interacts with SARS-CoV-2 accessory protein ORF3a, and report that this interaction leads to the 352 de-repression of HSPA6. HSPA6 is a HSP70 family molecular chaperone, which are known to be 353 involved in the entry, replication, assembly, and release of various viral pathogens. We 354 speculate that SARS-CoV-2 has evolved this activity to ensure sufficient levels of molecular 355 chaperones are available to assist with the production of viral proteins in cells. 356 Next, using this newly discovered SARS-CoV-2-human protein-protein interactome, we 357 performed drug repurposing and identified a list of top 21 drugs. We found that although some of 358 these drugs can directly target the host factors, most of them indirectly affect the host factors 359 through PPIs with their targets (Fig. 4b). Further, we have identified carvedilol and 360 hydrochlorothiazide as potential host-targeting treatments for COVID-19 supported by multiple 361 lines of evidence (strong network proximities to SARS-CoV-2 host factors, significantly reduced 362 SARS-CoV-2 positive test risks in patients using these drugs based on large-scale EHR data, 363 strong anti-SARS-CoV-2 profiles based on high-throughput drug screening, and experimental 364 validation of anti-SARS-CoV-2 activity). To understand the potential mechanisms of carvedilol’s 365 anti-SARS-CoV-2 activity, we examined the carvedilol’s mechanism-of-action impacted by SARS366 CoV-2 host factors using network analysis (Fig. S6 and Table S9). Among the 579 unique host 367 factors, 237 (41%) have PPIs with carvedilol targets. A large portion of the human proteins in the 368 enriched pathways (protein translation [26/37, 70%], mRNA splicing [14/21, 67%], glucose 369 metabolism [9/15, 60%], and neutrophil mediated immunity [14/27, 52%]) have PPIs with 370 carvedilol targets, suggesting a potential mechanism in which carvedilol inhibits SARS-CoV-2 371 replication through multiple important pathways such as protein translation and mRNA splicing. 372 At individual target level, several targets showed significant proximities, including VEGFA, 373 NDUFC2, GJA1, ABCB1, HIF1A, VCAM1, and KCNH2 (Fig. S6). These targets may play 374 potential roles in carvedilol’s anti-SARS-CoV-2 activity, which warrants future in-depth 375 mechanistic investigations. 376 We acknowledge several limitations. The network-based SARS-CoV-2 treatment 377 discovery may be affected by the incompleteness of the human protein-protein interactome and 378 drug-target network. Therefore, we relied not only on the network discoveries, but also 379 incorporated other types of evidence, such as EHR-based validation and experimental validation. 380 Our EHR-based validation is retrospective and can only be applied to commonly used drugs due 381 to data availability. Although we adjusted for several confounding factors, other unknown factors 382 may still have effect on the results of EHR-based validation. Therefore, the drugs identified in this 383 study must be validated using randomized clinical trials before they can be used in patients with 384 COVID-19. 385 Materials and Methods 386 SARS-CoV-2 ORF clones 387 ORF3b (plasmid no. 141384; Addgene), NSP4 (plasmid no. 141369; Addgene), NSP12 (plasmid 388 no. 141378; Addgene), NSP13 (plasmid no. 141379; Addgene), and NSP14 (plasmid no. 141380; 389 Addgene) were a gift from Nevan Krogan, University of California, San Francisco. NSP6 (plasmid 390 no. 149309; Addgene) and NSP16 (plasmid no. 141269; Addgene) were a gift from Fritz Roth, 391 University of Toronto, which we cloned into our pHAGE-CMV-GAW-3xMyc-IRES-PURO construct 392 using Gateway. E, M, N, NSP1, NSP2, NSP3, NSP5, NSP7, NSP8, NSP9, NSP10, NSP15, 393 ORF3a, ORF6, ORF7a, ORF7b, ORF8, ORF9b, ORF9c, ORF10, and S, cloned into pCAG-FLAG 394 and pcDNA6B-FLAG constructs, were a gift from Pei-Hui Wang, Shandong University. All SARS395 CoV-2 ORFs were codon-optimized and expressed in either pLVX-EF1alpha-eGFP-2xStrep396 IRES-Puro (plasmid no. 141395; Addgene), pHAGE-CMV-GAW-3xMyc-IRES-PURO, pCAG397 FLAG, or pcDNA6B-FLAG mammalian expression vectors. 398 Y2H 399 Y2H screens were carried out as previously described. In brief, viral ORFs were cloned 400 into pDEST-AD and pDEST-DB vectors using Gateway LR to generate N-terminal ORF 401 fusions. Similarly, human ORFeome 8.1 was cloned into pDEST-AD and pDEST-DB vectors. 402 All AD and DB expression clones were transformed into Y2H Saccharomyces cerevisiae 403 strains MATa Y8800 and MATα Y8930 (genotype: leu2-3, 112 trp1-901 his3Δ200 ura3-52 404 gal4Δ gal80Δ GAL2::ADE2 GAL1::HIS3@LYS2 GAL7::lacZ@MET2 cyh2R), respectively. To 405 screen out autoactivating DB-ORFs, all DB-ORF MATα Y8930 transformants were mated 406 pairwise against empty pDEST-AD MATa Y8800 transformants and scored for growth on SC407 Leu-Trp+3AT and SC-Leu-Trp-Ade plates, where DB-ORFs that triggered reporter activity were 408 removed from further experiments. To increase screening throughput, 24 human ORF AD or 409 DB clones were pooled into single human ORF AD or DB wells, respectively. Viral ORF AD 410 and DB clones were then mated pairwise against pools of human ORF DB and AD clones, 411 respectively. Mated transformants were incubated overnight at 30 °C before being plated onto 412 SC-Leu-Trp to select for mated diploid yeast. After another overnight incubation at 30 °C, 413 diploid yeast was plated onto SC-Leu-Trp-His+3AT and SC-Leu-Trp-Ade selection plates. After 414 another overnight incubation at 30 °C, plates were replica-cleaned and incubated again for 415 three days at 30 °C for final interaction calling. 416 PLATE-seq 417 Each colony was picked 6 times into 96-well plates containing 15 μL of 2.5 mg/mL Zymolyase 418 (catalog no. E1004; Zymo Research) and incubated for 45 min at 37 °C followed by 10 min at 419 95 °C to prepare yeast cell lysate used as PLATE-seq DNA template. PLATE-seq was carried 420 out as previously described. In brief, plasmid(s) from individual wells of 96-well plates were 421 PCR amplified using a plasmid-specific forward primer and a reverse primer consisting of a 422 well-position-specific barcode and TruSeq 3′ sequencing adapter. Amplicons derived from the 423 same 96-well plate were pooled and purified using QIAquick PCR Purification Kit (ca talog no. 424 28104, Qiagen). Each amplicon pool was subject to Tn5 tagmentation to fragment the 425 amplicons and append adapters consisting of a plate-specific barcode and TruSeq 5′ 426 sequencing adapter. Tagmented DNA was purified using QIAquick PCR Purification Kit 427 (catalog no. 28104, Qiagen) and pooled across all 96-well plates. These pools were then 428 subjected to low-cycle PCR both to extend the TruSeq end adapters with sequences 429 compatible for binding to the Illumina flow cell and to enrich for only DNA fragments consisting 430 of TruSeq adapter sequences on both ends of the plate specific and well-position-specific 431 barcodes. PLATE-seq libraries were paired-end sequenced on an Illumina MiSeq. 432 Affinity purification 433 Caco-2 (HTB-37; ATCC) cells were cultured in EMEM (catalog no. 30-2003; ATCC) with 15% 434 FBS (catalog no. 30-2020; ATCC) at 37 °C with 5% CO2. All 28 SARS-CoV-2 ORFs were codon435 optimized and cloned into mammalian expression vectors that contained Strep, Myc, or FLAG 436 affinity tags. SARS-CoV-2 ORF plasmids and corresponding empty vectors were individually 437 transfected in biological duplicates into Caco-2 cells using Lipofectamine 3000 Transfection 438 Reagent (catalog no. L3000001; Invitrogen) following manufacturer’s instructions. Cells were 439 harvested 72 hr post-transfection and lysed using RIPA lysis buffer (50 mM Tris-HCl [pH 7.5], 150 440 mM NaCl, 1% (v/v) Nonidet P 40 Substitute, 5 mM EDTA, phosphatase inhibitor (catalog no. 441 4906845001; Roche), and protease inhibitor cocktail (catalog no. 11873580001; Roche)). 442 Samples were incubated for 30 min at 4 °C and then centrifuged at 13,000 ×g for 15 min at 4 °C. 443 Supernatants were collected and incubate with either MagStrep "type3" XT beads (catalog no. 2444 4090-002; IBA Lifesciences), Myc-Trap Agarose (catalog no. yta-10; ChromoTek) or Anti-FLAG 445 M2 Affinity Gel (catalog no. A2220; Millipore) overnight at 4 °C. Strep-tagged samples were 446 washed with 10x Buffer W (catalog no. 2-1003-100; IBA Lifesciences) three times at 4°C. Myc447 and FLAG-tagged samples were washed with RIPA buffer. Strep-tagged samples were eluted 448 using 10x Buffer BXT (catalog no. 2-1042-025; IBA Lifesciences). Mycand FLAG-tagged 449 samples were eluted using IP elution buffer (100 mM Tris-HCl [pH 7.5], 1% (v/v) SDS) and 450 incubated for 15 min at 65 °C. Other primary antibodies used in this study include c-Myc 451 Monoclonal Antibody (catalog no. 13-2500; Invitrogen) and Monoclonal Anti-FLAG M2 Antibody 452 (catalog no. F3165; Sigma-Aldrich). 453 Proteomic sample preparation 454 IP eluates were reduced using 200 mM TCEP for 1 hr at 55 °C. Samples were then alkylated 455 using 375 mM iodoacetamide for 30 min at room temperature in the absence of light. Samples 456 were digested using Trypsin Gold, Mass Spectrometry Grade (catalog no. V5280; Promega) at 457 an enzyme-to-substrate ratio of 1:100 and incubated overnight with nutation at 37 °C. Peptide 458 concentrations were measured using Pierce Quantitative Colorimetric Peptide Assay (catalog no. 459 23275; Thermo Scientific). Samples were normalized and resuspended using 1M 460 Triethylammonium bicarbonate (TEAB) for TMT experiments (catalog no. 90114; Thermo 461 Scientific). Samples were labeled using TMT10plex Isobaric Mass Tagging Kit (catalog no. 90113; 462 Thermo Scientific) at a (w/w) label-to-peptide ratio of 10:1 for 1 hr at room temperature. Labeling 463 reactions were quenched by the addition of 5% hydroxylamine and immediately pooled and dried 464 using a SpeedVac. Labeled peptides were enriched and fractionated using Pierce High pH 465 Reversed-Phase Peptide Fractionation Kit according to the manufacturer’s protocol (catalog no. 466 84868; Thermo Scientific). 467 LC-MS/MS 468 Fractions were analyzed using an EASY-nLC 1200 System (catalog no. LC140; Thermo 469 Scientific) equipped with an in-house 3 μm C18 resin(Michrom BioResources) packed capillary 470 column (75 μm × 25 cm) coupled to an Orbitrap Fusion Lumos Tribrid Mass Spectrometer (catalog 471 no. IQLAAEGAAPFADBMBHQ; Thermo Scientific). The mobile phase and elution gradient used 472 for peptide separation were as follows: 0.1% formic acid in water as buffer A and 0.1% formic acid 473 in 80% acetonitrile as buffer B; 0-5 min, 5%-10% B; 5-65 min, 10-55% B; 66-67 min, 55%-95% B; 474 67-68 min, 2% B; 68-72 min, 95% B; 72-80 min, 5% B; with a flow rate set to 200 nL/min. MS1 475 precursors were detected at m/z = 375-1500 and resolution = 120,000. A CID-MS2-HCD-MS3 476 method was used for MS data acquisition. Precursor ions with charge of 2+ to 7+ were selected 477 for MS2 analysis at resolution = 30,000, isolation width = 0.4 m/z, maximum injection time = 50 478 ms, and CID collision energy at 35%. 6 SPS precursors were selected for MS3 analysis and ions 479 were fragmented using HCD collision energy at 65%. Spectra were recorded using Thermo 480 Xcalibur Software Version 4.1 (catalog no. OPTON-30965; Thermo Scientific) and Tune 481 application version 3.0 (Thermo Scientific). Raw data were searched using Proteome Discoverer 482 Software 2.3 (Thermo Scientific) against an UniProtKB human database containing all SARS483 COV-2 proteins. Search parameters specified precursor mass and fragment mass tolerance of 484 15 ppm. Peptide-spectrum matches (PSMs) were searched with SEQUEST HT and Percolator 485 and filtered at FDR < 1 were considered significantly. 486 Downstream proteomic analysis 487 We developed a novel pipeline using a customized linear model (inspired by MSstatsTMT) to 488 identify high-confidence viral-host interactions from TMT-AP/MS datasets. Briefly, PSMs filtered 489 at 1% FDR were selected for quantification by (1) the number of reporter intensity values per 490 fraction, (2) percent isolation interference, and (3) precursor intensity values to select for one 491 instance of a peptide peak. If more than one PSM passed these criteria, then the average of the 492 reporter ion intensities per channel of these PSMs were taken to represent the quantification of 493 the peptide peak. The reporter intensity values of selected PSMs were log transformed, weighed 494 with their respective precursor intensities, and averaged to obtain protein level quantification 495 values. Our pipeline’s novelty lies in its ability to retain useful information separated across 496 fractions at the PSM level while ensuring no violation of the assumption of independence, such 497 that our linear fixed-effects model with conditions (e.g., sample vs. control), as a fixed effect, can 498 be utilized. An improved p-value calculation was used through Empirical Bayes estimation of prior 499 variance as implemented in R limma package. 