Abstract Characterizing the interactions that SARS-CoV-2 viral RNAs make with host cell proteins during infection can improve our understanding of viral RNA functions and the host innate immune response. Using RNA antisense purification and mass spectrometry, we identified up to 104 human proteins that directly and specifically bind to SARS-CoV-2 RNAs in infected human cells. We integrated the SARS-CoV-2 RNA interactome with changes in proteome abundance induced by viral infection and linked interactome proteins to cellular pathways relevant to SARS-CoV-2 infections. We demonstrated by genetic perturbation that cellular nucleic acid-binding protein (CNBP) and La-related protein 1 (LARP1), two of the most strongly enriched viral RNA binders, restrict SARS-CoV-2 replication in infected cells and provide a global map of their direct RNA contact sites. Pharmacological inhibition of three other RNA interactome members, PPIA, ATP1A1, and the ARP2/3 complex, reduced viral replication in two human cell lines. The identification of host dependency factors and defence strategies as presented in this work will improve the design of targeted therapeutics against SARS-CoV-2. Subject terms: SARS-CoV-2, Virology, RNA metabolism, Proteomics __________________________________________________________________ Interactions between SARS-CoV-2 viral RNAs and host cell proteins during infection are evaluated to improve our understanding of viral RNA functions and the host innate immune response. Main The rapid spread of a new severe acute respiratory syndrome-related coronavirus (SARS-CoV-2) around the globe has led to a worldwide spike in a SARS-like respiratory illness termed coronavirus disease 2019 (COVID-19)^[64]1. To date, more than one million lives have been lost due to COVID-19. A detailed understanding of the molecular interactions and perturbations occurring during SARS-CoV-2 infection is required to understand the biology of SARS-CoV-2 and design therapeutic strategies. SARS-CoV-2 is an enveloped, positive-sense, single-stranded RNA virus that, upon infection of a host cell, deploys a ‘translation-ready’ RNA molecule, which uses the protein synthesis machinery of the host to express a set of viral proteins crucial for replication^[65]2. Replication of the full-length viral genome and transcription of subgenomic RNAs both involve the synthesis of negative-strand RNA intermediates^[66]3. In common with other RNA viruses, SARS-CoV-2 is dependent on effectively engaging host cell factors such as regulators of RNA stability, processing, localization and translation to facilitate replication and production of progeny. The host cell, on the other hand, must detect the pathogen and activate appropriate innate immune response pathways to restrict virus infection^[67]4. Studies on SARS-CoV-2-infected human cells to date have focused on characterizing expression or modification changes in the host cell transcriptome^[68]5,[69]6 or proteome^[70]7–[71]9. Further, interactions between recombinant viral proteins and host proteins have been identified in uninfected cells^[72]10,[73]11. Mapping of the interactions between viral and host proteins has revealed cellular pathways relevant to productive infection^[74]12. However, these studies cannot reveal how viral RNA is regulated during infection or how host cell RNA metabolism is remodelled to enable virus replication^[75]13. We sought to obtain an unbiased and quantitative picture of the cellular proteins that directly bind to SARS-CoV-2 RNAs in infected human cells. Recent RNA capture and quantitative mass spectrometry (MS) approaches^[76]14–[77]17 applied ultraviolet (UV) crosslinking to create covalent bonds between RNA molecules and the proteins they directly interact with. Unlike chemical crosslinking, UV irradiation does not stabilize protein–protein or RNA–RNA interactions, making it a preferable choice for dissecting direct RNA–protein interactions^[78]18,[79]19. RNA antisense purification and quantitative mass spectrometry (RAP–MS) combines UV crosslinking with a highly denaturing purification procedure and is ideally suited to capture and identify only those proteins that bind directly to SARS-CoV-2 RNAs^[80]14,[81]15. Results Capturing SARS-CoV-2 RNAs in infected human cells To purify SARS-CoV-2 RNAs and the complement of directly crosslinked cellular proteins from infected human cells, we designed a pool of biotinylated DNA oligonucleotides antisense to the positive-sense SARS-CoV-2 RNA and its subgenomic messenger RNAs. As a cellular system, we selected the human liver cell line Huh7, which is naturally permissive to both SARS-CoV-1 and SARS-CoV-2 replication^[82]20,[83]21. SARS-CoV-2 preferentially infects cells in the respiratory tract, but infection of multiple organs, including the liver, has been reported^[84]22. To test if our pool of antisense capture probes was suitable for the purification of SARS-CoV-2 RNAs from infected Huh7 cells, we performed RAP–MS 24 h after infection when viral replication levels were high^[85]21. We implemented a covalent protein capture step after the release of SARS-CoV-2 RNA-bound proteins, which enabled us to identify RNA sequences crosslinked to purified proteins (Fig. [86]1a and [87]Methods). Protein-crosslinked RNA fragments mapped to the entire length of the viral genome with near-complete sequence coverage, indicating that interactions across all viral RNA regions were captured (Extended Data Fig. [88]1a). Sequencing reads originating from SARS-CoV-2 RNA made up 93 and 92% of all mapped