key: cord-0904094-a8nztlth
authors: Kamel, Wael; Noerenberg, Marko; Cerikan, Berati; Chen, Honglin; Järvelin, Aino I.; Kammoun, Mohamed; Lee, Jeffrey Y.; Shuai, Ni; Garcia-Moreno, Manuel; Andrejeva, Anna; Deery, Michael J.; Johnson, Natasha; Neufeldt, Christopher J.; Cortese, Mirko; Knight, Michael L.; Lilley, Kathryn S.; Martinez, Javier; Davis, Ilan; Bartenschlager, Ralf; Mohammed, Shabaz; Castello, Alfredo
title: Global analysis of protein-RNA interactions in SARS-CoV-2 infected cells reveals key regulators of infection
date: 2021-05-24
journal: Mol Cell
DOI: 10.1016/j.molcel.2021.05.023
sha: a770fb30cc84a059ce8471f409f932ea15b4bfcd
doc_id: 904094
cord_uid: a8nztlth

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) causes COVID-19. SARS-CoV-2 relies on cellular RNA-binding proteins (RBPs) to replicate and spread, although which RBPs control its life cycle remains largely unknown. Here, we employ a multi-omic approach to identify systematically and comprehensively the cellular and viral RBPs that are involved in SARS-CoV-2 infection. We reveal that SARS-CoV-2 infection profoundly remodels the cellular RNA-bound proteome, which includes wide-ranging effects on RNA metabolic pathways, non-canonical RBPs and antiviral factors. Moreover, we apply a new method to identify the proteins that directly interact with viral RNA, uncovering dozens of cellular RBPs and six viral proteins. Amongst them, several components of the tRNA ligase complex, which we show regulate SARS-CoV-2 infection. Furthermore, we discover that available drugs targeting host RBPs that interact with SARS-CoV-2 RNA inhibit infection. Collectively, our results uncover a new universe of host-virus interactions with potential for new antiviral therapies against COVID-19.

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) emerged in Wuhan, In this study, we employ multiple proteome-wide approaches to discover the role of 26

RBPs in SARS-CoV-2 life cycle. We discover that the repertoire of cellular RBPs widely 27 remodels in response to SARS-CoV-2 infection, affecting proteins involved in RNA 28 metabolism, antiviral defences and other pathways. Moreover, we identify the cellular 29 and viral proteins that interact with SARS-CoV-2 RNAs employing a new approach Table S1 ). This suggests that early RBP responses are either subtle or are 26 variable across replicates. Conversely, 335 RBPs were significantly altered at 24 hpi. Of 27 these, 176 showed increased and 159 decreased RNA-binding activity ( Figure 1G and 28 Table S1 ). Importantly, SARS-CoV-2-regulation affects both classical RBPs and 29

unorthodox RBPs lacking known RNA-binding domains (RBDs) ( Figure 1F ). Moreover, 30 regulated RBPs, and especially those stimulated by SARS-CoV-2, include proteins 31 annotated by GO terms and KEGG pathways related to antiviral response and innate 32

The kinetics of RBP activation and inhibition can be informative for protein complex 23 dynamics and function. To further characterise RBP responses after SARS-CoV-2 24 infection, we clustered proteins based on their cRIC fold changes at 8 and 24 hpi. Our 25 analysis distinguishes eight RBP response profiles ( Figure 3A and Table S4 ). Clusters 2 26 and 7 are dominant, with 114 proteins in each group, reflecting that most RBPs changes 27 are only detected at 24 hpi. By contrast, 70 RBPs exhibited more complex RNA-binding 28 patterns, distributing across clusters 1, 3, 4, 5, 6 and 8. Table S4 ). Conversely, the 26 cap-and poly(A)-binding proteins eIF4E and PABPC1, as well as the other translation 27 initiation factors such as EIF4A1 and EIF4A2, EIF4B and EIF4G1 and EIF4G3, are 28 present in cluster 2, which is comprised of upregulated RBPs ( Figure 3A and B, and 29 Table S4 ). These opposed results support a model in which the cap-and poly(A)-binding 30 factors can interact with cellular mRNAs but cannot associate with EIF3 and the 31 ribosomal subunit 40S, which agrees with the reported action of NSP1 preventing 40S 