500 The fold change (FC) and p-values obtained from this linear model-based approach are 501 used to generate volcano plots for each viral bait protein compared to control and a hyperbolic 502 curve is optimized using the distribution of the log-transformed FCs of all identified proteins to 503 identify high-confidence interactors. A PSM cutoff, along with a peptide-coverage percent cutoff 504 (i.e., the percentage of all possible trypsin digested peptides, accounting for up to two missed 505 cleavages that can be found), based on the number of the viral protein’s PSMs and peptide506 coverage percentage, is also implemented prior to the optimization of this hyperbolic curve. 507 Co-immunoprecipitation 508 HEK 293T (CRL-3216; ATCC) cells were cultured in DMEM (catalog no. 30-2002; ATCC) 509 supplemented with 10% FBS (catalog no. 30-2020; ATCC) and incubated at 37 °C with 5% CO2. 510 Cells were seeded onto six-well plates and grown until reaching 70-80% confluency. SARS-CoV511 2 N, ORF3a, ORF7b, histone H1.4 or N+histone H1.4, Sec61 or ORF7b+Sec61, STT3A or 512 ORF7b+STT3A, and empty vector controls were transfected into cells by combining 2 μg of DNA 513 with 10 μL of 1 mg/mL PEI (catalog no. 23966; Polysciences Inc.) and 150 μL Opti-MEM (catalog 514 no. 31985062; Gibco). After 24 hr incubation, cells were gently washed three times with DPBS 515 (1X) (catalog no. 14040117; Gibco), resuspended with 200 μL cell lysis buffer (10 mM Tris-HCl 516 [pH 8.0], 137 mM NaCl, 1% (v/v) Triton X-100, 10% (v/v) glycerol, 2 mM EDTA, and protease 517 inhibitor cocktail (catalog no. 11873580001; Roche)) and incubated on ice for 30 min. Extracts 518 were then cleared by centrifugation at 16,000 ×g for 10 min at 4 °C. To perform co519 immunoprecipitation (co-IP), 100 μL cell lysate was incubated with 5 μL Red Anti-FLAG M2 Affinity 520 Gel (catalog no. F2426; Millipore) overnight at 4 °C under gentle rotation. Bound proteins were 521 then washed three times with cell lysis buffer, eluted with 50 μl elution buffer (10 mM Tris-HCl [pH 522 8.0], 1% (v/v) SDS) and incubated for 10 min at 65 °C. Cell lysates and co-IP samples were then 523 treated with 6X SDS protein loading buffer (1 M Tris-HCl [pH 6.8], 10% (v/v) SDS, 50% (v/v) 524 glycerol, 0.03% (v/v) bromophenol blue, and 10% (v/v) β-mercaptoethanol), subjected to SDS525 PAGE, and transferred onto PVDF membranes (catalog no. GE10600023; Amersham). For 526 immunoblotting analysis, V5 Tag Monoclonal Antibody (catalog no. R960-25; Invitrogen), c-Myc 527 Monoclonal Antibody (catalog no. 13-2500; Invitrogen), Monoclonal Anti-FLAG M2 Antibody 528 (catalog no. F1804; Sigma-Aldrich), or ZNF579 Polyclonal Antibody (catalog no. A303-275A; 529 Bethyl Laboratories) were used at 1:1,000 dilutions. 530 qPCR 531 293T cells were cultured as above, and ORFa-FLAG or empty vector were introduced with 532 Lipofectamine 2000 Transfection Reagent (catalog no. 11668030; Invitrogen) using manufacturer 533 instructions. Transfection experiments were performed in duplicate. Media was replaced 6 hours 534 after transfection, and RNA was harvested using TRIzol Reagent (catalog no. 15596018; 535 Invitrogen). Reverse transcription was performed with the Maxima First Strand cDNA Synthesis 536 Kit for RT-qPCR, with dsDNase (catalog no. K1671; Thermo Scientific). qPCR was performed on 537 a LightCycler 480 System using LightCycler FastStart DNA Master SYBR Green I (catalog no. 538 03003230001; Roche Diagnostics). We used two primer sets for HSPA6 (FWD1: 539 CAAGGTGCGCGTATGCTAC, REV1: GCTCATTGATGATCCGCAACAC, FWD2: 540 CATCGCCTATGGGCTGGAC, REV2: GGAGAGAACCGACACATCGAA), and performed 3 541 technical replicates of 3 concentrations of cDNA (1:10, 1:100, 1:1000) for each replicate, and then 542 compared expression levels normalized to GAPDH using the double-delta Cp method. 543 Interactome comparative analysis 544 We compared our SARS-CoV-2-human interactome to a collection of three previously reported 545 interactomes, and compared with ours in terms of the overlap (Fisher’s exact test) with the 546 differentially expressed genes in SARS-CoV-2 from several SARS-CoV-2 RNA-seq/proteomics 547 datasets. These datasets include: (1) a single-cell dataset that contains CD8, Epithelial (Epi) 548 Ciliated, Epi-Secretory, Epi-Squamous, Macro, Mono, and NK cells from BALF. We performed 549 comparisons of virus vs. virus cells for each cell type; (2) bulk RNA-seq of human bronchial 550 epithelial cells infected with SARS-CoV-2 (GSE147507), denoted as SARS2-DEG; (3) 551 proteomic dataset of human Caco-2 cells infected with SARS-CoV-2, denoted as SARS2-DEP. 552 (4) bulk RNA-seq of upper airway from COVID-19 patients vs. non-COVID-19 patients 553 (GSE156063) , denoted as DE-NS; (5) bulk RNA-seq of peripheral blood mononuclear cell 554 (PBMC) isolated from COVID-19 patients vs. non-COVID-19 patients (GSE157103) , denoted 555 as DE-PBMC. For differential expression analysis, a cutoff of |log2FC| > 0.5 and FDR < 0.05 was 556 considered significant. 557 Functional enrichment analysis 558 Functional enrichment of our SARS-CoV-2 host factors were analyzed using Enrichr against the 559 Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) biological process 560 data sets. Pathways and GO terms with FDR < 0.05 were considered significantly enriched. 561 Selective pressure and evolutionary rates 562 The nonsynonymous and synonymous substitution rate ratio (dN/dS ratio) and the evolutionary 563 rate ratio of our SARS-CoV-2 host factors were evaluated as described in a previous study. 564 For dN/dS ratio, dN/dS<1 was considered purifying selection; dN/dS=1 was considered neutral 565 evolution; and dN/dS>1 was considered positive Darwinian selection. The evolutionary rate ratio 566 >1 was regarded as a fast rate and <1 as a slow rate. 567 Tissue gene expression specificity 568 We evaluated the gene expression specificity of the SARS-CoV-2 host factors in 33 tissues using 569 the RNA-Seq data from GTEx V8 (https://www.gtexportal.org/home/). The expression specificity 570 of gene i in tissue t was defined as 571 zit = Eit − Ei σi 572 where Ei was the mean and σi was the standard deviation of gene i’s expression across all 573 considered tissues, and Eit was the mean expression of gene i in tissue t. 574 Construction of human protein-protein interactome and drug-target network 575 The human protein-protein interactome and the drug-target network were used to screen for drugs 576 against the SARS-CoV-2 host factors. The human protein-protein interactome, composed of 577 17,706 protein nodes and 351,444 unique PPI edges was constructed in our previous 578 studies. Briefly, several types of high-quality PPI evidence gathered from public 579 databases and datasets were considered, including: binary PPIs identified by high-throughput 580 yeast two-hybrid in three datasets; lowor high-throughput experimentally discovered 581 kinase-substrate interactions from KinomeNetworkX, PhosphoNetworks, Human Protein 582 Resource Database (HPRD), DbPTM 3.0, Phospho.ELM, and PhosphositePlus; 583 signaling networks identified using low-throughput experiments in SignaLink2.0; protein 584 complexes revealed by robust affinity purification-mass spectrometry in BioPlex V2.016; and 585 curated PPIs from Instruct, IntAct, BioGRID, MINT, PINA, and InnateDB that were 586 identified by yeast two-hybrid studies, affinity purification-mass spectrometry, protein three587 dimensional structures, or low-throughput experiments. 588 The drug-target network was constructed using several data sources as described in our 589 recent studies: DrugBank database (v4.3), BindingDB, ChEMBL (v20), Therapeutic 590 Target Database (TTD), PharmGKB database, and IUPHAR/BPS Guide to 591 PHARMACOLOGY. Binding affinities Ki, Kd, IC50 or EC50 ≤ 10 μM were used as cutoff for the 592 drug-target interactions. All networks were visualized using Cytoscape 3.8.0. 593 Network proximity-based drug and drug combination screening 594 The “closest” network proximity measure was used to screen for 2,938 FDA approved or 595 investigational drugs. The “closest” distance dAB for two gene/protein sets A (e.g., drug targets) 596 and B (e.g., SARS-CoV-2 host factors) was calculated as: 597 〈dAB〉 = 1 ‖A‖ + ‖B‖ (∑ minb∈B a∈A d(a, b) + ∑ mina∈A b∈B d(a, b)) 598 where d(a, b) is the shortest path length of a and b in the human protein-protein interactome. 599 Network proximity dAB was further normalized to obtain a Z score using a permutation test with 600 randomly selected proteins from the interactome with similar degree distributions to A and B. 601 Permutation tests were repeated 1,000 times. We prioritized drugs by Z < -2 and FDR < 0.05. 602 The antiviral profiles of the prioritized drugs were retrieved from NCATS 603 (https://opendata.ncats.nih.gov/covid19/assays). NCATS contains experimental high-throughput 604 screening results for drugs from a series of screenings (some accompanied by counterscreens) 605 to evaluate their anti-SARS-CoV-2 potential. We included the following screening results: SARS606 CoV-2 cytopathic effect (CPE) and its counterscreen SARS-CoV-2 cytopathic effect (host tox 607 Counter) / Cytotoxicity; human fibroblast toxicity (hCYTOX); spike-ACE2 protein-protein 608 interaction (AlphaLISA) and its counterscreen spike-ACE2 protein-protein interaction (TruHit 609 Counter); ACE2 enzymatic activity; SARS-CoV pseudotyped particle entry (CoV-PPE) and its 610 counterscreen SARS-CoV pseudotyped particle entry counter screen (CoV-PPE_cs); MERS-CoV 611 pseudotyped particle entry (MERS-PPE) and its counterscreen MERS-CoV pseudotyped particle 612 entry counter screen (MERS-PPE_cs); and 3CL enzymatic activity. Based on the NCATS SARS613 CoV-2 data, we further selected a list of top drugs from the network proximity-based prioritization 614 that show ideal activities in at least two of these screenings. 615 COVID-19 patient data observations 616 Two independent datasets revealed corroborating evidence for the drug carvedilol which was 617 identified by our interactome prioritization framework. The first dataset (discovery dataset) was 618 from the Northwestern Medicine Enterprise Data Warehouse (NMEDW). We first identified 619 512,198 patients who had SARS-CoV-2 reverse transcription-polymerase chain reaction (RT620 PCR) test results recorded in NMEDW. Patients with a positive RT-PCR test were considered 621 COVID-19 positive, where the earliest time of the test was recorded as the effective time. Patients 622 that did not have any positive or presumptive positive RT-PCR tests and the latest PCR test was 623 negative (excluding pending and undetermined results) were considered COVID-19 negative, 624 where the latest time of the test was recorded as the effective time. By these metrics, 29,224 625 patients with pending or undetermined results were removed, yielding 482,974 patients of interest. 626 We excluded patients without age or sex information yielding a cohort of 481,526 patients, 66,541 627 of which were COVID-19 positive. We then extracted the carvedilol (and other drugs) 628 administration information for all patients in the final cohort. If a patient had a carvedilol 629 administration record with an administration date in the 6-month time window leading to the 630 effective RT-PCR result date and an administered dose > 0, the patient was considered 631 carvedilol+. We also extracted comorbidity information of the cohort for propensity score (PS) 632 matching, for which we used the Charlson Comorbidity Index (CCI). All comorbidities and 633 corresponding patient numbers are listed in Table 1. 634 The second dataset (external validation dataset) was an institutional review board635 approved COVID-19 registry dataset that included 168,712 individuals tested for SARS-CoV-2 636 infection (83,340 of which were positive cases) from March 8 th to May 26, 2021, at the Cleveland 637 Clinic in Ohio and Florida, United States (Table S8). Pooled oropharyngeal and nasopharyngeal 638 swab specimens were used to test for SARS-CoV-2 by RT-PCR assay in the Cleveland Clinic 639 Pathology and Laboratory Medicine Institute. All SARS-CoV-2 testing followed the guidelines 640 established by the Centers for Disease Control and Prevention of United States. The dataset 641 included baseline demographic information, medications, and COVID-19 test results. We used 642 REDCap electronic data capture tools to extract the patient data from the electronic health 643 records (EPIC Systems), and the data were manually checked by a professional team trained on 644 uniform sources for the study variables. A carvedilol exposure group (carvedilol+) included 645 patients that were actively taking carvedilol at the time of SARS-CoV-2 testing. Positive laboratory 646 test results for COVID-19 were used as the primary outcome. PS was used to match age, sex, 647 and race to reduce various confounding factors. Odds ratio was used to evaluate the carvedilol 648 benefit to primary outcome. All analyses were conducted by matchit package in the R v4.1.0 649 platform. 