key: cord-0741087-o9e872a0
authors: Lee, Sungyul; Lee, Young-suk; Choi, Yeon; Son, Ahyeon; Park, Youngran; Lee, Kyung-Min; Kim, Jeesoo; Kim, Jong-Seo; Kim, V. Narry
title: The SARS-CoV-2 RNA interactome
date: 2021-04-27
journal: Mol Cell
DOI: 10.1016/j.molcel.2021.04.022
sha: b881cd2f96c6a432d566bfef32934970643a102f
doc_id: 741087
cord_uid: o9e872a0

SARS-CoV-2 is an RNA virus whose success as a pathogen relies on its abilities to repurpose host RNA-binding proteins (RBPs) and to evade antiviral RBPs. To uncover the SARS-CoV-2 RNA interactome, we here develop a robust ribonucleoprotein (RNP) capture protocol and identify 109 host factors that directly bind to SARS-CoV-2 RNAs. Applying RNP capture on another coronavirus HCoV-OC43 revealed evolutionarily conserved interactions between coronaviral RNAs and host proteins. Transcriptome analyses and knockdown experiments delineated 17 antiviral RBPs including ZC3HAV1, TRIM25, PARP12, and SHFL and 8 proviral RBPs such as EIF3D and CSDE1 which are responsible for co-opting multiple steps of the mRNA life cycle. This also led to the identification of LARP1, a downstream target of the mTOR signaling pathway, as an antiviral host factor that interacts with the SARS-CoV-2 RNAs. Overall, this study provides a comprehensive list of RBPs regulating coronaviral replication and opens new avenues for therapeutic interventions.

end sequences. The sgRNAs are generated via discontinuous transcription which leads to the fusion between the 5′ leader sequence and the "body" parts containing the downstream open reading frames (Sola et al., 2015) that encode structural proteins (S, E, M, and N) and accessory proteins (3a, 3c, 6, 7a, 7b, 8, and 9b) (Kim et al., 2020a) .

To accomplish this, coronaviruses employ unique strategies to evade, modulate, and utilize the host machinery (Fung and Liu, 2019) . For example, the gRNA molecules must be kept in an intricate balance between translation, transcription, and encapsulation by recruiting the right host RNA-binding proteins (RBPs) and forming specific ribonucleoprotein (RNP) complexes. As host cells counteract by launching RBPs such as RIG-I, MDA5, and Toll-like receptors (TLRs) to recognize and eliminate viral RNAs, the virus needs to evade the immune system using its components to win the arms race between virus and host. How such stealthy devices are genetically coded in this compact RNA genome is yet to be explored (Snijder et al., 2016) . Thus, the identification of the RBPs that bind to viral transcripts (or the SARS-CoV-2 RNA interactome) is key to uncovering the molecular rewiring of viral gene regulation and the activation of antiviral defense systems.

Biochemical techniques for studying RNA-protein interactions have been developed (Ramanathan et al., 2019) with the advancement in protein-centric methods such as CLIPseq (crosslinking immunoprecipitation followed by sequencing) . In CLIPseq experiments, to identify direct RNA-protein interactions, RNP complexes are crosslinked by UV irradiation within cells, after which the protein of interest is immunoprecipitated and the associated RNAs are sequenced (Lee and Ule, 2018; Van Nostrand et al., 2020) . More recently, RNA-centric methods have also been developed to profile the mRNA interactome and RNP complexes (Roth and Diederichs, 2015) . After UV irradiation, the RNA of interest is purified with oligonucleotide probes and the crosslinked proteins are identified by mass spectrometry. For example, RAP-MS exhibits compelling evidence of highly confident profiling of proteins that bind to a specific RNA owing to a combination of long hybridization probes and harsh denaturing condition (Engreitz et al., 2013; McHugh et al., 2015) .

In this study, we developed a robust RNP capture protocol to define the repertoire of viral and host proteins that associate with the transcripts of coronaviruses, namely SARS-CoV-2 and HCoV-OC43. Network and transcriptome analyses combined with knockdown experiments revealed host factors that link the viral RNAs to mRNA regulators and putative antiviral factors.

To identify the viral and host proteins that directly interact with the genomic and subgenomic RNAs of SARS-CoV-2, we modified the RNA antisense purification coupled with mass spectrometry (RAP-MS) protocol (McHugh and Guttman, 2018) which was developed to profile the interacting proteins of a particular RNA species ( Figure 1A ). Briefly, virus-infected cells were first detached from culture vessels and irradiated with 254 nm UV to induce RNAprotein crosslink while preserving RNA integrity. Crosslinked cells were treated with DNase and lysed with an optimized buffer condition to homogenize and denature the proteins in high concentration. Pools of biotinylated antisense 90-nt probes were used to capture the denatured RNP complexes in a sequence-specific manner. After stringent washing and detergent removal, the RNP complexes were released and digested by serial benzonase and on-bead trypsin treatment. These modifications to the RAP-MS protocol enabled robust and sensitive identification of proteins directly bound to the RNA target of interest (see Methods for detailed explanation). Of note, virus infected cells were trypsin-suspended before UV cross-linking to minimize RNA degradation to capture the intact viral RNAs (see Methods for details).

We designed two separate pools of densely overlapping 90-nt antisense probes to achieve an unbiased perspective of the SARS-CoV-2 RNA interactome ( Figure 1B and Table S1 ). The SARS-CoV-2 transcriptome consists of a genomic RNA (gRNA) encoding 16 nonstructural proteins (nsps) and multiple subgenomic RNAs (sgRNAs) that encode structural and accessory proteins (Sola et al., 2015) . The sgRNAs are more abundant than the gRNA (Kim et al., 2020a) . The first pool ("Probe I") consists of 707 oligos tiles every 30 nucleotides across the ORF1ab region (266:21553, NC_045512.2) and thus hybridizes specifically with the gRNA molecules ( Figure 1B) . The second pool of 275 oligos ("Probe II") covers the remaining region (21563:29872, NC_045512.2) which is shared by both the gRNA and sgRNAs.

