To identify protein–protein interactions of SARS-CoV-2 and SARS-CoV and cellular proteins, we transduced A549 lung carcinoma cells with lentiviruses expressing individual haemagglutinin-tagged viral proteins (Fig. 1a, Extended Data Fig. 1a, Supplementary Table 1). Statistical modelling of quantitative data from affinity purification followed by mass spectrometry (AP–MS) analysis identified 1, 801 interactions between 1, 086 cellular proteins and 24 SARS-CoV-2 and 27 SARS-CoV bait proteins (Fig. 1b, Extended Data Fig. 1b, Supplementary Table 2), substantially increasing the number of reported interactions of SARS-CoV-2 and SARS-CoV1,2,5,6,10,11 (Supplementary Table 10). The resulting virus–host interaction network revealed a wide range of cellular activities intercepted by SARS-CoV-2 and SARS-CoV (Fig. 1b, Extended Data Table 1, Supplementary Table 2). In particular, we observed that SARS-CoV-2 targets a number of key innate immunity regulators (ORF7b–MAVS and ORF7b–UNC93B1), stress response components (N–HSPA1A) and DNA damage response mediators (ORF7a–ATM and ORF7a–ATR) (Fig. 1b, Extended Data Fig. 1c–e). Additionally, SARS-CoV-2 proteins interact with molecular complexes involved in intracellular trafficking (for example, endoplasmic reticulum–Golgi trafficking) and transport (for example, solute carriers and ion transport by ATPases) as well as cellular metabolism (for example, mitochondrial respiratory chain and glycolysis) (Fig. 1b, Extended Data Table 1, Supplementary Table 2). Comparing the AP–MS data of homologous SARS-CoV-2 and SARS-CoV proteins identified differences in the enrichment of individual host targets, highlighting potential virus-specific interactions (Fig. 1b (edge colour), c, Extended Data Figs. 1f, 2a, b, Supplementary Table 2). For instance, we recapitulated the known interactions between SARS-CoV NSP2 and prohibitins12 (PHB and PHB2), but this interaction was not conserved with SARS-CoV-2 NSP2, suggesting that the two viruses differ in their ability to modulate mitochondrial function and homeostasis through NSP2 (Extended Data Fig. 2a). The exclusive interaction of SARS-CoV-2 ORF8 with the TGF-β1–LTBP1 complex is another interaction that potentially explains the differences in pathogenicity of the two viruses (Extended Data Figs. 1f, 2b). Notably, disbalanced TGF-β signalling has been linked to lung fibrosis and oedema, a common complication of severe pulmonary diseases including COVID-1913,14,15,16.

Fig. 1: Joint analysis of SARS-CoV-2 and SARS-CoV virus–host protein–protein interactomes.
figure 1

a, Systematic comparison of interactomes and host proteome changes (effectomes) of the 24 SARS-CoV-2 and 27 SARS-CoV viral proteins, using 3 homologues from human coronaviruses (HCoV-NL63 and HCoV-229E) as reference for pan-coronavirus specificity. b, Combined virus–host protein-interaction network of SARS-CoV-2 and SARS-CoV measured by AP–MS. Homologous viral proteins are displayed as a single node. Shared and virus-specific interactions are denoted by the edge colour. The edge intensity reflects the P-value for the interaction (with the smallest P-value represented by solid edges and the highest P-value (<0.001) represented by faded edges). ECM, extracellular matrix; ER, endoplasmic reticulum; GPCR, G-protein-coupled receptor; HOPS, homotypic fusion and protein-sorting; MHC, major histocompatibility complex; SNARE, soluble N-ethylmaleimide-sensitive factor attachment protein receptor; COG, conserved oligomeric Golgi. c, The numbers of unique and shared host interactions between the homologous proteins of SARS-CoV-2 and SARS-CoV.

To map the virus–host interactions to the functions of viral proteins, we conducted a study of total proteomes of A549 cells expressing 54 individual viral proteins comprising the ‘effectome’ (Fig. 1a, Supplementary Table 3). This dataset provides clear links between changes in protein expression and virus–host interactions, as exemplified by ORF9b, which leads to a dysregulation of mitochondrial functions and binds to TOMM70, a known regulator of mitophagy2,17 (Fig. 1b, Supplementary Tables 2, 3). Global pathway-enrichment analysis of the effectome dataset confirmed that ORF9b of both viruses led to mitochondrial dysregulation2,18 (Extended Data Fig. 2c, Supplementary Table 3) and further highlighted virus-specific effects, as exemplified by the upregulation of proteins involved in cholesterol metabolism (CYP51A1, DHCR7, IDI1 and SQLE) by SARS-CoV-2 NSP6 but not by SARS-CoV NSP6. Of note, cholesterol metabolism was recently shown to be implicated in SARS-CoV-2 replication and has been suggested as a promising target for drug development19,20,21. Besides perturbations at the pathway level, viral proteins also specifically modulated single host proteins, possibly explaining more specific molecular mechanisms involved in viral protein function. Focusing on the 180 most affected host proteins, we identified RCOR3, a putative transcriptional corepressor, as strongly upregulated by NSP4 of both viruses (Extended Data Figs. 2d, 3a). Notably, apolipoprotein B (APOB) was substantially regulated by ORF3 and NSP1 of SARS-CoV-2, suggesting that it has an important role in SARS-CoV-2 biology (Extended Data Fig. 3b).

Multi-omics profiling of virus infection

Although the interactome and the effectome provide in-depth information on the activity of individual viral proteins, we aimed to directly study their combined activities in the context of viral infection. To this end, we infected A549 cells expressing angiotensin-converting enzyme 2 (ACE2) (A549-ACE2 cells) (Extended Data Fig. 4a, b) with SARS-CoV-2 or SARS-CoV, and profiled the effects of viral infection on mRNA expression, protein abundance, ubiquitination and phosphorylation in a time-resolved manner (Fig. 2 a-b).

Fig. 2: Multi-level profiling of SARS-CoV-2 and SARS-CoV infection.
figure 2

a, Time-resolved profiling of parallel SARS-CoV-2 and SARS-CoV infection by multiple omics methods. The plot shows the mass spectroscopy (MS) intensity estimates for spike proteins of SARS-CoV-2 and SARS-CoV over time (n = 4 independent experiments). MOI, multiplicity of infection. b, The numbers of distinct transcripts, proteins, ubiquitination and phosphorylation sites that are significantly up- or downregulated at given time points after infection (relative to mock infection at the same time point). Transcripts, proteins or sites that are regulated similarly by SARS-CoV-2 and SARS-CoV infection are shown in grey, those regulated specifically by SARS-CoV-2 are in orange and those regulated by SARS-CoV are in brown. c, d, Comparison of host transcriptome 12 h (c) and ubiquitinome 24 h (d) after infection (hpi) with SARS-CoV-2 (x-axis) or SARS-CoV (y-axis) (log2 fold change in comparison to the mock infection samples at the same time point). Significantly regulated transcripts by moderated t-test with false discovery rate-corrected two-sided P-value ≤0.05 (c) and significantly regulated sites by Bayesian linear model-based unadjusted two-sided P-value ≤10−3, |log2 fold change| ≥0.5 (d) are coloured according to specificity as indicated. Diamonds indicate that the actual log2 fold change was truncated to fit into the plot. n = 3 independent experiments. e, Phosphorylation (purple squares) and ubiquitination (red circles) sites on EGFR that are regulated upon SARS-CoV-2 infection. The plot shows median log2 fold change of abundance compared with mock infection at 24 and 36 hpi. All identified phosphorylation sites have known regulatory function. f, Profile plots of time-resolved EGFR K754 ubiquitination, EGFR T693 and S991 phosphorylation, and total EGFR protein levels in A549-ACE2 cells infected with SARS-CoV-2 or SARS-CoV with indicated median (line), 50% (shaded region) and 95% (dotted line) confidence intervals. n = 3 (ubiquitination) or 4 (phosphorylation and total protein) independent experiments.

In line with previous reports9,22, we found that both SARS-CoV-2 and SARS-CoV can downregulate the type I interferon response and activate a pro-inflammatory signature at transcriptome and proteome levels (Fig. 2a–c, Extended Data Fig. 4c–f, i, Supplementary Tables 4, 8, Supplementary Discussion 1). However, SARS-CoV elicited a more pronounced activation of the NF-κB pathway, correlating with its higher replication rate and potentially explaining the lower severity of pulmonary disease in cases of SARS-CoV-2 infection23 (Supplementary Tables 4, 5). By contrast, SARS-CoV-2 infection led to higher expression of FN1 and SERPINE1, which may be linked to the specific recruitment of TGF-β factors (Fig. 1b), supporting regulation of TGF-β signalling by SARS-CoV-2.