reads in 2 highly correlated replicate experiments (r = 0.994; Extended Data Fig. [89]1b,c). Fig. 1. RNA–protein interactome of SARS-CoV-2 in infected human cells. [90]Fig. 1 [91]Open in a new tab a, Outline of the RAP–MS method to identify proteins bound to SARS-CoV-2 RNA and their crosslinked RNA sequences. b, Quantification of SARS-CoV-2 RNA-interacting proteins relative to RMRP-interacting proteins. The scatter plot of log[2]-transformed TMT ratios from two biological replicates is shown. The grey dots represent all proteins detected with two or more unique peptides. c, Proteins enriched in SARS-CoV-2 RNA purifications (Supplementary Table [92]1). Left: core SARS-CoV-2 RNA interactome (adjusted P < 0.05). Left and right: expanded SARS-CoV-2 RNA interactome. Significantly enriched proteins are highlighted in teal; SARS-CoV-2-encoded proteins are highlighted in magenta. Adjusted P value: two-tailed moderated t-test. Extended Data Fig. 1. Capturing SARS-CoV-2 RNAs and bound proteins with RAP-MS. [93]Extended Data Fig. 1 [94]Open in a new tab a, Alignment of protein-crosslinked RNA fragments to the SARS-CoV-2 genome following RNA antisense purification of SARS-CoV-2 RNAs from infected Huh7 cells. Two replicate experiments are shown. b, Fraction of crosslinked RNA fragments mapping to the human or SARS-CoV-2 genomes in pilot RAP-MS experiments. c, Correlation plot for two replicate RAP experiments. CPM values for SARS-CoV-2 genes are shown. CPM: counts per million. d, As in b, but for full-scale SARS-CoV-2 RNA RAP-MS and RMRP RAP-MS experiments. e, Western blot of two replicate SARS-CoV-2 RNA and RMRP RAP-MS experiments. Indicated antibodies were used for protein detection. [95]Source data To identify proteins that specifically interact with SARS-CoV-2 RNAs as opposed to non-specific background proteins, we compared the protein content of SARS-CoV-2 RNA purifications to that of an unrelated control ribonucleoprotein complex of known composition. As the control, we used the endogenously expressed human ribonuclease mitochondrial RNA processing (RMRP) RNA and purified both SARS-CoV-2 RNA and RMRP from infected Huh7 cells. RMRP was selected for several reasons: (1) RMRP interacts with approximately ten well-known proteins that serve as an internal control^[96]15,[97]23; (2) RMRP is not translated; and (3) RMRP does not globally bind to mRNA. Hence, RMRP-binding proteins are distinct from the group of proteins expected to bind to SARS-CoV-2 RNAs, making it an ideal control for the discovery of unknown interactors. Further, the purification of SARS-CoV-2 RNA and RMRP from infected cells avoids biases resulting from widespread changes in the host cell proteome induced by viral infection. On average, approximately 90% of all crosslinked RNA fragments originated from the SARS-CoV-2 genome in SARS-CoV-2 RNA purifications, while more than 99% of crosslinked RNA fragments from RMRP purifications mapped to the human genome (Extended Data Fig. [98]1d). Western blot analysis confirmed the specific capture of SARS-CoV-2 nucleocapsid protein only in SARS-CoV-2-purified samples (Extended Data Fig. [99]1e). The RMRP component POP1 was detected only in RMRP purifications. Together, these experiments verify the high specificity of our approach for capturing the desired RNAs and the proteins that directly bind to them. An atlas of SARS-CoV-2 RNA–protein interactions in human cells Next, we subjected proteins purified with RMRP and SARS-CoV-2 RNAs to tandem mass tag (TMT) labelling and relative quantification by liquid chromatography coupled with tandem mass spectrometry (LC–MS/MS). In two replicate experiments, we identified 699 proteins, of which 583 were detected with 2 or more unique peptides (Supplementary Table [100]1 and [101]Methods). As shown in Fig. [102]1b, we found five known RMRP components among the ten most significantly enriched proteins in RMRP purifications. Next, we analysed proteins enriched in SARS-CoV-2 RNA purifications and found 15 SARS-CoV-2 proteins, 6 of which were among the 20 most significantly enriched proteins (Fig. [103]1b,c). In addition to 5 viral proteins translated from distinct open reading frames (ORFs), 10 of the 16 non-structural proteins (NSPs), which are derived from a precursor polyprotein^[104]24, were detected by RAP–MS. As expected, the SARS-CoV-2 nucleocapsid protein, which binds the viral RNA, was one of the two most significantly enriched viral proteins, followed by several known viral RNA binders, such as the endoribonuclease NSP15 (ref. ^[105]25), the RNA-dependent RNA polymerase (RdRP) NSP12 (ref. ^[106]26), the methyltransferase NSP16 (ref. ^[107]27), the RNA-binding protein NSP9 (ref. ^[108]28), the capping factor NSP10 (ref. ^[109]27), the primase NSP8 (ref. ^[110]26), the 5′-UTR binder NSP1 (ref. ^[111]29) and the multifunctional protein NSP3 (ref. ^[112]30). Remarkably, NSP3 and the most strongly enriched protein in our data, NSP6, were required for the formation of double-membrane vesicles^[113]31 and both proteins are candidate constituents of a molecular pore complex involved in the export of RNA from coronavirus double-membrane vesicles^[114]32. We also found ORF3a, which binds the 5′-end of the SARS-CoV-1 genome^[115]33, as well as ORF9b and the S and M proteins among strongly enriched candidates. While M is known to interact with the nucleocapsid protein, a model for genomic RNA packaging further suggests a possible RNA-binding function for M^[116]34. An RNA-binding activity of S was not previously reported. While S covers the