key: cord-0719161-9ft2f7x6
authors: Pal, Lipika R.; Cheng, Kuoyuan; Nair, Nishanth Ulhas; Martin-Sancho, Laura; Sinha, Sanju; Pu, Yuan; Riva, Laura; Yin, Xin; Schischlik, Fiorella; Lee, Joo Sang; Chanda, Sumit K.; Ruppin, Eytan
title: Synthetic lethality-based prediction of anti-SARS-CoV-2 targets
date: 2022-04-27
journal: iScience
DOI: 10.1016/j.isci.2022.104311
sha: bc1bde077ef6cedebccd39ff722ce1f3b999ea48
doc_id: 719161
cord_uid: 9ft2f7x6

Novel strategies are needed to identify drug targets and treatments for the COVID-19 pandemic. The altered gene expression of virus-infected host cells provides an opportunity to specifically inhibit viral propagation via targeting the synthetic lethal and synthetic dosage lethal (SL/SDL) partners of such altered host genes. Pursuing this disparate antiviral strategy, here we comprehensively analyzed multiple in vitro and in vivo bulk and single-cell RNA-sequencing datasets of SARS-CoV-2 infection to predict clinically relevant candidate antiviral targets that are SL/SDL with altered host genes. The predicted SL/SDL-based targets are highly enriched for infected cell inhibiting genes reported in four SARS-CoV-2 CRISPR-Cas9 genome-wide genetic screens. We further selected a focused subset of 26 genes that we experimentally tested in a targeted siRNA screen using human Caco-2 cells. Notably, as predicted, knocking down these targets reduced viral replication and cell viability only under the infected condition without harming non-infected healthy cells.

The coronavirus disease 2019 pandemic caused by the novel coronavirus SARS-CoV-2 together with its emerging new variants in 2022 has resulted in hundreds of millions of infected people with millions of deaths worldwide (WHO Coronavirus (COVID-19) dashboard, 2022) . Up to this date, there are several thousand registered clinical trials of anti-COVID-19 therapies worldwide, (International Clinical Trials Registry Platform (ICTRP) , 2022), with remdesivir being the only drug approved by the United States Food and Drug Administration (FDA) and the drug regulatory agencies of several other countries (Beigel et al., 2020) . However, the clinical benefit of remdesivir in most COVID-19 patients is still modest (Pan et al., 2021) . A total of 14 therapies, including virus-neutralizing monoclonal antibodies bamlanivimab and etesevimab (Mahase, 2020) , sotrovimab, the combination of casirivimab and imdevimab, interleukin-6 receptor inhibitor Tocilizumab and Janus kinase (JAK) inhibitor baricitinib (in combination with remdesivir), molnupiravir, the combination of nirmatrelvir and ritonavir have obtained Emergency Use Authorization (EUA) from the U.S. FDA (U. S. Food and Drug Administration, 2021). Dexamethasone and other corticosteroids have been recommended for the control of COVID-19-related symptoms (Ledford, 2020) . Numerous preclinical studies have conducted anti-SARS-CoV-2 drug repurposing as well as genetic screening of varied scales, for example, (Daniloski et al., 2021; Riva et al., 2020; Wei et al., 2021) , supplemented by other studies providing computational target predictions, for example, (Cheng et al., 2021a; Cheng et al., 2021b; Tilocca et al., 2020; Zhou et al., 2020a; Zhou et al., 2020b) . Two mRNA vaccines from Pfizer-BioNTech and Moderna have been approved by the U.S. FDA (U. S. Food and Drug Administration, 2021) . Despite all these efforts, there is an urgent unmet need to identify additional anti-SARS-CoV-2 targets and drugs, to resolve the COVID-19 crisis (Jomah et al., 2020) .

Existing COVID-19 antiviral strategies are either directly targeting the virus or indirectly targeting the virus via host modulation. While the direct targeting of the viral life cycle to inhibit its proliferation remains challenging (Pan et al., 2021) , numerous recent studies have contributed to the understanding of the biology of SARS-CoV-2 infection, for example, (Bojkova et al., 2020; Forster et al., 2020; Paules and Fauci, 2021; V'Kovski et al., 2021) , to pursue host-directed therapies. A lot of interest has been focused on targeting ACE2 and TMPRSS2, the virus entry receptors (Muralidar et al., 2021; Ragia and Manolopoulos, 2020) and the search for drugs targeting these genes is underway Lin et al., 2021) . Majority of these hostdirected antiviral strategies aim to protect host cells from virus-induced death, including our J o u r n a l P r e -p r o o f recently published strategy of rescuing the infected host cells via targeting the host metabolism (Cheng et al., 2021a) . In contrast to these "host cell protection" approaches, a conceptually different "host cell inhibition" strategy has been proposed and demonstrated with poliovirus . In this latter strategy, the aim is to selectively inhibit the viability or even kill the virus-infected host cells (preferentially in the early stages of viral infection), thereby diminishing the capacity of these cells to sustain viral proliferation and propagation. This can be achieved utilizing synthetic lethality (SL), a type of genetic interaction where reduced viability or cell death occurs upon the inactivation of a pair of genes but not by either gene alone.