To better understand the mechanisms underlying perturbation of cellular signalling, we performed comparative ubiquitination and phosphorylation profiling following infection with SARS-CoV-2 or SARS-CoV. This analysis showed that 1,108 of 16,541 detected ubiquitination sites were differentially regulated by infection with SARS-CoV-2 or SARS-CoV (Fig. 2a, b, d, Extended Data Fig. 5a, Supplementary Table 6). More than half of the significant sites were regulated in a similar manner by both viruses. These included sites on SLC35 and SUMO family proteins, indicating possible regulation of sialic acid transport and the SUMO activity. SARS-CoV-2 specifically increased ubiquitination on autophagy-related factors (MAP1LC3A, GABARAP, VPS33A and VAMP8) as well as specific sites on EGFR (for example, K739, K754 and K970). In some cases, the two viruses targeted distinct sites on the same cellular protein, as exemplified by HSP90 family members (for example, K84, K191 and K539 on HSP90AA1) (Fig. 2d). Notably, a number of proteins (for example, ALCAM, ALDH3B1, CTNNA1, EDF1 and SLC12A2) exhibited concomitant ubiquitination and a decrease at the protein level after infection, pointing to ubiquitination-mediated protein degradation (Fig. 2d, Extended Data Figs. 4f, 5a, Supplementary Tables 5, 6). Among these downregulated proteins, EDF1 has a pivotal role in the maintenance of endothelial integrity and may be a link to endothelial dysfunctions described for COVID-1924,25. Profound regulation of cellular signalling pathways was also observed at the phosphoproteomic level: among 16,399 total quantified phosphorylation sites, 4,643 showed significant changes after infection with SARS-CoV-2 or SARS-CoV (Extended Data Fig. 5b, c, Supplementary Table 7). Highly regulated sites were identified for the proteins of the MAPK pathways (for example, MAPKAPK2, MAP2K1, JUN and SRC), and proteins involved in autophagy signalling (for example, DEPTOR, RICTOR, OPTN, SQSTM1 and LAMTOR1) and viral entry (for example, ACE2 and RAB7A) (Extended Data Fig. 5b, d). Notably, RAB7A was recently shown to be an important host factor for SARS-CoV-2 infection that assists endosomal trafficking of ACE2 to the plasma membrane26. We observed higher phosphorylation at S72 of RAB7A in SARS-CoV-2 infection compared with SARS-CoV or mock infection; this site is implicated in RAB7A intracellular localization and molecular association27. The regulation of known phosphorylation sites suggests an involvement of central kinases (cyclin dependent kinases, AKT, MAPKs, ATM, and CHEK1) linked to cell survival, cell cycle progression, cell growth and motility, stress responses and the DNA damage response; this was also supported by the analysis of enriched motifs (Extended Data Fig. 5e, f, Supplementary Tables 7, 8). Notably, SARS-CoV-2 infection, but not SARS-CoV infection, led to phosphorylation of the antiviral kinase EIF2AK2 (also known as PKR) at the critical regulatory residue S3328. This differential activation of EIF2AK2 could contribute to the difference in the growth kinetics of the two viruses (Supplementary Table 4, 5).

Our data clearly point to an interplay of phosphorylation and ubiquitination patterns on individual host proteins. For instance, EGFR showed increased ubiquitination on 6 lysine residues at 24 h post-infection (hpi) accompanied by increased phosphorylation of T693, S695 and S991 at 24 and 36 hpi (Fig. 2e, f). Ubiquitination of all six lysine residues on EGFR was more pronounced following infection with SARS-CoV-2 than with SARS-CoV. Moreover, vimentin, a central co-factor for coronavirus entry29 and pathogenicity30,31, displayed distinct phosphorylation and ubiquitination patterns on several sites early (for example, S420) or late (for example, S56, S72 and K334) in infection (Extended Data Fig. 6a, b). These findings underscore the value of testing different post-translational modifications simultaneously and suggest a concerted engagement of regulatory machineries to modify target protein functions and abundance.

Post-translational modification of viral proteins

The majority of viral proteins were also post-translationally modified. Of the 27 detected SARS coronavirus proteins, 21 were ubiquitinated. Nucleocapsid (N), spike (S), NSP2 and NSP3 were the most heavily modified proteins in both viruses (Extended Data Fig. 6c, Supplementary Table 6). Many ubiquitination sites were common to both viruses. Around half of the sites that were exclusively ubiquitinated in either virus were conserved between SARS-CoV and SARS-CoV-2. The remaining specifically regulated ubiquitination sites were unique to each virus, indicating that these acquired adaptations can be post-translationally modified and may recruit cellular proteins with distinct functions (Fig. 3a). Our interactome data identified several host E3 ligases (for example, we identified interactions between SARS-CoV-2 ORF3 and TRIM47, WWP1, WWP2 and STUB1; and between SARS-CoV-2 M and TRIM7) and deubiquitinating enzymes (for example, interactions between SARS-CoV-2 ORF3 and USP8; SARS-CoV-2 ORF7a and USP34; and SARS-CoV N and USP9X), suggesting crosstalk between ubiquitination and viral protein functions (Fig. 1b, Extended Data Fig. 6d, Supplementary Table 2). Of particular interest are extensive ubiquitination events on the S protein of both viruses (K97, K528, K825, K835, K921 and K947), which are distributed on functional domains (N-terminal domain, C-terminal domain (CTD), fusion peptide and heptad repeat 1 domain), potentially indicating critical regulatory functions that are conserved between the two viruses (Extended Data Fig. 6e). We observed phosphorylation of 5 SARS-CoV-2 proteins (M, N, S, NSP3 and ORF9b) and 8 SARS-CoV proteins (M, N, S, NSP1, NSP2, NSP3, ORF3 and ORF9b) (Extended Data Fig. 6f, Supplementary Table 7), on sites corresponding to known recognition motifs. In particular, CAMK4 and MAPKAPK2 potentially phosphorylate sites on S and N, respectively. Phosphorylation of cellular proteins suggested that the activities of these kinases were enriched in cells infected with SARS-CoV-2 or SARS-CoV (Extended Data Figs. 5e, f, 6e, g). Moreover, N proteins of both SARS-CoV-2 and SARS-CoV recruit GSK3, which could potentially be linked to phosphorylation events on these viral proteins (Fig. 1b, Extended Data Fig. 6g, Supplementary Table 7). Notably, we identified novel post-translationally modified sites located at functional domains of viral proteins; we detected ubiquitination at SARS-CoV-2 N K338 and phosphorylation on SARS-CoV-2 and SARS-CoV N S310 and S311 (Extended Data Fig. 6g). Mapping these sites to the atomic structure of the CTD32,33 highlights critical positions for the function of the protein (Fig. 3c, Extended Data Fig. 6h, Supplementary Discussion 2). Collectively, while the identification of differentially regulated sites may indicate pathogen-specific functions, insights from conserved post-translational modifications may also provide useful knowledge for the development of targeted pan-antiviral therapies.

Fig. 3: Phosphorylation and ubiquitination of SARS-CoV-2 and SARS-CoV viral proteins.
figure 3

a, Distribution of identified shared, differentially regulated and selectively encoded (sequence-specific) ubiquitination and phosphorylation sites on SARS-CoV-2 and SARS-CoV homologous proteins after infection of A549-ACE2 cells. PTM, post-translational modification. b, Mapping of the ubiquitination (red circles) and phosphorylation (purple squares) sites on an alignment of SARS-CoV-2 ORF3 and SARS-CoV ORF3a proteins, showing median log2 intensities in virus-infected A549-ACE2 cells at 24 hpi. Functional (blue) and topological (yellow) domains are mapped on each sequence. Ubiquitin-modifying enzymes binding to ORF3 and ORF3a as identified in our AP–MS experiments (Extended Data Fig. 1b) are indicated (green). TM, transmembrane domain. c, Surface and ribbon representation of superimposed SARS-CoV (Protein Data Bank (PDB) ID: 2CJR, brown) and SARS-CoV-2 (PDB ID: 6YUN, orange) N protein CTD dimers (r.m.s.d. values of 0.492 Å for 108 matching Cα atoms). Secondary structures are numbered in grey (prefixed with α for α-helix, β for β-strand and η for non-structured regions). Side chains are colour coded depending on whether they are in ubiquitinated (red), phosphorylated (purple) or unmodified (grey) sites. The K338 ubiquitination site unique to SARS-CoV-2 is shown as a close-up for both monomers (bottom right). Close-ups of inter-chain residue interactions established by non-phosphorylated (top right) and phosphorylated (middle right) SARS-CoV-2 S310 or SARS-CoV S311.

Viral perturbation of key cellular pathways

Our unified experimental design in a syngeneic system enabled direct time-resolved comparison of SARS-CoV-2 and SARS-CoV infection across different levels. Integrative pathway-enrichment analysis demonstrated that both viruses largely perturb the same cellular processes at multiple levels, albeit with distinct temporal patterns (Extended Data Fig. 7a). For instance, transcriptional downregulation of proteins involved in tau protein kinase activity and Fe ion sequestration at 6 hpi was followed by a decrease in protein abundance after 12 hpi (Supplementary Table 8). RHO GTPase activation, mRNA processing and the role of ABL in ROBO–SLIT signalling appeared to be regulated mostly through phosphorylation (Extended Data Fig. 7a). By contrast, processes connected to cellular integrity such as the formation of senescence-associated heterochromatin foci, apoptosis-induced DNA fragmentation and amino acid transport across the plasma membrane were modulated through concomitant phosphorylation and ubiquitination events, suggesting molecular links between these post-translational modifications. Ion transporters, especially the SLC12 family of cation-coupled chloride cotransporters—previously identified as cellular factors in pulmonary inflammation34—were also regulated at multiple levels, evidenced by reduced protein abundance as well as differential post-translational modifications (Extended Data Fig. 7a).