To first check whether our method specifically captures the viral RNP complexes, we compared the resulting purification from Vero cells infected with SARS-CoV-2 (BetaCoV/Korea/KCDC03/2020) at MOI 0.1 for 24 hours (Kim et al., 2020b ) by either Probe I or Probe II. As negative controls, we pulled-down without probes ("no probe" control) or with the control probes (for either 18S or 28S rRNA). Protein composition of each RNP sample J o u r n a l P r e -p r o o f was distinct as shown by silver staining and western blotting ( Figure S1A ) with prominent SARS-CoV-2 N protein associated with Probes I and II, as expected. Enrichment of SARS-CoV-2 RNAs were confirmed by RT-qPCR ( Figure S1B ), suggesting that our protocol purfies specific RNP complexes. Note that SARS-CoV-2 gRNA was not enriched in the Probe II experiment, hinting at the excess amount of sgRNAs over gRNA in our culture condition.

We conducted label-free quantification (LFQ) by liquid chromatography with tandem mass spectrometry (LC-MS/MS) and identified 429 host proteins and 9 viral proteins in total ( Figure 1C ). As highly abundant proteins may nonspecifically co-precipitate during the RNP capture experiment, we statistically modelled this protein background as a multinomial distribution and assessed the probability (i.e. p-value) of the quantity of the identified protein in the RNP capture (e.g. Probe I) experiment over the protein background of the control (e.g. no-probe) experiment (see Methods for details). This unweighted spectral count analysis resulted in 199 and 220 proteins that are overrepresented in the Probe I and Probe II sample, respectively (FDR < 10%, Table S2 ). Protein domain enrichment analysis revealed that these proteins indeed harbor RNA-binding domains such as RNA recognition motif (RRM) domain and K Homology (KH) domain ( Figure S1C ). Of note, unlike the cellular mRNA interactome (Castello et al., 2012; Gerstberger et al., 2014) , the RNA-binding repertoire of SARS-CoV-2 RNAs showed a depletion of DEAD/DEAH box helicase domains and an enrichment of KH domain.

As for viral proteins, the N protein was the most significantly enriched one, as expected ( Figure 1D ). The nsp1 protein was also statistically enriched in both Probe I and Probe II experiments ( Figure 1D ). Given that Probe II mainly precipitates sgRNAs ( Figure S1B ) owing to the abundance of sgRNAs in infected cells (Kim et al., 2020) , nsp1 is likely to interact with both gRNA and sgRNAs. Nsp12, S, M, and nsp9 were detected more with Probe I than with

Probe II, indicating their preferential interaction with gRNAs. Coronavirus nsp9 is a singlestrand RNA binding protein (Egloff et al., 2004; Sutton et al., 2004) essential for viral replication (Miknis et al., 2009) . Nsp1 is one of the major virulence factors that suppresses host translation by binding to the 40S ribosomal subunit (Thoms et al., 2020) . While nsp1 is mostly studied in the context of translational suppression of host genes (Narayanan et al., 2008) , our result hints at the direct role of nsp1 on the transcripts of SARS-CoV-2, corroborating the viral evasion models in which the amino-terminal part of nsp1 interacts with the cis-acting RNA hairpin SL1 in the 5′ UTR of SARS-CoV-2, which then dissociates nsp1 from 40S (Banerjee et al., 2020; Shi et al., 2020) . Furthermore, our result is not inconsistent J o u r n a l P r e -p r o o f with the ribosomal "gatekeeper" model where SL1 induces a structural rearrangement of nsp1 (Tidu et al., 2020) . It is also worth mentioning that the nsp2, nsp8, and ORF9b are not enriched in the viral probe experiments, suggesting that these viral proteins do not interact significantly with the viral RNAs.

To delineate the host proteins that are enriched in the SARS-CoV-2 RNP complex, we employed an additional negative control experiment with uninfected cells ( Figure 1E ). In effect, this control provides a conservative background of host proteins as shown by silver staining ( Figure S1D ). Distributions of peptide length were consistent across technical replicates ( Figure S1E ), demonstrating the robustness of the "on-bead" digestion step.

Spectral count analysis against the "uninfected probe control" resulted in 74 and 72 proteins that are enriched in the infected samples with Probe I and Probe II, respectively (FDR < 5%, Figure 1F -H). In combination, we define these 109 proteins as the "SARS-CoV-2 RNA interactome." In an independent RNA capture experiment with a control RNA probe targeting 7SL RNAs, the majority of the SARS-CoV-2 interactome (86 out of 109 proteins) were preferentially enriched in the viral probe experiments ( Figure S1F ), demonstrating the specificity of the viral RNA interactome compared to other abundant cytoplasmic RNAs. 37 host proteins such as CSDE1 (Unr), EIF4H, FUBP3, G3BP2, PABPC1, ZC3HAV1 were enriched in both the Probe I and Probe II RNP capture experiments on infected cells ( Figure   1F ), thus identifying a robust set of the "core SARS-CoV-2 RNA interactome." Gene ontology (GO) term enrichment analysis revealed that these host factors are involved in RNA stability control, mRNA function, and viral process ( Figure S1G ).

To investigate the evolutionary conservation of the RNA-protein interactions in coronaviruses, we conducted RNP capture on HCoV-OC43 that belongs to the lineage A of genus betacoronavirus (Figure 2A, S2A and S2B ). HCoV-OC43 shows 54.2% nucleotide homology to SARS-CoV-2 which belongs to lineage B. Two antisense probe sets (i.e. OC43 Probe I and OC43 Probe II) were designed in a similar manner as in SARS-CoV-2 experiments.

OC43 Probe I hybridizes only with the gRNA of HCoV-OC43, while Probe II can hybridize both gRNA and sgRNA molecules. As negative controls, the no-probe control on infected cells and the probe control on uninfected cells were used and confirmed by silver staining ( Figure S2C ). Of note, no substantial differences between time points were found except for increasing protein quantities, which is expected because these are mixed-stage conditions. J o u r n a l P r e -p r o o f Unweighted spectral count analysis against the no-probe control revealed 133, 167, 192 , and 160 proteins that are overrepresented in the OC43 Probe I experiment at 12, 24, 36, and 48 hours post-infection (hpi), respectively (FDR < 10%, Table S3 ). For the OC43 Probe II experiment, 119, 189, 194, and 185 proteins were overrepresented at each respective infection time point (FDR < 10%, Table S4 ). The analysis of all eight RNP capture experiments resulted in the enrichment of proteins containing canonical RNA binding domains such as the RRM domain and the KH domain ( Figure S2D ), indicating the specific precipitation of RBPs.

