Results

Phosphorylation Signaling Represents a Primary Host Response to SARS-CoV-2 Infection
To determine how SARS-CoV-2 hijacks host-protein signaling, a global phosphoproteomics experiment was performed in Vero E6 cells, a cell line originating from the kidney of a female African green monkey (Chlorocebus sabaeus) (Osada et al., 2014). This cell line was selected because of its high susceptibility to SARS-CoV-2 infection (Harcourt et al., 2020). Cells were harvested in biological triplicate at 6 time points after SARS-CoV-2 infection (0, 2, 4, 8, 12, or 24 h) or after mock infection at 0 or 24 h (Figure 1 A). Using a data-independent acquisition (DIA) proteomics approach, each sample was then partitioned and analyzed for changes in global protein abundance or phosphorylation (data available in Table S1). Chlorocebus sabaeus and human protein sequences were aligned, and phosphorylation sites and protein identifiers were mapped to their respective human protein orthologs. Phosphorylation fold changes calculated using the 0- or 24-h mock control were highly comparable (correlation coefficient r = 0.77); therefore, the 0-h mock control was used for all subsequent comparisons.
Figure 1 Global Proteomics of Phosphorylation and Abundance Changes upon SARS-CoV-2 Infection
(A) Vero E6 cells were infected with SARS-CoV-2 (MOI 1.0). After 1 h of viral uptake, cells were harvested (0 h) or, subsequently, after 2, 4, 8, 12, or 24 h. As a control, Vero E6 cells were also mock infected for 1 h and harvested immediately thereafter (0 h) or after 24 h of mock infection. All conditions were performed in biological triplicate. Following cell harvest, cells were lysed, and proteins were digested into peptides. Aliquots of all samples were analyzed by mass spectrometry (MS) to measure changes in protein abundance upon infection, whereas the remaining sample was enriched for phosphorylated peptides and subsequently analyzed to measure changes in phosphorylation signaling. A DIA approach was used for all MS acquisitions. Last, all phosphorylation sites and protein identifiers were mapped to their respective human protein orthologs.
(B) Principal-component analysis (PCA) of phosphorylation replicates after removing outliers. See also Figure S1.
(C) Correlation of protein abundance (AB) and phosphorylation sites (PHs) between replicates within a biological condition (red) and across biological conditions (black). Boxplots depict median (horizonal lines), interquartile range (boxes), maximum and minimum values (vertical lines), and outliers (solid circles).
(D) Median AB of individual SARS-CoV-2 proteins in the protein AB analysis.
(E) The number of significantly regulated PH groups across the infection time course.
(F) Volcano plot of PH group quantification 24 h after infection.
(G) The number of significantly regulated proteins across the infection time course.
(H) Volcano plot of protein AB quantification 24 h after infection.
(I) Gene Ontology enrichment analysis of all significantly changing proteins in terms of AB divided into two sets: downregulated (blue) and upregulated (red).
(J) Proportion of significantly regulated PH groups with a correlated (i.e., same direction, AB match) or anticorrelated (i.e., opposite direction, AB mismatch) significant or insignificant (gray) change in protein AB.
In (E)–(H), all infection time points are compared with the mock infection at 0 h, and significantly regulated proteins are defined as (absolute value of log2(infection/mock) > 1 and adjusted p < 0.05 or when only detected in infected or mock based on replicate and MS feature counts; STAR Methods).
See also Figure S1.
Quality control filtering of the data was performed, and two samples from each of the phosphorylation and protein abundance datasets were removed because of poor correlation with their respective replicates (Figures S1 A and S1B). Principal component analysis (PCA) of the remaining samples revealed good separation of mock and infected samples as well as high quantitative reproducibility between biological replicates (Figures 1B, 1C, and S1C). In total, high-quality quantification of 4,624 human-orthologous phosphorylation sites and 3,036 human-orthologous proteins was obtained (Figure S1D). Successful infection was confirmed by the observation of a dramatic increase in viral protein abundance over the course of a 24-h infection period (Figures 1D and S1E).
Figure S1 Proteomics: Quality Control (QC), Orthology, Enrichments, and Viral Proteins, Related to Figure 1
(A) Principal component analysis computed on intensities summarized by MSstats at the level of phosphorylation site groups within (from left to right) all runs, with one outlier run removed, and with two outlier runs removed. Outlier runs are labeled 00Hr.2 and 02.Hr.2. (B) Principal components analysis computed on protein intensities as summarized by MSstats (from left to right) within all runs, with one outlier run removed, and with two outlier runs removed. Outlier runs are both labeled 00Hr.2; one is mock and the other is infected. (C) Coefficient of variance boxplot for each condition. Black lines depict the median and their values are indicated above each boxplot. (D) Mapping detected and quantifiable proteins and phosphorylation sites from the green monkey (Chlorocebus sabaeus) protein sequences to human genes. Proteins and sites were considered quantifiable if MSstats computed a non-infinite fold change for any time point or if an infinite log2 fold change passes criteria for inclusion in any time point. (E) Intensities of viral proteins as summarized over all peptide ion fragments by MSstats, averaged across replicates. The MSstats summarization is based on the median intensity of all fragments after data pre-processing (STAR Methods). (F) Gene Ontology enrichment analysis for proteins significantly regulated in terms of abundance upon infection, separated by time point and direction of phosphorylation regulation. All terms with significant over-representation (adjusted p value < 0.01) in the regulated gene set are kept, and redundant terms are removed (see STAR Methods). Numbers in cells indicate the number of genes that match the term for a given time point and direction. (G) Gene Ontology enrichment analysis for significantly phosphorylated proteins upon infection, separated by time point and direction of protein regulation. Details same as for (F).