surface of the viral envelope, it has a transmembrane domain and an intracellular tail^[117]35, making it conceivable that S may indeed contact viral RNA. Discovery of 104 human proteins that bind SARS-CoV-2 RNA We next focused on the human proteins enriched in SARS-CoV-2 RNA purifications. We identified 276 proteins with a positive log[2] fold change. Of these, 57 were significantly enriched (adjusted P < 0.05, two-tailed t-test), which we subsequently defined as the set of core SARS-CoV-2 RNA interacting proteins (Fig. [118]1c). Additionally, we also defined an expanded SARS-CoV-2 RNA interactome using a relaxed false discovery rate (FDR) of less than 20% (Fig. [119]1c). The expanded SARS-CoV-2 RNA interactome encompassed 104 human proteins and included 13 SARS-CoV-2-encoded proteins. The vast majority of the human RNA interactome proteins (100 proteins, 96%) have been identified previously in system-wide studies aimed at capturing proteins that crosslink to RNA^[120]36 (Supplementary Table [121]2). Comparing this expanded SARS-CoV-2 RNA interactome with the poly(A)-RNA interactome in Huh7 cells^[122]37, revealed high overlap between both datasets (69 proteins, 66%) (Fig. [123]2a). Next, we compared our direct SARS-CoV-2 RNA interactome with proteins that directly or indirectly associate with the RNA genomes of Dengue and Zika viruses in Huh7.5 cells^[124]38. Sixty-six proteins (63%) of the expanded SARS-CoV-2 RNA interactome also associated with the Dengue and Zika virus RNAs, while 38 proteins (36.5%) were unique SARS-CoV-2 RNA binders (Fig. [125]2a). Since coronaviruses form replication/transcription complexes (RTCs), we also compared the expanded SARS-CoV-2 RNA interactome to the protein content of murine coronavirus RTCs^[126]39 and found 64 shared proteins (Supplementary Table [127]2). Fig. 2. Viral RNA contacts regulators of RNA metabolism and host response. [128]Fig. 2 [129]Open in a new tab a, Intersection of the expanded SARS-CoV-2 RNA interactome with the poly(A)-RNA interactome and the Dengue/Zika virus interactome in Huh7 cells (Supplementary Table [130]2). b, GO enrichment analysis of SARS-CoV-2 RNA interactome proteins. Circle sizes scale to the number of detected proteins. SRP, signal recognition particle. Statistical test: Fisher’s exact test with Benjamini–Hochberg adjustment. c, Protein–protein association network of the expanded SARS-CoV-2 RNA interactome. Published virus-associated proteins are highlighted. Proteins without connections are not shown. d, As in c but proteins undergoing dynamic phosphorylation upon SARS-CoV-2 infection^[131]7 are highlighted. e, As in c but proteins that overlap known drug target genes (Drug Gene Interaction Database) are highlighted. Finally, only 10 of the 332 human proteins that bound recombinant SARS-CoV-2 proteins in uninfected cells^[132]10 also bound directly to viral RNA in infected cells (Supplementary Table [133]2). These results highlight the importance of discriminating between protein–protein and RNA–protein interactions when dissecting the biology of SARS-CoV-2. Biological functions of SARS-CoV-2 RNA-binding proteins To analyse the biological functions of SARS-CoV-2 RNA binders, we performed a hypergeometric gene ontology (GO) enrichment analysis on the expanded SARS-CoV-2 RNA interactome. We observed strong enrichment for GO terms linked to translational initiation ([134]GO:0006413), nonsense-mediated decay ([135]GO:0000184), signal-recognition particle-dependent cotranslational protein targeting to the membrane ([136]GO:0006614) and viral transcription ([137]GO:0019083) (Fig. [138]2b and Supplementary Table [139]3). Consistent with the enrichment of these GO terms, the importance of subgenomic mRNA translation at the endoplasmic reticulum membrane is well established for coronaviruses^[140]40. Further, nonsense-mediated mRNA decay was recently described as an antiviral mechanism targeting coronavirus RNAs^[141]41. In agreement with the crucial role of mRNA translation, the expanded SARS-CoV-2 RNA interactome included 19 ribosomal proteins and 12 translation factors. Among the translation factors, the eukaryotic translation initiation factor 4F (EIF4F) components EIF4G1 and EIF4B are regulated by mammalian target of rapamycin (mTOR) signalling^[142]42,[143]43. EIF4B is important for recruiting the 40S subunit to mRNA and both the phosphatidylinositol-3-kinase (PI3K)/mTOR and mitogen-activated protein kinase (MAPK) pathways target EIF4B to control its activity^[144]43. Inhibition of PI3K/Akt/mTOR signalling has been demonstrated to suppress SARS-CoV-2 replication in Caco2 cells^[145]8. To examine the connectivity of the identified SARS-CoV-2 RNA-binding proteins and their relationship to virus-associated biological processes systematically, we constructed a protein–protein association network using our expanded RNA interactome (Fig. [146]2c and Supplementary Table [147]4). We observed a striking enrichment for physical interactions when comparing the total connectivity among RNA interactome proteins to the connectivity of equally sized networks sampled from expressed proteins (Extended Data Fig. [148]2 and [149]Methods; permutation test P < 2.2 × 10^−16). In addition to ribosomal proteins and translation factors, many virus-associated RNA-binding proteins are prominently represented in this network (Fig. [150]2c). Since RNA-binding proteins can be regulated by phosphorylation, we intersected our expanded SARS-CoV-2 RNA interactome with a recent phosphoproteomic dissection of SARS-CoV-2-infected