More specifically, viruses are known to alter the expression and activity of many host genes, and by identifying and targeting the "SL partner genes" of genes downregulated by the virus, the virusinfected cells can be specifically targeted, sparing non-infected healthy cells. A variant of SL, called synthetic dosage lethality (SDL) where decreased cell viability occurs only upon the overexpression of one gene and the inactivation of a second gene (Kroll et al., 1996) , can similarly be utilized to specifically target the infected cells based on the host genes that are upregulated by the virus. While we infer the SL and SDL partners of genes differentially expressed in infected vs control cells, we group them together (annotated as SL/SDL genes) in the following downstream analyses as we are searching for antiviral targets that are directly altered by the virus, whose inhibition (irrespective if they are SL or SDL) could selectively kill infected cells.

Adopting this distinct SL/SDL-based approach (see Figure 1a) , we systematically identify candidate targets for selective killing of host infected cells and further test the predicted candidate targets with in vitro assays.

Previously, we have developed a computational pipeline named ISLE (identification of clinically relevant synthetic lethality) to identify clinically relevant SL/SDL genetic interactions by mining large-scale genomic and transcriptomic data across human cancers (Lee et al., 2018) .

Applying ISLE, we have previously successfully identified a Gαq-driver mutation as marker for FAK inhibitor in uveal melanoma and a synergistic combination of asparaginase and MAPK inhibitors for treating melanoma and pancreatic cancer (Pathria et al., 2019) . Although the ISLE algorithm was originally designed to predict SL in cancers, we have recently found that some SL gene pairs predicted by ISLE are also functional in non-cancerous tissues (Cheng et al., 2021b) , reflecting the conservation of the wiring of molecular networks of human cells more generally. Therefore, we reasoned that ISLE can also be valuable in predicting SL gene pairs for non-cancer contexts. Adopting this conjecture, here we applied ISLE to a large cohort of both in vitro and in vivo SARS-CoV-2 RNA-sequencing datasets to identify SL/SDLbased anti-SARS-CoV-2 targets. We then set to combine these predictions with the results of J o u r n a l P r e -p r o o f large-scale SARS-CoV-2 in vitro genome-wide CRISPR and siRNA screens. Focusing on a selected list of 26 predicted targets, we tested them experimentally in a small-scale targeted siRNA screen and demonstrated their efficacy in inhibiting SARS-CoV-2 replication. As these targets are supported by analyzing in vivo data, these are more likely to be of translational relevance.

Our analysis proceeds in four steps: (A) First, we analyzed multiple published host cell gene expression datasets (Figure 1b) involving in vitro and in vivo human samples of SARS-CoV-2 infection to identify the strongest significantly differentially expressed (DE) genes in response to viral infection. (B) Second, applying the ISLE pipeline, we identified the most robust clinically relevant SL/SDL partner of these DE genes in human cells, as potential SL/SDL-based anti-SARS-CoV-2 targets. We show that these 454 predicted targets (Figure 1b , module 1) are enriched with genes whose knockout strongly reduced cell viability specifically in several SARS-CoV-2 infected CRISPR-Cas9 genetic screens, and thus could be potentially targeted to selectively kill infected cells and reduce viral spread. (C) We further filtered and prioritized a smaller list of 140 predicted targets based on support by several CRISPR screens (Figure 1b, module 2), and a high throughput genome-wide siRNA screen conducted and analyzed in our lab (Figure 1b, module 3). (D) Finally, we employed a network-based simulated annealing algorithm to further prioritize 26 highly ranked targets (Figure 1b , module 4), and conducted a small-scale targeted siRNA screen to experimentally test these targets in greater depth. A stepwise detailed description of the analysis performed, and the results obtained in each step follows.

To identify the host cell genes whose expression is altered upon SARS-CoV-2 infection, we collected and analyzed multiple transcriptomic datasets of SARS-CoV-2 infection (Figure 1b) .

These include bulk RNA-seq data of in vitro samples from Vero E6 cells (Riva et al., 2020) and human A549 cells with exogenous ACE2 expression (A549 ACE2 ) (Blanco-Melo et al., 2020) , as well as in vivo nasopharyngeal swab samples from COVID-19 patients (Butler et al., 2021; Lieberman et al., 2020) . For in vivo analysis we additionally used single-cell RNA-seq data of nasopharyngeal and bronchial samples (Chua et al., 2020) and bronchoalveolar lavage fluid J o u r n a l P r e -p r o o f (BALF) samples (Liao et al., 2020) . For each of these datasets, we identified the significantly differentially expressed (DE) genes in the SARS-CoV-2-infected samples vs matched virusnegative controls (FDR<0.05 with additional dataset-specific criteria, see STAR Methods). For the single-cell datasets, we performed the DE analysis only in the epithelial cells, as these represent a primary target for SARS-CoV-2 infection, thus being our focus for SL/SDL-based antiviral strategy. Comparing the identified DE genes across datasets with Fisher's exact tests (STAR Methods), we find varied but overall higher-than-random pairwise similarities (i.e. odds ratio >1, Figure 1c ). The lowest odds ratio (2.7, adjusted P=2.66E-03) is found between the in vitro A549 ACE2 cell dataset and the in vivo single-cell RNA-seq samples (the Chua et al. dataset) , and the highest odds ratio (12.4, adjusted P=1.33E-38) is seen between the in vivo single-cell (the Liao et al. dataset) and in vivo bulk swab datasets. Reassuringly, the DE genes identified from the in vivo bulk and single-cell RNA-seq datasets have reasonably good overlaps ( Figure   1c ). We performed pathway enrichment analysis of the DE genes from each dataset (STAR Methods; Table S1 ) and noticed that although these genes are different across datasets, they are enriched for functionally similar pathways (e.g., platelet aggregation plaque formation pathway in human A549 ACE2 dataset and platelet activation signaling and aggregation in in vivo single-cell Chua et al. dataset ; Table S1 ).

