Introduction

The ancestral SARS-CoV-2 virus and its variants of concern (e.g., Alpha B.1.1.7, Beta B.1.351, Gamma P.1, Delta B.1.617.2, Omicron B.1.1.529) or interest (e.g., Lambda C.37) causing coronavirus disease 2019 (COVID-19) have been detected in various organs (e.g., liver, lung, kidney) and bodily fluids (e.g., saliva)1,2, with organ-specific SARS-CoV-2 evolution observed in long COVID cases3. Studies using mass spectrometry (MS)-based proteomics and yeast-two hybrid (Y2H) assays highlight the roles for SARS-CoV-2 proteins in pathogenesis and host factors or processes targeted during infection4,5,6,7,8. Recent findings reveal that virus-host protein-protein interactions (PPIs) differ in variants, aiding immune evasion9. However, published studies focused on cell-based models or peripheral blood mononuclear cells, leaving gaps in understanding how SARS-CoV-2 and its variants remodel host responses and PPIs in various organs and fluids, particularly in saliva, a key site for SARS-CoV-2 infection and transmission2. While repurposed drugs and vaccines have reduced COVID-19 deaths, the need for improved therapeutics or organ-specific treatments continues. Targeting virus-host interface is a key strategy for antiviral development, with peptide-based inhibitors showing promise for their specificity and tolerability10. As new SARS-CoV-2 variants (e.g., KP.3, XEC, JN.1) continue to emerge, developing therapeutic peptides remains a priority. To address these gaps, we used affinity purification (AP) and MS to map PPIs between human proteins and ancestral SARS-CoV-2 proteins, including the spike (S) protein from SARS-CoV-2 and four variants (Alpha B.1.1.7, Beta B.1.351, Delta B.1.617.2, Lambda C.37) across eight cell lines from five mammalian organs and the immune system.

Biochemical fractionation and MS further provided a comparative analysis of human protein assemblies affected by the Alpha, Delta, and Omicron (B.1.1.529) variants compared to ancestral SARS-CoV-2 infection in COVID-19 patient saliva. Analysis of 173 predicted multiprotein complexes (MPCs) from the human-SARS-CoV-2 PPI network revealed the biological significance of the SARS-CoV-2 papain-like protease NSP3 (a non-structural protein), which binds fibrinogen subunits to promote coagulation abnormalities and associates with interferon (IFN)-induced proteins to evade immune responses. Additionally, our deep learning algorithm11 identified three receptor binding ___domain (RBD)-targeting peptides from the S protein that inhibited entry and replication of ancestral and variant strains by blocking S-RBD and ACE2 (angiotensin-converting enzyme 2) interactions in human liver and monkey kidney cells. These peptides restored altered host PPIs in liver cells infected with ancestral or variant SARS-CoV-2. Together, this resource offers a comprehensive view of SARS-CoV-2-human PPIs across organ- and immune-derived cell lines and variants, providing insights into disease mechanisms and pathobiological changes in ancestral SARS-CoV-2 and its variants within human hosts, including the physiological context of COVID-19 patients. Our findings also establish a foundation for developing host-targeting therapies and broad-spectrum antivirals to prepare for future pandemics.

Results

AP-MS reveals organ- and immune cell line-dependent SARS-CoV-2-host PPIs

To survey interactions between human and SARS-CoV-2 proteins across mammalian organ- and immune-derived cell lines, we examined all 29 viral proteins (4 structural, 16 non-structural, and 9 accessory; Fig. 1a). Of these, 26 proteins, a catalytic dead mutant (NSP5 C145A), and enhanced green fluorescent protein (eGFP, as a negative control) were expressed using a mammalian vector with a 2x Strep-epitope tag at the C-terminus5 for AP-MS. Strep-tagged constructs for the SARS-CoV-2 S structural protein, two non-structural proteins (NSP3 and NSP16), and S proteins of Alpha, Beta, Delta, and Lambda variants, previously unavailable in Addgene plasmid repository or analyzed5, were cloned with a Gateway cassette carrying a versatile epitope (VA) tag (3× FLAG, 6× His, 2× Strep12) at the C-terminus (Supplementary Note 1). After confirming affinity-tagged viral protein expression by immunoblotting with an anti-Strep antibody (Supplementary Fig. 1a), ancestral SARS-CoV-2 proteins with 2x Strep or VA-tags were stably expressed in eight mammalian cell lines representing five organs (Kidney: HEK293, Vero E6/81; Liver: Huh-7; Intestinal: Caco-2; Lung: HULEC-5a; Brain: N9 microglia) and the immune system (inactivated M0 and activated M1 macrophages from THP-1). Cultured cells expressing these viral proteins underwent Strep affinity purification in at least two biological replicates. Purified samples were trypsinized and analyzed by MS, resulting in 639 AP-MS analyses (Fig. 1b). High reproducibility (r = 0.94) of MS2 protein spectral counts between replicate AP-MS experiments for each cell line (Supplementary Fig. 1b) validated data quality.