cells^[151]7, highlighting 30 proteins that might be dynamically phosphorylated in response to SARS-CoV-2 infection (Fig. [152]2d). Extended Data Fig. 2. Connectivity in RAP-MS protein-protein association network. Extended Data Fig. 2 [153]Open in a new tab Total number of connections observed in protein-protein association network constructed based on expanded SARS-CoV-2 RNA interactome (red bar, 1,534 connections), compared to number of connections observed in random networks of equal size (grey bars, mean 60 connections, z-score 76) using random sampling of proteins detected in proteome measurements. We next integrated known drug–target interactions^[154]44 within this network and identified 23 SARS-CoV-2 RNA interactome proteins that can be targeted with existing compounds, including peptidyl-prolyl cis-trans isomerase A (PPIA), actin-related protein 2 (ACTR2; henceforth ARP2), sodium/potassium-transporting ATPase subunit alpha-1 (ATP1A1), annexin A1 (ANXA1), cofilin-1 (CFL1) and epidermal growth factor receptor (EGFR) (Fig. [155]2e). Notably, EGFR is a known target of compounds that inhibit SARS-CoV-2 replication^[156]7,[157]8,[158]10. Identification of activated host response pathways To gain deeper insight into host response pathways activated upon SARS-CoV-2 infection, we globally measured protein abundance changes in infected cells. We performed triplicate MS experiments on SARS-CoV-2-infected and uninfected Huh7 cells and identified 10,956 proteins with 2 or more unique peptides (Fig. [159]3a and Supplementary Table [160]5). Among the detected proteins, 4,578 proteins were regulated (adjusted P < 0.05, two-tailed t-test) after 24 h of SARS-CoV-2 infection, which is consistent with widespread proteome regulation and agrees well with previously published data (Extended Data Fig. [161]3a)^[162]8,[163]9. As expected, proteome samples clustered according to their infection status in a principal component analysis (Extended Data Fig. [164]3b). Among differentially expressed proteins, we detected 13 viral proteins and 56 proteins from our expanded SARS-CoV-2 RNA interactome (Fig. [165]3a). Fig. 3. Connecting the SARS-CoV-2 RNA interactome to perturbations in host cells. [166]Fig. 3 [167]Open in a new tab a, Volcano plot of proteome abundance measurements in SARS-CoV-2-infected and uninfected Huh7 cells 24 h post-infection (n = 3) (Supplementary Table [168]5). Adjusted P value: two-tailed moderated t-test. SARS-CoV-2-encoded proteins are shown in magenta; human SARS-CoV-2 RNA interactome proteins are shown in teal; interferon response-related proteins are shown in purple. b, GSEA for the global proteome abundance measurements shown in a. Selected gene sets are shown; the full table displaying additional enriched gene sets is provided in Extended Data Fig. [169]3c. Statistical test: Kolmogorov–Smirnov test with Benjamini–Hochberg adjustment. NES, normalized enrichment score. c, Protein–protein association network of core SARS-CoV-2 RNA interactome proteins and their connections to differentially regulated proteins in SARS-CoV-2-infected cells based on curated interactions in STRING v.11 (ref. ^[170]96). Upregulated proteins are shown in light grey; downregulated proteins are shown in dark grey. Circle sizes scale to the number of connections of each interactome protein. Selected GO enrichments for network communities are shown in the transparent circles ([171]Methods). Full GO term analysis is provided in Supplementary Table [172]8. Extended Data Fig. 3. Proteome abundance changes in SARS-CoV-2 infected cells. [173]Extended Data Fig. 3 [174]Open in a new tab a, Correlation of protein abundance measurements reported in Klann et al. and this study (r = 0.411). Proteins displayed are significant at an adjusted P value threshold of 0.01 in both studies (n = 712). b, Principle component analysis for proteome measurements of SARS-CoV-2 (SCoV2) infected or mock infected Huh7 cells. c, GSEA for proteins significantly regulated in global proteome measurements. Gene sets enriched in addition to those shown in Figure 3b are presented. Statistical test: Kolmogorov-Smirnov test with Benjamini-Hochberg adjustment. d, Protein-protein association network of expanded SARS-CoV-2 RNA interactome proteins (blue: interactome protein, not regulated; red: interactome protein, regulated) and their connections to differentially regulated proteins upon SARS-CoV-2 infection. Upregulated proteins are shown in light grey; downregulated proteins are shown in dark grey. Circle sizes scale to the number of connections of each interactome protein. We next performed gene set enrichment analysis (GSEA) using our proteome abundance measurements. Among the most significantly enriched hallmark gene sets were ‘TGF-β signalling’, ‘TNF-α signalling via NF-κB’, ‘interferon (IFN)-γ response’ and ‘IL-6 JAK STAT3 signalling’ (Fig. [175]3b and Extended Data Fig. [176]3c), which is consistent with the induction of broad pro-inflammatory and antiviral responses in infected cells. Further, we observed significant enrichment of the gene sets ‘GO regulation of MAPK cascade’, ‘GO positive regulation of MAPK activity’ and ‘GO response to type I interferon’ (Supplementary Table [177]6). Recent evidence indicates that these pathways are indeed highly relevant in the context of SARS-CoV-2 infections^[178]5,[179]7,[180]8. Inhibition of growth factor signalling through the MAPK pathway, which responds to and controls the production of pro-inflammatory cytokines, including TNFα and IL-6, was shown to modulate SARS-CoV-2 