As expected, an increase was observed in the number of significantly regulated phosphorylation sites and proteins over the infection time course, with the majority of regulation occurring at the level of phosphorylation (Figures 1E and 1F) as opposed to protein abundance (Figures 1G and 1H). Of the few proteins that significantly increased in abundance upon infection, the vast majority were SARS-CoV-2 viral proteins (Figure 1H). In contrast, the majority of host proteins decreased in abundance. This finding is consistent with mechanisms of host mRNA nuclear export and/or host mRNA translation inhibition, which are common in viral infections (Kuss et al., 2013; Walsh and Mohr 2011). Gene Ontology enrichment analysis of significantly downregulated proteins revealed several terms related to platelet regulation (Figures 1I and S1F). Several downregulated host proteins are known to be involved in platelet regulation, thrombosis, and prevention of blood coagulation, including APOH, CD9, TSPAN14, AHSG, SERPINA1, and A2M (Mather et al., 2016; Mangin et al., 2009; Taggart et al., 2000). The downregulation of these proteins suggests that they may mechanistically contribute to symptoms of blood coagulation and stroke in COVID-19 patients (Han et al., 2020).
Lastly, the contribution of protein abundance to phosphorylation level changes was evaluated. For nearly all cases of a significantly changed phosphorylation site, no corresponding significant change in protein abundance was observed (Figure 1J), further suggesting that phosphorylation signaling represents a primary host response over this time course of infection as opposed to transcriptional regulation, which would influence protein abundance.

Phosphorylation of SARS-CoV-2 Viral Proteins by the Host Proteome
Viral protein phosphorylation within the host cell may play a role in sensing and responding to cell state. We detected 25 phosphorylation sites in SARS-CoV-2 viral proteins that we combined with another proteomics dataset (Davidson et al., 2020) to amass a total of 49 sites detected across seven viral proteins (Table S2). Of note, this analysis does not distinguish cleaved from uncleaved viral proteins in the assignment of viral phosphorylation sites. The degree of conservation, indicative of functional constraint, was estimated for each residue position (Figure 2 A; Ng and Henikoff 2003), and the sites were mapped to positions within structured regions for five proteins, with the majority observed in accessible positions (i.e., loops) (Figure 2B). The top kinase families predicted by sequence to regulate these sites included casein kinase II (CK2), cyclin-dependent kinase (CDK), and protein kinase C (PKC), among others (Figure 2C), suggesting that these kinases may contribute to regulation of viral replication.
Figure 2 Overview of SARS-CoV-2 Viral Protein PHs in the Host Cell
(A) Localization of PHs across viral protein sequences from this study and a previous study (Davidson et al., 2020). Stem height indicates predicted deleteriousness of alanine substitutions. Dot color indicates whether the residue is (true) or is not (false) predicted to form part of an interaction interface based on SPPIDER analysis. Positions with no structural coverage are excluded from interface prediction.
(B) Distribution of secondary structure elements in which viral PHs were found, as classified by the define secondary structure of proteins (DSSP) tool.
(C) Distribution of top matching host kinases to viral PHs according to NetPhorest tool (Horn et al., 2014).
(D) Phosphorylation cluster in the C-terminal tail of the M protein (red residues) structure (Heo and Feig 2020) and associated sequence motif. Asterisks indicate PHs.
(E) Alignment of M protein phosphorylation clusters across different coronaviruses. Asterisks indicate PHs.
(F) Surface electrostatic potential of non-phosphorylated (left) and phosphorylated (right) RNA-binding domains of the N protein (PDB: 6M3M). Positions of PHs are indicated by arrows. Blue denotes a positive charge potential, and red indicates a negative charge potential. Electrostatic potential was computed with the Advanced Poisson-Boltzmann Solver (APBS) tool after preparation with the PDB2PQR tool.
Although it is unlikely that all phosphorylation sites on viral proteins play important functional roles, several sites in membrane (M) protein, Nsp9, and nucleocapsid (N) protein (Figures 2D–2F) suggest potential functionality. Five phosphorylation sites were detected in the M protein cluster within a short C-terminal region of the protein (207–215; Figure 2D). Although these acceptor residues are not predicted to be conserved, several are negatively charged residues in M proteins of other related viruses (Figure 2E). This evolutionary pattern suggests that a negative charge in this region may play a functional role, reminiscent of other multi-site phosphorylation events (Serber and Ferrell 2007).
To identify phosphorylation sites that may regulate protein-protein interactions, all sites were mapped to 3D structures, and solvent accessibility based protein-protein interface identification and recognition (SPPIDER) was used to assess whether sites resided within interface regions (Porollo and Meller 2007; Figure 2A; Table S2). The single phosphorylation site in Nsp9 was predicted to be at an interface region (“True”), which was supported by inspection of the homodimer structure (PDB: 6W4B). Additional phosphorylation sites were predicted to be at interface residues within the S protein (Figure 2A). However, inspection of S in complex with the ACE2 receptor (Shang et al., 2020; Lan et al., 2020) reveals some of these phosphorylation sites to be near but not at the interface region.