The pathway-enrichment analysis provided a global and comprehensive picture of how SARS-CoV-2 and SARS-CoV affect the host. We next applied an automated approach to systematically explore the underlying molecular mechanisms contained in the viral interactome and effectome data. We mapped the measured interactions and effects of each viral protein onto the global network of cellular interactions35 and applied a network diffusion approach36 (Fig. 4a). This type of analysis uses known cellular protein–protein interactions, signalling and regulation events to identify connection points between cellular proteins that interact with viral proteins and the proteins affected by the expression of these viral proteins (Extended Data Figs. 1b, 2d, Supplementary Tables 2, 3). The connections inferred from the real data were significantly shorter than for randomized data, validating the relevance of the approach and the quality of the data (Extended Data Fig. 8a, b). The findings from this approach include the potential mechanisms by which ORF3 and NSP6 may regulate autophagy, the modulation of innate immunity by M, ORF3 and ORF7b, and the perturbation of integrin–TGF-β–EGFR–receptor tyrosine kinase signalling by ORF8 of SARS-CoV-2 (Fig. 4b, Extended Data Fig. 8c, d). Enriching these subnetworks with data on SARS-CoV-2 infection-dependent mRNA abundance, protein abundance, phosphorylation and ubiquitination (Fig. 4a) provided insights into the regulatory mechanisms activated by SARS-CoV-2. For instance, the analysis confirmed a role of NSP6 in both SARS-CoV-2- and SARS-CoV-induced autophagy37 and revealed the SARS-CoV-2 specific inhibition of autophagic flux by ORF3 protein, leading to the accumulation of autophagy receptors (SQSTM1, GABARAPL2, NBR1, CALCOCO2, MAP1LC3A, MAP1LC3B and TAX1BP1), consistent with the accumulation of MAP1LC3B protein observed in cells infected with SARS-CoV-2 (Fig. 4c, Extended Data Fig. 8e, f). This inhibition may be a result of the interaction of the ORF3 protein with the HOPS complex (VPS11, VPS16, VPS18, VPS39 and VPS41), which is essential for autophagosome–lysosome fusion, as well as the differential phosphorylation of regulatory sites (for example, on TSC2, mTORC1 complex, ULK1, RPS6 and SQSTM1) and ubiquitination of key components (MAP1LC3A, GABARAPL2, VPS33A and VAMP8) (Fig. 4c, Extended Data Fig. 8g). This inhibition of autophagosome function may have direct consequences for protein degradation. The abundance of APOB, a protein that is degraded via autophagy38, was selectively increased after SARS-CoV-2 infection or expression of SARS-CoV-2 ORF3 (Extended Data Fig. 3b, 8h). Accumulating APOB levels could increase the risk of arterial thrombosis39, one of the main complications contributing to lung, heart and kidney failure in patients with COVID-1940. The inhibition of the interferon (IFN)-α and IFN-β response observed at transcriptional and proteome levels was similarly explained by the network diffusion analysis (Extended Data Fig. 8i), which implicated multiple proteins of SARS-CoV-2 in the disruption of antiviral immunity. Further experiments functionally corroborated the inhibition of IFN-α and IFN-β induction or signalling by ORF3, ORF6, ORF7a, ORF7b and ORF9b (Extended Data Fig. 8j). Upon virus infection, we observed the regulation of TGF-β and EGFR pathways modulating cell survival, motility and innate immune responses (Extended Data Fig. 9a–d). Specifically, our network diffusion analysis revealed a connection between the binding of the ORF8 and ORF3 proteins to TGF-β-associated factors (TGF-β1, TGF-β2, LTBP1, TGFBR2, FURIN and BAMBI), the differential expression of extracellular matrix regulators (FERMT2 and CDH1) and the virus-induced upregulation of fibrinogens (FGA, FGB), fibronectin (FN1) and SERPINE141 (Extended Data Fig. 9a, b). The increased phosphorylation of proteins involved in MAPK signalling (for example, SHC1 (on S139), SOS1 (S1134/S1229), JUN (S63/S73), MAPKAPK2 (T334) and p38 (T180/Y182)) and receptor tyrosine kinase signalling (for example, phosphorylation of PI3K complex members PDPK1 (S241) and RPS6KA1 (S380)) as well as a higher expression of JUN, FOS and EGR1 are further indications of regulation of TGF-β and EGFR pathways (Extended Data Fig. 9a, c, d). In turn, TGF-β and EGFR signalling are known to be potentiated by integrin signalling and activation of YAP-dependent transcription42, which we observed to be regulated in a time-dependent manner upon SARS-CoV-2 infection (Extended Data Fig. 9a). As well as promoting virus replication, activation of these pathways has been implicated in fibrosis13,14,15, one of the hallmarks of COVID-1916.

Fig. 4: A network diffusion approach enables identification of molecular pathways linking protein–protein interactions with downstream changes in the host proteome.
figure 4

a, The network diffusion approach to identify functional connections between the host targets of a viral protein and downstream proteome changes. The results of network diffusion are integrated with omics datasets of SARS coronavirus infection to streamline the identification of affected host pathways. b, Subnetworks of the network diffusion linking host targets of SARS-CoV-2 ORF3 to factors involved in autophagy. The thickness of directed edges is proportional to the random-walk transition probability. Black edges denote connections present in ReactomeFI. c, Overview of perturbations to host-cell autophagy induced by SARS-CoV-2. The pathway regulation is derived from the network diffusion model for SARS-CoV-2 ORF3 and NSP6 and is overlaid with the changes in protein levels, ubiquitination (Ubi) and phosphorylation (Pho) induced by SARS-CoV-2 infection. FC, fold change; PM, plasma membrane.

Data-guided drug identification and testing

Together, the viral protein–host protein interactions and regulation of pathways observed at multiple levels identify potential points for targeting SARS-CoV and SARS-CoV-2 using well-characterized selective antiviral drugs. To test antiviral efficacy, we used time-lapse fluorescent microscopy of infection with a GFP reporter SARS-CoV-243. Inhibition of virus replication by treatment with IFN-α corroborated previous conclusions that efficient SARS-CoV-2 replication involves the inactivation of this pathway at an early step9,44 and confirmed the reliability of this screening approach (Extended Data Fig. 10a). We tested a panel of 48 drugs that modulate the pathways perturbed by the virus for their effects on SARS-CoV-2 replication (Fig. 5a, Supplementary Table 9). Of note, inhibitors of B-RAF (sorafenib, regorafenib and dabrafenib), JAK1/2 (baricitinib) and MAPK (SB239063), which are commonly used to treat cancer and autoimmune diseases45,46,47, significantly increased virus growth in an in vitro model of infection (Fig. 5a, Extended Data Fig. 10b, Supplementary Table 9). By contrast, inducers of DNA damage (tirapazamine and rabusertib) or an mTOR inhibitor (rapamycin) suppressed virus growth. The highest antiviral activity was observed for gilteritinib (a designated inhibitor of FLT3 and AXL), ipatasertib (an AKT inhibitor), prinomastat and marimastat (matrix metalloprotease (MMP) inhibitors) (Fig. 5a, b, Extended Data Fig. 10c, Supplementary Table 9). These compounds profoundly inhibited replication of SARS-CoV-2 while having no effects or minor effects on cell growth (Extended Data Fig. 10b, Supplementary Table 9). Quantitative PCR analysis indicated antiviral activities for gilteritinib and tirapazamine against SARS-CoV-2 and SARS-CoV (Fig. 5c, Extended Data Fig. 10d, e). Notably, prinomastat and marimastat, specific inhibitors of MMP2 and MMP9, showed selective activity against SARS-CoV-2 but not against SARS-CoV (Fig. 5c, Extended Data Fig. 10f, g). Activities of MMPs have been linked to TGF-β activation and pleural effusions, alveolar damage and neuroinflammation (for example, Kawasaki disease), all of which are characteristic of COVID-1923,48,49,50,51.

Fig. 5: Identification of pathways targeted by SARS-CoV-2 using a multi-omics profiling approach enables systematic testing of candidate antiviral therapies.
figure 5

a, b, A549-ACE2 cells were treated with the indicated drugs 6 h before infection with SARS-CoV-2-GFP (MOI of 3). Changes in cell viability and virus growth (a) in drug-treated cells compared with untreated A549-ACE2 cells at 48 hpi. A confluence cut-off of −0.2 log2 fold change was applied to remove cytotoxic compounds. b, Time courses of virus replication after pre-treatment of cells with prinomastat or gilteritinib. n = 4 independent experiments; *P ≤ 0.01 compared with control treatment, unadjusted two-sided Wilcoxon test. Norm., normalized. c, Drugs potentially targeting pathways identified in our study. Colour indicates antiviral activity against SARS-CoV-2 or SARS-CoV (brown–orange gradient) or SARS-CoV-2 specifically (orange), as inferred from in vitro experiments.

This drug screen demonstrates the value of our combined dataset, which profiles SARS-CoV-2 infection at multiple levels. We expect that further exploration of these rich data by the scientific community and additional studies of the interplay between different omics levels will substantially advance our molecular understanding of coronavirus biology, including the pathogenicity associated with specific human coronaviruses, such as SARS-CoV-2 and SARS-CoV. Moreover, this resource, together with complementary approaches26,52,53,54, will streamline the search for antiviral compounds and serve as a base for rational design of combination therapies that target the virus from multiple synergistic angles, thus potentiating the effect of individual drugs while minimizing potential side effects on healthy tissues.