Fig. 1: Organ- and immune cell-specific SARS-CoV-2-host interactions by AP-MS.
figure 1

a SARS-CoV-2 genome showing ORF1a/1b regions encoding 4 structural proteins, 9 accessory, and 16 non-structural proteins (NSPs) used for affinity purification. b AP-MS workflow applied to 29 ancestral and 4 spike (S) variants (Alpha, Beta, Delta, Lambda) across organ- and immune-derived human cell lines to build viral-host PPI networks; 3 non-human cell lines used for validation. MS, mass spectrometry; ML, machine learning. c Performance of PPI predictions using a random forest (RF) classifier, assessed by sensitivity vs. 1-specificity (area-under-the-curve, AUC) against a curated reference PPI set from BioGRID. High-confidence SARS-CoV-2-host PPI detection in one or more human cell lines (d), overlap with BioGRID (e), validation in non-human cell lines with RF scores below (≤0.8) or above (>0.8) the cut-off (f), and subcellular co-localization in the same cellular compartment (g). SARS-CoV-2 localization data is from published studies15,61 (with additional references reviewed in Grand, 202361), while the human protein data is sourced from the Human Protein Atlas62. h Enrichment of shared biological processes among SARS-CoV-2-interacting human proteins across multiple human cell lines. Top enriched terms in two or more cell lines are shown. P-values from a one-sided hypergeometric test and adjusted for FDR ≤ 5e−2 with the Benjamini-Hochberg (BH) method. i Coverage (%) and significance (P = 1e−4, red dotted line) of SARS-CoV-2-interacting human proteins overlapping with those from other pathogens (Supplementary Data 5). P-values from one-sided hypergeometric test. KHSV, Kaposi’s sarcoma herpesvirus; SCoV1, Severe acute respiratory syndrome coronavirus 1; MERS, Middle East Respiratory Syndrome; WNV, West Nile virus; HIV, Human immunodeficiency virus; HCV, Hepatitis C virus; HPV, Human papillomavirus; Mtb, Mycobacterium tuberculosis. j Human proteins interacting with SARS-CoV-2 baits shared or not shared with the indicated viruses. Source data are provided as a Source Data file.

To define SARS-CoV-2-human PPIs, MSspectra from Strep-tagged viral bait purifications were mapped to SARS-CoV-2 and mammalian host protein sequences using SEQUEST/STATQUEST, MS-GF + , and MaxQuant search algorithms to improve peptide identification and protein coverage (Supplementary Note 2). For each cell line, outputs from these search engines were scored independently using CompPASS-Plus (Comparative Proteomic Analysis Software Suite13) and MiST (Mass spectrometry interaction STatistics14) scoring procedures to assign confidence scores for viral bait-human prey associations (Fig. 1b). To enhance prediction accuracy, CompPASS-Plus and MiST scores were combined into a single probabilistic score via a supervised random forest (RF) model trained on positive SARS-CoV-2-human PPIs from BioGRID database and a negative dataset of non-interacting protein pairs (Supplementary Note 3). The model’s performance for each cell line, evaluated through 10-fold cross-validation against a BioGRID reference set of curated PPIs withheld during training, achieved an average false discovery rate (FDR) of 4.4e−3 with an RF score threshold > 0.8 that recovered most reference PPIs as shown by area-under-the-curve (AUC) analysis (Fig. 1c and Supplementary Fig. 1c). This yielded a high-confidence network of 4158 interactions between 30 SARS-CoV-2 (including the NSP5-C145A mutant) and 1774 human proteins across five human cell lines (Fig. 1d and Supplementary Fig. 1d; Supplementary Data 1). These interactions were cross-validated with human orthologs of interacting proteins from non-human cell lines, such as N9 microglia (mouse brain) and Vero E6/81 (African Green monkey kidney).

The quality of AP-MS-derived interactions was supported by several lines of evidence. First, most (75%, 3136 of 4158) SARS-CoV-2-human PPIs were detected across multiple human cell lines, while a quarter (25%, 1022 of 4158) were specific to a single organ-derived cell line (Fig. 1d and Supplementary Fig. 1e; Supplementary Data 2). Over half (54%, 556 of 1022) of these were found in Huh-7 liver cells (Supplementary Fig. 1f). Although detecting PPIs across multiple cell lines reinforces confidence (Fig. 1d), we were interested in the organ-specific interactions identified. To assess whether this trend persists in the unthresholded ( <0.8) dataset, we conducted a comparative analysis and observed a similar pattern (Fig. 1d). This confirmed that the majority (64-75%) of PPIs occur across multiple cell lines, while a smaller fraction (25-36%) remains organ-specific, regardless of whether the dataset was thresholded or unthresholded.