Finally, phosphorylation sites in N protein, a structural protein that binds to and assists with packaging viral RNA, were investigated. Most sites occurred within the N-terminal portion of the protein, at or near the RNA binding region, but avoided the C-terminal dimerization domain. The cluster of phosphorylation sites within an arginine/serine (RS)-dipeptide rich region, C-terminal to the RNA binding region (Figure 2A), is conserved in other coronavirus N proteins. This region is phosphorylated in SARS-CoV by serine-arginine (SR) protein kinases, modulating the role of SARS-CoV N protein in host translation inhibition (Peng et al., 2008). It is likely that phosphorylation of this same region in SARS-CoV-2 plays a similar role. Interestingly, in vitro inhibition of SARS-CoV N protein phosphorylation at the RS-rich region results in reduced viral load and cytopathic effects (Wu et al., 2009), highlighting its importance for viral fitness. In addition, sites spanning the sequence of the RNA binding domain, which forms a claw-like structure, have been observed (Kang et al., 2020). Several phosphorylation sites cluster in the structural model, predicted to affect the surface charge of the so-called acidic wrist region (Figure 2F) but not the positive surface charge of the RNA binding pocket. We hypothesize that this surface charge difference may modulate N protein function, potentially via allosteric regulation of RNA binding capacity.

Phosphorylation of SARS-CoV-2 Host-Interacting Proteins during Infection
The recently published SARS-CoV-2 virus-human protein-protein interaction map identified 332 human proteins interacting with 27 (26 wild-type and 1 mutant) viral proteins (Gordon et al., 2020). Here we found some of these host proteins (40 of 332) to be significantly differentially phosphorylated upon infection (Figure 3 ). Virus-host protein-protein interactions could drive changes in phosphorylation by affecting host protein subcellular localization or by sterically blocking kinase access. Furthermore, phosphorylation of these proteins upon infection may signify an additional mode of functional control over these potential dependency and restriction factors.
Figure 3 Phosphorylation on SARS-CoV-2 Virus-Human Interacting Proteins
The SARS-CoV-2 virus-host protein-protein interaction map (Gordon et al., 2020) found 332 human proteins interacting with 27 (26 wild-type and 1 mutant) viral proteins. Here we found 40 of 332 proteins significantly differentially phosphorylated across at least two time points (adjusted p < 0.05 and absolute value of log2 fold change [abs(log2FC)] > 1). Viral proteins are shown as red diamonds. Interacting human proteins are shown as gray circles. PHs emanate from human proteins, colored by their log2 fold change compared with uninfected control samples (red, increase; blue, decrease) at each time point (0, 2, 4, 8, 12, and 24 h after infection) in a clockwise fashion. An interactive version of phosphorylation data can be found at https://kroganlab.ucsf.edu/network-maps.
The SARS-CoV-2 N protein is known to interact with several RNA-processing proteins that are differentially phosphorylated during infection, including LARP1 and RRP9. Here LARP1 phosphorylation decreases on several sites, which is known to consequently increase LARP1 affinity for 3′ untranslated regions (UTRs) of mRNAs encoding ribosomal proteins, driving inhibition of protein synthesis (Hong et al., 2017). This mechanism may be utilized by SARS-CoV-2 to prioritize synthesis of viral proteins over host proteins. In addition, ORF6 interacts with the NUP98/RAE complex, and NUP98 phosphorylation was observed to increase at S888, a site within its peptidase domain. NUP98 autocatalytic cleavage is required for localization to the nuclear pore; thus, it is possible that NUP98 interaction with ORF6 and/or its virus-induced phosphorylation prevents host mRNA export through the nuclear pore (Krull et al., 2010; Hodel et al., 2002). A similar mechanism is employed by vesicular stomatitis virus (VSV) matrix protein to block host mRNA export by targeting the NUP98/RAE complex, leading to exclusive translation of cytoplasmic VSV mRNAs (Quan et al., 2014).
For Nsp12, the majority of its protein interactors displayed decreased phosphorylation during infection. Because Nsp12 is known to encode the RNA-dependent RNA polymerase, responsible for replicating the viral genome, and several of these interacting proteins are related to RNA processing (LARP4B and CRTC3), their regulation may possess functional implications for Nsp12 in viral RNA replication. In addition, Nsp8 interacts with several proteins whose phosphorylation increases (LARP7 and MPHOSPH10) and decreases (CCDC86) on several sites. Notably, LARP7 and MEPCE are important regulators of RNA polymerase II-mediated transcription elongation as part of the 7SK small nuclear ribonucleoprotein particle (snRNP) complex. Regulation of these phosphorylation sites may contribute to the regulation of positive transcription elongation factor b (P-TEFb [CDK9]) and transcriptional regulation of the virus, similar to how these proteins are regulated during HIV infection (Mbonye et al., 2015).