To determine whether the changes that SARS-CoV-2 induces in the cellular RBPome 30 are shared with other viruses, we compared the SARS-CoV-2 cRIC data to that of SINV 31 J o u r n a l P r e -p r o o f 9 (Garcia-Moreno et al., 2019). SINV is a positive stranded virus from the alphavirus 1 genus. As SARS-CoV-2, SINV genome is capped and polyadenylated, although it is 2 substantially smaller (~11kb vs ~30kb). Moreover, both viruses produce subgenomic 3

RNAs and replicate in the cytoplasm. Strikingly, nearly 40% of the changes in RBP 4 activity observed in SARS-CoV-2 were also present in the SINV cRIC dataset ( Figure  5 4A-C). This exciting result indicates that even if these viruses belong to different families 6 and have little or no sequence homology, they cause similar alterations in the RBPome 7 that are consistent for both upregulated and downregulated RBPs ( Figure 4A and B). 8

Several antiviral factors were noticeable amongst the 93 RBPs with consistent 9 responses, TRIM25, TRIM56, ZC3HAV1 (also ZAP), DHX36 and GEMIN5 ( Figure 4D  10 and S4A). These antiviral RBPs are upregulated in both datasets, suggesting that they 11 are likely involved in the antiviral response against both SARS-CoV-2 and SINV. 12 TRIM25 is an E3 ubiquitin ligase whose catalytic activity is triggered by RNA binding and 13 Interestingly, 12% of the proteins exhibited opposite behaviour in the two viral models. 30

Many of these can be traced back to membraneless organelles such us paraspeckles 31 and stress granules. The core paraspeckle components NONO, PSPC1, SFPQ and 32 MATR3 display opposite trends, being repressed by SINV and stimulated or unaffected 33 J o u r n a l P r e -p r o o f 10 by SARS-CoV-2 ( Figure 4D and S4D) . It is proposed that paraspeckles are critical to 1 sequester proteins and/or mRNAs to regulate gene expression, although the importance 2 of paraspeckle proteins in virus infection remains poorly understood (Fox et al., 2018). 3

Similar anticorrelation was observed with the stress granule proteins G3BP1 and G3BP2 4 ( Figure 4D and S4D) . Stress granules play a defensive role against viruses by CoV-2 infected cells may thus reflect an opposite outcome, i.e. lower association with 10 RNA due to the induction of stress granules. 11

The SARS-CoV-2 RNA interactome 12 cRIC captures both SARS-CoV-2 and cellular mRNAs, which represent 14-19% and 80-13 84% of the eluted RNA, respectively ( Figure 2D and S2F). Therefore, it is not possible to 14 know a priori which of the observed protein-RNA interactions are driven by viral RNA. To 15 systematically identify the RBPs that interact directly with SARS-CoV-2 RNAs, we 16 applied a newly developed approach that we named viral RNA interactome capture 17 (vRIC) ( Figure 5A and B and S5A). In brief, SARS-CoV-2-infected and uninfected Calu-3 18 cells are treated with the RNA polymerase II (RNAPII) specific inhibitor flavopiridol (Fvo), 19

followed by a pulse with the photoactivatable nucleotide analogue 4-thiouridine (4SU). 20

As viral RNA polymerases are insensitive to Fvo, temporal inhibition of RNAPII causes 21 4SU to be predominantly incorporated into nascent viral RNAs. Cells are then UV 22 irradiated at 365 nm to induce crosslinks between viral RNA and proteins placed at a 23 'zero distance' from the 4SU molecules. As natural nucleotide bases do not absorb UV 24 at 365 nm, protein-RNA crosslinking is restricted to 4SU-containing viral RNA. Cells are 25 then lysed under denaturing conditions and poly(A)-containing RNA is captured with 26 oligo(dT) following a previously designed robust procedure (Castello et al., 2012). After 27 elution, proteins co-purified with the viral RNA are analysed by proteomics. 28