Second, over half (56%, 2296 of 4101) of SARS-CoV-2-human PPIs, excluding the NSP5-C145A mutant, were previously reported in BioGRID, while the remaining (44%, 1805 of 4101) have not been reported earlier (Fig. 1e). Among these, less than one-third (31%, 557 of 1805) were confirmed in multiple human cell lines in the unthresholded data (Supplementary Data 1). Additionally, more than a dozen were validated by co-immunoprecipitation (co-IP) in specific cell lines (Supplementary Fig. 1g) using protein-specific antibodies, emphasizing the biological relevance of our predicted PPIs. The detection of these newly identified PPIs may have been influenced by the choice of cell lines, methodologies, and the reliance on a single cell line in earlier SARS-CoV-2 interactome studies. Third, in the thresholded dataset, less than half (48%, 2002 of 4158) of SARS-CoV-2 interactors were identified in at least one of the three holdout non-human cell lines (Fig. 1f). However, this overlap increased by an average of 4.4-fold when using the unthresholded data, indicating that the chosen cut-off is highly stringent. Fourth, in addition to identifying SARS-CoV-2 interactors that co-localize within the same cellular compartment (Fig. 1g), 166 host factors interacting with SARS-CoV-2 proteins from human cell lines were detected in genetic screens15,16,17 as enhancers or inhibitors of SARS-CoV-2 or common cold coronavirus infections (Supplementary Fig. 1h). For example, Rab GTPases (RAB2A/7 A/ 10), interacting with NSP6/7 in HULEC-5a lung endothelial cells, were identified as key host-dependency factors for SARS-CoV-2 progression18. Next, protein abundance changes during SARS-CoV-2 infection19 was examined by calculating correlations between the abundance of viral proteins and their human interaction partners across human cell lines. Interacting partners of viral bait proteins exhibited stronger correlations in protein level changes during SARS-CoV-2 infection compared to non-interacting protein pairs (1.0e−4  <  P  >  1.9e−2; Supplementary Fig. 1i). Additionally, these SARS-CoV-2-human interacting proteins were significantly (P  =  5e−2) enriched in tissues such as salivary glands, pancreas, heart, and lungs (Supplementary Fig. 1j; Supplementary Data 3) based on protein abundance across human tissues20, suggesting that PPIs relevant to target tissue and infection are context-dependent.

Among SARS-CoV-2 binding partners, we identified significant enrichment (FDR ≤ 5e−2) of shared bioprocesses across human cell lines derived from various organs and immune system (Fig. 1h; Supplementary Data 4). These include host processes such as endoplasmic reticulum (ER)-Golgi vesicle transport, mRNA translation, and the unfolded protein response, which SARS-CoV-2 exploits for infection21,22,23. Comparing SARS-CoV-2 host interaction partners with other pathogens revealed the greatest similarity with SARS-CoV-1 (SCoV1; Fig. 1i; Supplementary Data 5), reflecting their clinical and genetic similarities. SARS-CoV-2 also shared more PPIs with SCoV1 than with Middle East Respiratory Syndrome Coronavirus (MERS-CoV; Fig. 1j). Notably, human proteins interacting with SARS-CoV-2 NSP7/13 and ORF7A/9B/9 C are also targeted by SCoV1, similar to the shared interactors between MERS-CoV and SARS-CoV-2 NSP2/4/7/8 (Supplementary Fig. 1k), highlighting these PPIs as potential antiviral targets. Together, these findings confirm that our AP-MS network, derived from diverse organ and immune cell lines, significantly expands the number of reported PPIs, offering avenues for therapeutic development.

Ancestral and variant SARS-CoV-2 S-host PPIs vary in organ- and immune-derived cells

Since enhanced fitness of circulating SARS-CoV-2 variants is linked to mutations in the S protein (the primary target of COVID-19 vaccines), it is critical to comprehend how mutations in the S protein of the variants, compared to ancestral SARS-CoV-2 influence virus-host protein assemblies. Therefore, we examined the S regions of three variants of concern (Alpha, Beta, Delta) and one variant of interest (Lambda), which significantly affect virulence, transmission, and COVID-19 severity24. Using PCR mutagenesis, we created 42 hotspot non-synonymous mutations or deletions in the S protein of Alpha (9), Beta (9), Delta (11), and Lambda (13) variants (Fig. 2a; Supplementary Data 6; Supplementary Note 1) as opposed to single mutations that do not mimic all mutated residues. Sequence-verified plasmids were then C-terminally tagged with a VA epitope containing 2x Strep using Gateway cloning and stably expressed in HEK293 cells through lentiviral infection (Supplementary Note 1). The specificity of expressed variant and wildtype S proteins was confirmed using an anti-Strep antibody (Supplementary Fig. 1a). Strep-tagged S proteins were then affinity-purified from selected human cell lines for MS analysis.

Fig. 2: Comparative analysis on SARS-CoV-2 ancestral and variant S protein interactomes.
figure 2

a Schematic showing substitutions/deletions in S protein variants analyzed by AP-MS. Human proteins (n = 668) interacting with ancestral and variant S proteins across viral strains (b, c) and human cell lines (d). b shows non-redundant interactions independent of cell type, while panel c includes redundant interactions accounting for cell types. e Localization of human prey proteins (from the Human Protein Atlas62) with ancestral or variant S bait proteins (from this study and a published report15) in the same subcellular compartment (left). Workflow (top right) for subcellular localization of VA-tagged (2x Strep) S proteins in HEK293 cells. Representative fluorescence images (bottom right; n = 5) showing co-localization of VA-tagged ancestral or variant S proteins (anti-Strep, green) with ER tracker (blue). Additional markers is shown in Supplementary Fig. 2a. Scale bar, 20 μm. Bar plot (far right) showing consistent S protein expression in HEK293 cells (normalized to GAPDH loading control), based on immunoblot (IB) quantification from Supplementary Fig. 1a. Band intensity measurements (n = 10) were obtained from representative IB from the same experiment. Data are shown as mean ± standard error of the mean (SEM). f Heatmap of unique/shared human prey proteins among ancestral and variant S baits in human cell lines, filtered by random forest (RF) and first principal component analysis (PCA) scores > 0.8. g Enrichment of annotated complexes/processes (CORUM, Gene Ontology, Reactome, WikiPathways) for S-interacting host proteins, including the components of RNP, spliceosomal, Drosha-DGCR8, and small nuclear RNP complexes. P-values from one-sided hypergeometric test and adjusted for FDR ≤ 5e−2 with the BH method. h Pie chart of host factors interacting with ancestral and/or variant S proteins. i Comparison of ancestral or variant S-host PPIs unique to this study vs. a recent report9. Source data are provided as a Source Data file.