650 Anti-SARS-CoV-2 activity assay for carvedilol 651 A549 (CCL-185; ATCC) cells exogenously expressing angiotensin-converting enzyme 2 (ACE2) 652 (A549-ACE2) were a gift from Benjamin R. Tenoever, Icahn School of Medicine at Mount Sinai. 653 A549-ACE2 cells were cultured in DMEM (catalog no. 11965092; ThermoFisher) with 10% FBS 654 (catalog no. 100-106; GeminiBio) and used for SARS-CoV-2 infection. SARS-CoV-2 virus 655 (nCoV/Washington/1/2020) was provided by the Biocontainment Laboratory–University of Texas 656 Medical Branch Galveston National Laboratory, Texas, United States. Vero E6 (CRL-1586; 657 ATCC) cells were used to propagate and titer SARS-CoV-2. SARS-CoV-2 infections were 658 performed under biosafety level 3 conditions at the Biocontainment Laboratory–University of 659 Chicago Howard T. Ricketts Laboratory, Illinois, United States. A549-ACE2 cells cultured in 660 DMEM with 2% FBS were treated with carvedilol for 2 hours at the indicated concentrations. Cells 661 were infected with an MOI of 0.5 in media containing the appropriate concentration of drug. 48 hr 662 post-infection, cells were fixed with 10% formalin (catalog no. 305-510; Fisherbrand), blocked, 663 and probed with mouse anti-SARS-CoV-2-spike antibody (catalog no. GTX632604; GeneTex) 664 diluted 1:1,000 for 4 hr, rinsed, and probed with anti-mouse-HRP (catalog no. MP7401; Vector 665 Labratories) for 1 hr, washed, and then developed with DAB substrate (catalog no. 34065; 666 ThermoScientifc) for 10 min. Spike positive cells (n>40) were quantified by light microscopy as 667 blinded samples. A sigmoid fit was used to extract EC50 values using MATLAB. 668 Data availability 669 We downloaded the GTEx v8 dataset from https://gtexportal.org/home/. The human protein670 protein interactome and drug-target network are found in https://github.com/ChengF-Lab/COVID671 19_Map. An interactive version of Fig. 1b can be found in https://github.com/ChengF-Lab/COVID672 19_PPI. All other data can be found in the supplementary tables. We accessed ZNF579 ChIP673 seq data from the ENCODE portal under accession ENCSR018MQH. 674 Code availability 675 The network proximity framework can be found in https://github.com/ChengF-Lab/COVID676 19_Map. 677 Acknowledgements 678 This work was primarily supported by the National Institute of Aging (NIA) under Award Number 679 U01AG073323 and R01AG066707 to F.C. and the National Institute of General Medical Sciences 680 R01GM124559 to H. Y. This work was supported in part by NIA grants 3R01AG066707-01S1, 681 3R01AG066707-02S1, and R56AG074001 to F.C., the National Institute of General Medical 682 Sciences (R01GM125639, R01GM130885, RM1GM139738), The National Institute of Diabetes 683 and Digestive and Kidney Diseases (R01DK115398) to HY. 684 Competing interests 685 The authors declare that there are no competing interests. 686 References 687 1 Phillips, N. The coronavirus is here to stay here's what that means. Nature 590, 382384, doi:10.1038/d41586-021-00396-2 (2021). 2 Veldhoen, M. & Simas, J. P. Endemic SARS-CoV-2 will maintain post-pandemic immunity. Nat Rev Immunol 21, 131-132, doi:10.1038/s41577-020-00493-9 (2021). 3 Dolgin, E. COVID vaccine immunity is waning how much does that matter? Nature 597, 606-607, doi:10.1038/d41586-021-02532-4 (2021). 4 Kozlov, M. COVID-vaccine booster shot shows promise in Israeli study. Nature, doi:10.1038/d41586-021-02516-4 (2021). 5 Goldberg, Y. et al. Waning immunity of the BNT162b2 vaccine: A nationwide study from Israel. medRxiv, 2021.2008.2024.21262423, doi:10.1101/2021.08.24.21262423 (2021). 6 Schmidt, F. et al. High genetic barrier to SARS-CoV-2 polyclonal neutralizing antibody escape. Nature 600, 512-516, doi:10.1038/s41586-021-04005-0 (2021). 7 Callaway, E. COVID super-immunity: one of the pandemic's great puzzles. Nature 598, 393-394, doi:10.1038/d41586-021-02795-x (2021). 8 Durmus Tekir, S. D. & Ulgen, K. O. Systems biology of pathogen-host interaction: networks of protein-protein interaction within pathogens and pathogen-human interactions in the post-genomic era. Biotechnol J 8, 85-96, doi:10.1002/biot.201200110 (2013). 9 Gordon, D. E. et al. A SARS-CoV-2 protein interaction map reveals targets for drug repurposing. Nature 583, 459-468, doi:10.1038/s41586-020-2286-9 (2020). 10 Gordon, D. E. et al. Comparative host-coronavirus protein interaction networks reveal pan-viral disease mechanisms. Science 370, doi:10.1126/science.abe9403 (2020). 11 Li, J. et al. Virus-Host Interactome and Proteomic Survey Reveal Potential Virulence Factors Influencing SARS-CoV-2 Pathogenesis. Med (N Y) 2, 99-112 e117, doi:10.1016/j.medj.2020.07.002 (2021). 12 Stukalov, A. et al. Multilevel proteomics reveals host perturbations by SARS-CoV-2 and SARS-CoV. Nature 594, 246-252, doi:10.1038/s41586-021-03493-4 (2021). 13 Rajagopala, S. V. Mapping the Protein-Protein Interactome Networks Using Yeast TwoHybrid Screens. Adv Exp Med Biol 883, 187-214, doi:10.1007/978-3-319-23603-2_11 (2015). 14 Causier, B. Studying the interactome with the yeast two-hybrid system and mass spectrometry. Mass Spectrom Rev 23, 350-367, doi:10.1002/mas.10080 (2004). 15 Yu, H. et al. High-quality binary protein interaction map of the yeast interactome network. Science 322, 104-110, doi:10.1126/science.1158684 (2008). 16 Murphy, J. P. et al. Multiplexed Relative Quantitation with Isobaric Tagging Mass Spectrometry Reveals Class I Major Histocompatibility Complex Ligand Dynamics in Response to Doxorubicin. Anal Chem 91, 5106-5115, doi:10.1021/acs.analchem.8b05616 (2019). 17 Santin, Y. G. Uncovering the in vivo proxisome using proximity-tagging methods. Bioessays 41, e1900131, doi:10.1002/bies.201900131 (2019). 18 Berggard, T., Linse, S. & James, P. Methods for the detection and analysis of proteinprotein interactions. Proteomics 7, 2833-2842, doi:10.1002/pmic.200700131 (2007). 19 ten Have, S., Boulon, S., Ahmad, Y. & Lamond, A. I. Mass spectrometry-based immunoprecipitation proteomics the user's guide. Proteomics 11, 1153-1159, doi:10.1002/pmic.201000548 (2011). 20 Zhang, G., Annan, R. S., Carr, S. A. & Neubert, T. A. Overview of peptide and protein analysis by mass spectrometry. Curr Protoc Protein Sci Chapter 16, Unit16 11, doi:10.1002/0471140864.ps1601s62 (2010). 21 Gingras, A. C. & Raught, B. Beyond hairballs: The use of quantitative mass spectrometry data to understand protein-protein interactions. FEBS Lett 586, 2723-2731, doi:10.1016/j.febslet.2012.03.065 (2012). 22 Asara, J. M., Christofk, H. R., Freimark, L. M. & Cantley, L. C. A label-free quantification method by MS/MS TIC compared to SILAC and spectral counting in a proteomics screen. Proteomics 8, 994-999, doi:10.1002/pmic.200700426 (2008). 23 Collier, T. S. et al. Direct comparison of stable isotope labeling by amino acids in cell culture and spectral counting for quantitative proteomics. Anal Chem 82, 8696-8702, doi:10.1021/ac101978b (2010). 24 Megger, D. A. et al. Comparison of label-free and label-based strategies for proteome analysis of hepatoma cell lines. Biochim Biophys Acta 1844, 967-976, doi:10.1016/j.bbapap.2013.07.017 (2014). 25 Li, Z. et al. Systematic comparison of label-free, metabolic labeling, and isobaric chemical labeling for quantitative proteomics on LTQ Orbitrap Velos. J Proteome Res 11, 1582-1590, doi:10.1021/pr200748h (2012). 26 Stepath, M. et al. Systematic comparison of label-free, SILAC, and TMT techniques to study early adaption toward Inhibition of EGFR signaling in the colorectal cancer cell line DiFi. J Proteome Res 19, 926-937, doi:10.1021/acs.jproteome.9b00701 (2020). 27 Team, M. G. C. P. et al. The completion of the Mammalian Gene Collection (MGC). Genome Res 19, 2324-2333, doi:10.1101/gr.095976.109 (2009). 28 Das, J. et al. Cross-species protein interactome mapping reveals species-specific wiring of stress response pathways. Sci Signal 6, ra38, doi:10.1126/scisignal.2003350 (2013). 29 Vo, T. V. et al. A Proteome-wide Fission Yeast Interactome Reveals Network Evolution Principles from Yeasts to Human. Cell 164, 310-323, doi:10.1016/j.cell.2015.11.037 (2016). 30 Fragoza, R. et al. Extensive disruption of protein interactions by genetic variants across the allele frequency spectrum in human populations. Nat Commun 10, 4141, doi:10.1038/s41467-019-11959-3 (2019). 31 Wierbowski, S. D. et al. A massively parallel barcoded sequencing pipeline enables generation of the first ORFeome and interactome map for rice. Proc Natl Acad Sci U S A 117, 11836-11842, doi:10.1073/pnas.1918068117 (2020). 32 Hoffmann, M. et al. SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor. Cell 181, 271-280 e278, doi:10.1016/j.cell.2020.02.052 (2020). 33 Cinatl, J. et al. Treatment of SARS with human interferons. Lancet 362, 293-294, doi:10.1016/s0140-6736(03)13973-6 (2003). 34 Hoehl, S. et al. Evidence of SARS-CoV-2 Infection in Returning Travelers from Wuhan, China. N Engl J Med 382, 1278-1280, doi:10.1056/NEJMc2001899 (2020). 35 Lamers, M. M. et al. SARS-CoV-2 productively infects human gut enterocytes. Science 369, 50-54, doi:10.1126/science.abc1669 (2020). 36 Xiao, F. et al. Evidence for Gastrointestinal Infection of SARS-CoV-2. Gastroenterology 158, 1831-1833 e1833, doi:10.1053/j.gastro.2020.02.055 (2020). 37 Huang, Y., Yang, C., Xu, X. F., Xu, W. & Liu, S. W. Structural and functional properties of SARS-CoV-2 spike protein: potential antivirus drug development for COVID-19. Acta Pharmacol Sin 41, 1141-1149, doi:10.1038/s41401-020-0485-4 (2020). 38 Xavier, C. P. et al. Structural and functional diversity of novel coronin 1C (CRN2) isoforms in muscle. J Mol Biol 393, 287-299, doi:10.1016/j.jmb.2009.07.079 (2009). 39 Martina, J. A., Bonangelino, C. J., Aguilar, R. C. & Bonifacino, J. S. Stonin 2: an adaptorlike protein that interacts with components of the endocytic machinery. J Cell Biol 153, 1111-1120, doi:10.1083/jcb.153.5.1111 (2001). 40 Ren, X. et al. COVID-19 immune features revealed by a large-scale single-cell transcriptome atlas. Cell 184, 1895-1913 e1819, doi:10.1016/j.cell.2021.01.053 (2021). 41 Blanco-Melo, D. et al. Imbalanced host response to SARS-CoV-2 drives development of COVID-19. Cell 181, 1036-1045 e1039, doi:10.1016/j.cell.2020.04.026 (2020). 42 Bojkova, D. et al. Proteomics of SARS-CoV-2-infected host cells reveals therapy targets. Nature 583, 469-472, doi:10.1038/s41586-020-2332-7 (2020). 43 Mick, E. et al. Upper airway gene expression reveals suppressed immune responses to SARS-CoV-2 compared with other respiratory viruses. Nat Commun 11, 5854, doi:10.1038/s41467-020-19587-y (2020). 44 Overmyer, K. A. et al. Large-scale multi-omic analysis of COVID-19 severity. Cell Syst 12, 23-40 e27, doi:10.1016/j.cels.2020.10.003 (2021). 45 Cheng, F. et al. A genome-wide positioning systems network algorithm for in silico drug repurposing. Nat Commun 10, 3476, doi:10.1038/s41467-019-10744-6 (2019). 46 Cheng, F. et al. Network-based approach to prediction and population-based validation of in silico drug repurposing. Nat Commun 9, 2691, doi:10.1038/s41467-018-05116-5 (2018). 47 Luo, Y. et al. A multidimensional precision medicine approach identifies an autism subtype characterized by dyslipidemia. Nat Med 26, 1375-1379, doi:10.1038/s41591020-1007-0 (2020). 48 GTEx Consortium. The Genotype-Tissue Expression (GTEx) project. Nat Genet 45, 580585, doi:10.1038/ng.2653 (2013). 49 Jing, Y. et al. Potential influence of COVID-19/ACE2 on the female reproductive system. Mol Hum Reprod 26, 367-373, doi:10.1093/molehr/gaaa030 (2020). 50 Rozenblatt-Rosen, O. et al. Interpreting cancer genomes using systematic host network perturbations by tumour virus proteins. Nature 487, 491-495, doi:10.1038/nature11288 (2012). 51 Zhong, Q. et al. An inter-species protein-protein interaction network across vast evolutionary distance. Mol Syst Biol 12, 865, doi:10.15252/msb.20156484 (2016). 52 Ren, Y. et al. The ORF3a protein of SARS-CoV-2 induces apoptosis in cells. Cell Mol Immunol 17, 881-883, doi:10.1038/s41423-020-0485-9 (2020). 53 Xia, H. et al. Evasion of type I interferon by SARS-CoV-2. Cell Rep 33, 108234, doi:10.1016/j.celrep.2020.108234 (2020). 