Next, we predicted the SL/SDL partners of the identified DE genes, as they constitute potential antiviral targets that can selectively impair the function of infected host cells and thus limit their capacity to support viral proliferation (list of SL/SDL pairs of each dataset are given in Table S2 ).

Specifically, we applied the ISLE algorithm (Lee et al., 2018) to predict the SL partners of the genes significantly down-regulated by SARS-CoV-2 infection in the datasets described above.

For the up-regulated genes, we analogously predicted their SDL partners with a modified version of ISLE (STAR Methods). Interestingly, the SL/SDL partners identified across datasets with FDR cutoff 0.1 exhibit overall higher similarities than those similarities observed among the DE genes, with a minimum pairwise odds ratio of 4.3 (adjusted P=1.48E-118; Figure 1d ). In particular, the overlap in the SL/SDL partner genes between the in vitro A549 ACE2 cell dataset and the in vivo single-cell RNA-seq samples (in Chua et al. dataset) is higher than the overlap observed between their corresponding DE genes (SL odds ratio of 4.3 compared to corresponding DE odds ratio of 2.7). This suggests that the DE genes identified across the datasets may be functionally similar, resulting in higher chances of forming SL/SDL interactions with the same genes, which is indeed supported by the pathway enrichment results of the DE genes as described above (Table S1 ).

Reassured by the overall similarity of the SL/SDL partner genes across datasets, we then selected a subset of 454 SL/SDL partner genes that is common to all in vitro and in vivo datasets (Table S2F , see details in STAR Methods; illustrated in Figure 1b , module 1), which we focus on further analysis.

To further filter the 454 predicted SL/SDL-based targets, we collected data from two published CRISPR-Cas9 genetic screens that aim to identify host genes modulating SARS-CoV-2 replication. These include a screen in the Vero E6 cell line (Wei et al., 2021) and another in the human A549 ACE2 cell line at two different multiplicities of infections (MOIs; Daniloski et al., 2021) .

We note that these studies seek to identify targets whose knockout promotes cell survival by reducing the virus-induced cytopathic effects, while our strategy aims to identify genes whose knockout specifically reduces the viability of infected cells. Thus, we expect that our predictions would be enriched in genes whose knockout results in strong cell dropouts in these genetic screens. Indeed, we find that our 454 predicted SL/SDL-based target genes are significantly enriched for cell viability-reducing genes (adjusted P values from Fisher's exact tests range from 2.38E-02 to 4.00E-04, Figure 1e ; STAR Methods). The 454 SL/SDL-based targets are also enriched for host proteins that interact with SARS-CoV-2 viral proteins (Fisher's exact test adjusted P=1.10E-02, Figure 1e ; data of virus-interacting proteins is from (Gordon et al., 2020; Stukalov et al., 2021) , indicating that the predicted targets are closely relevant to SARS-CoV-2 biology and infection. Among the 454 predicted targets, 140 genes are significantly strong cell viability-reducing genes (Wei et al., 2021, P=1E-05; Daniloski et al., 2021, P=0 .002 for MOI1 and P=0.02 for MOI3) when depleted in both CRISPR-Cas9 screen datasets (illustrated in Figure 1b , module 2; Figure 1e ). These genes are also significantly enriched amongst the genes whose knockout reduced cell viability in two other recently published genome-wide CRISPR-Cas9 knockout screens in human hepatoma cells P=1.09E-04; and (Schneider et al., 2021) P=3.73E-03 for the screen at 33º C and P=4.95E-04 for the screen at 37º C; Figure   1f ). We thus focused our further analysis on these 140 candidate targets as listed in Table S3 , ranked by a weighted DE score (WDS) that considers both the number of DE genes genetically interacting with each SL/SDL partner and the DE fold change (STAR Methods).

Among those 140 targets, those that are inhibited by existing drugs are listed in Figure 2a ( for Covid-19 patients. A pathway enrichment analysis of these 140 SL/SDL-based targets (STAR Methods, Table S3E ) shows enrichment in pathways including cell cycle processes, stress responses, DNA break repair processes, RNA metabolic processes and RNA polymerase II transcriptions (Figure 2b) . The DE genes are enriched in pathways including complement cascade, ion channel transport, innate immune processes, cytokine signaling, interferon signaling and neutrophil degranulation (Figure 2c , Table S3F ). A pairwise pathway enrichment analysis (STAR Methods; Figure 2d ) shows that many SL/SDL interactions include cell-cell communications, innate immune system, post translational protein modification (i.e., columns with high numbers of significant P values in Figure 2d , Table S3G ). The pathway combination involving the largest number of gene pairs is between the innate immune system (DE genes) and the cell cycle (SL/SDL partners) (involving 9 pairs, adjusted P=5.63E-04). These results suggest that the cellular immune responses induced by SARS-CoV-2 infection can potentially control the cells fate when targeted in conjunction with the central pathways regulating cell proliferation and nucleic acid/protein metabolism.