No statistical methods were used to predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

Cell lines and reagents

HEK293T, A549, Vero E6 and HEK293-R1 cells were authentified by PCR-single-locus-technology and their respective culturing conditions were described previously55. All cell lines were tested to be mycoplasma free. Expression constructs for C-terminal HA-tagged viral open reading frames were synthesized (Twist Bioscience and BioCat) and cloned into pWPI vector as described previously56 with the following modifications: a starting ATG codon was added, internal canonical splicing sites were replaced with synonymous mutations and a C-terminal HA-tag, followed by an amber stop codon, was added to individual viral open reading frames. A C-terminally HA-tagged ACE2 sequence was amplified from an ACE2 expression vector (provided by S. Pöhlmann)57 into the lentiviral vector pWPI-puro. A549 cells were transduced twice, and A549-ACE2 cells were selected with puromycin. Lentivirus production, transduction of cells and antibiotic selection were performed as described previously52. RNA isolation (Macherey–Nagel NucleoSpin RNA plus), reverse transcription (TaKaRa Bio PrimeScript RT with gDNA eraser) and quantitative PCR with reverse transcription (RT–qPCR) (Thermo-Fisher Scientific PowerUp SYBR green) were performed as described previously54. RNA isolation for next generation sequencing applications was performed according to the manufacturer’s protocol (Qiagen RNeasy mini kit, RNase free DNase set). For detection of protein abundance by western blotting, HA–horseradish peroxidase (HRP) (Sigma-Aldrich; H6533; 1:2,500 dilution), ACTB–HRP (Santa Cruz; sc-47778; 1:5,000 dilution), MAP1LC3B (Cell Signaling; 3868; 1:1,000 dilution), MAVS (Cell Signaling; 3993; 1:1,000 dilution), HSPA1A (Cell Signaling; 4873; 1:1,000 dilution), TGF-β (Cell Signaling; 3711; 1:1,000 dilution), phospho-p38 (T180/Y182) (Cell Signaling; 4511; 1:1,000 dilution), p38 (Cell Signaling; 8690; 1:1,000 dilution) and SARS-CoV-2 or SARS-CoV N protein (Sino Biological; 40143-MM05; 1:1,000 dilution) antibodies were used. Secondary antibodies detecting mouse (Cell Signaling; 7076; 1:5,000 dilution; Jackson ImmunoResearch; 115-035-003; 1:5,000 dilution), rat (Invitrogen; 31470; 1:5,000 dilution), and rabbit IgG (Cell Signaling; 7074; 1:5,000 dilution) were coupled to HRP. For AP–MS and affinity purification–western blotting applications, HA beads (Sigma-Aldrich and Thermo Fisher Scientific) and Streptactin II beads (IBA Lifesciences) were used. Imaging of western blots was performed as described58. Recombinant human IFN-α used for stimulation of cells in the reporter assay was a gift from P. Stäheli (Institute of Virology, University of Freiburg), recombinant human IFN-γ was purchased from PeproTech, and IVT4 was produced as described before59. All compounds tested in the viral inhibitor assay are listed in Supplementary Table 9.

Virus strains, stock preparation, plaque assay and in vitro infection

SARS-CoV-Frankfurt-1, SARS-CoV-2-MUC-IMB-1 and SARS-CoV-2-GFP strains43 were produced by infecting Vero E6 cells cultured in DMEM medium (10% FCS, 100 μg ml−1 Streptomycin, 100 IU ml−1 penicillin) for 2 days (MOI of 0.01). Viral stock was collected and spun twice (1,000g for 10 min) before storage at −80 °C. Titre of viral stock was determined by plaque assay. Confluent monolayers of Vero E6 cells were infected with serial fivefold dilutions of virus supernatants for 1 h at 37 °C. The inoculum was removed and replaced with serum-free MEM (Gibco, Life Technologies) containing 0.5% carboxymethylcellulose (Sigma-Aldrich). Two days after infection, cells were fixed for 20 min at room temperature with formaldehyde added directly to the medium to a final concentration of 5%. Fixed cells were washed extensively with PBS before staining with water containing 1% crystal violet and 10% ethanol for 20 min. After rinsing with PBS, the number of plaques was counted and the virus titre was calculated.

A549-ACE2 cells were infected with either SARS-CoV-Frankfurt-1 or SARS-CoV-2-MUC-IMB-1 strains (MOI of 2) for the subsequent experiments. At each time point, the samples were washed once with 1× TBS buffer and collected in sodium deoxycholate (SDC) lysis buffer (100 mM Tris HCl pH 8.5; 4% SDC) for proteome-phosphoproteome-ubiquitinome analysis, sodium dodecyl sulfate (SDS) lysis buffer (62.5 mM Tris HCl pH 6.8; 2% SDS; 10% glycerol; 50 mM DTT; 0.01% bromophenol blue) for western blot, or buffer RLT (Qiagen) for transcriptome analysis. The samples were heat-inactivated and frozen at −80 °C until further processing.

Affinity purification and mass spectrometric analyses of SARS-CoV-2, SARS-CoV, HCoV-229E and HCoV-NL63 proteins expressed in A549 cells

To determine the interactomes of SARS-CoV-2 and SARS-CoV and the interactomes of an accessory protein (encoded by ORF4 or ORF4a of HCoV-229E or ORF3 of HCoV-NL63) that presumably represents a homologue of the ORF3 and ORF3a proteins of SARS-CoV-2 and SARS-CoV, respectively, four replicate affinity purifications were performed for each HA-tagged viral protein. A549 cells (6 × 106 cells per 15-cm dish) were transduced with lentiviral vectors encoding HA-tagged SARS-CoV-2, SARS-CoV or HCoV-229E/NL63 proteins and protein lysates were prepared from cells collected 3 days after transduction. Cell pellets from two 15-cm dishes were lysed in lysis buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 1.5 mM MgCl2, 0.2% (v/v) NP-40, 5% (v/v) glycerol, cOmplete protease inhibitor cocktail (Roche), 0.5% (v/v) 750 U/μl Sm DNase) and sonicated (5 min, 4 °C, 30 s on, 30 s off, low settings; Bioruptor, Diagenode). Following normalization of protein concentrations of cleared lysates, virus protein-bound host proteins were enriched by adding 50 μl anti-HA-agarose slurry (Sigma-Aldrich, A2095) with constant agitation for 3 h at 4 °C. Non-specifically bound proteins were removed by four subsequent washes with lysis buffer followed by three detergent-removal steps with washing buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 1.5 mM MgCl2, 5% (v/v) glycerol). Enriched proteins were denatured, reduced, alkylated and digested by addition of 200 μl digestion buffer (0.6 M guanidinium chloride, 1 mM tris(2-carboxyethyl)phosphine (TCEP), 4 mM chloroacetamide (CAA), 100 mM Tris-HCl pH 8, 0.5 μg LysC (WAKO Chemicals) and 0.5 μg trypsin (Promega) at 30 °C overnight. Peptide purification on StageTips with three layers of C18 Empore filter discs (3M) and subsequent mass spectrometry analysis was performed as described previously55,56. In brief, purified peptides were loaded onto a 20-cm reverse-phase analytical column (75 μm diameter; ReproSil-Pur C18-AQ 1.9 μm resin; Dr Maisch) and separated using an EASY-nLC 1200 system (Thermo Fisher Scientific). A binary buffer system consisting of buffer A (0.1% formic acid (FA) in H2O) and buffer B (80% acetonitrile (ACN), 0.1% FA in H2O) with a 90-min gradient (5–30% buffer B (65 min), 30–95% buffer B (10 min), wash out at 95% buffer B (5 min), decreased to 5% buffer B (5 min), and 5% buffer B (5 min)) was used at a flow rate of 300 nl per min. Eluting peptides were directly analysed on a Q-Exactive HF mass spectrometer (Thermo Fisher Scientific). Data-dependent acquisition included repeating cycles of one MS1 full scan (300–1 650 m/z, R = 60,000 at 200 m/z) at an ion target of 3 × 106, followed by 15 MS2 scans of the highest abundant isolated and higher-energy collisional dissociation (HCD) fragmented peptide precursors (R = 15,000 at 200 m/z). For MS2 scans, collection of isolated peptide precursors was limited by an ion target of 1 × 105 and a maximum injection time of 25 ms. Isolation and fragmentation of the same peptide precursor was eliminated by dynamic exclusion for 20 s. The isolation window of the quadrupole was set to 1.4 m/z and HCD was set to a normalized collision energy of 27%.