SARS-CoV-2 Infection Regulates Host Kinase Signaling
To study global changes in kinase signaling and their effect on host protein phosphorylation, regulated phosphorylation sites were grouped in five clusters based on their dynamics using a data-driven clustering approach (Figure 4 A; STAR Methods). For each of the groups, an enrichment analysis was performed for functions and pathways (Figure 4A; Table S3). The dynamics of these changes can be linked to the viral life cycle: entry (0–2 h), replication (4–12 h), and egress (24 h). Clusters 1 and 2 include phosphorylation sites that are, on average, upregulated during infection. Cluster 1 sites tended to be upregulated within 2 h (i.e., linked to viral entry) and were enriched in mRNA processing, cell cycle, apoptosis, and proteins involved in HIV infection. Cluster 2 included apoptosis proteins with later onset of phosphorylation, associated with replication and/or egress. Phosphorylation sites in clusters 3 and 4 were downregulated and enriched in RNA-processing functions. Sites within cluster 5 possessed a dynamic response to infection, with immediate downregulation followed by a rise during the middle and renewed downregulation at late time points. This cluster was enriched for DNA replication and the cell cycle, among others. These observations are corroborated by standard Gene Ontology (GO) enrichment analyses of biological processes regulated by phosphorylation (Figure S1G; Table S3; STAR Methods).
Figure 4 Signaling Changes in Host Cells in Response to SARS-CoV-2 Infection
(A) Clusters of significantly changing PHs (abs(log2FC) > 1 and adjusted p < 0.05) across the time course of infection with non-redundant enriched Reactome pathway terms (adjusted p value [q] < 0.01) for each cluster. Horizontal red lines below each pathway term correspond to phosphorylated proteins belonging to the pathway, and a black-bordered rectangle is indicative of a significantly enriched term.
(B) Kinases depicting a strong change in activity upon infection (abs(log10(p)) > 2.5) in at least one time point, with predicted activity in at least 5 of 6 time points.
(C) Schematic representation of interaction between host kinases and SARS-CoV-2 viral proteins from Gordon et al. (2020). Substrate PHs for each kinase are color-coded as blue (down) and red (up) based on the direction of change during infection. Only PHs corresponding to the kinase activity direction are shown.
(D) Correlation of kinase activity profiles of each time point with other biological conditions with at least one significantly changing kinase (abs(log10(p)) > 2.5) and having significant correlation with at least one time-point (false discovery rate [FDR] < 5%).
(E) Overall phosphorylation change (−log10(p)) of a protein complex, estimated as the change in phosphorylation on member proteins. Only non-redundant protein complexes with a significant change in phosphorylation (abs(log10(p)) > 2.5) in at least one time point are shown.
See also Figure S2.
We estimated activity regulation for 97 kinases based on the regulation of their known substrates (Ochoa et al., 2016; Hernandez-Armenta et al., 2017; Table S4), with the strongest regulation linked to viral entry (0–2 h) and late replication/egress (24 h). The kinases predicted to be most strongly activated (Figures 4B and S2A) include several members of the p38 pathway, including p38ɣ (MAPK12), CK2 (CSNK2A1/2), Ca2+/calmodulin-dependent protein kinase (CAMK2G), and the guanosine monophosphate (GMP)-dependent protein kinases PRKG1/2, which can inhibit Rho signaling. Kinases predicted to be downregulated include several cell cycle kinases (CDK1/2/5 and AURKA), cell growth-related signaling pathway kinases (PRKACA, AKT1/2, MAPK1/3, and PIM1), and the cytoskeleton regulators (PAK1), among others. Kinase activity estimates based on the 24-h mock control gave highly correlated results (r = 0.81), identifying the same set of highly regulated kinases (Figure S2D). Some of the changes in kinase activity can be directly linked to host-viral protein interactions (Figure 4C). Among the 10 interacting kinases detected in a virus-host protein-protein interaction map (Gordon et al., 2020), an increase in activity for CK2 and a decrease for MARK2 and PRKACA were observed (Figure 4C). Of note, although we predict decreased activity for PRKACA based on phosphorylation of its substrates, we simultaneously detected a significant increase in T198 phosphorylation (8, 12, and 24 h post infection) within its activation loop, suggesting an increase in PRKACA activity. It is possible that Nsp13 is sequestering active PRKACA away from its typical substrates.
Figure S2 Full Kinase Activities, Correlated Conditions, and Regulated Complexes, Related to Figure 4
(A) Changes in predicted kinase activities across different time points post-infection. (B) Correlation of kinase activity profiles of each time point with other biological conditions. Kinase activities were estimated for a wide-range of biological conditions obtained from previously published phosphoproteomics datasets (Ochoa et al., 2016). (C) Changes in phosphorylation in protein complexes. Overall phosphorylation change (-log10 (p value) of a protein complex was derived from change in phosphorylation of sites in member proteins. (D) Kinase activity estimates when using either the 0- or 24-h mock controls for those top regulated kinase activities from the 0-h control comparison.
To better understand the signaling states of cells over the course of infection, we compared our data with a compilation of public phosphoproteomics datasets of other conditions (Ochoa et al., 2016; Figures 4D and S2B). The first and last time point of infection resembled a kinase activation state induced by inhibition of mTOR, ERK, AKT, and EGFR, consistent with the estimated kinase activities of these growth-related pathways. Between 2 and 12 h after infection, kinase activity states resembling inhibition of phosphatidylinositol 3-kinase (PI3K), p70S6K, and Rho-associated protein kinases (ROCKs) were observed. Finally, several of the time points resembled S/G2 cell cycle state, suggestive of a cell cycle block. Conversely, some conditions were anticorrelated with kinase activity profiles (Figures 4D and S2B). In line with an S/G2 cell-cycle block, infection signaling appeared opposite to that of a mitotic cell. In addition, inhibitors of histone deacetylases (HDACs) (scriptaid and trichostatin A), the proteasome (bortezomib), Hsp90 (geldanamycin), and voltage-gated sodium channels (valproic acid) were also anticorrelated. These drugs, or drugs targeting these protein activities, could induce a signaling state that inhibits viral replication.