In the screens analyzed above, the cell viability-decreasing effects observed are potentially confounded by virus-induced cytopathic effects. We hence additionally analyzed a very recent imaging-based genome-wide siRNA screen that we performed previously in human Caco-2 cells (data deposited at https://figshare.com/s/4117ac39b1d21b56f5e6), where SARS-CoV-2 replication and cell number after each individual target gene knockdown were measured independently via anti-SARS-CoV-2 antibody and DAPI staining, respectively. 135 of our 140 predicted candidate genes were covered within the siRNA library of this screen (Table S3H ). First, across these genes, we found that knockdowns resulting in lower cell numbers tend to also have lower percentage of SARS-CoV-2-positive cells (Spearman's ρ=0.16, P<2.2e-16), supporting the notion that reducing the viability of the virus-infected cells can indeed inhibit virus replication.

Second, we find that the knockdown of these 135 predicted genes indeed results in significantly lower cell numbers compared to the knockdown of the rest of the genes in the screen (one-sided Wilcoxon rank-sum test P=0.002, Figure 2e ). Reassuringly, within the 135 SL/SDL candidates, the candidate genes with lower cell viability (relative cell number <1.0) have higher weighted DE J o u r n a l P r e -p r o o f score (one-sided Wilcoxon rank-sum test P=0.032), and similarly, the candidates with lower percentage of SARS-CoV-2-positive cells (z score<0) also have higher weighted DE score (onesided Wilcoxon rank-sum test P=0.057; STAR Methods). These results suggest that the SL/SDLbased predictions and the associated weighted DE score measuring the strength of SL/SDL effects can facilitate the identification of anti-SARS-CoV-2 targets that function via specifically reducing the viability of the infected host cells.

Next, we aimed to further narrow down the list of targets that we should further test in an additional, dedicated genomic screen. To this end, we set to first identify the subset of SL/SDL interactions with higher relevance to SARS-CoV-2 infection. Using simulated annealing we identified an optimal SARS-CoV-2 specific SL subnetwork that maximizes the correlation between viral inhibition scores measured in the genome-wide siRNA screen conducted in house and the SL-based weighted DE scores of the target genes it includes (see STAR Methods). To experimentally validate the effect of the resulting 26 candidates on SARS-CoV-2 replication and cell viability, we performed an additional imaging-based siRNA assay in human intestinal epithelial Caco-2 cells. These cells were selected for these assays as they harbor endogenous expression on ACE2 and TMPRSS2 and are highly permissible to SARS-CoV-2 infection, they are amenable to high-throughput siRNA transfection, and the gastrointestinal epithelium is a target for SARS-CoV-2 infection in vivo (Xiao et al., 2020) . The individual knockdown of each of the 26 genes was performed in a 384-well plate format using four replicates per target (Figure 3a) . The percentage of virus-infected cells after target knockdown (reflecting viral load) was quantified via immunofluorescence staining with anti-SARS-CoV-2 antibody. Non-targeting (scrambled) siRNAs were used as negative control, and siRNAs targeting ACE2 and TMPRSS2 (two genes essential for viral entry) as positive controls (Figure 3b ; STAR Methods). The total cell number after gene knockdown (reflecting cell viability) was measured based on DAPI staining, using siRNAs targeting essential genes for the survival of the host cells (Toxic siRNA) as positive controls for cell viability (Figure 3c) .

We first established the antiviral efficacy of the predicted targets -individual knockdown of the 26 predicted SL/SDL targets resulted in an overall reduction in viral replication (P=6.57E-07; Figure 3b ) and reduced cell viability (P=0.047; Figure 3c ) compared to scrambled siRNAs.

Of note, the knockdown of ACE2 and TMPRSS2 also markedly reduced viral replication, consistent with their previously recognized potential as effective drug targets (Hoffmann et al., J o u r n a l P r e -p r o o f 2021; Lin et al., 2021) . Second, we tested if knockdown of the predicted targets indeed selectively reduces cell viability in infected cells but not in non-infected cells. To this end we quantified the number of DAPI-positive cells after knockdown of the 26 predicted SL/SDL targets in both infected and non-infected conditions (Figure 3a) . Reassuringly, we found that the cell numbers after knockdown of the SL/SDL targets, including VKORC1, MED8, and EIF4G1 but not of the non-SL target ACE2, were significantly reduced in the infected compared to non-infected conditions (P=3.302E-11, Figure 3d -e), testifying to the selective killing induced by the predicted SL/SDL targets. Overall, 24 out of 26 targets showed reduced viral load compared to the control and reduced cell viability in the infected condition compared to non-infected state (Figure 3e ; Table   S4A ).

Among our 26 predicted targets, the top three emerging in our screen are VKORC1, an anti-coagulant that binds to ORF7a of SARS-CoV-2 (Holcomb et al., 2020; Janssen and Walk, 2020) and targeted by warfarin (Irwin et al., 2021) ; MED8, a gene required for the activation of transcription (Boube et al., 2002) and EIF4G1, known for its role in cell growth, proliferation and differentiation (Ramírez-Valle et al., 2008) (Figure 3d) . Some of the validated targets are known to be involved in the context of other viral infections (Table S4B )such as, CDC25B promotes influenza A virus replication (Cui et al., 2018) .

Given the current lack of effective treatment options for COVID-19, identifying new drug targets for the virus causing this rapidly spreading disease with emerging new variants is highly critical.

In this work, we predicted potential clinically relevant SL/SDL-based anti-SARS-CoV-2 targets.