Using the same approach as for ancestral SARS-CoV-2 (Fig. 1b), we employed CompPASS-Plus and MiST algorithms to score interactions for S protein variants. Due to limited training data for the variants, outputs from both scoring metrics were merged and analyzed using first principal component analysis (PCA14; Supplementary Note 3). PCA scores for individual cell lines were averaged into a composite score and subjected to a stringent threshold cut-off > 0.8 to define high-confidence interactions. The resulting network identified 668 host proteins, with a quarter (25%, 165 of 668) shared across two or more SARS-CoV-2 S strains when thresholded, and capturing an additional quarter (28%, 190 of 668) of human proteins interacting with the S viral protein when unthresholded (Fig. 2b). The remaining (47%, 313 of 668) were distributed differently, with a fifth (21%, 142) constituting the ancestral SARS-CoV-2 S protein, and a smaller fraction (5-8%, 33-54) were exclusive to each variant (Fig. 2b; Supplementary Data 7). Further analysis identified 65 human proteins interacting with one or more SARS-CoV-2 variants across 2-4 human cell lines (Supplementary Data 7). Among these, ENO2, a glycolysis marker linked to increased SARS-CoV-2 load and inflammation25, and PARD6G, a cellular polarity regulator exploited by SARS-CoV-2 to evade immune responses26, were consistently targeted by one or more variants in four (Caco-2, HEK293, HULEC-5A, THP1-M1) of six human cell lines tested. On average, the ancestral SARS-CoV-2 S protein interacted with 1.4-times more host factors than variants (Fig. 2c), with interactions most frequent in lung endothelial cells (HULEC-5a; Fig. 2d). Examining the localization of S protein variants in HEK293 cells revealed their presence in the ER, Golgi, and plasma membrane, similar to the ancestral S protein15, but sometimes to a lesser extent (Fig. 2e and Supplementary Fig. 2a). While the ancestral S protein showed consistent localization across cell types (data shown for HEK293), variant S proteins exhibited mutation-driven localization changes despite unchanged S protein abundance (Fig. 2e) in HEK293 cells expressing Strep-tagged S protein from each SARS-CoV-2 strain.

The localization of the ancestral viral S protein aligns with that of its human binding partners, but mutations in variant S proteins affecting localization may influence virus-host PPIs (Fig. 2e; Supplementary Data 8), though do not fully account for the PPI differences observed across cell types. Consistently, enriched ( | Z-score | ≥ 1.96; P  ≤  0.05) human interacting proteins associated with each viral S protein differ significantly between compartments and across variants and the ancestral strain (Supplementary Fig. 2b; Supplementary Data 9). Human host factors also showed reconfiguration, with some proteins uniquely bound and others shared among variant and ancestral S proteins (Fig. 2f; Supplementary Data 10). For example, host factors interacting with S proteins were enriched in cohesive subunits linked to annotated MPCs or biological terms from CORUM, Reactome, Gene Ontology (GO), and WikiPathways (Fig. 2g; Supplementary Data 11). These included S proteins interacting with the components of ribonucleoprotein (RNP), spliceosomal, Drosha-DGCR8, and small nuclear RNP complexes, with co-IP confirming that different S variants bind to unique host partners in specific human cell lines (Supplementary Fig. 2c). This suggests that, like SARS-CoV-227, variant S proteins can target the same cellular pathways, yet they exploit these host factors through specific interactions, and crucially, distinct protein partners, facilitating replication and functioning as viral RNA sensors in innate immunity27.

A small subset (7%, 48 of 668) of host proteins interacting with all five SARS-CoV-2 S strains was primarily found in HULEC-5a lung endothelial cells (Fig. 2h and Supplementary Fig. 2d). These proteins were enriched (FDR ≤ 5e−2) in functions related to the mitochondrial respiratory chain, cytokine production, and cytoskeleton organization (Supplementary Fig. 2e), suggesting they are critical host ‘hub’ proteins potentially exploited by the virus for replication, transmission, and immune evasion. A sizable portion (40%, 267 of 668) of PPIs was both variant- and cell type-specific, with variability observed in shared PPIs across variants (Supplementary Fig. 2f). Inactivated THP-1 M0 macrophages and HEK293 cells showed more shared interactions among variants, while HULEC-5a, Caco-2, and Huh-7 cells exhibited fewer (Supplementary Fig. 2g). Notably, over three-fourths (79%, 526 of 668) of interactions between ancestral or variant S proteins and host factors across various human cell lines (Fig. 2i; Supplementary Data 12) were previously unreported9, suggesting cell type-specific PPIs may underlie these differences.