Fourteen viral proteins including N, M, and S were detected within the HCoV-OC43 RNP complexes ( Figure 2B , Figure S2E ). HCoV-OC43 2a, an accessory protein unique to betacoronavirus lineage A, was also detected, indicating that this protein of unknown function may act as an RBP. The RdRP nsp12 and the papain-like protease nsp3 also appeared along with the other nsps identified in this experiment. Only a marginal amount of the HCoV-OC43 nsp1 was detected ( Figure S2E ), implying the functional divergence of nsp1 in betacoronavirus lineages A and B.

Next, we compared the host factors that form the viral RNA interactome of HCoV-OC43 and SARS-CoV-2. All 107 proteins from the SARS-CoV-2 interactome were also detected in the HCoV-OC43 interactome throughout multiple infection timepoints, except for RBMS1 and DDX3Y ( Figure 2C ), suggesting a large overlap of RBPs that are common among betacoronavirus. To identify a confident set of the host factors that bind to HCoV-OC43 transcripts, we applied our spectral count analysis on the HCoV-OC43 experiment of 36 hpi as a representative ( Figure S2A ) and conducted statistical analysis in comparison to the no-probe control ( Figure 2D , FDR < 10%) and the uninfected-probe control ( Figure 2E , FDR < 5%). We identified 67 and 70 host proteins for the HCoV-OC43 Probe I and Probe II experiments, respectively (Table S5) . 38 proteins were statistically enriched in both probe sets and showed GO term enrichment related to mRNA regulation ( Figure S2F ). Together, these 99 proteins were defined as "HCoV-OC43 RNA interactome." Among these 99 proteins, 52 proteins were also identified as the SARS-CoV-2 RNA interactors ( Figure 2F ), and thus are conserved host components of the betacoronavirus RNA interactome. Among these, 14 proteins (CELF1, EIF4H, ELAVL1, FAM120A, FUBP3, IGF2BP3, MATR3, MOV10, NONO, PABPC1, PABPC4, PTBP3, RALY, SND1 and ZC3HAV1) were detected with both probe sets (I/II) and from both viruses ( Figure 2F ).

To understand the regulatory significance of the SARS-CoV-2 RNA interactome, we compiled a list of "neighboring" proteins that are known to physically interact with the factors identified in our study (see Methods for details). In particular, we generated a physical interaction network centered (or seeded) by the core SARS-CoV-2 interactome ( Figure S3A ).

Network analysis revealed several network hubs (e.g. NPM1 and PABPC1) and two highly connected network modules: the ribosomal subunits and the EIF3 complex. GO term enrichment analysis resulted in translation-related biological processes ( Figure S3B ), most likely due to the overrepresentation of ribosomal proteins and subunits of the EIF3 complex, which reflects the active translational status of viral mRNPs.

To achieve a more indepth functional perspective of the RNA interactome, we reconstructed the physical interaction network with the SARS-CoV-2 RNA interactome but excluding ribosomal proteins and EIF3 proteins ( Figure 3A ). This analysis identified additional hub proteins such as TRIM25, SQSTM1, and KHDRBS1. GO term enrichment analysis revealed multiple steps of the mRNA life cycle such as mRNA splicing, mRNA export, mRNA stability, and stress granule assembly ( Figure 3B ), suggesting these mRNA regulators are co-opted to assist the viral life cycle. Interestingly, we also found GO terms related to viral processes and innate immune response. In terms of intracellular localization, the SARS-CoV-2 RNA interactome is enriched by proteins localized in the paraspeckle and cytoplasmic RNP granule (e.g. stress granule) compared to the cellular mRNA interactome 

To investigate the regulation of the SARS-CoV-2 RNA interactome, we utilized published transcriptome data of SARS-CoV-2-infected cells (Blanco-Melo et al., 2020) .

Transcriptome analysis of ACE2-expressing A549 cells revealed host factors of SARS-CoV-2 RNA interactome that are differentially expressed after infection ( Figure 4A ). Specifically, PARP12, SHFL, CELF1, and TRIM25 are up-regulated upon infection. Treatment of ruxolitinib, a JAK1 and 2 inhibitor, in infected cells suppressed the expression of 5 host factors (MOV10, PARP12, SHFL, TRIM25, and ZC3HAV1) ( Figure 4B ). TRIM25 and PARP12 are part of the 62 vertebrate core ISGs (Shaw et al., 2017) . Interferon-beta (IFNß) treatment on normal human bronchial epithelial (NHBE) cells induces PARP12, SHFL, and TRIM25 ( Figure 4C ). Consistently, proteome analysis of IFNß-treated cells (Kerr et al., 2020) exhibited an upregulation of PARP12, TRIM25, and ZC3HAV1 ( Figure S3C 

To measure the impact of these host proteins on coronavirus RNAs, we conducted knockdown experiments ( Figure S4A -B, Table S1 ) and infected Calu-3 cells with SARS-CoV-2 ( Figure 5A and Figure S5A ). Calu-3 cells are human lung epithelial cells and often used as a model system for coronavirus infection (Sims et al., 2008) . Strategically, we selected a subset of the SARS-CoV-2 RNA interactome that covers a broad range of functional modules that we identified above: JAK-STAT signaling, mRNA transport, mRNA stability, and translation.