54 Rui, Y. et al. Unique and complementary suppression of cGAS-STING and RNA sensingtriggered innate immune responses by SARS-CoV-2 proteins. Signal Transduct Target Ther 6, 123, doi:10.1038/s41392-021-00515-5 (2021). 55 Wang, R. et al. ORF3a Protein of Severe Acute Respiratory Syndrome Coronavirus 2 Inhibits Interferon-Activated Janus Kinase/Signal Transducer and Activator of Transcription Signaling via Elevating Suppressor of Cytokine Signaling 1. Front Microbiol 12, 752597, doi:10.3389/fmicb.2021.752597 (2021). 56 Consortium, E. P. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57-74, doi:10.1038/nature11247 (2012). 57 Davis, C. A. et al. The Encyclopedia of DNA elements (ENCODE): data portal update. Nucleic Acids Res 46, D794-D801, doi:10.1093/nar/gkx1081 (2018). 58 Kelleher, D. J. & Gilmore, R. An evolving view of the eukaryotic oligosaccharyltransferase. Glycobiology 16, 47R-62R, doi:10.1093/glycob/cwj066 (2006). 59 Ruiz-Canada, C., Kelleher, D. J. & Gilmore, R. Cotranslational and posttranslational Nglycosylation of polypeptides by distinct mammalian OST isoforms. Cell 136, 272-283, doi:10.1016/j.cell.2008.11.047 (2009). 60 Lindenbach, B. D. & Rice, C. M. trans-Complementation of yellow fever virus NS1 reveals a role in early RNA replication. J Virol 71, 9608-9617, doi:10.1128/JVI.71.12.9608-9617.1997 (1997). 61 Beatty, P. R. et al. Dengue virus NS1 triggers endothelial permeability and vascular leak that is prevented by NS1 vaccination. Sci Transl Med 7, 304ra141, doi:10.1126/scitranslmed.aaa3787 (2015). 62 Lu, H., Cherepanova, N. A., Gilmore, R., Contessa, J. N. & Lehrman, M. A. Targeting STT3A-oligosaccharyltransferase with NGI-1 causes herpes simplex virus 1 dysfunction. FASEB J 33, 6801-6812, doi:10.1096/fj.201802044RR (2019). 63 Marceau, C. D. et al. Genetic dissection of Flaviviridae host factors through genomescale CRISPR screens. Nature 535, 159-163, doi:10.1038/nature18631 (2016). 64 Puschnik, A. S. et al. A Small-Molecule Oligosaccharyltransferase Inhibitor with Panflaviviral Activity. Cell Rep 21, 3032-3039, doi:10.1016/j.celrep.2017.11.054 (2017). 65 Parnas, O. et al. A Genome-wide CRISPR Screen in Primary Immune Cells to Dissect Regulatory Networks. Cell 162, 675-686, doi:10.1016/j.cell.2015.06.059 (2015). 66 Mohd Ropidi, M. I., Khazali, A. S., Nor Rashid, N. & Yusof, R. Endoplasmic reticulum: a focal point of Zika virus infection. J Biomed Sci 27, 27, doi:10.1186/s12929-020-0618-6 (2020). 67 Linxweiler, M., Schick, B. & Zimmermann, R. Let's talk about Secs: Sec61, Sec62 and Sec63 in signal transduction, oncology and personalized medicine. Signal Transduct Target Ther 2, 17002, doi:10.1038/sigtrans.2017.2 (2017). 68 Heaton, N. S. et al. Targeting Viral Proteostasis Limits Influenza Virus, HIV, and Dengue Virus Infection. Immunity 44, 46-58, doi:10.1016/j.immuni.2015.12.017 (2016). 69 Iwasa, A. et al. Contribution of Sec61alpha to the life cycle of Ebola virus. J Infect Dis 204 Suppl 3, S919-926, doi:10.1093/infdis/jir324 (2011). 70 Cascarina, S. M. & Ross, E. D. A proposed role for the SARS-CoV-2 nucleocapsid protein in the formation and regulation of biomolecular condensates. FASEB J 34, 98329842, doi:10.1096/fj.202001351 (2020). 71 Dutta, N. K., Mazumdar, K. & Gordy, J. T. The Nucleocapsid Protein of SARS-CoV-2: a Target for Vaccine Development. J Virol 94, doi:10.1128/JVI.00647-20 (2020). 72 Zeng, W. et al. Biochemical characterization of SARS-CoV-2 nucleocapsid protein. Biochem Biophys Res Commun 527, 618-623, doi:10.1016/j.bbrc.2020.04.136 (2020). 73 Grifoni, A. et al. A Sequence Homology and Bioinformatic Approach Can Predict Candidate Targets for Immune Responses to SARS-CoV-2. Cell Host Microbe 27, 671680 e672, doi:10.1016/j.chom.2020.03.002 (2020). 74 Buttinelli, M., Panetta, G., Rhodes, D. & Travers, A. The role of histone H1 in chromatin condensation and transcriptional repression. Genetica 106, 117-124, doi:10.1023/a:1003745315540 (1999). 75 Ye, X. et al. Linker Histone in Diseases. Int J Biol Sci 13, 1008-1018, doi:10.7150/ijbs.19891 (2017). 76 Liu, X. et al. HIST1H1C Regulates Interferon-beta and Inhibits Influenza Virus Replication by Interacting with IRF3. Front Immunol 8, 350, doi:10.3389/fimmu.2017.00350 (2017). 77 Conn, K. L., Hendzel, M. J. & Schang, L. M. Linker histones are mobilized during infection with herpes simplex virus type 1. J Virol 82, 8629-8646, doi:10.1128/JVI.0061608 (2008). 78 Zhou, Y. et al. Network-based drug repurposing for novel coronavirus 2019nCoV/SARS-CoV-2. Cell Discov 6, 14, doi:10.1038/s41421-020-0153-3 (2020). 79 Zhou, Y. et al. A network medicine approach to investigation and population-based validation of disease manifestations and drug repurposing for COVID-19. PLoS Biol 18, e3000970, doi:10.1371/journal.pbio.3000970 (2020). 80 Jeon, S. et al. Identification of Antiviral Drug Candidates against SARS-CoV-2 from FDA-Approved Drugs. Antimicrob Agents Chemother 64, e00819-00820, doi:10.1128/AAC.00819-20 (2020). 81 Dyall, J. et al. Repurposing of clinically developed drugs for treatment of Middle East respiratory syndrome coronavirus infection. Antimicrob Agents Chemother 58, 48854893, doi:10.1128/AAC.03036-14 (2014). 82 Cong, Y. et al. MERS-CoV pathogenesis and antiviral efficacy of licensed drugs in human monocyte-derived antigen-presenting cells. PLoS One 13, e0194868, doi:10.1371/journal.pone.0194868 (2018). 83 Johansen, L. M. et al. FDA-approved selective estrogen receptor modulators inhibit Ebola virus infection. Sci Transl Med 5, 190ra179, doi:10.1126/scitranslmed.3005471 (2013). 84 Martin, W. R. & Cheng, F. Repurposing of FDA-Approved Toremifene to Treat COVID19 by Blocking the Spike Glycoprotein and NSP14 of SARS-CoV-2. J Proteome Res 19, 4670-4677, doi:10.1021/acs.jproteome.0c00397 (2020). 85 Kc, G. B. et al. A machine learning platform to estimate anti-SARS-CoV-2 activities. Nature Machine Intelligence 3, 527-535, doi:10.1038/s42256-021-00335-w (2021). 86 Daly, J. L. et al. Neuropilin-1 is a host factor for SARS-CoV-2 infection. Science 370, 861-865, doi:10.1126/science.abd3072 (2020). 87 Cantuti-Castelvetri, L. et al. Neuropilin-1 facilitates SARS-CoV-2 cell entry and infectivity. Science 370, 856-860, doi:10.1126/science.abd2985 (2020). 88 Wang, K. et al. CD147-spike protein is a novel route for SARS-CoV-2 infection to host cells. Signal Transduct Target Ther 5, 283, doi:10.1038/s41392-020-00426-x (2020). 89 Shimada, H. et al. Epiplakin modifies the motility of the HeLa cells and accumulates at the outer surfaces of 3-D cell clusters. J Dermatol 40, 249-258, doi:10.1111/13468138.12076 (2013). 90 Saadi, I. et al. Deficiency of the cytoskeletal protein SPECC1L leads to oblique facial clefting. Am J Hum Genet 89, 44-55, doi:10.1016/j.ajhg.2011.05.023 (2011). 91 Zeng, C. et al. SARS-CoV-2 Spreads through Cell-to-Cell Transmission. bioRxiv, 2021.2006.2001.446579, doi:10.1101/2021.06.01.446579 (2021). 92 Wan, Q., Song, D., Li, H. & He, M. L. Stress proteins: the biological functions in virus infection, present and challenges for target-based antiviral drug development. Signal Transduct Target Ther 5, 125, doi:10.1038/s41392-020-00233-4 (2020). 93 Huang, T. et al. MSstatsTMT: Statistical Detection of Differentially Abundant Proteins in Experiments with Isobaric Labeling and Multiple Mixtures. Mol Cell Proteomics 19, 17061723, doi:10.1074/mcp.RA120.002105 (2020). 94 Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43, e47, doi:10.1093/nar/gkv007 (2015). 95 Kuleshov, M. V. et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res 44, W90-97, doi:10.1093/nar/gkw377 (2016). 96 Hirsh, A. E., Fraser, H. B. & Wall, D. P. Adjusting for selection on synonymous sites in estimates of evolutionary distance. Mol Biol Evol 22, 174-177, doi:10.1093/molbev/msh265 (2005). 97 Bezginov, A., Clark, G. W., Charlebois, R. L., Dar, V. U. & Tillier, E. R. Coevolution reveals a network of human proteins originating with multicellularity. Mol Biol Evol 30, 332-346, doi:10.1093/molbev/mss218 (2013). 98 Cheng, F. et al. Studying tumorigenesis through network evolution and somatic mutational perturbations in the cancer interactome. Mol Biol Evol 31, 2156-2169, doi:10.1093/molbev/msu167 (2014). 99 Cheng, F., Kovacs, I. A. & Barabasi, A. L. Network-based prediction of drug combinations. Nat Commun 10, 1197, doi:10.1038/s41467-019-09186-x (2019). 100 Smith, I. N., Thacker, S., Seyfi, M., Cheng, F. & Eng, C. Conformational Dynamics and Allosteric Regulation Landscapes of Germline PTEN Mutations Associated with Autism Compared to Those Associated with Cancer. Am J Hum Genet 104, 861-878, doi:10.1016/j.ajhg.2019.03.009 (2019). 101 Rolland, T. et al. A proteome-scale map of the human interactome network. Cell 159, 1212-1226, doi:10.1016/j.cell.2014.10.050 (2014). 102 Rual, J. F. et al. Towards a proteome-scale map of the human protein-protein interaction network. Nature 437, 1173-1178, doi:10.1038/nature04209 (2005). 103 Cheng, F., Jia, P., Wang, Q. & Zhao, Z. Quantitative network mapping of the human kinome interactome reveals new clues for rational kinase inhibitor discovery and individualized cancer therapy. Oncotarget 5, 3697-3710, doi:10.18632/oncotarget.1984 (2014). 104 Hu, J. et al. PhosphoNetworks: a database for human phosphorylation networks. Bioinformatics (Oxford, England) 30, 141-142, doi:10.1093/bioinformatics/btt627 (2014). 105 Keshava Prasad, T. S. et al. Human Protein Reference Database--2009 update. Nucleic Acids Res 37, D767-772, doi:10.1093/nar/gkn892 (2009). 106 Lu, C. T. et al. DbPTM 3.0: an informative resource for investigating substrate site specificity and functional association of protein post-translational modifications. Nucleic Acids Res 41, D295-305, doi:10.1093/nar/gks1229 (2013). 107 Dinkel, H. et al. Phospho.ELM: a database of phosphorylation sites--update 2011. Nucleic Acids Res 39, D261-267, doi:10.1093/nar/gkq1104 (2011). 108 Hornbeck, P. V. et al. PhosphoSitePlus, 2014: mutations, PTMs and recalibrations. Nucleic Acids Res 43, D512-520, doi:10.1093/nar/gku1267 (2015). 109 Csabai, L., Olbei, M., Budd, A., Korcsmaros, T. & Fazekas, D. SignaLink: Multilayered Regulatory Networks. Methods in molecular biology (Clifton, N.J.) 1819, 53-73, doi:10.1007/978-1-4939-8618-7_3 (2018). 110 Huttlin, E. L. et al. The BioPlex Network: A Systematic Exploration of the Human Interactome. Cell 162, 425-440, doi:10.1016/j.cell.2015.06.043 (2015). 111 Meyer, M. J., Das, J., Wang, X. & Yu, H. INstruct: a database of high-quality 3D structurally resolved protein interactome networks. Bioinformatics (Oxford, England) 29, 1577-1579, doi:10.1093/bioinformatics/btt181 (2013). 112 Orchard, S. et al. The MIntAct project--IntAct as a common curation platform for 11 molecular interaction databases. Nucleic Acids Res 42, D358-363, doi:10.1093/nar/gkt1115 (2014). 113 Oughtred, R. et al. The BioGRID interaction database: 2019 update. Nucleic Acids Res 47, D529-D541, doi:10.1093/nar/gky1079 (2019). 114 Licata, L. et al. MINT, the molecular interaction database: 2012 update. Nucleic Acids Res 40, D857-861, doi:10.1093/nar/gkr930 (2012). 115 Cowley, M. J. et al. PINA v2.0: mining interactome modules. Nucleic Acids Res 40, D862-865, doi:10.1093/nar/gkr967 (2012). 116 Breuer, K. et al. InnateDB: systems biology of innate immunity and beyond--recent updates and continuing curation. Nucleic Acids Res 41, D1228-1233, doi:10.1093/nar/gks1147 (2013). 117 Law, V. et al. DrugBank 4.0: shedding new light on drug metabolism. Nucleic Acids Res 42, D1091-1097, doi:10.1093/nar/gkt1068 (2014). 118 Liu, T., Lin, Y., Wen, X., Jorissen, R. N. & Gilson, M. K. BindingDB: a web-accessible database of experimentally determined protein-ligand binding affinities. Nucleic Acids Res 35, D198-201, doi:10.1093/nar/gkl999 (2007). 119 Gaulton, A. et al. ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Res 40, D1100-1107, doi:10.1093/nar/gkr777 (2012). 120 Yang, H. et al. Therapeutic target database update 2016: enriched resource for bench to clinical drug target and targeted pathway information. Nucleic Acids Res 44, D10691074, doi:10.1093/nar/gkv1230 (2016). 121 Whirl-Carrillo, M. et al. Pharmacogenomics knowledge for personalized medicine. Clin Pharmacol Ther 92, 414-417, doi:10.1038/clpt.2012.96 (2012). 122 Pawson, A. J. et al. The IUPHAR/BPS Guide to PHARMACOLOGY: an expert-driven knowledgebase of drug targets and their ligands. Nucleic Acids Res 42, D1098-1106, doi:10.1093/nar/gkt1143 (2014). 123 Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 13, 2498-2504, doi:10.1101/gr.1239303 (2003). 124 Harris, P. A. et al. Research electronic data capture (REDCap)--a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform 42, 377-381, doi:10.1016/j.jbi.2008.08.010 (2009).