To further link kinase activities to downstream protein complexes, enrichment of up- or downregulated phosphorylation sites was determined within a curated set of human protein complexes defined by CORUM (Giurgiu et al., 2019; Figures 4E and S2C). This analysis revealed significant changes in phosphorylation of splicing related complexes (spliceosome), the proteasome (PA700-20S-PA28), and chromatin remodeling complexes (HuCHRAC and MLL2). In addition, a subset of regulated phosphorylation sites were detected that have known regulatory functions or high predicted functional scores (Ochoa et al., 2020) that are linked to regulation of protein activities (Table S5). Consistent with the observed signaling changes described above, these regulatory phosphorylation sites are involved in activation of chaperones (including HSP90), proteasome activity, inhibition of the anaphase-promoting complex (APC), and regulation of HDACs and cytoskeleton proteins, among others.

CK2 and N Co-localize at Virus-Induced Filopodial Protrusions
The phosphoproteomics data indicated regulation of several kinases and effector proteins related to cytoskeleton organization upon SARS-CoV-2 infection. Kinases downstream of the Rho/Rac/Cdc42 GTPases (PAK1/2 and ROCK1/2) and several well-characterized phosphorylation site targets of PAK1/2 kinase in vimentin (VIM S39 and S56) and stathmin (STMN1 S16 and S25) were found to be downregulated during infection (Figures 5A and 5B). The interaction of Nsp7 with RHOA (Gordon et al., 2020) may contribute to this downregulation. In contrast, signaling via CK2 is strongly upregulated, as determined by the increase in phosphorylation of well-characterized target sites (Figures 5A and 5B). Among the many roles of this kinase, we noted increased phosphorylation of cytoskeleton protein targets such as ɑ-Catenin (CTNNA1 S641) and the heavy chain of the motor protein Myosin IIa (MYH9 S1943). In addition to these kinase-mediated effects, the Nsp2 protein of SARS-CoV-2 also interacts directly with strumpellin (WASHC5), a subunit of the actin assembly-inducing WASH complex (Gordon et al., 2020), further implicating cytoskeleton regulation during infection. To study the relevance of these observations in a human infection model, high-resolution immunofluorescence imaging of fixed Caco-2 human colon epithelial cells was performed 24 h after infection (STAR Methods).
Figure 5 Colocalization of CK2 and Viral Proteins at Actin Protrusions
(A) Pathway of regulated PHs and SARS-CoV-2 interaction partners involved in cytoskeletal reorganization. Dashed lines indicate downregulation of activity, while solid lines indicate upregulation of activity.
(B) Regulation of individual kinase activity or PHs depicted in (A).
(C) Caco-2 cells infected with SARS-CoV-2 at an MOI of 0.1 for 24 h prior to immunostaining for F-actin and M protein, as indicated. Shown is a confocal section revealing M protein localization along and to the tip of filopodia (left) and magnification of the dashed box (right).
(D) Dot plot quantification of the number and length of filopodia in untreated (mock) or infected Caco-2 cells for 24 h with SARS-CoV-2. Filopodium length was measured from the cortical actin to the tip of the filopodium. Error bars represent SD. Statistical testing by Mann-Whitney test.
(E) Caco-2 cells infected with SARS-CoV-2 at an MOI of 0.01 for 24 h prior to immunostaining for F-actin, N protein, and casein kinase II (CK2) as indicated (left). Shown is magnification of the dashed box as single channels (right).
(F) Magnification of the dashed box from (E) with quantification of colocalization between CK2 and N protein throughout infected Caco-2 cells. Displayed is the proportion of N protein-positive particles colocalizing with CK2. Error bars represent SD.
(G and H) Scanning electron microscopy (G) and transmission electron microscopy (H) images of SARS-CoV-2 budding from Vero E6 cell filopodia (black arrow in H).
See also Figure S3.
SARS-CoV-2 infected Caco-2 cells were imaged for filamentous actin and the SARS-CoV-2 M protein, revealing prominent M protein clusters, possibly marking assembled SARS-CoV-2 viral particles, localized along the shafts and at the tips of actin-rich filopodia (Figures 5B and S3 B). SARS-CoV-2 infection induced a dramatic increase in filopodial protrusions, which were significantly longer and more branched than in uninfected cells (Figure 5D). Uninfected cells also exhibited filopodial protrusions, but their frequency and shape were dramatically different (Figure S3A). Reorganization of the actin cytoskeleton is a common feature of many viral infections and is associated with different stages of the viral life cycle (Taylor et al., 2011).