When we depleted RBPs that are induced by SARS-CoV-2 infection or IFNß treatment, namely PARP12, TRIM25, ZC3HAV1, CELF1, and SHFL, the viral RNA levels increased ( Figure 5B ), which suggests that these RBPs may directly suppress coronaviruses. For instance, ZC3HAV1 (ZAP/PARP13) is an ISG and known to restrict the replication of many RNA viruses such as HIV-1 (Retroviridae), Sindbis virus (Togaviridae), and Ebola (Filoviridae) (Goodier et al., 2015) by promoting RNA degradation (Zhu et al., 2011) and translational repression (Zhu et al., 2012) . ZC3HAV1 recognizes CpG and recruits decay factors to degrade HIV RNAs (Takata et al., 2017) . TRIM25 is required for activation of ZC3HAV1 (Zheng et al., 2017) by ubiquitin ligation which depends on RNA binding activity (Choudhury et al., 2017) . Our knockdown results indicate that both ZC3HAV1 and TRIM25 J o u r n a l P r e -p r o o f may act as antiviral factors against SARS-CoV-2 ( Figure 5B ). Consistently, ectopic expression of the short isoform of ZC3HAV1 (ZC3HAV1-S) showed an antiviral effect against SARS-CoV-2 ( Figure 5D ). The long isoform (ZC3HAV1-L) which is known to have a stronger effect against some RNA viruses (Kerns et al., 2008) did not have a significant effect on SARS-CoV-2, indicating a difference in specificity between the isoforms ( Figure 5D and Figure S5B ). Further investigation is needed to understand how ZC3HAV1 and TRIM25 recognize and suppress SARS-CoV-2 transcripts and if SARS-CoV-2 counteracts these antiviral factors.

Other interferon-stimulated RBPs may also be involved in host defense against SARS-CoV-2. SHFL (Shiftless/RyDEN) was induced upon viral infection and interferon treatment, and suppressed by JAK inhibitor (Figure 4 ). SHFL is known to inhibit the translation of diverse RNA viruses, including dengue virus (Flaviviridae) and HIV (Retroviridae) (Balinsky et al., 2017; Suzuki et al., 2016; Wang et al., 2019) . Under our experimental condition, upregulation of viral RNA was modest in SHFL-depleted cells, but further examination is needed as the knockdown efficiency of ISGs were low in infected cells ( Figure S5A ). Notably, ectopic expression of SHFL significantly downregulated viral growth ( Figure 5D and Figure   S5B ), demonstrating that SHFL indeed has an antiviral activity against SARS-CoV-2, presumably by blocking ribosomal frameshift in ORF1 translation as reported recently by dual fluorescence frameshift reporters (Schmidt et al., 2021) .

Further analyses on other ISGs will also be important for future studies. For instance, PARP12, a cytoplasmic mono-ADP-ribosylation (MARylation) enzyme, is known to have broad antiviral activity against RNA viruses by multiple mechanisms including blocking cellular RNA translation (Atasheva et al., 2014; Welsby et al., 2014) or triggering proteasome-mediated destabilization of viral proteins (Li et al., 2018) . Of note, coronavirus nsp3 carries a conserved macrodomain that can remove ADP-ribose to reverse the activity of PARP enzymes (Fehr et al., 2015) . Knockdown of PARP12 and PARP14 was shown to increase the replication of the macrodomain-deficient mouse hepatitis virus (MHV) which belongs to the lineage A of genus betacoronavirus (Grunewald et al., 2019) , which is in line with our knockdown results ( Figure 5B ). Our RNA interactome data suggests that the RNAbinding activity of PARP12 may help explain the underlying molecular mechanism of its antiviral activity against SARS-CoV-2 transcripts.

We noticed that the coronaviral RNA interactomes are enriched with RBPs with KH domains ( Figure S1C and Figure Probe I/II and OC43 Probe I/II) ( Figure 2C ). FUBP3 binds to the 3′ UTR of cellular mRNAs regulating mRNA localization (Blichenberg et al., 1999; Mukherjee et al., 2019) . Its connection to the life cycle of coronavirus is unknown to our knowledge.

Apart from the above RBPs, we identified multiple host factors that have not been previously described in the context of viral infection. Most notably, LARP1 depletion resulted in a substantial upregulation of viral RNAs ( Figure 5B ). Consistently, ectopic expression of LARP1 led to a reduction of viral growth ( Figure 5D and Figure S5B ), indicating that LARP1 may have an antiviral function. LARP1 is known to recognize the 5′ terminal oligopyrimidine (5′ TOP) motif which is frequently found in mRNAs encoding ribosomal proteins and translation factors (Fonseca et al., 2015; Tcherkezian et al., 2014) . LARP1 represses the translation of 5′ TOP mRNAs in response to metabolic stress, and this repression is relieved by mTORC1-catalyzed phosphorylation of LARP1 (Hong et al., 2017; Lahr et al., 2017) .

reported that LARP1 interacts with the SARS-CoV-2 N protein (Gordon et al., 2020) , although the significance of this interaction remains unknown. SARS-CoV-2 transcripts do not carry the 5′ TOP motif, suggesting that the mechanism of the antiviral activity of LARP1 may be different from that of 5′ TOP mRNA regulation. During the review process of this work, an independent group also confirmed the antiviral function of LARP1 and performed CLIP-seq on LARP1 which revealed that LARP1 may interact with multiple internal pyrimidine-rich sites throughout the body of viral transcripts (Schmidt et al., 2021) , complementing the identification of LARP1 in our RNP capture analysis ( Figure 1H ).

As LARP1 is known to be controlled by mTORC1 (Hong et al., 2017; Lahr et al., 2017) , we examined the effect of mTOR inhibitors on SARS-CoV-2 replication. Lately, mTOR inhibitors such as everolimus and omipalsib were proposed as therapeutic options to mitigate the cytokine storm or to protect from lung fibrosis in severe COVID-19 patients (Karam et al., 2020; Klann et al., 2020; Terrazzano et al., 2020) . A human cell study reported the inhibition of viral replication by sapanisertib which targets both mTORC1 and mTORC2 (Schmidt et al., 2021) . However, in contrast to previous proposals, we found an increase in SARS-CoV-2 replication after treatment with rapamycin ( Figure S5C ) which is the prototypic inhibitor specific to mTORC1 (Jacinto et al., 2004) . These seemingly opposing results begs the question on how LARP1 regulates viral transcripts without the 5′ TOP motif in a J o u r n a l P r e -p r o o f mTORC1-dependent manner. It also needs to be understood how the single/simultaneous inhibition of mTORC1 and mTORC2 affects viral life cycle at the molecular level. In the meantime, a call for caution and rational thinking is needed in terms of the usage of mTOR inhibitors for the treatment of COVID-19 patients (Husain and Byrareddy, 2020).