Our control experiments showed that Fvo strongly abrogates RNAPII transcription from 29 a strong tetracycline-inducible cytomegalovirus promoter, and that neither Fvo nor 4SU 30 interfered with SARS-CoV-2 replication ( Figure 5C and S5A-C). In mock cells, 4SU 31 incorporation followed by 365nm UV crosslinking and oligo (dT) capture led to the 32 J o u r n a l P r e -p r o o f 11 isolation of the steady state RBPome ( Figure S5E -I). However, when 4SU was omitted 1 or Fvo was added, the amount of protein co-isolated with RNA was massively reduced in 2 both silver staining and proteomic analyses ( Figure S5D , F and G). These results show 3 that active RNAPII is required in uninfected cells to achieve efficient 4SU-dependent 4 protein-RNA UV crosslinking. Conversely, when cells were infected with SARS-CoV-2, 5 efficient protein isolation was observed despite Fvo treatment ( Figure 5D -E and S5E-I). proteins at 10% FDR ( Figure 5E , Table S5 ). The SARS-CoV-2 mRNA interactome is 13 enriched in proteins annotated by the GO term 'RNA binding' (89%) and harbouring 14 known RBDs (65%) ( Figure 5F -G), supporting the capacity of vRIC to identify bona fide 15 protein-RNA interactions. The SARS-CoV-2 RNA interactome is enriched in GO terms 16 associated RNA metabolism (RNA splicing, transport, stability, silencing and translation), 17 antiviral response (e.g. RIGI pathway), cytoplasmic granule assembly (stress granules 18 and P-bodies), and virus biology (e.g. viral process, dsRNA binding, IRES-dependent 19 viral RNA translation) ( Figure 5H ). Notably, 8 and 9 proteins were annotated by innate 20 immunity related terms in KEGG and GO, respectively ( Figure 5I) . 21

Recently, a complementary SARS-CoV-2 RNA interactome has been generated in 22 SARS-CoV-2 infected hepatoma (Huh-7) cells using RAP-MS, which combines UV 23 crosslinking and specific antisense probes (Schmidt et al., 2020). Gratifyingly, this 24 dataset overlaps well with our vRIC data despite being generated with different cell 25 types (hepatocytes versus lung epithelial cells) and methods (RAP-MS versus vRIC) 26 ( Figure S5J ). However, vRIC identified substantially more RBPs than RAP-MS at all 27 FDR cut-offs tested, providing additional SARS-CoV-2 RNA interactors. 28

To determine to what extent the SARS-CoV-2 RNA interactome harbours cellular RBPs 29 that are also present in the RNPs of other viruses, we compared the SARS-CoV-2 vRIC 30 to a SINV vRIC dataset generated in a parallel study (Kamel et al., In preparation). The 31 SARS-CoV-2 vRIC dataset is smaller than the SINV counterpart, likely due to the limited 32 starting material available ( Figure S5K ). Nevertheless, 60% of the RBPs within the 1 SARS-CoV-2 RNA interactome were also present in that of SINV ( Figure 5J ). These 2 striking results suggest that viral RNPs may share a larger proportion of cellular factors 3 than previously anticipated, opening the possibility to target commonly used RBPs in 4 broad-spectrum therapeutic approaches. 5

The cRIC analysis revealed global alterations of the translation machinery ( Figure 3B EIF4G3, EIF4A1, EIF4A2, EIF4B and PABPC1 ( Figure 5E and Table S5 ). However, one 10 of the critical components is missing: the cap-binding protein EIF4E. While we cannot 11 rule out that this missing protein is a false negative, other capped RNA viruses such as 12 EIF3 subunits in the cRIC analysis ( Figure 5E and Table S5 ). These results suggest that 18 even though EIF3 subunits C and D have an overall reduced association with RNAs 19 likely due to NSP1 action, they do interact with SARS-CoV-2 RNA to enable viral protein 20

cRIC revealed an upregulation of many HNRNPs ( Figure S3F ). To test if viral RNA is 22 involved in these alterations, we examined the vRIC dataset. Notably, 10 HNRNPs proportion of the protein that engages in RNA binding. We can thus suggest that 18