We further tested our initial list of 454 emerging predictions using two published CRISPR-Cas9 genetic screens in two different cell lines (Vero E6 and human A549 with exogenous ACE2 expression) infected with SARS-CoV-2. Intersecting our predicted targets with the hits from the CRISPR screens, we further narrowed down this initial list to that of 140 clinically relevant candidate targets. A network-based analysis of these 140 targets using the results of a genomewide siRNA screen that we have previously performed, led to the selection of a smaller subset of 26 genes, which we further experimentally tested and validated via a small-scale targeted siRNA screen in human Caco-2 cells, identifying targets that selectively reduce cell viability only in infected cells, demonstrating the utility of our synthetic lethality approach. Our analysis involves the integration of multiple published gene expression datasets of SARS-CoV-2 infection, both in J o u r n a l P r e -p r o o f vitro and in vivo, finding overall a moderate but robust level of concordance of virus-induced DE genes, especially and importantly across the in vivo datasets. This analysis encompasses some of the largest available transcriptomic datasets of COVID-19 patients (Butler et al., 2021; Lieberman et al., 2020) , as well as single-cell datasets, which adds to the clinical relevance of our predicted targets. Many additional in vitro transcriptomics datasets on SARS-CoV-2 infection are available (like in human hepatoma cells, or in non-small-cell lung cancer cells), but we have chosen to build up our 140 targets list based on the two particular datasets of Vero E6 and A549 ACE2 cells, as the same cell lines were used in the two CRISPR-Cas9 genetic screens that we used for validation. Reassuringly, we identified the significantly differentially expressed (DE) genes in the SARS-CoV-2 infected samples vs matched controls in Calu3 cell lines (available from (Blanco-Melo et al., 2020) , as we did for A549 ACE2 and Vero E6 cells. We then did an enrichment test for the overlapping DE genes between Calu3 and A549 ACE2 , and Calu3 and Vero E6 cell lines (odds ratio and adjusted p-values computed using the Fisher exact test; similar to the analysis done in Fig. 1c) and found higher-than-random pairwise enrichment. Therefore, the relevant transcriptomic analyses are comparable between these cell lines.

Our study is motivated by the SL-based antiviral strategy proposed by with preliminary proof-of-concept application for poliovirus demonstrated in . This strategy aims to specifically reduce the viability or impair the normal functioning of the infected host cells, thereby suppressing their capacity to support viral proliferation . To tease apart these two factors, we specifically employed an imaging-based siRNA assay where both viral replication and cell counts after gene knockdown were measured. In our previously reported genome-wide screen using this method, a positive correlation between these two variables was observed, supporting the inhibition of host cell viability as a potential antiviral strategy. In the targeted siRNA screen conducted in this study, a similar correlation was observed, where the knockdown of our 26 candidate targets significantly reduced viral replication accompanied by a reduction in cell number. However, importantly, knocking down these genes did not harm the non-infected cells. This reduction in cell viability in infected cells is likely not driven by virus-induced cytopathic effects (CPE), as cells treated with the negative control scrambled siRNA showed significantly higher rates of infection and higher cell number counts than cells treated with targeting siRNAs (Figure 3b-d) . The Pearson correlation value between log2FC infection and cell number in infected cells was in the positive range (p value = 0.1235).

Further in vivo validations are required to translate these in vitro validated predictions to the clinic.

In summary, our study represents a first of its kind proof-of-concept application of a synthetic lethality inference approach to predict anti-SARS-CoV-2 targets in a systematic genome-wide manner. We provide a list of potential targets that are supported by multiple lines of evidence, importantly, emerging from the integrative analysis of both in vitro and in vivo data, giving rise to the hope that they then are more likely to be translationally relevant and help in the development of new drugs for treating COVID-19.

To predict the SL/SDL partner genes, we applied the ISLE algorithm which was originally designed to identify synthetic lethal partners in the cancer setting (Lee et al., 2018) . Therefore, an important assumption underlying our approach is that many of the SL/SDL interactions identified by ISLE for pan-cancer analysis (i.e., SL/SDLs which are functional across many cancer types) are also functionally relevant in normal non-cancerous cells due to conservation in the cellular network. We have demonstrated evidence supporting this assumption in a recent study (Cheng et al., 2021b) , although not all partners identified by ISLE are functional outside the context of cancer. At least partially mitigating this limitation, the final 140 candidates target list we reported represents a high-confidence subset filtered based on COVID-19 patient data supported by the CRISPR-Cas9/siRNA genetic screens in the SARS-CoV-2 setting, and these candidates also have considerable literature support for their relevance to SARS-CoV-2. Of 

ER is a co-founder of Medaware, Metabomed, and Pangea Therapeutics (divested from the latter). ER serves as a non-paid scientific consultant to Pangea Therapeutics, a company developing a precision oncology SL-based multi-omics approach. The other authors declare no competing interests. 

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Eytan Ruppin (eytan.ruppin@nih.gov).

This study did not generate new unique reagents.