The SARS-CoV-2 RNA interactome includes specific components of the 40S and 60S ribosomal subunits and translational initiation factors ( Figure 1F-H) . Knockdown experiments indicated that ribosomal proteins (RPS9 and RPS3) and translation initiation factor EIF4H may have antiviral activities ( Figure 5B ). EIF4H along with EIF4B is a cofactor for RNA helicase EIF4A (Rogers et al., 2001 ) whose depletion results in RNA granule formation (Tauber et al., 2020) . EIF4H and EIF4B were both identified as the core SARS-CoV-2 RNA interactome ( Figure 1F ). EIF4H was also reported to interact with SARS-CoV-2 nsp9 (Gordon et al., 2020) Together, our observations implicate that SARS-CoV-2 infection may be closely intertwined with the regulation of ribosome biogenesis, metabolic rewiring, and global translational control.

Regarding proviral factors that may be hijacked by SARS-CoV-2, certain translation factors EIF3A, EIF3D, and CSDE1 exhibited proviral effects ( Figure 5B ). EIF3A is the RNAbinding component of the mammalian EIF3 complex and evolutionarily conserved along with EIF3B and EIF3C (Masutani et al., 2007) . EIF3D is known to interact with mRNA cap and is required for specialized translation initiation . CSDE1 (Unr) is required for IRES-dependent translation in human rhinovirus (Picornaviridae) and poliovirus (Picornaviridae) (Anderson et al., 2007; Boussadia et al., 2003) . In all, our finding suggests that SARS-CoV-2 may recruit EIF3D and CSDE1 to respectively regulate cap-dependent and IRES-dependent translation initiation (Lee et al., 2017) of SARS-CoV-2 gRNA and sgRNAs.

Our current study reveals a broad-spectrum of known antiviral factors such as TRIM25, ZC3HAV1, PARP12, HDLBP, and SHFL and also many RBPs whose antiviral functions are unknown such as LARP1, FUBP3, FAM120A/C, EIF4H, RPS3, RPS9, SND1, CELF1, RALY, and CNBP. Conversely, knockdown of 8 host proteins led to a statistically significant decrease in viral RNAs ( Figure 5B ), suggesting that these host proteins may be proviral (Kamel et al., 2020) . In sum, this list of proteins reflects constant host-pathogen interactions and opens new avenues to explore unknown mechanisms of viral life cycle and immune evasion.

Along with proteins regulating RNAs, it would also be interesting to consider the possibility of 'riboregulation' (Hentze et al., 2018) in which RNA controls its interacting proteins. Dengue virus, for example, uses its subgenomic RNA called sfRNA to sequester TRIM25 (Chapman et al., 2014) . The sgRNA/gRNA ratio is a critical determinant of epidemic potential of dengue virus (Manokaran et al., 2015) . Notably, coronaviruses including SARS-CoV-2 produces substantial amounts of noncanonical sgRNAs that may serve as noncoding decoys to interact with host RBPs to modulate host immune responses (Kim et al., 2020a) . , RNA structure (Lan et al., 2020) , genome-wide CRISPR screen , and off-label drug screening (Riva et al., 2020) have all provided invaluable insights of the underlying biology of this novel human coronavirus. In line with these efforts, our SARS-CoV-2 RNA interactome data, together with the related works reported recently (Flynn et al., 2021; Kamel et al., 2020; Schmidt et al., 2021) , will offer insights into the host-viral interaction that regulate the life cycle of coronaviruses. Data interpretation in the context of publicly available orthogonal information has enabled the identification of proviral and antiviral protein candidates. We expect that further efforts to generate and integrate system-level data will elucidate the pathogenicity of SARS-CoV-2 and introduce new strategies to combat COVID-19. 

We have filed a patent relevant to this paper.

One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in science. Proteins were expressed 24 hours followed by virus infection for another 24 hours. Data are represented as mean ± s.e.m. (n = 3 independent experiments). RNA levels are shown relative to that in HEK293T cells transfected with FLAG-tagged GFP plasmids. 28S rRNA was used for total RNA normalization. * P < 0.05; one-sided Student's t-test.

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, V. Narry Kim (narrykim@snu.ac.kr) Material Availability

All reagents generated in this study are available from the Lead Contact upon request. All 4 cell lines used in this study (HCT-8, male; Lenti-X 293T, female; Calu-3, male; Vero) (Shah et al., 2014) were maintained in culture media supplemented with 10% FBS and 1×

Antibiotic-Aintimycotic solution (Gibco) and routinely cultured at 37℃ with 5% CO 2 . FBS were thawed in a 37℃ water bath and were not heat-inactivated. RPMI 1640 with HEPES (Welgene, LM011-03) were used when culturing HCT-8 while DMEM with high glucose (Welgene, LM001-05) were used for Vero, Calu-3, and Lenti-X 293T cells. During virus infection for both SARS-CoV-2 and HCoV-OC43, the serum concentration was reduced to final 2%. During the infection of HCoV-OC43, the temperature for HCT-8 was lowered to 35℃. PCR test results for mycoplasma contamination were negative for all 4 cell lines (Test order through Bionics, Korea). Cell line authentication for Lenti-X 293T by short tandem repeat (STR) analysis reported 100% match to 293T (CRL-3216), according to the service requested through ATCC (ATCC sales order: SO0054768). 3 other purchased cell lines were morphologically evaluated for their cell identity without STR tests.

Human samples RNA sequencing data from human patients were downloaded from publicly available database from the published work (Blanco-Melo et al., 2020) . No samples from human subjects were used in this study.

No animal experiments were performed in this study.