ORF1a/b and NCAP establish optimal and stable interactions with RNA, while M, 19

ORF9b and, especially, S mediate shorter-lived and/or geometrically less favourable 20 interactions for crosslinking. However, the high protein sequence coverage and peptide 21 intensity in both vRIC and cRIC experiments strongly support that all these proteins 22 vRIC, which supports that NSP13 only interacts with viral RNA. 8

The proteins M and S also reliably and robustly co-purify with RNA upon cRIC and vRIC 9

( Figure 1E potential role for RNA in the process of particle formation. 28

The viral protein ORF9b was also consistently identified by both cRIC and vRIC, 29

supporting that it is a novel RNA-binding protein ( Figure 1E there is a discrete region in ORF9b that generates a positively charged surface with high 4 probability to interact with nucleic acids ( Figure 6G and S6C and D). Further work is 5 required to define the role of the RNA-binding activity of ORF9b in SARS-CoV-2 6 infection. 7

Therefore, our data reveal seven viral proteins that harbour RNA-binding activity, six of 8 which interact with SARS-CoV-2 RNA. Amongst these, M, S, ORF9b emerge as novel 9

RBPs based on both our study and (Schmidt et al., 2020). 10

To determine if our study has potential for discovery of new regulators of SARS-CoV-2 12 infection, we assessed the incidence of vRIC and cRIC identified proteins in genome 13 wide screens with other viruses. The superset includes studies using RNA interference 14 (RNAi), CRISPR-Cas9, and haploid line screens for 36 viruses (Table S7 ). This analysis 15 revealed that cRIC and vRIC identified 47 RBPs linked to phenotypes in functional 16 screenings (>3 studies; Figure 7A , B, S7A, Table S7 ). Moreover, we used an automated 17 To determine the biomedical potential of cellular RBPs for COVID-19 treatment, we 23 compared the subset of RBPs stimulated by SARS-CoV-2 infection and the subset of 24 proteins that interact with SARS-CoV-2 RNA to drug databases ( Figure S7C ). 25

Importantly, 54 proteins within these datasets have potential inhibitors available ( Figure  26 S7C). To prove the value of these RBPs as therapeutic targets, we tested five drugs in 27

Calu-3 cells infected with SARS-CoV-2 ( Figure 7C and S7D). Our results show that two 28 of these compounds targeting HSP90 and IGF2BP1 (IMP1) cause a strong inhibition of 29 SARS-CoV-2 protein production, with two additional drugs targeting ELAVL1 (HuR) and 30 MSI2 causing moderate effects and one compound targeting PKM having slight effects. 31

The anti-SARS-CoV-2 effects of HSP90 inhibitors have been recently confirmed by an 1 independent study (Wyler et al., 2021). These results reflect the potential of RBPs as 2 targets for antiviral drugs. Figure S7E ). Importantly, silencing of DDX1 caused 2 a strong reduction of intracellular SARS-CoV-2 RNA that correlates with a parallel 3 reduction of NCAP ( Figure 7E -F and S7E). FAM98A KD led to milder effects in both viral 4 RNA levels and NCAP accumulation ( Figure 7F -G and S7E). Since DDX1 is a core 5 subunit of the tRNA-LC and FAM98A is secondary, these differential effects are 6 expected. To provide further insights into the effects of DDX1 KD, we generated 7

RNAseq data. We observed that DDX1 KD equally affected all viral transcripts, despite 8 having no effect on cell viability ( Figure 7H ). Conversely, DDX1 KD had no detectable 9 effects in xbp1 mRNA expression and splicing and, with few exceptions, most UPR 10 response genes remained unaltered ( Figure 7H and S7G-J). This indicates that either 11 DDX1 KD leaves sufficient tRNA-LC for the xbp1 mRNA splicing to occur or, 12 alternatively, that SARS-CoV-2 does not produce a strong UPR response in A549-ACE2 13 cells. These results suggest that tRNA-LC plays a role in SARS-CoV-2 replication. 14