Figure S3 Microscopy Images Showing Response to SARS-CoV-2 Infection, Related to Figure 5
(A) Non-infected Caco2 cells co-stained for F-actin, CK2 and nuclei (DAPI). Magnification of the indicated area is displayed as a single channel and merged images on the right panels. (B) Caco2 cells infected with SARS-CoV-2 at an MOI of 0.1 for 24 h prior to immunostaining for F-actin and M-protein, as indicated. See lower (1) and right (2) panel for magnification of regions indicated by dashed boxes. (C) Scanning electron microscopy and (D) transmission electron microscopy image of SARS-CoV-2 budding from Vero E6 cell filopodia. (E) N protein was found to physically interact with casein kinase II subunits (cartoon, left), CSNK2B and CSNK2A2 (Gordon et al., 2020). To test whether N protein could directly control CK2 activity, N protein was transduced via lentivirus in Vero E6 cells and stably induced via doxycycline for 48 hours followed by phosphoproteomics analysis. Kinase activities were calculated as before (STAR Methods) and top up- (> 1.5, red) and downregulated (< 1.5, blue) kinases are shown. See Table S1 for full phosphoproteomics data and Table S4 for full list of predicted kinase activities.
We hypothesize that induction of virus-containing filopodia could be important for SARS-CoV-2 egress and/or cell-to-cell spread within epithelial monolayers. Given that Rho/PAK/ROCK signaling is downregulated, we next asked whether CK2 could play a role in this process. At 24 h, infected cells showed CK2 expression along the thin filopodial protrusions (Figure 5E), partially co-localized with SARS-CoV-2 N protein (Figure 5F). Scanning and transmission electron microscopy were used (Figures 5G, 5H, S3C, and S3D) to image the cellular protrusions at higher resolution. Assembled viral particles are clearly visible along these filopodia (Figure 5G), with instances where the viral particles appear to be budding from the protrusions (Figure 5H). Finally, we performed a global phosphoproteomics analysis of Vero E6 cells overexpressing N protein and observed CK2 activity to be significantly upregulated (Figure S3E; Tables S1 and S4). Because CK2 activity can promote actin polymerization (D’Amore et al., 2019), we hypothesize that N protein may allosterically control CK2 activity and regulate cytoskeleton organization.

SARS-CoV-2 Infection Promotes p38/MAPK Signaling Activity and Cell Cycle Arrest
Kinase activity analysis of SARS-CoV-2 phosphorylation profiles predicted upregulation of several components of the p38/mitogen-activated protein kinase (MAPK) signaling pathway, including MAP2K3, MAP2K6, MAPK12, MAPKAPK2 (MK2), and MAPKAPK3 (Figures 6A and 6B). Immunoblotting for activated phospho-p38 (T180/Y182), phospho-MK2 (T334), and phospho-cAMP response element-binding protein (CREB) and phospho-ATF-1 at their respective MAPKAPK2 sites (S133 in both) confirmed activation of the p38/MAPK pathway during SARS-CoV-2 infection in ACE2-expressing A549 human lung carcinoma cells (ACE2-A549) (Figure 6C). Furthermore, phosphoproteomics data depict increased phosphorylation of p38 pathway substrates such as negative elongation factor E (NELFE), heat shock protein beta-1 (HSPB1), and signal transducer and activator of transcription 1-alpha/beta (STAT1), among others (Figure 6D). Regulation of these sites occurs late in the time course (24 h after infection), likely reflecting a more advanced stage of viral infection, replication, and egress.
Figure 6 SARS-CoV-2 Activates the p38/MAPK Signaling Pathway and Causes Cell Cycle Arrest
(A) Diagram of the p38/MAPK signaling pathway.
(B) Kinase activity analysis for kinases in the p38/MAPK pathway.
(C) Western blot analysis of phosphorylated p38/MAPK signaling components in mock- and SARS-CoV-2-infected ACE2-A549 cells 24 h after infection.
(D) Log2 fold change profiles of indicated p38/MAPK substrates during SARS-CoV-2 infection in Vero E6 cells.
(E) Transcription factor activity analysis of SARS-CoV-2-infected A549, Calu-3, and NHBE cells, comparing p38/MAPK transcription factors with transcription factors not associated with the p38/MAPK pathway. Statistical test: Mann-Whitney test.
(F) qRT-PCR analysis of the indicated mRNA from ACE2-A549 cells pre-treated with the p38 inhibitor SB203580 at the indicated concentrations for 1 h prior to infection with SARS-CoV-2 for 24 h. Statistical test: Student’s t test. See also Figures S4 and S5.
(G) Heatmap of Pearson’s correlation coefficients comparing SARS-CoV-2-infected Vero E6 phosphorylation profiles with profiles of cells with induced DNA damage and cells arrested at the indicated cell cycle stages.
(H) Log2 fold change profiles of the indicated cell cycle and DNA damage substrates during SARS-CoV-2 infection in Vero E6 cells.
(I) DNA content analysis of cells infected with SARS-CoV-2 for 24 h compared with mock-infected cells.
The p38/MAPK pathway mediates the cellular response to environmental stress, pathogenic infection, and pro-inflammatory cytokine stimulation, whereas downstream effectors of the pathway include transcription factors and RNA binding proteins that promote inflammatory cytokine production (Cuadrado and Nebreda 2010; Wen et al., 2010). Analysis of estimated transcription factor activity from gene expression data (STAR Methods; Table S6) derived from the infection of a human lung carcinoma cell line (A549), a human epithelial lung cancer cell line (Calu3), and primary human bronchial epithelial (NHBE) cells demonstrated that transcription factors regulated by the p38/MAPK pathway were among the most highly activated upon infection (Figure 6E; Blanco-Melo et al., 2020).