By scanning the genomic RNAs of SARS-CoV-2 (NCBI RefSeq accession NC_045512.2) and HCoV-OC43 (GenBank accession AY391777.1) from head to tail, partially overlapping 90 nt tiles were enumerated. These tiles were designed to have 30 nt spacing, so adjacent tiles share a subsequence of 60 nt. To avoid ambiguous targeting, tiles were aligned to the human transcriptome (version of Oct 14, 2019) using bowtie 2 (Langmead and Salzberg, 2012 ) and multi-mapped sequences were discarded. To prepare biotinylated antisense oligonucleotides (ASOs) in bulk, the sequence elements for in vitro transcription (IVT), reverse transcription (RT) and PCR were added to the 90 nt tiles. The T7 promoter (5′-TAA TAC GAC TCA CTA TAG GG-3′) and a pad for RT priming (5′-TGG AAT TCT CGG GTG CCA AGG-3′) were added to the head and tail of each tile, respectively. We grouped ASOs into two sets for each viral genome: "Probe I" targets the unique region of genomic RNA Table S1 .

Mass production of biotin-labeled ASO ASO templates were amplified using KAPA HiFi HotStart ReadyMix (Roche) and PCR primers for an ASO pool. PCR products were purified by QIAquick PCR purification kit (QIAGEN). RNA intermediates were then transcribed using MEGAscript T7 transcription kit (Invitrogen), and DNA templates were degraded by TURBO DNase (Invitrogen). To clean up enzymes and other reagents, 1.8X reaction volume of AMPure XP (Beckman) was applied and polyethylene glycol was added to be final 20%. The size selection was carried out according to the manufacturer's protocol. Biotinylated ASOs were synthesized by RevertAid Reverse Transcriptase (Thermo Scientific) and 5′ biotin-TEG primer. RNA intermediates were hydrolyzed at 0.1 M NaOH and neutralized with acetic acid. Finally, ASO purification was performed in the same manner as IVT RNA selection. The primer sequences used for PCR and reverse transcription are listed in Table S1 .

The Uniprot reference proteome sets for human (UP000005640, canonical, SwissProt) and

African green monkey (Chlorocebus sabaeus; UP000029965, canonical, SwissProt and TrEMBL) were used to identify host proteins in each mass spectrometry experiment (version 03/21/2020) (UniProt Consortium, 2019). The reference proteome set for the Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was manually curated largely based on the NCBI Reference Sequence (NC_045512.2) and related literature of other accessory proteins (e.g. ORF3b, ORF9b and ORF9c). The reference proteome set for the Human coronavirus OC43 (HCoV-OC43) was compiled based on the Uniprot Swiss-Prot proteins for HCoV-OC43 (taxonomy:31631) except for HCoV-OC43 Protein I which was separated into Protein Ia and Protein Ib (or N2) (Vijgen et al., 2005) .

Virus experiments were carried out in accordance with the biosafety guideline by the Korea respectively, both with 1× Antibiotic-Antimycotic (Gibco) and 10% FBS (Gibco) and cultured in CO 2 incubator with 5% CO 2 at 37 C. For SARS-CoV-2 infection, 7 × 10 6 Vero cells were plated in T-175 flasks 24 hours before infection. Cells were washed with serum-free media and incubated with 5 mL virus-diluted media for 30 minutes at 0.1 MOI, as determined by plaque assay. After infection, virus containing media was replaced with reduced-serum media (2% FBS) and cultured until the harvest. For HCoV-OC43 infection, a similar protocol was used except for incubation temperature lowered to 33 C. For siRNA transfection, 3.5 × 10 5 Calu-3 cells, maintained in DMEM with 1X Antibiotic-Antimycotic and 10% FBS in CO 2 incubator with 5% CO 2 at 37 C, were plated in 12 well plate and final 50 nM siRNAs were reverse-transfected using Lipofectamine RNAiMAX (Invitrogen) and ON-TARGETplus SMARTpool siRNAs (Horizon Discovery). Cell viability after siRNA knockdown was measured by splitting 1/100 th of cells from uninfected cells, 48 hours after transfection into 96 well plates in triplicates and cell number was measured by MTT assay (Promega) at 4 hours after addition of tetrazolium dye.

For overexpression experiments, CDS sequences were PCR amplified from Calu-3 cDNA and cloned into pCI4-FLAG vector for SHFL, and LARP1. Plasmids for two isoforms of ZC3HAV1 (p3XFLAG-CMV14 ZC3HAV1-S or ZC3HAV1-L) were kindly provided by Gwangseog Ahn's lab at SNU. 2 × 10 5 293T cells were plated in 24 well plate 18 hours before transfection and 1 ug plasmid DNA was mixed with 3 uL of FuGENE HD (Promega).

After 24 hours for protein expression, cells were infected with SARS-CoV-2 at 0.5 MOI.

For total RNA purification from virus-infected cells, 1 mL TRIzol LS (Invitorgen) were added to media-removed cell monolayers per single well of 12 well plates followed by on-column DNA digestion and purification (Zymo Research). For RNA purification from RNP capture sample, bead-captured RNAs were digested with 100 ng Proteinase K (PCR grade, Roche) and incubated at 37˚C for 1 hour, followed by RNA isolation by TRIzol LS with GlycoBlue (Invitrogen). 1~5 µg RNA were reverse-transcribed using RevertAid transcriptase (Thermo Scientific) and random hexamer. qPCR was performed with primer pairs listed in Table S1 and PowerSYBR Green (Applied Biosystems) and analyzed with QuantStudio 5 (Thermo Scientific). Mass spectrometric raw data files were processed for Label-Free Quantification with MaxQuant (version 1.6.15.0) (Cox and Mann, 2008) using the built-in Andromeda search engine (Cox et al., 2011) at default settings with a few exceptions. Briefly, for peptidespectrum match (PSM) search, cysteine carbamidomethylation was set as fixed modifications, and methionine oxidation and N-terminal acetylation were set as variable modifications. Tolerance for the first and main PSM search were 20 and 4.5 ppm, respectively. Peptides from common contaminant proteins were identified by utilizing the contaminant database provided by MaxQuant. FDR threshold of 1% was used for both the peptide and protein level. The match-between-runs option was enabled with default parameters in the identification step. Finally, LFQ was performed for those with a minimum ratio count of 1.