replication^[181]7,[182]8. In agreement with recent transcriptome studies^[183]5,[184]6, our proteome data suggest activation of interferon signalling upon SARS-CoV-2 infection. Among interferon-related genes, we observed significant upregulation of several major components of IFN signalling pathways, including STAT1 and IRF9, which together with STAT2 make up the interferon stimulated gene factor 3 (ISGF3) complex, their upstream components TYK2 and JAK1, as well as their downstream targets IFIT1, IFIT3, IFITM3, OAS2 and ISG15 (Fig. [185]3a). Other strongly upregulated IFN-related genes include BST2, SP110, UBE2L6, ADAR and TGIF1 (Supplementary Table [186]5). Notably, many SARS-CoV-2 RNA interactome members are linked to the IFN response. These include the strongly enriched PUM1 (ref. ^[187]45), YBX1 (ref. ^[188]46), SYNCRIP^[189]47, G3BP1 (refs. ^[190]48,[191]49), G3BP2 (refs. ^[192]48,[193]49), EIF4B^[194]50, MOV10 (ref. ^[195]51), CAPRIN1 (ref. ^[196]49), DDX3X^[197]52, LSM14A^[198]53, RyDEN^[199]54,[200]55, STRAP^[201]56, ANXA1 (ref. ^[202]57), DDX1 (ref. ^[203]58), PCBP2 (ref. ^[204]59), HNRNPA2B1 (ref. ^[205]60) and YWHAZ^[206]61. In conclusion, our proteome analysis verifies the induction of an appropriate host response in SARS-CoV-2-infected Huh7 cells and further supports an important role for IFN and MAPK signalling in SARS-CoV-2 infection. Interplay between SARS-CoV-2 RNA binders and host cell proteins As an RNA-based obligate intracellular parasite, SARS-CoV-2 must effectively interface with the host cell and rewire RNA metabolism and RNA-associated regulatory processes. In addition to controlling the RNA life cycle^[207]62, host RNA-binding proteins are an integral part of regulatory circuits that participate in host defence mechanisms^[208]63,[209]64. To examine the interplay and connectivity between direct SARS-CoV-2 RNA binders and the host cell proteome, we used curated protein–protein interaction data to build a network that visualizes interactions between SARS-CoV-2 RNA binders and regulated host proteins (Fig. [210]3c, Extended Data Fig. [211]3d and Supplementary Table [212]7). We considered the connectivity among all differentially expressed host proteins and those that were detected in our core RNA interactome. Interactome proteins had a greater than twofold enrichment for network connections (mean 108) when compared to proteins not detected by RAP–MS (mean 45), indicating a significant enrichment in connectivity (Wilcoxon test, P = 8.92 × 10^−08). To further contextualize this network, we overlaid biological processes that were enriched among regulated protein communities that associate with SARS-CoV-2 RNA binders (Fig. [213]3c and Supplementary Table [214]8). This analysis highlighted several cellular pathways and processes emerging as highly relevant in the context of SARS-CoV-2 infections, including myeloid-mediated immunity^[215]65, receptor signalling^[216]8, protein phosphorylation^[217]7,[218]8, vesicle transport^[219]8,[220]10, protein folding^[221]6,[222]7 and translational regulation^[223]8,[224]66. Taken together, our network analysis connects RNA interactome proteins to emerging SARS-CoV-2 biology and provides a map of putative regulatory hubs in SARS-CoV-2 infections. Genetic screens identify functional SARS-CoV-2 RNA binders To functionally stratify our direct RNA binders, we intersected the SARS-CoV-2 RNA interactome with a recent genome-wide CRISPR perturbation screen designed to identify host factors that affect cell survival after SARS-CoV-2 infection^[225]67. Out of 104 human proteins in our expanded RNA interactome, we obtained CRISPR z-scores for 94 proteins^[226]67; depletion of 11 of these proteins had a statistically significant effect on SARS-CoV-2-induced cell death (Fig. [227]4a). Strikingly, cellular nucleic acid-binding protein (CNBP), the human protein most significantly enriched in RAP–MS, also had the most significant effect on virus-induced cell death among all SARS-CoV-2 RNA interactome members (Fig. [228]4a). In addition to the 11 aforementioned proteins, the direct SARS-CoV-2 RNA binders cold shock domain-containing protein E1 (CSDE1)^[229]68, polyadenylate-binding protein 1 (PABPC1) (refs. ^[230]11,[231]68) and Ras-related protein Rab-7a (RAB7A)^[232]11 were also identified as host factors with functional relevance in SARS-CoV-2 infections by genetic screening approaches. Fig. 4. CNBP contacts SARS-CoV-2 viral RNA. [233]Fig. 4 [234]Open in a new tab a, SARS-CoV-2 RNA interactome proteins overlaid on genome-wide CRISPR perturbation data from SARS-CoV-2-infected Vero E6 cells^[235]67. Members of the expanded RNA interactome with significant (adjusted P < 0.05, two-sided z-test with Benjamini–Hochberg correction) changes in CRISPR z-scores are highlighted in magenta. The y axis is capped at 1 × 10^−19, excluding 4 genes. b, Western blot of Huh7 CNBP knockout and control cell lines (top). RT–qPCR measurements of intracellular SARS-CoV-2 RNA (RdRP gene) at 48 h post-infection in Huh7 CNBP knockout and control cells (bottom). Quantification relative to 18S rRNA and control cells is shown. Values are the mean ± s.d. (n = 3 independent infections). P values were determined using an unpaired two-tailed t-test. ****P < 0.0001. c, Distribution of CNBP eCLIP peaks to different RNA types and transcript regions. d, Meta-gene analysis of CNBP eCLIP signal across mature mRNAs. e, CNBP eCLIP data aligned to the SARS-CoV-2 RNA genome. The fold change relative to the size-matched input is shown. MACS2-enriched peaks are shown below