Proteome analyses of cells expressing SARS-CoV-2, SARS-CoV, HCoV-229E or HCoV-NL63 proteins

For the determination of proteome changes in A549 cells expressing SARS-CoV-2, SARS-CoV, HCoV-229E or HCoV-NL63 proteins, a fraction of 1 × 106 lentivirus-transduced cells from the affinity purification samples were lysed in guanidinium chloride buffer (6 M guanidinium chloride, 10 mM TCEP, 40 mM CAA, 100 mM Tris-HCl pH 8), boiled at 95 °C for 8 min and sonicated (10 min, 4 °C, 30 s on, 30 s off, high settings). Protein concentrations of cleared lysates were normalized to 50 μg, and proteins were pre-digested with 1 μg LysC at 37 °C for 1 h followed by a 1:10 dilution (100 mM Tris-HCl pH 8) and overnight digestion with 1 μg trypsin at 30 °C. Peptide purification on StageTips with three layers of C18 Empore filter discs (3M) and subsequent mass spectrometry analysis was performed as described previously55,56. In brief, 300 ng of purified peptides were loaded onto a 50-cm reversed-phase column (75 μm inner diameter, packed in house with ReproSil-Pur C18-AQ 1.9 μm resin (Dr Maisch)). The column temperature was maintained at 60 °C using a homemade column oven. A binary buffer system, consisting of buffer A (0.1% FA) and buffer B (80% ACN, 0.1% FA), was used for peptide separation, at a flow rate of 300 nl min−1. An EASY-nLC 1200 system (Thermo Fisher Scientific), directly coupled online with the mass spectrometer (Q Exactive HF-X, Thermo Fisher Scientific) via a nano-electrospray source, was employed for nano-flow liquid chromatography. Peptides were eluted by a linear 80 min gradient from 5% to 30% buffer B (0.1% v/v FA, 80% v/v ACN), followed by a 4 min increase to 60% B, a further 4 min increase to 95% B, a 4 min plateau phase at 95% B, a 4 min decrease to 5% B and a 4 min wash phase of 5% B. To acquire MS data, the data-independent acquisition (DIA) scan mode operated by the XCalibur software (Thermo Fisher) was used. DIA was performed with one full MS event followed by 33 MS/MS windows in one cycle resulting in a cycle time of 2.7 s. The full MS settings included an ion target value of 3 × 106 charges in the 300–1,650 m/z range with a maximum injection time of 60 ms and a resolution of 120,000 at m/z 200. DIA precursor windows ranged from 300.5 m/z (lower boundary of first window) to 1,649.5 m/z (upper boundary of 33rd window). MS/MS settings included an ion target value of 3 × 106 charges for the precursor window with an Xcalibur-automated maximum injection time and a resolution of 30,000 at m/z 200.

To generate the proteome library for DIA measurements purified peptides from the first and the fourth replicates of all samples were pooled separately and 25 μg of peptides from each pool were fractionated into 24 fractions by high pH reversed-phase chromatography as described earlier60. During each separation, fractions were concatenated automatically by shifting the collection tube every 120 s. In total 48 fractions were dried in a vacuum centrifuge, resuspended in buffer A* (0.2% trifluoroacetic acid (TFA), 2% ACN) and subsequently analysed by a top-12 data-dependent acquisition (DDA) scan mode using the same LC gradient and settings. The mass spectrometer was operated by the XCalibur software (Thermo Fisher). DDA scan settings on full MS level included an ion target value of 3 × 106 charges in the 300–1,650 m/z range with a maximum injection time of 20 ms and a resolution of 60,000 at m/z 200. At the MS/MS level the target value was 105 charges with a maximum injection time of 60 ms and a resolution of 15,000 at m/z 200. For MS/MS events only, precursor ions with 2–5 charges that were not on the 20-s dynamic exclusion list were isolated in a 1.4 m/z window. Fragmentation was performed by higher-energy C-trap dissociation with a normalized collision energy of 27 eV.

Infected time-course proteome–phosphoproteome–diGly proteome sample preparation

Frozen lysates of infected A549-ACE2 cells collected at 6, 12 and 24 hpi (and 36 hpi for the phosphoproteomics study) were thawed on ice, boiled for 5 min at 95 °C and sonicated for 15 min (Branson Sonifierer). Protein concentrations were estimated by tryptophan assay61. To reduce and alkylate proteins, samples were incubated for 5 min at 45 °C with TCEP (10 mM) and CAA (40 mM). Samples were digested overnight at 37 °C using trypsin (1:100 w/w, enzyme/protein, Sigma-Aldrich) and LysC (1:100 w/w, enzyme/protein, Wako).

For proteome analysis, 10 μg of peptide material were desalted using SDB-RPS StageTips (Empore)61. In brief, samples were diluted with 1% TFA in isopropanol to a final volume of 200 μl and loaded onto StageTips, subsequently washed with 200 μl of 1% TFA in isopropanol and 200 μl 0.2% TFA/ 2% ACN. Peptides were eluted with 75 μl of 1.25% ammonium hydroxide (NH4OH) in 80% ACN and dried using a SpeedVac centrifuge (Eppendorf, Concentrator Plus). They were resuspended in buffer A* (0.2% TFA, 2% ACN) before LC–MS/MS analysis. Peptide concentrations were measured optically at 280 nm (Nanodrop 2000, Thermo Scientific) and subsequently equalized using buffer A*. One microgram of peptide was analysed by LC–MS/MS.

The rest of the samples were diluted fourfold with 1% TFA in isopropanol and loaded onto SDB-RPS cartridges (Strata-X-C, 30 mg per 3 ml, Phenomenex), pre-equilibrated with 4 ml 30% MeOH/1% TFA and washed with 4 ml 0.2% TFA. Samples were washed twice with 4 ml 1% TFA in isopropanol, once with 0.2% TFA/2% ACN and eluted twice with 2 ml 1.25% NH4OH/80% ACN. Eluted peptides were diluted with ddH2O to a final ACN concentration of 35%, snap frozen and lyophilized.

For phosphopeptide enrichment, lyophilized peptides were resuspended in 105 μl of equilibration buffer (1% TFA/80% ACN) and the peptide concentration was measured optically at 280 nm (Nanodrop 2000, Thermo Scientific) and subsequently equalized using equilibration buffer. The AssayMAP Bravo robot (Agilent) performed the enrichment for phosphopeptides (150 μg) by priming AssayMAP cartridges (packed with 5 μl Fe3+-NTA) with 0.1% TFA in 99% ACN followed by equilibration in equilibration buffer and loading of peptides. Enriched phosphopeptides were eluted with 1% ammonium hydroxide, which was evaporated using a Speedvac for 20 min. Dried peptides were resuspended in 6 μl buffer A* and 5 μl was analysed by LC–MS/MS.

For diGly peptide enrichment, lyophilized peptides were reconstituted in IAP buffer (50 mM MOPS, pH 7.2, 10 mM Na2HPO4, 50 mM NaCl) and the peptide concentration was estimated by tryptophan assay. K-ɛ-GG remnant containing peptides were enriched using the PTMScan Ubiquitin Remnant Motif (K-ɛ-GG) Kit (Cell Signaling Technology). Cross-linking of antibodies to beads and subsequent immunopurification was performed with slight modifications as previously described62. In brief, two vials of cross-linked beads were combined and equally split into 16 tubes (~31 μg of antibody per tube). Equal peptide amounts (600 μg) were added to cross-linked beads, and the volume was adjusted with IAP buffer to 1 ml. After 1 h of incubation at 4 °C and gentle agitation, beads were washed twice with cold IAP and 5 times with cold ddH2O. Thereafter, peptides were eluted twice with 50 μl 0.15% TFA. Eluted peptides were desalted and dried as described for proteome analysis with the difference that 0.2% TFA instead of 1%TFA in isopropanol was used for the first wash. Eluted peptides were resuspended in 9 μl buffer A* and 4 μl was subjected to LC–MS/MS analysis.

DIA measurements

Samples were loaded onto a 50-cm reversed-phase column (75 μm inner diameter, packed in house with ReproSil-Pur C18-AQ 1.9 μm resin (Dr Maisch)). The column temperature was maintained at 60 °C using a homemade column oven. A binary buffer system, consisting of buffer A (0.1% FA) and buffer B (80% ACN plus 0.1% FA) was used for peptide separation, at a flow rate of 300 nl min−1. An EASY-nLC 1 200 system (Thermo Fisher Scientific), directly coupled online with the mass spectrometer (Orbitrap Exploris 480, Thermo Fisher Scientific) via a nano-electrospray source, was employed for nano-flow liquid chromatography. The FAIMS device was placed between the nanoelectrospray source and the mass spectrometer and was used for measurements of the proteome and the PTM-library samples. Spray voltage was set to 2,650 V, RF level to 40 and heated capillary temperature to 275 °C.

For proteome measurements we used a 100 min gradient starting at 5% buffer B followed by a stepwise increase to 30% in 80 min, 60% in 4 min and 95% in 4 min. The buffer B concentration stayed at 95% for 4 min, decreased to 5% in 4 min and stayed there for 4 min. The mass spectrometer was operated in data-independent mode (DIA) with a full scan range of 350–1,650 m/z at 120,000 resolution at 200 m/z, normalized automatic gain control (AGC) target of 300% and a maximum fill time of 28 ms. One full scan was followed by 22 windows with a resolution of 15,000, normalized AGC target of 1,000% and a maximum fill time of 25 ms in profile mode using positive polarity. Precursor ions were fragmented by HCD (NCE 30%). Each of the selected compensation voltage (CV) (−40, −55 and −70 V) was applied to sequential survey scans and MS/MS scans; the MS/MS CV was always paired with the appropriate CV from the corresponding survey scan.

For phosphopeptide samples, 5 μl were loaded and eluted with a 70-min gradient starting at 3% buffer B followed by a stepwise increase to 19% in 40 min, 41% in 20 min, 90% in 5 min and 95% in 5 min. The mass spectrometer was operated in DIA mode with a full scan range of 300–1,400 m/z at 120,000 resolution at 200 m/z and a maximum fill time of 60 ms. One full scan was followed by 32 windows with a resolution of 30,000. Normalized AGC target and maximum fill time were set to 1,000% and 54 ms, respectively, in profile mode using positive polarity. Precursor ions were fragmented by HCD (NCE stepped 25–27.5–30%). For the library generation, we enriched A549 cell lysates for phosphopeptides and measured them with 7 different CV settings (−30, −40, −50, −60, −70, −80 or −90 V) using the same DIA method. The noted CVs were applied to the FAIMS electrodes throughout the analysis.