Host responses remodeled by ancestral and variant infections in saliva

Prompted by notable changes in virus-host PPIs specific to organ-specific cell lines or shared among variants compared to the ancestral virus, we examined how these reconfigured protein assemblies manifest in the saliva of COVID-19 patients infected with SARS-CoV-2 and its variants. This analysis helps characterize viral-host PPIs in the native physiological state during viral infection, particularly given SARS-CoV-2’s ease of transmission through saliva2. Since co-fractionation MS (CF-MS) can determine native MPC membership in lysates from human cells28 and model organisms29, we performed biochemical fractionation with size-exclusion chromatography (SEC) and MS profiling to separate native macromolecules in soluble protein extracts prepared from the saliva of healthy individuals and patients infected with ancestral or variant (Alpha, Delta, Omicron) SARS-CoV-2 strains (Fig. 3a). A total of 1128 biochemical fractions from 12 fractionation experiments, including replicates, were analyzed by MS, with spectra processed using multiple search algorithms, as in AP-MS, identifying 9247 human proteins (Supplementary Data 13). Protein elution patterns and the average correlation of proteins based on MS-detected spectral counts (r = 0.7) were reproducible across replicate experiments (Supplementary Fig. 3a, b). PCA analysis (Supplementary Fig. 3c) revealed clear separation of samples positive and negative for SARS-CoV-2 or its variants in multidimensional space.

Fig. 3: Viral-host PPIs and host responses in saliva from SARS-CoV-2 ancestral vs. variant infections.
figure 3

a CF-MS scoring pipeline identified reconfigured host-viral or host-host PPIs from 1128 SEC (size-exclusion chromatography) fractions using soluble protein extracts from saliva of healthy individuals (n = 3) and patients infected with ancestral (n = 3), Alpha (n = 2), Delta (n = 2), and Omicron (n = 2) variants. b Overlap of high-confidence (HC) host-viral PPIs detected via AP-MS in human cell lines and CF-MS in ancestral SARS-CoV-2-infected patient saliva. Host-viral PPIs also include those detected as high-confidence in the AP-MS dataset but below the CF-MS threshold, and vice versa. c Distribution of protein abundance (MS2 spectral counts) and viral-host PPIs across SARS-CoV-2 proteins detected by CF-MS. d Viral-host PPIs identified in SARS-CoV-2-infected patient saliva via CF-MS, and supported by AP-MS in human cell lines, analyzed with and without (i.e., All) RF score thresholds. e Scatterplot (left) showing Pearson correlation coefficients (PCCs) for co-eluting human protein pairs in saliva from SARS-CoV-2-infected vs. healthy or variant-infected patients at various P-value thresholds based on PCC score differences and two-sided Z-score transformation. Histogram (right) displays significantly differential (DF) human protein pairs at P ≤ 1.0e−5 and BH-adjusted FDR ≤ 5.0e−4 in the indicated comparisons. f Heatmap of positive and negative DF correlations in co-eluted human protein pairs from SARS-CoV-2- vs. healthy or variant-infected saliva. g Bar plot of positive and negative DF correlations of human protein pairs between SARS-CoV-2- and variant-infected individuals. h Specificity of human protein interaction pairs across all three variants. i GO biological term enrichment for human proteins interacting more or less specifically with variants. P-value from a one-sided Fisher exact test, and adjusted for FDR ≤ 5e−2 with the BH method. j Correlation profiles (autocorrelation) of co-eluted human protein interactions between SARS-CoV-2 and variants, categorized as high (>0.25) or low (<0.25), with zoom-in examples. Source data are provided as a Source Data file.

Stably associated proteins within a MPC are expected to co-elute, and their associations can be assessed by pairwise similarity of their co-elution profiles30. Thus, we computed nine correlation metrics (i.e., Apex, Bayes correlation, cosine correlation, Jaccard index, mutual information, Pearson correlation coefficient (PCC), PCC adjusted for Poisson noise, P-value derived from PCC, weighted cross correlation; Supplementary Note 3) for each search engine, capturing distinct features of profile similarity in SARS-CoV-2 patients (Fig. 3a). After combining the scores from all metrics, protein pairs were input into a RF classifier, trained using positive PPIs from BioGRID, while non-interacting protein pairs from the training set were used as the negative dataset. Predicted PPIs from each search algorithm were combined, and their RF scores averaged to generate a unified PPI network. The predictive accuracy of these measurements, assessed using 10-fold cross-validation against a curated subset of BioGRID PPIs to train the RF model in AP-MS, achieved high sensitivity and specificity with an AUC of 0.90 (Supplementary Fig. 3d).