This paper analyzes existing, publicly available data. These accession numbers for the datasets are listed in the key resources 

Caco-2 (ATCC HTB-37) are epithelial cells isolated from colon tissue from a 72-year-old, white, male with colorectal adenocarcinoma. Cells were cultured in Eagle's Minimum Essential Medium 

Several RNA-sequencing datasets of SARS-CoV-2-infected and matched control samples were obtained as follows: Vero E6 cells (Riva et al., 2020) (Chua et al., 2020; Liao et al., 2020) , we only focused on epithelial cells (based on the cell type annotation provided by the authors) as the virus-infected cells relevant to our target prediction approach and performed DE analysis for virus-positive vs negative samples using the Seurat R package (Stuart et al., 2019) , specifically, for nasopharyngeal swab and bronchial basal cells from (Chua et al., 2020) and bronchoalveolar lavage fluid epithelial cells from (Liao et al., 2020) . Significant DE genes (FDR<0.05) with absolute log fold-change values greater than 0.5 were selected, resulting in 166 down and 181 up-regulated genes from the (Chua et al., 2020) dataset and 196 down and 320 up-regulated genes from the (Liao et al., 2020) dataset.

For the significant DE genes identified from each of the above datasets, we identified their SL partners (for down-regulated genes) using the ISLE algorithm (Lee et al., 2018) (Table S2) . ISLE was originally designed for SL prediction in the cancer setting, but as discussed in the main text here we apply it to identify potential SLs for antivirus. ISLE involves four steps analyzing gene essentiality data in cancer cell lines, tumor molecular profiles, cancer patient survival data, and gene phylogeny data to identify clinically relevant SL interactions (described in detail in Lee et al., 2018) . For the up-regulated DE genes, we instead identified their SDL partner genes also using ISLE but with minor modifications to account for the directional difference between SL and SDL.

For example, in the third step of ISLE analyzing cancer patient survival data, while the original ISLE algorithm tests for the association of concomitant down-regulation of a pair of genes with better patient survival as a feature of SL, in the modification for SDL we instead tested for the association of down-regulation of one gene and the up-regulation of another gene in a gene pair with better patient survival, which corresponds to the SDL definition. We used FDR cutoff 0.1 to select significant SL/SDL partners. We ranked the ISLE-predicted partner genes by their P values from the patient survival analysis. Significant partner genes (a total of 454 genes, Table S2F ) that are common to all analyzed datasets were taken for further downstream analysis.

We obtained two currently available genome-wide CRISPR-Cas9 genetic screening data aiming to identify host genes regulating SARS-CoV-2 infection, one from Vero E6 cell line (Wei et al., 2021) and the other from human alveolar basal epithelial carcinoma A549 cell line with exogenous ACE2 expression (A549 ACE2 ; Daniloski et al., 2021) . The complete results of sgRNA differential expression were obtained from the Supplementary Materials of the two publications. To validate that our predicted targets are indeed SL with SARS-CoV-2 infection, i.e., the inhibition of these genes can lead to more cell dropout in the virus-infected cells than in control cells, we tested for enrichment of genes showing negative log fold-changes in the CRISPR-Cas9 screens with the GSEA method (Subramanian et al., 2005) as implemented in the R package fgsea (Korotkevich et al., 2021) . Specifically, the log fold-change values across genes from each CRISPR-Cas9 screen were ranked, and GSEA was applied to the ranked list with the "gene set" being the 454 predicted SL/SDL-based targets common to all datasets. To obtain our final list of 140 candidate targets, we intersect the 454 consensus predictions with the genes showing the strongest negative log fold-changes in the CRISPR-Cas9 screens. These strongest CRISPR hits were selected as follows: In the Vero E6 dataset, we selected 680 genes with negative mean log2 foldchanges with FDR<0.1. The A549 ACE2 dataset contains results from two screens with different multiplicities of infections (MOIs), and we selected 2811 genes from the lower MOI screen and 1653 genes from the higher MOI screen whose absolute log fold-changes are lesser than -0.5.

The intersection of all selected CRISPR genes with the 454 consensus SL/SDL-based targets gives the final list of 140 candidate target genes ( Table S3) .

Each of 140 candidates have the same or different significant DE genes from six different transcriptomic datasets -in vitro Vero E6 (Riva et al., 2020) , in vitro A549 ACE2 cells (Blanco-Melo et al., 2020) , two in vivo Bulk Swab datasets (Butler et al., 2021; Lieberman et al., 2020) and two in vivo single-cell RNA-seq datasets (Chua et al., 2020; Liao et al., 2020) . We hypothesized that the effect of knockdown of these SL/SDL-based targets is dependent upon the synthetic lethality strength, as reflected by the extent of perturbation of DE genes by the virus (log fold change) in their respective transcriptomic datasets. We considered the median absolute log fold change of a DE gene in all six datasets (if it is available). If a DE gene is down-regulated in the dataset where it is considered to be significant, then all down-regulated log fold changes in other datasets will contribute to the final median value, and similar will be the treatment for up-regulated cases.

For a candidate gene g, its weighted DE score (Sg) was calculated as in equation 1, J o u r n a l P r e -p r o o f

We also tested for the enrichment between the 454 predicted targets with human proteins reported to interact with SARS-CoV-2 viral proteins (Gordon et al., 2020; Stukalov et al., 2021) (Fabregat et al., 2018) collected from MSigDB v7.2 (Liberzon et al., 2011) . Pathways with more than three genes overlapping with the candidate target set were considered. P values across pathways were corrected for multiple testing with the Benjamini-Hochberg method.