For the analysis of K-ɛ-GG peptide samples, half of the samples were loaded. We used a 120-min gradient starting at 3% buffer B followed by a stepwise increase to 7% in 6 min, 20% in 49 min, 36% in 39 min, 45% in 10 min and 95% in 4 min. The buffer B concentration stayed at 95% for 4 min, decreased to 5% in 4 min and stayed there for 4 min. The mass spectrometer was operated in DIA mode with a full scan range of 300–1,350 m/z at 120,000 resolution at m/z 200, normalized AGC target of 300% and a maximum fill time of 20 ms. One full scan was followed by 46 windows with a resolution of 30,000. Normalized AGC target and maximum fill time were set to 1,000% and 54 ms, respectively, in profile mode using positive polarity. Precursor ions were fragmented by HCD (NCE 28%). For K-ɛ-GG peptide library, we mixed the first replicate of each sample and measured them with eight different CV setting (−35, −40, −45, −50, −55, −60, −70 or −80 V) using the same DIA method.

Processing of raw MS data

AP–MS data

Raw MS data files of AP–MS experiments conducted in DDA mode were processed with MaxQuant (version 1.6.14) using the standard settings and label-free quantification (LFQ) enabled (LFQ min ratio count 1, normalization type none, stabilize large LFQ ratios disabled). Spectra were searched against forward and reverse sequences of the reviewed human proteome including isoforms (UniprotKB, release 2019.10) and C-terminally HA-tagged SARS-CoV-2, SARS-CoV and HCoV proteins by the built-in Andromeda search engine63.

In-house Julia scripts ( were used to define alternative protein groups: only the peptides identified in AP–MS samples were considered for being protein group-specific, protein groups that differed by the single specific peptide or had less than 25% different specific peptides were merged to extend the set of peptides used for protein group quantitation and reduce the number of protein isoform-specific interactions.

Viral protein overexpression and DIA MS data

Spectronaut version 13 (Biognosys) with the default settings was used to generate the proteome libraries from DDA runs by combining files of respective fractionations using the human fasta file (Uniprot, 2019.10, 42 431 entries) and viral bait sequences. Proteome DIA files were analysed using the proteome library with the default settings and disabled cross run normalization.

SARS-CoV-2/SARS-CoV-infected proteome/PTM DIA MS data

Spectronaut version 14 (Biognosys)64 was used to generate the libraries and analyse all DIA files using the human fasta file (UniprotKB, release 2019.10) and sequences of SARS-CoV-2/SARS-CoV proteins (UniProt, release 2020.08). Orf1a polyprotein sequences were split into separate protein chains according to the cleavage positions specified in the UniProt. For the generation of the PTM-specific libraries, the DIA single CV runs were combined with the actual DIA runs and either phosphorylation at serine, threonine or tyrosine, or GlyGly at lysine, was added as variable modification to default settings. The maximum number of fragment ions per peptide was increased to 25. The proteome DIA files were analysed using direct DIA approach with default settings and disabled cross run normalization. All post-translational modification DIA files were analysed using their respective hybrid library and either phosphorylation at Serine/Threonine/Tyrosine or GlyGly at Lysine was added as an additional variable modification to default settings with LOESS normalization and disabled PTM localization filter.

A collection of in-house Julia scripts( were used to process the elution group (EG) -level Spectronaut reports, identify PTMs and assign EG-level measurements to PTMs. The PTM was considered if at least once it was detected with ≥0.75 localization probability in EG with q-value ≤10−3. For further analysis of given PTM, only the measurements with ≥0.5 localization probability and EG q-value ≤10−2 were used.

Bioinformatic analysis

Unless otherwise specified, the bioinformatic analysis was done in R (version 3.6), Julia (version 1.5) and Python (version 3.8) using a collection of in-house scripts ( and


The following public datasets were used in the study: Gene Ontology and Reactome annotations (; IntAct Protein Interactions (, v2019.12); IntAct Protein Complexes (, v2019.12); CORUM Protein Complexes (, v2018.3); Reactome Functional Interactions (; Human (v2019.10), Human-CoV, SARS-CoV-2 and SARS-CoV (v2020.08) protein sequences:

Statistical analysis of MS data

MaxQuant and Spectronaut output files were imported into R using in-house maxquantUtils R package ( For all MS datasets, the Bayesian linear random effects models were used to define how the abundances of proteins change between the conditions. To specify and fit the models we used the msglm R package (, which uses the rstan package (version 2.19)65 for inferring the posterior distribution of the model parameters. In all the models, the effects corresponding to the experimental conditions have regularized horseshoe+ priors66, whereas the batch effects have normally distributed priors. Laplacian distribution was used to model the instrumental error of MS intensities. For each MS instrument used, the heteroscedastic intensities noise model was calibrated with the technical replicate MS data of the instrument. These data were also used to calibrate the logit-based model of missing MS data (the probability that the MS instrument will fail to identify the protein given its expected abundance in the sample). The model was fit using unnormalized MS intensities data. Instead of transforming the data by normalization, the inferred protein abundances were scaled by the normalization multiplier of each individual MS sample to match the expected MS intensity of that sample. This allows taking the signal-to-noise variation between the samples into account when fitting the model. Due to high computational intensity, the model was applied to each protein group separately. For all the models, 4,000 iterations (2,000 warmup + 2,000 sampling) of the no-U-turn Markov Chain Monte Carlo were performed in 7 or 8 independent chains, every 4th sample was collected for posterior distribution of the model parameters. For estimating the statistical significance of protein abundance changes between the two experimental conditions, the P-value was defined as the probability that a random sample from the posterior distribution of the first condition would be smaller (or larger) than a random sample drawn from the second condition. No-multiple hypothesis testing corrections were applied, since this is handled by the choice of the model priors.

Statistical analysis of AP–MS data and filtering for specific interactions

The statistical model was applied directly to the MS1 intensities of protein group-specific LC peaks (evidence.txt table of MaxQuant output). In R GLM formula language, the model could be specified as

$$\log ({\rm{I}}{\rm{n}}{\rm{t}}{\rm{e}}{\rm{n}}{\rm{s}}{\rm{i}}{\rm{t}}{\rm{y}})\approx 1+{\rm{A}}{\rm{P}}{\rm{M}}{\rm{S}}+{\rm{B}}{\rm{a}}{\rm{i}}{\rm{t}}+{\rm{B}}{\rm{a}}{\rm{i}}{\rm{t}}:{\rm{V}}{\rm{i}}{\rm{r}}{\rm{u}}{\rm{s}}+{\rm{M}}{\rm{S}}1{\rm{p}}{\rm{e}}{\rm{a}}{\rm{k}}+{\rm{M}}{\rm{S}}{\rm{b}}{\rm{a}}{\rm{t}}{\rm{c}}{\rm{h}},$$

where the APMS effect models the average shift of intensities in AP–MS data in comparison to full proteome samples, Bait is the average enrichment of a protein in AP–MS experiments of homologous proteins of both SARS-CoV and SARS-CoV-2, and Bait:Virus corresponds to the virus-specific changes in protein enrichment. MS1peak is the log ratio between the intensity of a given peak and the total protein abundance (the peak is defined by its peptide sequence, PTMs and the charge; it is assumed that the peak ratios do not depend on experimental conditions67), and MSbatch accounts for batch-specific variations of protein intensity. APMS, Bait and Bait:Virus effects were used to reconstruct the batch effect-free abundance of the protein in AP–MS samples.

The modelling provided the enrichment estimates for each protein in each AP experiment. Specific AP–MS interactions had to pass the two tests. In the first test, the enrichment of the candidate protein in a given bait AP was compared against the background, which was dynamically defined for each interaction to contain the data from all other baits, where the abundance of the candidate was within 50–90% percentile range (excluding top 10% baits from the background allowed the protein to be shared by a few baits in the resulting AP–MS network). The non-targeting control and Gaussian luciferase baits were always preserved in the background. Similarly, to filter out any potential side-effects of very high bait protein expression, the ORF3 homologues were always present in the background of M interactors and vice versa. To rule out the influence of the batch effects, the second test was applied. It was defined similarly to the first one, but the background was constrained to the baits of the same batch, and 40–80% percentile range was used. In both tests, the protein has to be fourfold enriched over the background (16 fold for highly expressed baits: ORF3, M, NSP13, NSP5, NSP6, ORF3a, ORF7b, ORF8b and HCoV-229E ORF4a) with P-value ≤ 10−3.

Additionally, we excluded the proteins that, in the viral protein expression data, have shown upregulation, and their enrichment in AP–MS data was less than 16 times stronger than observed upregulation effects. Finally, to exclude the carryover of material between the samples sequentially analysed by MS, we removed the putative interactors, which were also enriched at higher levels in the samples of the preceding bait, or the one before it.

For the analysis of interaction specificity between the homologous viral proteins, we estimated the significance of interaction enrichment difference (corrected by the average difference between the enrichment of the shared interactors to adjust for the bait expression variation). Specific interactions have to be fourfold enriched in comparison to the homologue with P-value ≤ 10−3.

Statistical analysis of DIA proteome effects upon viral protein overexpression

The statistical model of the viral protein overexpression dataset was similar to AP–MS data, except that protein-level intensities provided by Spectronaut were used. The PCA analysis of the protein intensities has identified that the second principal component is associated with the batch-dependent variations between the samples. To exclude their influence, this principal component was added to the experimental design matrix as an additional batch effect.