23
Physical interactions between viral and host proteins are responsible for almost all aspects of the 24 viral life cycle and the host's immune response. Studying viral-host protein-protein interactions is 25 thus crucial for identifying strategies for treatment and prevention of viral infection. Here, we use 26 high-throughput yeast two-hybrid and affinity purification followed by mass spectrometry to 27 generate a comprehensive SARS-CoV-2-human protein-protein interactome network consisting 28 of both binary and co-complex interactions. We report a total of 739 high-confidence interactions,

29
showing the highest overlap of interaction partners among published datasets as well as the   [3][4][5] , and 53 the emergence of numerous viral variants raise concern of a perennial selection for more 54 infectious or virulent mutants to again sweep through the globe. More recently, engineering of 55 SARS-CoV-2 spike (S) protein chimeras containing several mutations of concern found a variant 56 that completely evaded the immune response of all except those that were afforded protection 57 with recovery from natural infection followed by vaccination 6,7 . This, along with other elusive 58 phenomena, highlight the gaps in our understanding of the interplay between this virus and its 59 host upon natural infection and immunization, and thus, much work is still to be done to elucidate 60 the pathobiology of SARS-CoV-2, especially as the maintenance of immunity against this 61 pathogen remains at utmost interest to global public health.

62
Viruses interface with host cell surfaces to gain entry, wherein they interact with 63 intracellular proteins to hijack host mechanisms that facilitate viral replication and evasion of an 64 immune response 8 . Studying viral-host protein-protein interactions (PPIs) is therefore pivotal for quantification. We filtered for interactions that met stringent fold change and p-value based cutoffs 123 (see Methods). In all, we report a total of 472 high-confidence co-complex SARS-CoV-2-human 124 PPIs via AP/MS, 440 of which were unique to this assay (Table S1). Altogether, our orthogonal 125 approaches generated a network composed of 739 interactions among 28 viral and 579 host 126 proteins (Table S1).