The final network comprises 2019 interactions between 17 SARS-CoV-2 and 880 host proteins, with each interaction having an RF score > 0.75 in co-elution profiles (Fig. 3b and Supplementary Fig. 3d; Supplementary Data 14). While some viral proteins went undetected in the saliva of SARS-CoV-2-infected patients, over half (56%, 17 of 30) of the expressed proteins showed no bias in protein abundance relative to their identified human interacting partners by MS (Fig. 3c). Our analysis also revealed only 80 overlapping high-confidence PPIs between the two methods. However, an additional 519 PPIs were found in the AP-MS high-confidence dataset but fell below the CF-MS threshold, or vice versa (Fig. 3b). Regardless of the methods used, the overlap of high-confidence viral-host PPIs between saliva and individual cell lines remained modest (Fig. 3d). This aligns with the limited detection of host factors observed between Y2H and AP-MS methods4, suggesting that the modest overlap may stem from underestimations in predicted interaction scores due to constraints in the reference dataset used for training. Additionally, methodological differences, such as higher viral protein expression during natural infections compared to affinity-tagged protein studies, may contribute to these discrepancies.

To explore the divergence in host responses to ancestral vs. variant viruses, we built a differential network from the saliva of infected patients by comparing the ancestral virus to healthy subjects and the ancestral virus to each SARS-CoV-2 variant. This was achieved by calculating co-elution profile similarity using PCC. For each pair of co-eluting human proteins, a P-value was assigned based on differences in correlation scores across the comparative ‘input’ samples. By applying differential score thresholds with a FDR ≤ 5e−4 (Fig. 3e), we identified 15,659 significant differential human PPIs when comparing the ancestral virus to healthy subjects, and to variants (Fig. 3f and Supplementary Fig. 3e; Supplementary Data 15). Of these, over one-third of human protein pairs showed either positive (38%, 5968 of 15,659) or negative (35%, 5535 of 15,659) differential correlations with at least one variant (Fig. 3g). A smaller subset were more (219 of 15,659) or less (129 of 15,659) specific across all variants (Fig. 3h; Supplementary Data 16).

Enrichment analysis (FDR ≤ 5e−2) revealed co-eluted human proteins from the haptoglobin-hemoglobin complex and/or the immunoglobulin kappa variable cluster as the most correlated across variants compared to the ancestral SARS-CoV-2 strain (Fig. 3i and Supplementary Fig. 3f), consistent with observed changes in these protein levels in COVID-19 patients31,32. In contrast, the least correlated processes among variants involved host interacting proteins associated with Ca2+ signaling and other cellular functions (Fig. 3i), indicating distinct cellular responses between variants and the ancestral virus. Further analysis of differential networks revealed 224 human proteins in a variant exhibiting varying degree of autocorrelation with the ancestral virus (Fig. 3j). For instance, fewer than half of these proteins (44%, 99 of 224), including those involved in epigenetics and immunoglobulin heavy or kappa variable regions, showed high autocorrelation ( >0.25) with both the ancestral virus and all three variants (Supplementary Data 17). Conversely, fewer than one-tenth (8%; 17 of 224) of proteins, such as copper-containing ceruloplasmin and cardiac titin, exhibited low autocorrelation ( < 0.25), showing marked differences between the ancestral virus and its variants (Fig. 3j). These findings suggest that ancestral and variant SARS-CoV-2 strains may elicit similar or distinct host responses during infection. Supporting this, large variations in viral protein abundance were observed in patients’ saliva during infections with variants compared to the ancestral virus (Supplementary Fig. 3g). For instance, ORF3A levels were higher in Delta than in Omicron, while the innate immune antagonists ORF6 and N were elevated in Omicron compared to other variants or the ancestral virus, aligning with reported Omicron infection patterns9. A comparison of immune, epithelial, myeloid, and connective tissue compartments in host salivary glands revealed that each SARS-CoV-2 strain induced 143 unique host interacting protein pairs during infection (Supplementary Fig. 3h, i; Supplementary Data 18). Among these, 31 host proteins from compartments such as immune (Tc subtype 2), epithelial (duct, mucous acini, myoepithelia), myeloid (M1/M2-macrophages), and connective (fibroblasts, arterial, venules, capillaries) tissues showed stronger responses during variant infections compared to the ancestral SARS-CoV-2. Notably, Omicron responses closely resembled Alpha strain, highlighting differences in how these variants regulate host inflammatory and immune responses.

Merged network links SARS-CoV-2 proteins to neurological and druggable host factors

To understand the shared and unique interactions of host proteins with each SARS-CoV-2 viral proteins in infected human cell lines and patient saliva, we integrated AP-MS and CF-MS datasets into a ‘merged’ network, identifying 6097 viral-host PPIs (Supplementary Data 19). The merged network exhibited a power-law degree distribution with fewer degrees of separation (Supplementary Fig. 4a), indicative of a scale-free network. Integration of CF-MS data with AP-MS improved overlap with BioGRID interactors from 2296 to 2512 (Fig. 1e and Supplementary Data 20), confirming the robustness of our findings across organ cell lines and SARS-CoV-2-infected patient saliva. Analysis of host interactions with each viral protein revealed that, consistent with BioGRID data, SARS-CoV-2 proteins NSP6, S, and ORF3A/7 A ranked in the top 30% for interactions with human proteins (Supplementary Fig. 4b), underscoring their critical roles in the viral life cycle. Beyond co-IP (Supplementary Fig. 1g), we independently confirmed one-tenth (10%, 637 of 6097) of the SARS-CoV-2-human PPIs from the merged network in blood plasma from four SARS-CoV-2-infected individuals using CF-MS (Supplementary Fig. 4c and Supplementary Data 19). Among these, 139 PPIs were also detected in various human cell lines or in saliva from SARS-CoV-2-infected patients, strengthening the clinical relevance of these interactions in a physiological context and aiding in the refinement of therapeutic target selection.