the fold change track. [236]Source data CNBP functions as an antiviral regulator CNBP is required to activate the innate immune response and has been linked to regulating the expression of pro-inflammatory cytokines in response to foreign nucleic acid sensing^[237]69,[238]70. Notably, CNBP-deficient animals were highly susceptible to infections with different pathogens^[239]69,[240]70. These findings are consistent with CNBP-depleted cells being sensitized to virus-induced cell death, which suggests that CNBP may act as an antiviral regulator. To corroborate the functional importance of CNBP in SARS-CoV-2 infections, we generated polyclonal Huh7 CNBP knockout cell lines using CRISPR–Cas9 (Fig. [241]4b). We infected CNBP knockout cells with SARS-CoV-2 and noted significantly elevated levels of intracellular viral RNA compared to matched Huh7 control cells (Fig. [242]4b). Thus, CNBP is indeed a functionally important SARS-CoV-2 RNA interactor. To confirm the direct physical engagement of SARS-CoV-2 RNAs by CNBP, we performed enhanced crosslinking and immunoprecipitation (eCLIP) in SARS-CoV-2-infected Huh7 cells and quantified the enrichment of CNBP peaks relative to size-matched input libraries^[243]71. First, we analysed CNBP binding to the human transcriptome. Consistent with earlier reports^[244]72, CNBP bound to protein-coding transcripts and displayed a preference for binding within the coding sequence (CDS) of mRNAs (Fig. [245]4c,d). A large number of transcripts bound by CNBP in SARS-CoV-2-infected cells were previously reported as CNBP targets (approximately 46%; Supplementary Table [246]9). We next analysed CNBP binding to SARS-CoV-2 RNA and observed several strongly enriched binding sites in the viral genome (Fig. [247]4e). These data provide strong evidence for a direct interaction between CNBP and SARS-CoV-2 RNAs in infected cells and validate that RAP–MS indeed identifies direct RNA binders. Further, the finding that CNBP preferentially associates with the CDS of mature mRNAs lends credibility to its previously proposed role as a translational regulator^[248]72 in addition to its function in regulating pro-inflammatory cytokines. LARP1 binds genomic and subgenomic SARS-CoV-2 RNAs Other than CNBP, two members of the La-related protein (LARP) family, namely LARP1 and LARP4, were strongly enriched in SARS-CoV-2 RNA purifications. While LARP1 did not quite meet our significance cut-off, both LARP1 and LARP4 were among the 15 host proteins with the strongest enrichment based on overall effect size, indicating that LARP1 is very likely a SARS-CoV-2 RNA binder. Additionally, LARP1 was detected among protein–protein interactors of the nucleocapsid protein in uninfected cells^[249]10. Given that LARP1 is a major downstream target of mammalian target of rapamycin complex 1 (mTORC1) (refs. ^[250]73,[251]74) and inhibition of PI3K/Akt/mTOR was recently shown to inhibit SARS-CoV-2 replication in Caco2 cells^[252]8, we sought to characterize the LARP1-SARS-CoV-2 axis in greater detail. We performed eCLIP^[253]71 to map direct physical interactions between LARP1 and its RNA targets. LARP1 predominantly bound protein-coding transcripts and we observed most of the enriched peaks in the CDS, followed by 5′-UTR and 3′-UTR sequences (Fig. [254]5a). Previous work suggested that LARP1 binds the 7-methylguanosine triphosphate (m^7Gppp) moiety of the cap and the adjacent 5′-terminal oligopyrimidine (5′TOP) motif of mRNAs to regulate their translation^[255]75. Consistent with this finding, our eCLIP data revealed a strong enrichment of 5′-proximal nucleotides in 5′-UTR sequences and we recovered an oligopyrimidine motif reminiscent of TOP-like sequences in approximately 30% of all bound 5′-UTRs (Fig. [256]5b,c). Out of 112 mRNAs that are regulated by LARP1 downstream of mTOR^[257]76, we observed LARP1 binding to 84 mRNAs (75%; Supplementary Table [258]10). In line with the known regulatory functions of LARP1 (ref. ^[259]76), LARP1 target transcripts were most strongly enriched for GO terms linked to translational regulation (Supplementary Table [260]11). Together, these data demonstrate that our eCLIP experiments recovered known regulatory interactions of LARP1. Fig. 5. LARP1 binds SARS-CoV-2 RNAs and restricts viral replication. [261]Fig. 5 [262]Open in a new tab a, Distribution of LARP1 eCLIP peaks to different RNA types and transcript regions. b, Meta-gene analysis of LARP1 eCLIP signal across mature mRNAs. c, Oligopyrimidine-rich sequence motif discovered de novo in LARP1 peaks mapping to 5′-UTRs ([263]Methods). d, LARP1 eCLIP data aligned to the SARS-CoV-2 RNA genome. The fold change relative to the size-matched input is shown. MACS2-enriched peaks are shown above the fold change track. Oligopyrimidine-rich sequences that coincide with strongly enriched LARP1 peaks are highlighted. A zoom-in to the SARS-CoV-2 5′-leader sequence is shown below the genomic alignment. e, Left: RT–qPCR measurements of intracellular SARS-CoV-2 RNA at 24 h post-infection in WT HEK293 cells or 4 different LARP1 knockout cell lines. Quantification relative to 18S rRNA and WT cells is shown. Right: Infectious viral titres in the supernatants of infected cells quantified by plaque assays at 24 h post-infection. P values were determined using an unpaired two-tailed t-test. f, Left: RT–qPCR measurements of intracellular SARS-CoV-2 RNA at 24 h post-infection in HEK293 cells transiently overexpressing GFP or LARP1–GFP proteins. Quantification relative to 18S rRNA and GFP-overexpressing cells is shown. Right: Infectious viral titres in the supernatants of infected cells quantified by plaque assays at 24 h post-infection. P values were determined using an unpaired one-tailed t-test. g, RT–qPCR measurements of intracellular SARS-CoV-2 RNA at 24 h post-infection in LARP1 knockout cells complemented with either GFP or LARP1–GFP plasmids. Quantification relative to 18S rRNA and GFP-transfected WT cells is shown. P values were determined using an unpaired two-tailed t-test. e–g, All values are the mean ± s.d. (n = 3 independent infections) h, Quantification of ribosomal frameshifting efficiency using a dual-fluorescence translation reporter (Extended Data Fig. [264]4d) in HEK293 cells is shown. Data were normalized to cells transfected with eCFP (n = 6 independent transfections, except for control RNA n = 4). Values are the mean ± s.d. ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05; NS, not significant; FSE, frameshift element. Having confirmed the quality of our eCLIP experiment on host RNAs, we next characterized LARP1 binding to SARS-CoV-2 RNAs and found several regions of enrichment that coincided with oligopyrimidine sequences (Fig. [265]5d). Notably, we observed LARP1 binding to the first 70 nucleotides at the 5′-end of the SARS-CoV-2 genome, which corresponds to the viral 5′-leader sequence^[266]77 and contains a TOP-like motif instance (Fig. [267]5d). Binding to the 5′-leader, which is present in all viral subgenomic mRNAs, suggests a direct association of LARP1 with subgenomic mRNAs. LARP1 represses SARS-CoV-2 replication To determine the impact of LARP1 depletion on SARS-CoV-2 replication, we generated four clonal LARP1 knockout cell lines using CRISPR–Cas9 in HEK293 cells (Extended Data Fig. [268]4a). We infected cells with SARS-CoV-2 and measured intracellular viral RNA levels and the production of infectious virus. Compared to wild-type (WT) cells, LARP1 knockout cells displayed approximately fivefold higher levels of intracellular viral RNA and a similar increase in the production of infectious virus (Fig. [269]5e). Conversely, transient overexpression of LARP1 fused to green fluorescent protein (GFP) in WT cells led to a significant reduction of viral RNA and infectious virus when compared to GFP expression alone (Fig. [270]5f and Extended Data Fig. [271]4b). Next, we complemented LARP1 knockout cells with transiently expressed LARP1–GFP proteins (Fig. [272]5g, Extended Data Fig. [273]4c). In all knockout cell lines, we observed a clear reduction in intracellular viral RNA that approached WT levels when compared to cells transfected with GFP alone. These experiments established that LARP1 functions as a repressor of SARS-CoV-2 replication in infected human cells. Extended Data Fig. 4. Functional validation of SARS-CoV-2 RNA binders. [274]Extended Data Fig. 4 [275]Open in a new tab a, Western blot of WT HEK293 cells and four different HEK293 LARP1 knockout (KO) cell lines generated with CRISPR-Cas9 (see Methods). Expression of LARP1 was evaluated relative to Tubulin. b, Western blot of HEK293 cells transiently overexpressing (OE) GFP or LARP1-GFP proteins at 48 h post transfection. Arrows indicate endogenous LARP1 proteins and GFP-tagged LARP1. c, Western blot of four different HEK293 LARP1 knockout cell lines transiently transfected with plasmids encoding GFP or LARP1-GFP proteins at 48 h post transfection. Experiments were repeated at least two times. d, Schematic of dual-fluorescence translation reporter to quantify ribosomal frameshifting efficiency. The depicted control construct contains enhanced GFP (eGFP) and mCherry in an in-frame orientation, leading to the production of both fluorescent proteins separated by a self-cleaving 2A peptide when the 0 reading frame is translated. In the frameshift construct depicted below, eGFP and mCherry are separated by an in-frame stop codon, preventing the production of mCherry when the 0 reading frame is translated. −1FS leads to the production of eGFP and mCherry and the ratio between both fluorescent proteins is a direct measure of frameshifting efficacy. −1FS: –1 ribosomal frameshifting. [276]Source data RyDEN suppresses ribosomal frameshifting during SARS-CoV-2 RNA translation LARP1 interacts with PABPC1 and both LARP1 and PABPC1 have been proposed to reside in the same ribonucleoprotein complex with RyDEN^[277]54, all of which were enriched in RAP–MS experiments. In addition to being an IFN-induced protein, RyDEN suppresses Dengue virus production in infected cells^[278]54 and inhibits programmed -1 ribosomal frameshifting (-1FS) in human immunodeficiency virus type 1 (HIV-1) infections^[279]55. In coronaviruses, production of RdRP requires translation of the ORF1b gene, which is controlled by -1FS. For SARS-CoV-2, it is presently unknown if the efficiency of -1FS is important for the viral life cycle^[280]78. To dissect if RyDEN can modulate the frequency of -1FS in SARS-CoV-2, we generated a dual-colour fluorescence reporter system to quantify frameshifting efficiency in response to RyDEN induction, as seen upon SARS-CoV-2 infection (Extended Data Fig. [281]4d and [282]Methods). Using a reporter containing the HIV-1 frameshift element as a positive control, we confirmed that overexpression of RyDEN fused to enhanced cyan fluorescent protein (eCFP) suppressed -1FS when compared to eCFP expression alone (Fig. [283]5h). Importantly, overexpression of RyDEN also led to a significant reduction of -1FS during translation of the SARS-CoV-2 frameshift element (Fig. [284]5h). Together, our results show that