We provide a systematic and comprehensive analysis of protein-RNA interactions in 16

remodelling of the RBPome which involves the upregulation and downregulation of more 18 than three hundred RBPs. We also discovered dozens of cellular proteins that interact 19

with SARS-CoV-2 RNAs, which are promising for the development of new therapeutic 20 approaches. Importantly, we find shared host-virus interactions between the viruses 21 SARS-CoV-2 and SINV, which reflect the existence of cellular RBPs with 'master' 22 regulatory roles in virus infection. Similar work with other viruses and cell types will 23 expand our knowledge on these critical protein-RNA interactions. The relevance and 24 complementarity of our datasets is illustrated by the discovery of the tRNA-LC as a key 25 regulator of SARS-CoV-2 as well as RBP-targeting compounds with antiviral activity. 26

Our study also discovers novel viral RNPs, including S, M and ORF9b, opening new 27 angles to investigate their roles in SARS-CoV-2 infection. 28

In the future, cRIC and vRIC could be extended to other coronaviruses and other 29 biological models such as primary cells and organoids. Generating additional time points 30 and using replication inhibitors such as Remdesivir, it will be possible to study the 31 dynamics of viral RNPs throughout the infection. Moreover, combining such approaches 32 with CLIP-based methods will make it possible to identify the motifs that cellular RBPs 1 recognise in viral RNAs and will provide new insights into their function in infection. We 2 are hopeful that this work will further shed light on the pathogenesis of SARS-CoV-2 and 3 accelerate the discovery of therapies for COVID 19. 4

As any proteomic approach, RIC and vRIC have a bias related to protein abundance, 6 size, and physicochemical properties of their tryptic peptide sequences. UV irradiation 7

induces RNA-to-protein crosslinks in a very specific manner as it requires zero 8 'distances'. However, the higher specificity comes at a price of lower efficiency when 9 compared to chemical crosslinkers such as formaldehyde. UV underperforms with 10 transitory interactions and contacts with the ribose-phosphate backbone as the hpi for SINV. * FDR < 20% ; ** FDR < 10% and *** FDR < 1%. 20 

Further information and requests for resources and reagents should be directed to and 4 will be fulfilled by the Lead Contact, Alfredo Castello (alfredo.castello@glasgow.ac.uk). 5

Material is available upon request from the authors. 7

The mass spectrometry proteomics data have been deposited to the ProteomeXchange 9

Consortium via the PRIDE partner repository with the dataset identifier PXD023418. The replaced with primary antibodies specific for SARS-CoV NCAP (Table S8) x:y:z and images were deconvolved with maximum likelihood algorithm using cellSens 6 (5 iterations, default PSF, no noise reduction, Olympus). Background subtraction was 7 performed on all channels using rolling ball subtraction method (radius = 250 px) in 8 ImageJ (National Institutes of Health). Fluorescence intensity profiles were obtained 9

using ImageJ "Plot profile" tool across 8 µm regions on 0.4 µm max intensity z-projected 10 images. Voxel intensities were normalized to maximum intensity value obtained from 11 'SARS-CoV-2 infected' condition. 12