Correlated profiles of stably associated components can predict functional links between proteins30. We therefore calculated correlations among all SARS-CoV-2 viral protein pairs based on their interactions with human proteins in the merged network, identifying several pairs with positively correlated profiles. Notably, strong correlations (average r = 0.6) were observed, with high similarity in the host proteins shared between NSP5/11/12/15 and nucleocapsid (N) proteins, between NSP4/6 and ORF7A (average r = 0.7), and between NSP7/10, S and ORF9C (average r = 0.7; Fig. 4a, b and Supplementary Data 21). These correlations align with intraviral PPIs of SARS-CoV-2 identified through compartmentalization-aided interaction screening in cells33 and Y2H6 methods. Analysis of host proteins interacting with these correlated viral protein pairs revealed significant enrichment (FDR ≤ 5e−2) in bioprocesses like glycolysis, respiration, and ER-to-Golgi vesicle transport, among others (Fig. 4c). These findings highlight that many SARS-CoV-2 proteins target overlapping human proteins or those involved in shared bioprocesses4,15.

Fig. 4: Merged network links SARS-CoV-2 to neurological and druggable host factors.
figure 4

a Spearman correlation between SARS-CoV-2 viral protein pairs based on their shared human proteins in the merged (AP-MS and CF-MS) network. b Sankey plot showing shared host proteins among the correlated SARS-CoV-2 viral proteins from a. c GO biological term enrichment for host factors shared among correlated viral proteins in b. P-values from one-sided Fisher’s exact test and adjusted for FDR ≤ 5e−2 with the BH method. d Analysis of SARS-CoV-2-host PPIs (n = 1279) in M0 and M1 macrophages and N9 mouse microglia (human orthologs) vs. SARS-CoV-2 patient saliva. e Host-viral PPIs detected in M1 macrophages and/or microglia but absent in SARS-CoV-2 patient saliva are linked to neuroinflammation and neurological disorders. f Predicted MPCs (n = 173) showing SARS-CoV-2 (yellow rhombus) and human (gray circle) proteins, with potential drug targets highlighted in red circle or rhombus. Viral-host PPIs are represented by green lines. Zoomed-in views of selected MPCs illustrate both established (dashed lines) and new (thick lines) interactions. g Human MPC subunits interacting with SARS-CoV-2 proteins are targeted by at least one of 7626 drugs from DrugBank. h Heatmap of 13 drugs targeting 10 or more human host factors across 52 MPCs, with each MPC subunit targeted by at least two drugs. Source data are provided as a Source Data file.

In COVID-19 or long COVID patients, SARS-CoV-2 infection can impair neurological and cognitive issues by targeting microglia, which release M1-like proinflammatory cytokines, causing inflammation and neurotoxicity34,35. Another critical immune component affected are macrophages, where functional changes, such as cytokine storms are harmful to the host36. To understand how SARS-CoV-2 alters protein assemblies in these immune cells, we analyzed 1279 host interacting proteins from 30 Strep-tagged SARS-CoV-2 viral proteins (including the NSP5-C145A mutant) expressed in inactivated (M0) or activated (M1) macrophages and microglia, compared to SARS-CoV-2 patient saliva (Supplementary Data 22). This analysis identified proteins specific to macrophages (M0, M1), microglia, or saliva, as well as those shared by M1 macrophages and/or microglia but not in SARS-CoV-2-infected patient saliva (Fig. 4d and Supplementary Fig. 4d). Notably, 34 host factors linked to M1 macrophages and microglia, involved in neuroinflammation (e.g., ACSL4, NEK7, ATP13A1) or neurological disorders (e.g., AIFM1, eEF1A2, EIF4G1, CCT2, MFN2, TUBB4A, UBQLN4, VAPB) were absent in SARS-CoV-2 patient saliva (Fig. 4e). Additionally, 6-28 SARS-CoV-2 viral proteins interacted with 19 host factors in M1 macrophages and/or microglia, including proteins involved in RNA metabolism (DDX3Y), translation (EIF4A2, EEF1A2), cytoskeleton (actin, ß-Tubulin), cell adhesion (RAP1A), and hub functions (HSPA2, NPM1) (Supplementary Fig. 4e, f). These interaction patterns suggest SARS-CoV-2 are likely to leverage these host factors to enhance infection, suppress IFNs to inhibit host innate immunity, and exacerbate disease severity.