To investigate the contribution of the p38/MAPK pathway to cytokine production, SARS-CoV-2-infected ACE2-A549 cells were treated with the p38 inhibitor SB203580. The mRNA of the inflammatory cytokines interleukin-6 (IL-6), tumor necrosis factor alpha (TNF-α), and others increased during infection and were inhibited by p38 inhibition in a dose-dependent manner (Figures 6F, right, and S4A). Interestingly, p38 inhibition also reduced SARS-CoV-2 subgenomic mRNA (Figure 6F, left) in the absence of major cellular toxicity (Figure S5 ), indicative of reduced viral replication. The SB203580-induced decrease in virus production was further confirmed using an anti-SARS-CoV-2 N protein (anti-NP) antibody-based assay (Figure S5, New York Vero E6). Multiplexed ELISA analysis of supernatants of cells from the same experiment demonstrated strong upregulation of inflammatory cytokines at the protein level, including IL-6, CXCL8, CCL20, and CCL2, which were decreased upon p38 inhibition (Figure S4 B; Table S7). However, because SARS-CoV-2 replication is also inhibited by SB203580, we cannot deconvolve the contributions of p38/MAPK pathway activity and SARS-CoV-2 virus presence on cytokine production.
Figure S5 Pharmacological Profiling for Viral Titers and Cell Viability, Related to Figure 7
Dose response of phosphoproteomics-informed drugs and compounds. Assays performed in New York (N; red, anti-NP; blue TCID50) and Paris (P; red, RT-qPCR; purple, plaque assays) across two cell lines (A549-ACE2 and Vero E6). Cell viability shown in black. Mean of three biological replicates is shown. Error bars are SEM.
Figure S4 Cytokine Profiling upon Infection, p38 Inhibition, and Cell Cycle Analysis, Related to Figure 6
(A) RT-qPCR analysis of indicated mRNA from A549-ACE2 cells pre-treated with p38 inhibitor SB203580 at indicated concentrations for one hour prior to infection with SARS-CoV-2 for 24 hours. Statistical test is Student’s t test. Error bars are SD. (B) Same as in (A) but a Luminex-based quantification of indicated cytokines. Error bars are SD. (C) Cell cycle analysis of Vero E6 cells (same as in Figure 6I) upon SARS-CoV-2 infection at an MOI of 1. Cell stained with DAPI DNA stain prior to flow cytometry analysis. Statistical test is Mann-Whitney test. Error bars are SD.
Comparing phosphoproteomics profiles of SARS-CoV-2-infected cells with a database of phosphorylation profiles collected at specific cell cycle stages, viral infection was most highly correlated with cells arrested at the S/G2 transition and was negatively correlated with profiles of cells in mitosis (Figure 6G). We also observed SARS-CoV-2-dependent regulation of CDK2 T14/Y15 phosphorylation, initially increased in response to SARS-CoV-2 infection at 2 h, followed by a decrease over the remainder of the infection time course (Figure 6H, left). CDK2 activity promotes transition from the G2 phase of the cell cycle into mitosis and is inhibited by phosphorylation at positions T14 and Y15 by kinases WEE1 and MYT1, preventing premature entry into mitosis (Parker and Piwnica-Worms 1992; Mueller et al., 1995). CDK2 can also become phosphorylated when the cell cycle is arrested because of checkpoint failure or DNA damage. In addition, H2AX S140 phosphorylation (i.e., γ-H2AX), a hallmark of the DNA damage response, exhibited a profile similar to CDK2, suggesting that the DNA damage response may become activated early during infection (Rogakou et al., 1998; Figure 6H, right).
To more directly test whether SARS-CoV-2 infection affects cell cycle progression, cells were infected with SARS-CoV-2 for 24 h, and their DNA content was measured using DAPI DNA staining and flow cytometry. A significant increase in the fraction of cells in S phase and at the G2/M transition and a decrease in the fraction of cells in G0/G1 phase were observed (Figures 6I and S4C). This observation is consistent with arrest between S and G2 phases of the cell cycle. A relationship between p38 activity and cell cycle arrest has been described previously, and the two could be linked mechanistically during SARS-CoV-2 infection (Lee et al., 2002; Yee et al., 2004).

Mapping Kinase Activities to Pharmacological Modulators Identifies SARS-CoV-2 Therapies
To identify effective therapies for SARS-CoV-2 infection, kinase inhibitors were mapped to the most differentially regulated kinase activities (Figure 7 A) and to specific phosphorylation sites (Table S8; STAR Methods). This resulted in a list of 87 drugs and compounds: 10 FDA-approved, 53 undergoing clinical testing, and 24 pre-clinical. Many of the drugs and compounds identified were reported to target several host kinases in cell-free assays at a minimum, but many have been observed to hit targets in cellular assays as well (Figure 7A). We reasoned that testing molecules with both overlapping and unique targets would help specify the molecular targets of greatest importance for SARS-CoV-2. Here, 68 total drugs and compounds were tested for antiviral efficacy (via qRT-PCR, anti-NP antibody, plaque assay, and/or TCID50) and cellular toxicity at two different institutions (in New York [Mount Sinai, 25 drugs/compounds] and Paris [Institut Pasteur, 62]) and in two cell lines (Vero E6 [68] and A549-ACE2 [61]). All pharmacological profiling results can be found in Figure S5 and Table S8.