RyDEN is induced upon SARS-CoV-2 infection, associates with the SARS-CoV-2 RNA in infected cells and modulates the efficiency of SARS-CoV-2 -1FS. Pharmacological inhibition of interactome proteins restricts viral replication Next, we tested if targeting the SARS-CoV-2 RNA interactome and its associated pathways with known inhibitors is effective in restricting viral replication. We selected four inhibitors that target components of our expanded RNA interactome: PPIA; ARP2; ATP1A1; and DDX3X. While DDX3X is a DEAD-box RNA helicase and canonical RNA-binding protein, PPIA, ARP2 and ATP1A1 are non-classical RNA binders that are nonetheless robustly detected among RNA-binding proteins in Huh7 cells^[285]36,[286]37,[287]79. In addition to Huh7 cells, we evaluated all inhibitors in Calu3 cells, a human lung epithelial cell line that is naturally susceptible to SARS-CoV-2 infection. We observed a dose-dependent inhibition of intracellular viral RNA expression accompanied by a reduction in the production of infectious virus for the PPIA inhibitor cyclosporin A (Extended Data Fig. [288]5a,b), the ARP2/3 complex inhibitor CK-548 and the ATP1A1 inhibitor ouabain (Fig. [289]6a,b). The observed effect was highly consistent between Calu3 and Huh7 cells (Fig. [290]6a,b). While CK-548 treatment reduced cell viability at the highest concentration in Huh7 cells, we did not observe such effects at identical concentrations in Calu3 cells. All other efficacious inhibitors had no apparent effect on cell viability (Extended Data Fig. [291]5c,d). Unlike the three aforementioned compounds, inhibition of DDX3X only led to a moderate reduction of intracellular viral RNA and infectious virus in Calu3 cells at the highest concentration (Fig. [292]6a,b). Extended Data Fig. 5. Pharmacological inhibition of SARS-CoV-2 RNA interactome proteins. [293]Extended Data Fig. 5 [294]Open in a new tab a, RT-qPCR measurements of intracellular SARS-CoV-2 RNA (RdRP gene) in infected Calu3, Huh7 and A549-ACE2 cells after inhibitor treatment. Inhibitors were used at indicated concentration (left to right). Calu3 cells were assayed 24 h post-infection, Huh7 and A549-ACE2 cells were assayed 48 h post-infection. Values are normalized to 18S rRNA measurements and compared to untreated or DMSO treated cells. b, Infectious viral titers in the supernatants of infected Calu3, Huh7 and A549-ACE2 cells after inhibitor treatment. Inhibitors were used at indicated concentration (left to right). Calu3 cells were assayed 24 h post-infection, Huh7 and A549-ACE2 cells were assayed 48 h post-infection. All values in a-b are mean ± s.d. (n = 3 independent infections) c-d, Cell viability assay in inhibitor-treated and untreated cells. Values are the mean ± s.d. (n = 3 independent treatments). P values determined in unpaired two-tailed t-test. ***P < 0.001; **P < 0.01; *P < 0.05; ns, not significant. Fig. 6. RNA interactome inhibitors reduce virus replication. [295]Fig. 6 [296]Open in a new tab a, Top: RT–qPCR measurements of intracellular SARS-CoV-2 RNA at 24 h post-infection in Calu3 cells after inhibitor treatment. Inhibitors were used at the indicated concentrations (left to right). Values were normalized to 18S rRNA measurements and compared to DMSO-treated cells. Bottom: Infectious viral titres in the supernatants of infected Calu3 cells quantified by plaque assays at 24 h post-infection. b, As in a but for Huh7 cells and at 48 h post-infection. All values are the mean ± s.d. (n = 3 independent infections). P values were determined using an unpaired two-tailed t-test. ***P < 0.001, **P < 0.01, *P < 0.05. Beyond inhibiting direct RNA binders, we also targeted mTORC1, the upstream regulatory complex that controls LARP1 activity. Consistent with LARP1 restricting SARS-CoV-2 replication, we observed that inhibiting mTORC1/2 resulted in reduced viral replication in Huh7 and Calu3 cells (Fig. [297]6a,b). These findings agree well with previous results showing that mTORC1 phosphorylates LARP1, which leads to a translational de-repression of LARP1 target mRNAs^[298]76. Indeed, recent phosphoproteomic surveys demonstrate that LARP1 undergoes dynamic phosphorylation in response to SARS-CoV-2 infection^[299]7,[300]8. Inhibition of another upstream regulator, TANK-binding kinase 1, which interacts with the SARS-CoV-2 RNA binders DDX3X^[301]52 and ANXA1 (ref. ^[302]57), increased the levels of viral RNA and infectious virus in A549-ACE2 cells, but did not show a consistent effect in Huh7 or Calu3 cells (Extended Data Fig. [303]5a,b). Together, our experiments demonstrate that RNA interactome proteins represent viable targets for inhibiting SARS-CoV-2 replication. The SARS-CoV-2 RNA interactome provides valuable starting points for future mechanistic studies and may help developing new antiviral approaches for COVID-19. Discussion Decoding how the RNA genomes of pathogenic RNA viruses interface with the host cell proteome has been a long-standing challenge. In this study, we provide detailed molecular insights into the identity of host factors and cellular machinery that directly and specifically bind SARS-CoV-2 RNAs during infection of human cells. We integrate CRISPR perturbation data and perform genetic and pharmacological validation experiments that together suggest functional roles for 18 RNA interactome proteins in SARS-CoV-2 infections. Beyond identifying proteins that bind SARS-CoV-2 RNAs, we globally mapped where CNBP and LARP1 contact viral and human RNA and report binding preferences that are consistent with previously described