Fisher's exact tests were used to test the enrichment of the SL/SDL gene pairs in each possible pathway combination (i.e. "pathway pair") ( Table S3G) . Specifically, pathway annotation of single genes from the Reactome database (Fabregat et al., 2018) collected from MSigDB v7.2 (Liberzon et al., 2011 ) was used to create all possible pathway pairs and their member gene pairs. The "background" set for the enrichment was the gene pairs formed between the virus-induced DE genes and all genome-wide genes. P values were corrected with the Benjamini-Hochberg method.

The predicted targets (140 candidates) were mapped to known drugs (clinically approved and experimental/ investigational) that target (i.e. inhibit) them using drug target annotation from DrugBank (Wishart et al., 2018) . We manually searched the literature on each of the 26 candidate genes to assess their relevance to SARS-CoV-2 or other viral infection based on prior knowledge.

Detailed annotation with regard to literature support for the candidate genes are provided in Table   S4B .

We used our previously performed genome-wide siRNA screen, carried out in human Caco-2 cells to identify host cell factors that affect the replication of SARS-CoV-2 (data deposited at https://figshare.com/s/4117ac39b1d21b56f5e6).

Out of 140 candidate genes, we have cell viability and viral replication information for 135 SL/SDLbased targets. The size of our synthetic lethal network for 135 candidates including DE genes from all six different transcriptomic datasets was 854. We used the simulated annealing method to maximize the correlation between viral inhibition from our genome-wide siRNA screen and the SL/SDL-based weighted DE scores, which resulted in an optimal SARS-CoV-2 specific SL subnetwork. We finally selected 26 target genes from this subnetwork where percentage of SARS-CoV-2-positive cells was reduced (z-score < 0) in the previously performed genome-wide siRNA screen.

'Optim' function in R was used to optimize the whole SL network, using variable network sizes and seed and using 10,000 iterations for each run. All other parameters were used as default in the function. Different sizes of networks were optimized for minimization of correlation (Spearman's ρ in R) between weighted DE score and cell viability from siRNA screen for candidate genes present in that network. The resulting SL network was used to calculate the correlation between weighted DE score and percentage of SARS-CoV-2-positive cells of those candidate genes in the selected network. The aim of this in silico experiment was to extract a SARS-CoV-2 specific SL network where cell viability and viral replication will be lower upon knocking down the candidate genes in that network.

Screen. To evaluate the effect of the 26 predicted targets on viral replication and cell viability, a targeted siRNA screen was conducted using human Caco-2 cells (ATCC). siRNAs targeting these genes were obtained from the Dharmacon ON-TARGETplus SMARTpool library and individually arrayed in 384-well plates at a concentration of 12.5 nM per well. First, siRNAs were incubated with 0.1 L Lipofectamine RNAiMAX transfection reagent (ThermoFisher) diluted in 9.90 l Opti-MEM media (ThermoFisher) for 20 min at room temperature. Then, 3,000 Caco-2 cells diluted in cell growth media (see below) were seeded on top and incubated for 48 h at 37°C with 5% CO2

conditions. Cells were then either infected with SARS-CoV-2 (isolate USA-WA1/2020) at a multiplicity of infection (MOI) 0.1 or left non-infected. After an incubation period of 48 h at 37°C with 5% CO2 conditions, cells were fixed with 4% PFA (Boston BioProducts) for 4 h at room temperature. Cells were then washed twice with PBS prior to 20 min permeabilization with 0.5% Triton X-100 and 1 h blocking with 3% BSA (Sigma) at room temperature. Cells were then incubated for 2 h at room temperature with primary anti-SARS-CoV-2 N protein rabbit polyclonal J o u r n a l P r e -p r o o f antibody (kind gift from Dr. Adolfo Garcia-Sastre), followed by three PBS washes and 1 h incubation with Alexa Fluor 488-conjugated anti-rabbit secondary antibody (A-11008, ThermoFisher) diluted in 3% BSA. Cells were washed three times with PBS and then stained with DAPI (4,6-diamidine-2-phenylindole, KPL). Plates were sealed and stored at 4°C until imaging.

Caco-2 cells growth media: 40 L DMEM media (Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS, Gibco), and 50 U/mL penicillin -50 µg/mL streptomycin (Fisher Scientific).

Analysis. The resulting number of cells and SARS-CoV-2 replication after each individual target knockdown was measured using high-content imaging. Images were acquired using the IC200

imaging system (Vala Sciences), and then analyzed using the analysis software Columbus v2.5

(Perkin Elmer). The number of DAPI + objects was used to calculate the number of cells in the well, and the number of SARS-CoV-2 + to calculate the percentage of infected cells. The average percentage of infected cells for each target knockdown (n=4) was normalized to that of the negative control scrambled and used to calculate the log2FC infection.

Differential expression analyses were performed using R library DEseq2 (Love et al., 2014) . Gene enrichment analysis was done using R package fgsea (Korotkevich et al., 2021) . Fisher's exact test was performed using the fisher.test function from the R software. P values were corrected with the Benjamini-Hochberg method. 'Optim' function in R was used to optimization of the network. Correlation between two data were calculated using Spearman's ρ in R software. Figure 2 ). Figure 1 ). Figure 3 ).