Figure 7 Mapping Regulated Kinases to Kinase Inhibitors Identifies SARS-CoV-2 Therapies
(A) Kinase inhibitors (left) mapped to kinases (right) whose activity was regulated by SARS-CoV-2 infection. Lines connecting them indicate known kinase targets for each drug/compound.
(B) Vero E6 cells pre-treated with remdesivir at the indicated doses, followed by SARS-CoV-2 infection for 48 h. Percent viral titer compared with mock drug treatment (anti-NP antibody; red line, dots, and text) and cell viability (black) is depicted. Error bars represent SD.
(C) As in (B). Vero E6 cells were treated with the CK2 inhibitor silmitasertib. Physical interactions between N protein and the CSNK2A2 and CSNK2B CK2 subunits were observed in a prior study (Gordon et al., 2020).
(D) Predicted increased kinase activity for the p38 signaling pathway and drugs/compounds targeting pathway members (ralimetinib, MAPK13-IN-1, and ARRY-797) and upstream drivers (gilteritinib).
(E–G) As in (B). Vero E6 cells treated with the AXL inhibitor gilteritinib (E), the MAPK11/14 inhibitor ralimetinib (F), or the MAPK13 inhibitor MAPK13-IN-1 (G) prior to SARS-CoV-2 infection.
(H) A549-ACE2 lung epithelial cells were treated with the MAPK14 inhibitor ARRY-797 prior to SARS-CoV-2 infection.
(I) Small interfering RNA (siRNA) knockdown of p38 pathway genes in A549-ACE2 leads to a significant decrease in SARS-CoV-2 viral replication (red), as assessed by qRT-PCR in the absence of effects on cell viability (black). ACE2 and non-targeting siRNAs are included as positive and negative controls, respectively.
(J and K) Vero E6 or A549-ACE2 cells were treated with PIKFYVE inhibitor apilimod (J) or the CDK inhibitor dinaciclib (K) prior to SARS-CoV-2 infection.
See also Figure S6.
We found pharmacological inhibitors of CK2, p38 MAPK signaling, PIKFYVE, and CDKs to possess strong antiviral efficacy. Cells were pre-treated with inhibitor molecules, followed by SARS-CoV-2 infection (STAR Methods), and virus quantity (anti-NP antibody against SARS-CoV-2) and cell viability were quantified 48 h after infection. As a positive control and for comparison, remdesivir was tested, and the expected favorable antiviral activity was observed (half maximal inhibitory concentration [IC50 ] = 1.28 μM; Figure 7B). Silmitasertib, an inhibitor of CSNK2A1 and CSNK2A2, was found to possess antiviral activity (IC50 = 2.34 μM; Figures 7C and S5). In conjunction with data supporting physical interaction (Gordon et al., 2020) and co-localization with N protein (Figure 5F), as well as a potential role in remodeling extracellular matrix upon infection (Figures 5 and S3), CK2 signaling appears to be an important pathway hijacked by SARS-CoV-2. Furthermore, silmitasertib is currently being considered for human testing as a potential treatment for COVID-19.
To probe SARS-CoV-2 dependence on MAPK signaling, SARS-CoV-2 replication was measured in response to pharmacological and genetic perturbation of MAPK components that were upregulated during infection (Figure 7D). Potent antiviral activity was observed for gilteritinib (Figure 7E; IC50 = 0.807 μM), an inhibitor of AXL kinase, upstream of p38; ralimetinib (Figure 7F; IC50 = 0.873 μM), an inhibitor of MAPK11 (p38ɑ) and MAPK14 (p38β); MAPK13-IN-1 (Figure 7G; IC50 = 4.63 μM), an inhibitor of MAPK13 (p38-δ); and ARRY-797 (Figure 7H; IC50 = 0.913 μM) in A549-ACE2 cells, a MAPK14 inhibitor. To further probe the dependence of SARS-CoV-2 on p38 pathway members, small interfering RNA (siRNA)-mediated knockdown of MAP2K3, p38-δ (MAPK13), and p38-ɣ (MAPK12) was performed in A549-ACE2 cells, and a significant decrease in SARS-CoV-2 replication was observed for all three, with little to no effect on cell viability (Figure 7I).
In addition, we noted marked regulation of phosphatidylinositol enzyme activities for PIK3CA, PLCB3, and PIKFYVE, suggesting a potential role for the appropriate balance of phosphatidylinositol species. To target this process, apilimod, a small-molecule inhibitor of PIKFYVE, was tested and found to possess strong antiviral activity in two cell lines (Vero E6, IC50 < 0.08 μM; A549-ACE2, IC50 = 0.007 μM), corroborated by a recent study (Ou et al., 2020; Figure 7J). Lastly, we noted pronounced regulation of CDK signaling pathways (Figure 4B) and cell cycle stage (Figure 6I) during viral infection, suggesting that the virus may regulate the cell cycle to enhance viral replication. Accordingly, strong antiviral activity for the CDK inhibitor dinaciclib was observed across two cell lines (Vero E6, IC50 = 0.127 μM; A549-ACE2, IC50 = 0.032 μM) (Figure 7K).