Statistical analysis for RNP capture experiment To identify host and viral proteins that interact with the particular RNA species of interest (e.g. sgRNA or gRNA), we utilized the results from the "bead only" and "probe only" samples as technical backgrounds. Specifically, the "bead only" (or no-probe) experiment in infected cells was used to account for non-specific interactors and biotin-containing carboxylases (e.g. PCCA, ACACA, and ACACB) and determine the set of host and viral proteins that in a broad sense bind to the RNA, which we call Probe I/II "binding" proteins. The probe experiment in uninfected cells (i.e. "probe only") was then used as the technical background against target J o u r n a l P r e -p r o o f RNA-independent interactors and determine the set of host proteins that are enriched for the target RNA, which we call Probe I/II "enriched" proteins.

To accomplish this, we considered the protein spectra count data as a multinomial distribution and applied a statistical test for spectra count enrichment. This is analogous to modelling a bag of skittles and statistically identifying whether you received statistically more "green" skittles than random. In this study, the control samples were used to estimate the parameters of the multinomial distribution of the null hypothesis. In turn, this is normalizing the spectra count data by the total counts, and the basis for this is our multinomial modelling of the data generative process.

Specifically, let N_p be the number of identified spectra counts for protein group p from the case experiment (e.g. Probe I experiment in infected cells), and M_p be the respective count number from the control experiment (e.g. no-probe experiment in infected cells). For each protein with ≥ 1, the statistical significance of enrichment is:

where = ∑ is the total spectra count, is the background probability, and ( ; , ) is the binomial distribution of successes in trails with success probability .

Finally, the Benjamini-Hochberg method was used to adjust the p-values and control the false discovery rate.

We conducted enrichment analyses of Gene Ontology (GO) terms (Gene Ontology Consortium, 2001) by means of summarizing the function of tens of host proteins identified in the RNP capture experiment. In general, Fisher's exact test is used to estimate the statistical significance of the association (i.e. contingency) between a particular GO term and the gene set of interest. To improve the explanatory power of this analysis, we used the weight01 algorithm (Alexa et al., 2006) from the topGO R package which accounts for the GO graph structure and reduces local dependencies between GO terms. Detailed information of the Gene Ontology was from the GO.db R package (version 3.8.2), and GO gene annotations were from the org.Hs.eg.db R package (version 3.8.2).

We integrated protein-protein interaction data from the BioGRID database (Release 3.5.187) (Stark et al., 2006) and retrieved other proteins that do not necessarily bind to the SARS-CoV-2 RNA but form either transient or stable physical interactions with the host proteins identified from the RNP capture experiments. In detail, we considered only human proteinprotein interactions that were (1) found from at least two different types of experiments and (2) reported by at least three publication records which resulted in a total of 65,625 interactions covering 12,143 human proteins. Physical interactions between SARS-CoV-2 proteins and human proteins were by affinity capture and mass spectrometry in SARS-CoV-2 protein expressing cells (Gordon et al., 2020) . The network R package and the ggnet2 function of the GGally R package was used for graph visualization.

Pfam database (version 33.1) (El-Gebali et al., 2019) was used for protein domain enrichment analysis. Taxon 9606 (human) and Taxon 60711 (green monkey) protein domain annotations were used to analyze RNP capture results of HCoV-OC43 and SARS-CoV-2, respectively. One-sided Fisher's exact test was applied to estimate the statistical enrichment of a particular protein domain for the specific gene set (e.g. SARS-CoV-2 Probe I binding proteins). We utilized the set of all proteins identified in the RNP capture experiments and all protein domains annotated to those proteins as the statistical background of the enrichment analysis.

To investigate the subcellular localizations of the SARS-CoV-2 interactome, we leveraged the protein subcellular localization information from the Human cell map database v1 (Go et al., 2019) . Information from the SAFE algorithm was used primarily but then supplemented by information from the NMF algorithm in case of "no prediction" or "-" localizations.

Localization terms of the NMF algorithm were matched to terms of the SAFE algorithm in general, but few were mapped to the higher term of the SAFE algorithm. For example, the "cell junction" term of the NMF algorithm was merged to the "cell junction, plasma membrane" term of the SAFE algorithm.

J o u r n a l P r e -p r o o f Supplementary Table Legends   Table S1 . Oligonucleotide sequences, Related to Figure 1B , 5B and 5C. Table S2 . SARS-CoV-2-RNA-binding, Related to Figure 1C and S1C. Table S3 . HCoV-OC43-gRNA-binding, Related to Figure 2C and S2D. Table S4 . HCoV-OC43-sgRNA-binding, Related to Figure 2C and S2D. 

Improved scoring of functional groups from gene expression data by decorrelating GO graph structure

Internal initiation of translation from the human rhinovirus-2 internal ribosome entry site requires the binding of Unr to two distinct sites on the 5' untranslated region

Interferon-stimulated poly(ADP-Ribose) polymerases are potent inhibitors of cellular translation and virus replication

IRAV (FLJ11286), an Interferon-Stimulated Gene with Antiviral Activity against Dengue Virus, Interacts with MOV10

The mRNA-bound proteome and its global occupancy profile on protein-coding transcripts

SARS-CoV-2 Disrupts Splicing, Translation, and Protein Trafficking to Suppress Host Defenses

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

Identification of a cis-acting dendritic targeting element in MAP2 mRNAs

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

The Global Phosphorylation Landscape of SARS-CoV-2 Infection

Unr is required in vivo for efficient initiation of translation from the internal ribosome entry sites of both rhinovirus and poliovirus

In vitro reconstitution of SARS-coronavirus mRNA cap methylation

Insights into RNA biology from an atlas of mammalian mRNA-binding proteins

The structural basis of pathogenic subgenomic flavivirus RNA (sfRNA) production

RNA-binding activity of TRIM25 is mediated by its PRY/SPRY domain and is required for ubiquitination