127
We visualized the SARS-CoV-2-human protein-protein interactome through a network 128 shown in Fig. 1b. The colors of the edges between the viral proteins (represented as diamond 129 nodes) and the host proteins (represented as circle nodes) indicate the methods that detected 130 the interaction. Host proteins that interact with a single viral protein are shown in boxes connected 131 to their interacting partner. Several human proteins interact with multiple SARS-CoV-2 proteins, 132 such as ACTN4, ITGB1BP2, TRIM27, and ACTN1 (n=7, 6, 5, and 5, respectively), while the 133 majority of human proteins (469, 81%) interact with only one SARS-CoV-2 protein (Fig. S1a).

144
Proteins including ALG5, G3BP1, CLCC1, VPS39, SIGMAR1, G3BP2, and RAP1GDS1 are 145 identified in all four interactomes (Fig. 2a). Importantly, our interactome offers 361 (62%) newly 146 discovered human host factors which in total interact with SARS-CoV-2 proteins in 493 novel 147 interactions. For S protein, which plays a key role in the entry of SARS-CoV-2 into host cells 37 , we identified 24 novel interacting partners. Among these interacting partners of S protein, we 149 found that CORO1C 38 and STON2 39 express on the cell membrane, suggesting potential cell entry 150 of SARS-CoV-2 through these human proteins in addition to known mechanisms.