As with AP–MS data, the two statistical tests were used to identify the significantly regulated proteins (column ‘is_change’ in Supplementary Table 3). First, the absolute value of median log2-fold change of the protein abundance upon overexpression of a given viral protein in comparison to the background had to be above 1.0 with P-value ≤ 10−3. The background was individually defined for each analysed protein. It was composed of experiments, where the abundance of given protein was within the 20–80% percentile range of all measured samples. Second, the protein had to be significantly regulated (same median log2-fold change and P-value thresholds applied) against the batch-specific background (defined similarly to the global background, but using only the samples of the same batch).

An additional stringent criterion was applied to select the most significant changes (column ‘is_top_change’ in Supplementary Table 3; Extended Data Fig. 1i).

For each protein we classified bait-induced changes as: ‘high’ when |median log2 fold-change| ≥ 1 and P-value ≤ 10−10 both in background and batch comparisons; ‘medium’ if 10−10 < P-value ≤ 10−4 with same fold-change requirement; and ‘low’ if 10−4 < P-value ≤ 10−2 with the same fold-change requirement. All other changes were considered non-significant.

We then required that ‘shared’ top-regulated proteins should have exactly one pair of SARS-CoV-2 and SARS-CoV high- or medium-significant homologous baits among the baits with either up- or downregulated changes and no other baits with significant changes of the same type.

We further defined ‘SARS-CoV-2-specific’ or ‘SARS-CoV-specific’ top-regulated proteins to be the ones with exactly one high-significant change, and no other significant changes of the same sign. For ‘specific’ hits we additionally required that in the comparison of high-significant bait to its homologue |median log2 fold-change| ≥ 1 and P-value ≤ 10−3. When the homologous bait was missing (SARS-CoV-2 NSP1, SARS-CoV ORF8a and SARS-CoV ORF8b), we instead required that in the comparison of the high-significant change to the background |median log2 fold-change| ≥ 1.5.

The resulting network of most affected proteins was imported and prepared for publication in Cytoscape v.3.8.168.

Statistical analysis of DIA proteomic data of SARS-CoV-2 and SARS-CoV-infected A549-ACE2 cells

Similarly to the AP–MS DDA data, the linear Bayesian model was applied to the EG-level intensities. To model the protein intensity, the following linear model (in R notation) was used:

$$\log ({\rm{Intensity}}(t))\sim 1+\sum _{{t}_{i}\le t}({\rm{after}}({t}_{i})+({\rm{infection}}\,\pm {\rm{CoV2}}):{\rm{after}}({t}_{i}))+\,{\rm{EG}},$$

where the after(ti) effect corresponds to the protein abundance changes in mock-infected samples that happened between ti − 1 and ti after infection and it is applied to the modelled intensity at all time points starting from ti; infection:after(ti) (ti = 6, 12, 24) is the common effect of SARS-CoV-2 and SARS-CoV infections occurring between ti-1 and ti; CoV2:after(ti) is the virus-specific effect within ti - 1 and ti hpi that is added to the log intensity for SARS-CoV-2-infected samples and subtracted from the intensity for SARS-CoV ones; EG is the elution group-specific shift in the measured log-intensities.

The absolute value of median log2 fold change between the conditions above 0.25 and the corresponding unadjusted P-value ≤ 10−3 were used to define the significant changes at a given time point in comparison to mock infection. We also required that the protein group is quantified in at least two replicates of at least one of the compared conditions. Additionally, if for one of the viruses (for example, SARS-CoV-2) only the less stringent condition (|median log2 fold-change| ≥ 0.125, P-value ≤ 10−2) was fulfilled, but the change was significant in the infection of the other virus (SARS-CoV), and the difference between the viruses was not significant, the observed changes were considered significant for both viruses.

Statistical analysis of DIA phosphoproteome and ubiquitinome data of SARS-CoV-2 and SARS-CoV infections

The data from single- double- and triple-modified peptides were analysed separately and, for a given PTM, the most significant result was reported.

The data were analysed with the same Bayesian linear model as proteome SARS-CoV and SARS-CoV-2 infection data. In addition to the intensities normalization, for each replicate sample the scale of the effects in the experimental design matrix was adjusted, so that on average the correlation between log fold changes of the replicates was 1:1. The same logic as for the proteome analysis, was applied to identify significant changes, but the median log2 fold change had to be larger than 0.5, or 0.25 for the less stringent test. We additionally required that the PTM peptides are quantified in at least two replicates of at least one of the compared conditions. To ignore the changes in PTM site intensities that are due to proteome-level regulation, we excluded PTM sites on significantly regulated proteins if the directions of protein and PTM site changes were the same and the difference between their median log2 fold changes was less than two. Phosphoproteomics data were further analysed with Ingenuity Pathway Analysis software (Qiagen;

Transcriptomic analysis of SARS-CoV-2 and SARS-CoV infected A549-ACE2 cells

For the analysis of the transcriptome data, Gencode gene annotations v28 and the human reference genome GRCh38 were derived from the Gencode homepage (EMBL-EBI). Viral genomes were derived from GenBank (SARS-CoV-2 - LR824570.1, and SARS-CoV - AY291315.1). Dropseq tool v1.12 was used for mapping raw sequencing data to the reference genome. The resulting UMI filtered count matrix was imported into R v3.4.4. CPM (counts per million) values were calculated for the raw data and genes having a mean cpm value less than 1 were removed from the dataset. A dummy variable combining the covariates infection status (mock, SARS-CoV, SARS-CoV-2) and time point was used for modelling the data within Limma (v3.46.0)69.

Data were transformed with the Voom method69 followed by quantile normalization. Differential testing was performed between infection states at individual time points by calculating moderated t-statistics and P-values for each host gene. A gene was considered to be significantly regulated if the false discovery rate-adjusted P-value was below 0.05.

Gene set enrichment analysis

We used Gene Ontology, Reactome and other EnrichmentMap gene sets of human proteins (version 2020.10)70 as well as protein complexes annotations from IntAct Complex Portal (version 2019.11)71 and CORUM (version 2019)72. PhosphoSitePlus (version 2020.08) was used for known kinase-substrate and regulatory sites annotations, Perseus (version was used for annotation of known kinase motifs. For transcription factor enrichment analysis (Extended Data Fig. 2e) the significantly regulated transcripts were submitted to ChEA3 web-based application74 and ENCODE data on transcription factor–target gene associations were used75.

To find the non-redundant collection of annotations describing the unique and shared features of multiple experiments in a dataset (Fig. 1d, Extended Data Fig. 2l, m), we used in-house Julia package OptEnrichedSetCover.jl (, which employs evolutionary multi-objective optimization technique to find a collection of annotation terms that have both significant enrichments in the individual experiments and minimal pairwise overlaps.

The resulting set of terms was further filtered by requiring that the annotation term has to be significant with the specified unadjusted Fisher’s exact test P-value cut-off in at least one of the experiments or comparisons (the specific cut-off value is indicated in the figure legend of the corresponding enrichment analysis).

The generation of diagonally-split heat maps was done with the VegaLite.jl package (

Viral PTMs alignment

For matching the PTMs of SARS-CoV-2 and SARS-CoV the protein sequences were aligned using the BioAlignments.jl Julia package (v.2.0; with the Needleman–Wunsch algorithm using BLOSUM80 substitution matrix, and applying −5 and −3 penalties for the gap and extension, respectively.

For the cellular proteins, we required that the viral phosphorylation or ubiquitination site is observed with q-value ≤ 10−3 and localization probability ≥ 0.75. For the PTMs with lower confidence (q-value ≤ 10−2 and localization probability ≥ 0.5) we required that the same site is observed with high confidence at the matching position of the orthologous protein of the other virus.

Network diffusion analysis

To systematically detect functional interactions, which may connect the cellular targets of each viral protein (interactome dataset) with the downstream changes it induces on proteome level (effectome dataset), we have used the network diffusion-based HierarchicalHotNet method36 as implemented in Julia package HierarchicalHotNet.jl ( Specifically, for network diffusion with restart, we used the ReactomeFI network (version 2019)35 of cellular functional interactions, reversing the direction of functional interaction (for example, replacing kinase→substrate interaction with substrate→kinase). The proteins with significant abundance changes upon bait overexpression (|median(log2 fold change)| ≥ 0.25, P ≤ 10−2 both in the comparison against the controls and against the baits of the same batch) were used as the sources of signal diffusion with weights set to \({w}_{i}=\sqrt{|{\rm{m}}{\rm{e}}{\rm{d}}{\rm{i}}{\rm{a}}{\rm{n}}{\log }_{2}({\rm{f}}{\rm{o}}{\rm{l}}{\rm{d}}\,{\rm{c}}{\rm{h}}{\rm{a}}{\rm{n}}{\rm{g}}{\rm{e}})|\cdot |{\log }_{10}P\text{-}{\rm{v}}{\rm{a}}{\rm{l}}{\rm{u}}{\rm{e}}\,|}\), otherwise the node weight was set to zero. The weight of the edge gi→gj was set to wi,j = 1 + wj. The restart probability was set to 0.4, as suggested in the original publication, so that the probability of the random walk to stay in the direct neighbourhood of the node is the same as the probability to visit more distant nodes. To find the optimal cutting threshold of the resulting hierarchical tree of strongly connected components (SCCs) of the weighted graph corresponding to the stationary distribution of signal diffusion and to confirm the relevance of predicted functional connections, the same procedure was applied to 1 ,000 random permutations of vertex weights as described in Reyna et al.36 (vertex weights are randomly shuffled between the vertices with similar in and out degrees). Since cutting the tree of SCCs at any threshold t (keeping only the edges with weights above t) and collapsing each resulting SCC into a single node produces the directed acyclic graph of connections between SCCs, it allowed efficient enumeration of the paths from the ‘source’ nodes (proteins strongly perturbed by viral protein expression with vertex weight w, w ≥ 1.5) to the ‘sink’ nodes (interactors of the viral protein). At each threshold t, the average inverse of the path length from source to sink nodes was calculated as:

$${L}_{{\rm{avg}}}^{-1}(t)=\frac{1}{{N}_{{\rm{source}}}\cdot {N}_{{\rm{sink}}}}\sum _{p}{L}_{{\rm{SCC}}}^{-1}(p),$$

where Nsource is the number of sources, Nsink is the number of sinks, LSCC(p) is the number of SCCs that the given path p from source to sink goes through, and the sum is for all paths from sources to sinks. The metric changes from 1 (all sources and sinks in the same SCC) to 0 (no or infinitely long paths between sources and sinks). For the generation of the diffusion networks we were using the topt threshold that maximized the difference between \({L}_{{\rm{avg}}}^{-1}(t)\) for the real data and the third quartile of \({L}_{{\rm{avg}}}^{-1}(t)\) for randomly shuffled data.