MaxQuant enables high peptide identification rates, individualized ppb-range mass accuracies and proteome-wide protein quantification

Andromeda: a peptide search engine integrated into the MaxQuant environment

The severe acute respiratory syndrome-coronavirus replicative protein nsp9 is a single-stranded RNA-binding subunit unique in the RNA virus world

The Pfam protein families database in 2019

The Xist lncRNA exploits threedimensional genome architecture to spread across the X chromosome

The nsp3 macrodomain promotes virulence in mice with coronavirus-induced encephalitis

The coding capacity of SARS-CoV-2

Discovery and functional interrogation of SARS-CoV-2 RNA-host protein interactions

La-related Protein 1 (LARP1) Represses Terminal Oligopyrimidine (TOP) mRNA Translation Downstream of mTOR Complex 1 (mTORC1)

Human Coronavirus: Host-Pathogen Interaction

Creating the gene ontology resource: design and implementation

A census of human RNA-binding proteins

A proximity biotinylation map of a human cell

The Broad-Spectrum Antiviral Protein ZAP Restricts Human Retrotransposition

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

Commentary: Middle East Respiratory Syndrome Coronavirus (MERS-CoV): Announcement of the Coronavirus Study Group

The coronavirus macrodomain is required to prevent PARP-mediated inhibition of virus replication and enhancement of IFN expression

Impaired type I interferon activity and inflammatory responses in severe COVID-19 patients

A new virus isolated from the human respiratory tract

A brave new world of RNAbinding proteins

LARP1 functions as a molecular switch for mTORC1-mediated translation of an essential class of mRNAs

Mammalian TOR complex 2 controls the actin cytoskeleton and is rapamycin insensitive

Global analysis of protein-RNA interactions in SARS-CoV-2 infected cells reveals key regulators of infection

Positive selection and increased antiviral activity associated with the PARP-containing isoform of human zinc-finger antiviral protein

Dynamic rewiring of the human interactome by interferon signaling

The Architecture of SARS-CoV-2 Transcriptome

Identification of Coronavirus Isolated from a Patient in Korea with COVID-19

La-related protein 1 (LARP1) binds the mRNA cap

The Molecular Biology of Coronaviruses

Comparative analysis of RNA genomes of mouse hepatitis viruses

Structure of the full SARS-CoV-2 RNA genome in infected cells

Fast gapped-read alignment with Bowtie 2

Advances in CLIP Technologies for Studies of Protein-RNA Interactions

eIF3d is an mRNA capbinding protein that is required for specialized translation initiation

Regulation Mechanisms of Viral IRES-Driven Translation

PARP12 suppresses Zika virus infection through PARP-dependent degradation of NS1 and NS3 viral proteins

Dengue subgenomic RNA binds TRIM25 to inhibit interferon expression for epidemiological fitness

Reconstitution reveals the functional core of mammalian eIF3

RAP-MS: A Method to Identify Proteins that Interact Directly with a Specific RNA Molecule in Cells

The Xist lncRNA interacts directly with SHARP to silence transcription through HDAC3

β-Actin mRNA interactome mapping by proximity biotinylation

Severe acute respiratory syndrome coronavirus nsp1 suppresses host gene expression, including that of type I interferon, in infected cells

An RNA-centric dissection of host complexes controlling flavivirus infection

Coronavirus as a possible cause of severe acute respiratory syndrome

The PRIDE database and related tools and resources in 2019: improving support for quantification data

Coronaviruses post-SARS: update on replication and pathogenesis

Methods to study RNA-protein interactions

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

Modulation of the helicase activity of eIF4A by eIF4B, eIF4H, and eIF4F

Molecular biology: Rap and chirp about X inactivation

Interplay between SARS-CoV-2 and the type I interferon response

The SARS-CoV-2 RNA-protein interactome in infected human cells

Do you know the sex of your cells?

Fundamental properties of the mammalian innate immune system revealed by multispecies comparison of type I interferon responses

SARS-CoV-2 Nsp1 suppresses host but not viral translation through a bipartite mechanism

SARS-CoV replication and pathogenesis in an in vitro model of the human conducting airway epithelium

The Nonstructural Proteins Directing

Continuous and Discontinuous RNA Synthesis in Coronaviruses

BioGRID: a general repository for interaction datasets

A SARS-CoV-2 BioIDbased virus-host membrane protein interactome and virus peptide compendium: new proteomics resources for COVID-19 research

The nsp9 replicase protein of SARS-coronavirus, structure and functional insights

Characterization of RyDEN (C19orf66) as an Interferon-Stimulated Cellular Inhibitor against Dengue Virus Replication

CG dinucleotide suppression enables antiviral defence targeting nonself RNA

Modulation of RNA Condensation by the DEAD-Box Protein eIF4A

Proteomic analysis of cap-dependent translation identifies LARP1 as a key regulator of 5'TOP mRNA translation

Structural basis for translational shutdown and immune evasion by the Nsp1 protein of SARS-CoV-2

The viral protein NSP1 acts as a ribosome gatekeeper for shutting down host translation and fostering SARS-CoV-2 translation

The Future of Cross-Linking and Immunoprecipitation (CLIP)

UniProt: a worldwide hub of protein knowledge

A large-scale binding and functional map of human RNA-binding proteins

Complete genomic sequence of human coronavirus OC43: molecular clock analysis suggests a relatively recent zoonotic coronavirus transmission event

Regulation of HIV-1 Gag-Pol Expression by Shiftless

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

PARP12, an interferon-stimulated gene involved in the control of protein translation and inflammation

Discovery of seven novel Mammalian and avian coronaviruses in the genus deltacoronavirus supports bat coronaviruses as the gene source of alphacoronavirus and betacoronavirus and avian coronaviruses as the gene source of gammacoronavirus and deltacoronavirus

TRIM25 Is Required for the Antiviral Activity of Zinc Finger Antiviral Protein

A pneumonia outbreak associated with a new coronavirus of probable bat origin

Zinc-finger antiviral protein inhibits HIV-1 infection by selectively targeting multiply spliced viral mRNAs for degradation

Translational repression precedes and is