To induce shRNA expression A549-Ace2 cells and the derived shRNA lines were 14 cultured in doxycycline containing media (1 µg/ml) for >14 days. Peptides were loaded onto a trap-column (Thermo Scientific PepMap 100 C18, 5 µm 7 particle size, 100A pore size, 300 µm i.d. x 5mm length) and separation of peptides was 8 performed by C18 reverse-phase chromatography at a flow rate of 300 nL/min and a 9 reverse-phase nano Easy-Spray column (Thermo Scientific PepMap C18, 2µm particle 10 size, 100A pore size, 75µm i.d. x 50cm). WCP peptides were acquired in a 120 min run Peptides were searched against reference Uniport datasets: human proteome 32 (Uniprot_id: UP000005640, downloaded Nov2016) and SARS-CoV-2 (Uniprot_id: 1 UP000464024, downloaded 24June2020). False discovery rate (FDR) was set at 1% for 2 both peptide and protein identification. For cRIC and WCP samples, MaxQuant search 3 was performed with "match between run" activated. For vRIC samples, since each 4 sample was analyzed on both Eclipse and QExactive mass spectrometers, raw spectra 5 form both runs were combined as separate fractions in the MaxQuant search (the 6 spectra from the Eclipse was assigned fraction 1 and the spectra from the QExactive is 7 assigned fraction 5, and each sample was as independent experiment). 

Using the GO annotation available via the GO.db R package (3.11.4), GO terms 5 including the term 'RNA binding' (to annotate RNA-binding related functions, processes, 6 or compartments) or term 'immun' or exact terms 'immune response' and 'innate immune 7 response' (to annotate immunity related functions, processes, or compartments) were 8 selected. The full list of terms is provided as a supplementary table (Table S9 ). The R 9 package org.Hs.eg.db (3.11.4) was used to identify the genes (proteins) in our dataset 10 that are annotated to these GO terms using the cross-database id mapping functionality. 11 GO enrichment analysis was performed using PANTHER classification system 12 (http://www.pantherdb.org) (Mi et al., 2019) . 13

KEGG pathways under the 'Immune system' category in the high-level KEGG hierarchy 15 available via the R package "KEGGREST" (1.28.0) were selected (see tableS9) and 16 genes mapping to these pathways were identified using "org.Hs.eg.db." Table S9 . The proteins containing these domains were 25 identified using org.Hs.eg.db. 26

To automatically query the NCBI Entrez Utilities REST API, the R package "rentrez" 28 ). This corrects for the fact that in RNA-seq data, variance grows with the mean and 24 therefore, without suitable correction, only the most highly expressed genes drive the 25 clustering. The 500 genes showing the highest variance were used to perform PCA 26 using the "prcomp" function implemented in the base R package "stats" (4.0.2). Finally, 27 differential expression analysis was performed using the R package "DESeq2" (1.28.1). 28 "DESeq2" estimates variance-mean dependence in count data from high-throughput 29 sequencing data and tests for differential expression based on a model using the transcript isoforms between SARS2-infected and uninfected cells (three replicates 3 each). Briefly, we created a flattened exon annotation from protein coding transcripts of 4 genes and lncRNAs using the dexseq_prepare_annotation.py python script 5 accompanying the package. We then assigned the reads into this simplified annotation 6 using dexseq_count.py provided with DEXSeq in a strand-specific fashion. Data was 7 then tested for differential exon usage and estimated exon fold changes using the R 8 package. DEXSeq models count data using a negative binomial (NB) distribution and (resulting in long protein isoform) and those that were not (shorter protein isoform). 27

The "tidyverse suite" (1.3.0) was used for data wrangling in R, and "rtracklayer" • Inhibition of these cellular RBPs hampers SARS-CoV-2 infection.

• The tRNA ligase complex is a key regulator of SARS-CoV-2. 

TRIM25 Enhances the Antiviral 2 Action of Zinc-Finger Antiviral Protein (ZAP)

Moderated estimation of fold change and 4 dispersion for RNA-seq data with DESeq2

Emerging Roles of

Gemin5: From snRNPs Assembly to Translation Control

Translation inhibition and stress granules 8 in the antiviral immune response

Protein Synthesis Initiation in Eukaryotic Cells

Protocol Update for large-scale genome and gene function analysis with the 13 PANTHER classification system (v.14.0)

An RNA-centric dissection of 16 host complexes controlling flavivirus infection

Genetic mechanisms 19 of critical illness in Covid-19

Sequestration of G3BP coupled with efficient translation inhibits 22 1. Early inhibition, late recovery/stimulation