In the generated SCC networks, the direction of the edges was reverted back, and the results were exported as GraphML files using in-house Julia scripts ( The catalogue of the networks for each viral bait is available as Supplementary Data 1.

To assess the significance of edges in the resulting network, we calculated the P-value of the edge gi→gj as the probability that the transition probability between the given pair of genes based on permuted data is higher than the transition probability based on the real data:

$$P({w}_{{\rm{real}}}({g}_{i},{g}_{j})\le {w}_{{\rm{perm}}}({g}_{i},{g}_{j})).$$

This P-value was stored as the ‘prob_perm_walkweight_greater’ edge attribute of GraphML output. The specific subnetworks predicted by the network diffusion (Fig. 4b–d) were filtered for edges with P ≤ 0.05.

When the gi→gj connection was not present in the ReactomeFI network, to recover the potential short pathways connecting gi and gj, ReactomeFI was searched for intermediate gk nodes, such that the edges gi→gk and gk→gj are present in ReactomeFI. The list of these short pathways is provided as the ‘flowpaths’ edge attribute in GraphML output.

The GraphML output of network diffusion was prepared for publication using yEd (v.3.20;

Intersection with other SARS coronavirus datasets

The intersection between the data generated by this study and other publicly available datasets was done using the information from respective supplementary tables. When multiple viruses were used in a study, only the comparisons with SARS-CoV and SARS-CoV-2 were included. For time-resolved data, all time points up to 24 hpi were considered. The dataset coverage was defined as the number of reported distinct protein groups for proteomic studies and genes for transcriptomic studies. Confident interactions or significant regulations were filtered according to the criteria specified in the original study. A hit was considered as ‘confirmed’ when it was significant both in the present study and the external data and showed the same trend.

qRT–PCR analysis

RNA isolation from SARS-CoV and SARS-CoV-2 infected A549-ACE2 cells was performed as described above (Qiagen). Five hundred nanograms total RNA was used for reverse transcription with PrimeScript RT with gDNA eraser (Takara). For relative transcript quantification PowerUp SYBR Green (Applied Biosystems) was used. Primer sequences can be provided upon request.

Co-immunoprecipitation and western blot analysis

HEK293T cells were transfected with pWPI plasmid encoding single HA-tagged viral proteins, alone or together with pTO-SII-HA expressing host factor of interest. 48 h after transfection, cells were washed in PBS, flash frozen in liquid nitrogen and kept at −80 °C until further processing. Co-immunoprecipitation experiments were performed as described previously55,56. In brief, cells were lysed in lysis buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 1.5 mM MgCl2, 0.2% (v/v) NP-40, 5% (v/v) glycerol, cOmplete protease inhibitor cocktail (Roche), 0.5% (v/v) 750 U μl−1 Sm DNase) and sonicated (5 min, 4 °C, 30 s on, 30 s off, low settings; Bioruptor, Diagenode SA). HA or Streptactin beads were added to cleared lysates and samples were incubated for 3 h at 4 °C under constant rotation. Beads were washed six times in the lysis buffer and resuspended in 1× SDS sample buffer 62.5 mM Tris-HCl pH 6.8, 2% SDS, 10% glycerol, 50 mM DTT, 0.01% bromophenol blue). After boiling for 5 min at 95 °C, a fraction of the input lysate and elution were loaded on NuPAGE Novex 4–12% Bis-Tris (Invitrogen), and further submitted to western blotting using Amersham Protran nitrocellulose membranes. Imaging was performed by HRP luminescence (ECL, Perkin Elmer).

SARS-CoV-2 infected A549-ACE2 cell lysates were sonicated (10 min, 4 °C, 30 s on, 30 s off, low settings; Bioruptor, Diagenode). Protein concentration was adjusted based on Pierce660 assay supplemented with ionic detergent compatibility reagent. After boiling for 5 min at 95 °C and brief centrifugation at maximum speed, the samples were loaded on NuPAGE Novex 4–12% Bis-Tris (Invitrogen), and blotted onto 0.22 μm Amersham Protran nitrocellulose membranes (Merck). Primary and secondary antibody stainings were performed according to the manufacturer’s recommendations. Imaging was performed by HRP luminescence using Femto kit (ThermoFischer Scientific) or Western Lightning PlusECL kit (Perkin Elmer).

Mapping of post-translational modification sites on the N CTD structure

N CTD dimers of SARS-CoV-2 (PDB: 6YUN) and SARS-CoV (PDB: 2CJR) were superimposed by aligning the α-carbons backbone over 111 residues (from position 253/254 to position 364/365 following SARS-CoV-2/SARS-CoV numbering) by using the tool MatchMaker76 as implemented in the Chimera software77. Ubiquitination sites were visually inspected and mapped by using the PyMOL software ( Phosphorylation on Ser310/311 was simulated in silico by using the PyTMs plugin as implemented in PyMOL78. Inter-chain residue contacts, dimer interface area, free energy and complex stability were comparatively analysed between non-phosphorylated and phosphorylated SARS-CoV-2 and SARS-CoV N CTD by using the PDBePISA server79. Poisson–Boltzmann electrostatic surface potential of native and post-translationally modified N CTD was calculated by using the PBEQ Solver tool on the CHARMM-GUI server by preserving existing hydrogen bonds80. Molecular graphics depictions were produced with the PyMOL software.

Reporter assay and IFN bioassay

The following reporter constructs were used in this study: pISRE-luc was purchased from Stratagene, EF1-α-ren was obtained from E. Gürlevik (Department of Gastroenterology, Hepatology and Endocrinology, Hannover Medical School, Germany), pCAGGS-Flag-RIG-I was obtained from C. Basler (Department of Microbiology, Mount Sinai School of Medicine, USA), pIRF1-GAS-ff-luc, pWPI-SMN1-flag and pWPI-NS5 (ZIKV)-HA was described previously56,81.

For the reporter assay, HEK293-R1 cells were plated in 24-well plates 24 h before transfection. Firefly reporter and Renilla transfection control were transfected together with plasmids expressing viral proteins using polyethylenimine (PEI, Polysciences) for untreated and treated conditions. In 18 h cells were stimulated for 8 h with a corresponding inducer and collected in the passive lysis buffer (Promega). Luminescence of Firefly and Renilla luciferases was measured using dual-luciferase-reporter assay (Promega) according to the manufacturer’s instructions in a microplate reader (Tecan).

Total amounts of IFN-α and IFN-β in cell supernatants were measured by using 293T cells stably expressing the firefly luciferase gene under the control of the mouse Mx1 promoter (Mx1-luc reporter cells)82. In brief, HEK293-R1 cells were seeded, transfected with pCAGGS-flag-RIG-I plus viral protein constructs and stimulated as described above. Cell supernatants were collected in 8 h. Mx1-luc reporter cells were seeded into 96-well plates in triplicates and were treated 24 h later with supernatants. At 16 h after incubation, cells were lysed in the passive lysis buffer (Promega), and luminescence was measured with a microplate reader (Tecan). The assay sensitivity was determined by a standard curve.

Viral inhibitor assay

A549-ACE2 cells were seeded into 96-well plates in DMEM medium (10% FCS, 100 μg ml−1 streptomycin, 100 IU ml−1 penicillin) one day before infection. Six hours before infection, or at the time of infection, the medium was replaced with 100 μl of DMEM medium containing either the compounds of interest or DMSO as a control. Infection was performed by adding 10 μl of SARS-CoV-2-GFP (MOI of 3) per well and plates were placed in the IncuCyte S3 Live-Cell Analysis System (Essen Bioscience), where whole well real-time images of mock (phase channel) and infected (GFP and phase channel) cells were captured every 4 h for 48 h. Cell viability (mock) and virus growth (mock and infected) were assessed as the cell confluence per well (phase area) and GFP area normalized by cell confluence per well (GFP area/phase area) respectively using IncuCyte S3 Software (Essen Bioscience; version 2019B rev2).

For comparative analysis of antiviral treatment activity against SARS-CoV and SARS-CoV-2, A549-ACE2 cells were seeded in 24-well plates, as previously described. Treatment was performed for 6 h with 0.5 ml of DMEM medium containing either the compounds of interest or DMSO as a control, and infected with SARS-CoV-Frankfurt-1 or SARS-CoV-2-MUC-IMB-1 (MOI of 1) for 24 h. Total cellular RNA was collected and analysed by RT–qPCR, as previously described.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.