151
For the entire interactome, functional enrichment analysis revealed significantly 152 overrepresented biological processes (Fig. S2a), including protein translation, transcription, and 153 neutrophil-mediated immunity (highlighted with yellow background in Fig. 1b). Pathway 154 enrichment analysis show top enriched pathways including protein processing in the endoplasmic 155 reticulum, tight junction, glycolysis, ribosome, and protein export (Fig. S2b). For individual SARS-

156
CoV-2 proteins, many pathways and biological processes are shared in these viral proteins (

161
Overall, our interactome is comprised of abundant information that can be utilized for the 162 identification of novel pathobiology and host-targeting therapies. We also developed an interactive 163 visualization tool for our interactome which can be accessed from https://github.com/ChengF-

165
This SARS-CoV-2-human interactome is of high coverage and quality

166
To ensure the authenticity when applying our interactome for downstream studies, we first 167 evaluated the quality through several means. We examined three previously published SARS-

168
CoV-2-human protein-protein interactome networks 9,11,12 . Importantly, all three of these 169 interactomes were generated using AP/MS-based methods alone. We found that only a few 170 interactions overlapped among these datasets ( Fig. 2a

177
We next performed several comparisons using these interactomes and our own. First, we 178 examined whether these datasets contained interaction partners that coincided with genes that 179 had expression changes in response to SARS-CoV-2 infection. To this end, we performed 180 differential expression analysis for several bulk and single-cell RNA-seq datasets (see Methods).

181
For the single-cell dataset 40 which we compared the gene expression in SARS-CoV-2 + and SARS-

182
CoV-2 -cells, we found that our interactome showed significant overlap (Fisher's exact test, P < 183 0.01) with the differentially expressed genes (DEGs) in more cell types than that of other 184 interactomes ( Fig. 2b and Table S2). Using four bulk RNA-seq datasets that contained samples 185 such as upper airway and bronchial epithelial cells [41][42][43][44] , we found that our interactome had a 186 comparable number of significant overlaps to other datasets and showed the highest overall 187 Jaccard index with the bulk RNA-seq datasets ( Fig. 2c and Table S2). These results suggest that 188 our interactome is highly enriched in genes differentially expressed in response to SARS-CoV-2 189 infection.

190
Gene expression patterns in disease-related tissues carry important information for 191 revealing the pathogenesis of the disease and identifying potential treatments [45][46][47] .We therefore 192 examined the expression of the human host factors in different tissues ( Fig. S4 and Table S3) 193 using the GTEx data 48 . By normalizing the expression of each gene across different tissues (tissue 194 specificity, see Methods), we found that lung ranked the 7 th out of 33 tissues in terms of the 195 number of host factors with positive tissue specificity (Fig. 2d), suggesting that lung is one of the 196 tissues where these host factors have high expression 49 .