International Clinical Trials Registry Platform (ICTRP) (2022). COVID-19 trials

Coronavirus (COVID-19) Update: FDA Authorizes Monoclonal Antibodies for Treatment of COVID-19

Remdesivir for the Treatment of Covid-19 -Final Report

Cellular poly(c) binding proteins 1 and 2 interact with porcine reproductive and respiratory syndrome virus nonstructural protein 1β and support viral replication

Imbalanced Host Response to SARS-CoV-2 Drives Development of COVID-19

Proteomics of SARS-CoV-2-infected host cells reveals therapy targets

Evidence for a mediator of RNA polymerase II transcriptional regulation conserved from yeast to man

Shotgun transcriptome, spatial omics, and isothermal profiling of SARS-CoV-2 infection reveals unique host responses, viral diversification, and drug interactions

Blocking eIF4E-eIF4G interaction as a strategy to impair coronavirus replication

Potential Role of Cellular Senescence on Coronavirus Infections

Serum Cystatin C and Coronavirus Disease 2019: A Potential Inflammatory Biomarker in Predicting Critical Illness and Mortality for Adult Patients

Genome-scale metabolic modeling reveals SARS-CoV-2-induced metabolic changes and antiviral targets

Synthetic lethality across normal tissues is strongly associated with cancer risk, onset, and tumor suppressor specificity

COVID-19 severity correlates with airway epithelium-immune cell interactions identified by single-cell analysis

CDC25B promotes influenza A virus replication by regulating the phosphorylation of nucleoprotein

Identification of Required Host Factors for SARS-CoV-2 Infection in Human Cells

The Nucleocapsid Protein of SARS-CoV-2: a Target for Vaccine Development

The Reactome Pathway Knowledgebase

A Platform of Synthetic Lethal Gene Interaction Networks Reveals that the GNAQ Uveal Melanoma Oncogene Controls the Hippo Pathway through FAK

Phylogenetic network analysis of SARS-CoV-2 genomes

A SARS-CoV-2 protein interaction map reveals targets for drug repurposing

Camostat mesylate inhibits SARS-CoV-2 activation by TMPRSS2-related proteases and its metabolite GBPA exerts antiviral activity

Potential impact on coagulopathy of gene J o u r n a l P r e -p r o o f variants of coagulation related proteins that interact with SARS-CoV-2

Warfarin Dose Requirements in Adults Hospitalized With COVID-19 Infection: A Retrospective Case Series

Vitamin K epoxide reductase complex subunit 1 (VKORC1) gene polymorphism as determinant of differences in Covid-19-related disease severity

Clinical efficacy of antivirals against novel coronavirus (COVID-19): A review

Fast gene set enrichment analysis. bioRxiv

Establishing genetic interactions by a synthetic dosage lethality phenotype

Coronavirus breakthrough: dexamethasone is first drug shown to save lives

Harnessing synthetic lethality to predict the response to cancer treatment

Single-cell landscape of bronchoalveolar immune cells in patients with COVID-19

Molecular signatures database (MSigDB) 3.0

In vivo antiviral host transcriptional response to SARS-CoV-2 by viral load, sex, and age

Ceftazidime is a potential drug to inhibit SARS-CoV-2 infection in vitro by blocking spike protein-ACE2 interaction

Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2

Covid-19: FDA authorises neutralising antibody bamlanivimab for nonadmitted patients

Crippling life support for SARS-CoV-2 and other viruses through synthetic lethality

Targeting the viral-entry facilitators of SARS-CoV-2 as a therapeutic strategy in COVID-19

Viral protein engagement of GBF1 induces host cell vulnerability through synthetic lethality

Cellular senescence as a potential mediator of COVID-19 severity in the elderly

Repurposed Antiviral Drugs for Covid-19 -Interim WHO Solidarity Trial Results

Translational reprogramming marks adaptation to asparagine restriction in cancer

COVID-19: The therapeutic landscape

Inhibition of SARS-CoV-2 entry through the ACE2/TMPRSS2 pathway: a promising approach for uncovering early COVID-19 drug therapies

eIF4GI links nutrient sensing by mTOR to cell proliferation and inhibition of autophagy

Discovery of SARS-CoV-2 antiviral drugs through large-scale compound repurposing

SARS-CoV-2 SPIKE PROTEIN: an optimal immunological target for vaccines

Genome-Scale Identification of SARS-CoV-2 and Pan-coronavirus Host Factor Networks

Comprehensive Integration of Single-Cell Data

Multilevel proteomics reveals host perturbations by SARS-CoV-2 and SARS-CoV

Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles

Computational Immune Proteomics Approach to Target COVID-19

Perversely expressed long noncoding RNAs can alter host response and viral proliferation in SARS-CoV-2 infection

Coronavirus biology and replication: implications for SARS-CoV-2

Genetic Screens Identify Host Factors for SARS-CoV-2 and Common Cold Coronaviruses

Genome-wide CRISPR Screens Reveal Host Factors Critical for SARS-CoV-2 Infection

The Cancer Genome Atlas Pan-Cancer analysis project

DrugBank 5.0: a major update to the DrugBank database

Transcriptional and proteomic insights into the host response in fatal COVID-19 cases

Evidence for Gastrointestinal Infection of SARS-CoV-2

Disruption of Phosphatidylserine Synthesis or Trafficking Reduces Infectivity of Ebola Virus

Flaviviruses Exploit the Lipid Droplet Protein AUP1 to Trigger Lipophagy and Drive Virus Production

Network-based drug repurposing for novel coronavirus 2019-nCoV/SARS-CoV-2

This research was supported in part by the Intramural Research Program of the National Institutes of Health, NCI, CCR; and used the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). We acknowledge and thank the National Cancer Institute for providing financial and infrastructural support. The content of this publication does not necessarily reflect