197
We next inspected the evolutionary factors of the SARS-CoV-2 human host factors ( Fig.   198 2e-f and Table S4). Our SARS-CoV-2-human interactome showed more purifying selection

208
ORF3a is a SARS-CoV-2 accessory protein that has been reported to induce apoptosis in 293T 209 cells 52 and to suppress the innate immune response 53-55 via unclear molecular mechanisms. Our 210 interactome revealed that ORF3a physically interacts with ZNF579, a previously uncharacterized 211 human protein likely to be a transcription factor. We were able to validate this interaction using 212 co-immunoprecipitation (co-IP) western blotting (Fig. 3a). Furthermore, we found that the level of 213 ZNF579 protein is decreased after overexpression of ORF3a in 293T cells (Fig. 3b). As a result, 214 we hypothesized that the presence of ORF3a in cells might trigger changes in the transcriptional 215 state of human genes that are normally regulated by ZNF579. Using ENCODE ChIP-seq data 56,57 , 216 we identified that ZNF579 is bound strongly to the promoter of HSPA6 (Fig. 3c). We found that 217 overexpression of ORF3a in 293T cells causes massive induction of HSPA6 using qPCR (Fig.   218   3d). These results indicate that the multifunctional SARS-CoV-2 accessory protein ORF3a can 219 induce expression of HSPA6, presumably by disrupting ZNF579, which is likely to normally exert 220 repressive activity at the HSPA6 promoter. This represents an additional previously unknown 221 activity of this multifunctional viral accessory protein.

222
The oligosaccharyltransferase (OST) complex catalyzes the N-glycosylation of nascent 223 polypeptides in the endoplasmic reticulum 58 . Glycoproteins are critical for normal cell-cell interactions, RNA replication and pathogenesis [59][60][61] . Interestingly, OST inhibitor has been shown 225 to have activity against Dengue virus, Zika virus, West Nile virus, yellow fever viruses, and 226 HSV1 62-64 by affecting the viral replication. The OST complex was also found to be crucial for 227 innate immune responses triggered by lipopolysaccharide 65 . Notably, the OST complex subunits 228 STT3A/B, RPN1/2, and DDOST 66 were all shown to be present in our Y2H and AP/MS 229 interactome datasets, which we further validated using co-IP (Fig. 3e). Additionally, we also found 230 Sec61 ( Fig. 3f-g), which is a major component of the ER translocon that facilitates the entry of 231 nascent polypeptides into the ER lumen for protein processing 67

236
The SARS-CoV-2 nucleocapsid (N) protein binds to the viral RNA genome and is 237 multifunctional in viral RNA transcription, replication, and genome condensation [70][71][72] . N protein is 238 conserved and stable with ~90% amino acid homology to the SARS-COV N protein 73 . From our 239 dataset, we confirmed known interactions, including the stress granule core protein G3BP1/2 also 240 found in three other interactome datasets. In addition to these known interactions, we identified a 241 novel interaction between histone H1.4 and N protein. To validate this histone H1.4 and N protein 242 interaction, we overexpressed both N protein and histone H1.4 to perform co-IP, confirming their 243 interaction (Fig. 3h). Histone H1, also known as linker histone, mainly functions in chromatin 244 condensation and transcriptional repression 74,75 . Accumulating evidence suggests that linker 245 histone is essential in the pathogenesis of several diseases, particularly for viral infection 75 . There 246 is also evidence that Histone H1 could regulate IFN and inhibit influenza replication 76 , in addition 247 to playing a role in the regulation of viral gene expression 77 . Thus, we hypothesize that this novel 248 viral-host interaction could also be involved in mediation viral replication and/or gene expression.

250
In our previous studies, we demonstrated that by using the SARS-CoV-2 host factors and the 251 human protein-protein interactome, we could prioritize existing drugs for their anti-SARS-CoV-2 252 potential 78,79 . Using our newly discovered SARS-CoV-2-human protein-protein interactome 253 network, we performed network-based drug screening for more than 2,900 FDA-254 approved/investigational drugs (Fig. 4a). We obtained a list of 189 drugs with significantly small

262
Overall, these top drugs fall into several major categories, including anti-infective  Table S6). Other drugs, such as carvedilol and hydrochlorothiazide, may directly inhibit 272 viral entry by disrupting the Spike-ACE2 PPI ( Fig. S5 and Table S6).

273
In our previous efforts using existing SARS-CoV-2 interactomes and literature evidence,

294
To identify the drug-outcome relationships of these drugs, we used a state-of-the-art active 295 user-design approach 46 Table S7). The effect of carvedilol 302 was consistent for different race and sex subgroups ( Fig. 4c and Table S7). To validate these 303 observations, we used a second EHR database as an external validation set (168,712 total 304 individuals, 83,340 SARS-CoV-2 positive cases, Table S8). We found that carvedilol had a 305 sufficient number of usage cases for the drug-outcome evaluation. By comparing individuals with 306 and without carvedilol usages (PS-matched by age, sex, race, and/or comorbidities), we found 307 that carvedilol usage was associated with a 17% (OR = 0.83, 95% CI 0.78-0.88, P < 0.001) 308 significantly lowered risk of COVID-19 positive test (Fig. 4e). This effect was also consistent when 309 we examined subgroups from the registry in terms of race and sex (Fig. 4e).

310
We found that carvedilol not only showed favorable results in the EHR-based validation, 311 but also has a promising antiviral profile from NCATS, showing high potencies for multiple desired 312 activities (Fig. 4f). The NCATS profile of carvedilol is comparable to that of remdesivir, whose 313 profile was deemed highly desirable 85 (Fig. 4f). We then investigated carvedilol's anti-SARS-CoV-

324
In this study, we leveraged high-throughput Y2H and TMT-AP/MS to generate the first binary and   (Table S3). STON2 is ubiquitously 337 expressed and involved in endocytic machinery 39 . It is possible that SARS-CoV-2 can enter host 338 cells through binding of S protein not only to ACE2, NRP1 86,87 and BSG 88 , but also other 339 (unknown) factors such as CORO1C and STON2. We also noticed two proteins, EPPK1 89 and 340 SPECC1L 90 , that both express on the cell junctions. It has been suggested that SARS-CoV-2 341 could spread through cell-to-cell transmission 91 . These cell junction proteins that can be targeted 342 by SARS-CoV-2 S protein may facilitate its cell-to-cell transmission.

343
We noticed an overall high functional overlap of the SARS-CoV-2 proteins in terms of both 344 shared interacting partners (Fig. S1c) and shared enriched pathways and biological processes 345 (Fig. S3). This observation shows a high redundancy of SARS-CoV-2 proteins, which further 346 suggests that SARS-CoV-2 therapies that targets only specific viral protein or its host factors may 347 not be sufficient. This observation also justified the advantage and necessity of using network-348 based approach of drug discovery for SARS-CoV-2. As shown by the drug repurposing results, 349 the top drugs can potentially affect multiple SARS-CoV-2 host factors (Fig. 4b).

350
We identified a previously uncharacterized human transcription factor, ZNF579, that 351 interacts with SARS-CoV-2 accessory protein ORF3a, and report that this interaction leads to the de-repression of HSPA6. HSPA6 is a HSP70 family molecular chaperone, which are known to be 353 involved in the entry, replication, assembly, and release of various viral pathogens 92 . We 354 speculate that SARS-CoV-2 has evolved this activity to ensure sufficient levels of molecular 355 chaperones are available to assist with the production of viral proteins in cells.

356
Next, using this newly discovered SARS-CoV-2-human protein-protein interactome, we 357 performed drug repurposing and identified a list of top 21 drugs. We found that although some of 358 these drugs can directly target the host factors, most of them indirectly affect the host factors 359 through PPIs with their targets (Fig. 4b). Further, we have identified carvedilol and 360 hydrochlorothiazide as potential host-targeting treatments for COVID-19 supported by multiple

376
We acknowledge several limitations. The network-based SARS-CoV-2 treatment 377 discovery may be affected by the incompleteness of the human protein-protein interactome and 378 drug-target network. Therefore, we relied not only on the network discoveries, but also 379 incorporated other types of evidence, such as EHR-based validation and experimental validation.

380
Our EHR-based validation is retrospective and can only be applied to commonly used drugs due 381 to data availability. Although we adjusted for several confounding factors, other unknown factors 382 may still have effect on the results of EHR-based validation. Therefore, the drugs identified in this 383 study must be validated using randomized clinical trials before they can be used in patients with

562
The nonsynonymous and synonymous substitution rate ratio (dN/dS ratio) 96

575
The human protein-protein interactome and the drug-target network were used to screen for drugs 576 against the SARS-CoV-2 host factors.

588
The drug-target network was constructed using several data sources as described in our 589 recent studies 45

593
Network proximity-based drug and drug combination screening

594
The "closest" network proximity measure was used to screen for 2,938 FDA approved or 595 investigational drugs. The "closest" distance for two gene/protein sets (e.g., drug targets) 596 and (e.g., SARS-CoV-2 host factors) was calculated as: where ( , ) is the shortest path length of and in the human protein-protein interactome.

599
Network proximity was further normalized to obtain a Z score using a permutation test with 600 randomly selected proteins from the interactome with similar degree distributions to and .

602
The antiviral profiles of the prioritized drugs were retrieved from NCATS    Table 1.

634
The second dataset (external validation dataset) was an institutional review board-

685
The authors declare that there are no competing interests.