To define MPC membership, we used the ClusterONE clustering method to partition the merged network, identifying 173 distinct MPCs with 29 viral and 2002 unique human proteins (Fig. 4f and Supplementary Data 23). ClusterONE parameters were optimized using a composite score (maximal matching ratio, overlap, accuracy) by comparing MPCs to CORUM protein complex subunits interacting with SARS-CoV-2 proteins in BioGRID, selecting the highest-scoring parameters (Supplementary Fig. 4g). Notably, two-thirds (65%, 113 of 173) of the MPCs contain 15 or fewer subunits (Supplementary Fig. 4h). Of these, 166 include 1273 BioGRID-documented host-viral PPIs, while seven MPCs remain unreported (Supplementary Fig. 4i), offering avenues for SARS-CoV-2 biology exploration. To identify druggable host factors for drug repurposing in COVID-19, we analyzed MPC components against 7626 FDA-approved, experimental, and investigational drugs in DrugBank. Notably, one-fifth (24%, 490 of 2002) of human MPC subunits interacting with viral proteins are targeted by at least one drug (Fig. 4g and Supplementary Data 24). For example, fostamatinib, a tyrosine kinase inhibitor effective in phase II (NCT04579393) and III (NCT04629703) COVID-19 trials, targets host kinase factors associated with NSPs, ORFs, M, and N proteins (Supplementary Fig. 4j). Using co-IP, fostamatinib was shown to disrupt NSP3 interaction with MAP2K2 (mitogen-activated protein kinase kinase 2) in Caco-2, and MAST4 (microtubule-associated serine/threonine kinase family member 4) in HEK293 cells. Similarly, MARK1 (microtubule affinity regulating kinase 1), which interacts with ORF9B in M1 macrophages, was disrupted by fostamatinib but not by vehicle (Supplementary Fig. 4k). These results suggest that fostamatinib disrupts SARS-CoV-2 protein-host kinase binding to bolster host immunity and support its therapeutic potential in SARS-CoV-2 infection.

Conversely, we identified 55 drugs targeting four or more distinct human host factors within 98 MPCs, with 13 targeting over 10 different human proteins (Fig. 4h and Supplementary Data 25). For example, fostamatinib targets Src (intracellular protein-tyrosine kinase) association with NSP11, and BTK (Bruton’s tyrosine kinase) interaction with NSP9 (Supplementary Fig. 4j), consistent with prior studies highlighting these as key anti-SARS-CoV-2 and antiviral targets37,38. Among drugs (binimetinib, bosutinib, fostamatinib, selumetinib, trametinib) targeting MEK2/ MAP2K2 interactions with NSP3, selumetinib, trametinib, and binimetinib act as MEK inhibitors. Selumetinib reduces SARS-CoV-2-induced lung damage39, and trametinib blocks influenza A virus propagation and cytokine expression40. These findings suggest host-directed drug targets from MPCs offer a viable strategy for prioritizing therapeutics for SARS-CoV-2 treatment.

Viral-human MPCs from the merged network reveal new insights into SARS-CoV-2 biology

From the merged network, we explored two overlooked virus-host MPCs, and discussed below in terms of the molecular mechanisms involving previously unrecognized roles of viral proteins interacting with host machinery to reveal functional insights and host-targeting mechanisms.

NSP3-fibrinogen complex confers coagulation defects

We identified unexpected interactions within MPC 172 involving NSP3/10, ORF7A, and S proteins associated with the fibrinogen complex, comprising fibrinogen alpha (FGA), beta (FGB), and gamma (FGG), in SARS-CoV-2 patient saliva (Fig. 4f). Specifically, NSP3’s interaction with fibrinogen subunits, essential for blood clotting and thrombus formation41, led us to hypothesize that secreted NSP3 may disrupt normal coagulation by promoting abnormal clot formation. In-silico analysis revealed that NSP3 contains a secretion signal peptide (Fig. 5a), N- or O-link glycosylation sites (Fig. 5b), and four transmembrane helices (Fig. 5c) necessary for secretion. Consistent with these predictions, NSP3 was secreted in HepG2 culture supernatants (Fig. 5d), similar to the ORF8 SARS-CoV-2 protein42. Since fibrinogen is mainly produced and secreted into the bloodstream by liver hepatocytes, we investigated NSP3’s binding to fibrinogen in HepG2 human hepatoma cells. Co-IP confirmed NSP3 interaction with FGA in these fibrinogen-expressing cells, but not in non-fibrinogen-expressing HEK293 cells (Fig. 5e). This interaction was further validated independently by co-IP in blood plasma from SARS-CoV-2-infected patients and by immunofluorescence, which showed NSP3 co-localizing with fibrinogen in SARS-CoV-2-infected Huh-7 cells (Fig. 5f, g). To examine whether secreted NSP3 with protease activity might impair fibrin production and contribute to coagulation defects, we performed an optical density-based coagulation kinetic assay43. Incubating varying concentrations of purified NSP3 with recombinant fibrinogen showed that NSP3 impaired fibrin generation in a time- and concentration-dependent manner, albeit less effectively than thrombin, which served as a positive control (Fig. 5h). Elevated fibrinogen levels were also detected in the supernatants of SARS-CoV-2-infected Huh-7 cells (Fig. 5i) and in the saliva of individuals infected with ancestral and variant SARS-CoV-2 strains (Fig. 5j), aligning with increased fibrinogen levels reported in severe COVID-19 cases44. Notably, NSP3 when transfected in HepG2 cells contributed to elevated fibrinogen levels (Fig. 5k), reinforcing its dependence on fibrinogen. Furthermore, pro-inflammatory cytokines (IL-6, CCL2) were elevated in M1 macrophages (Fig. 5l), correlating with severe COVID-19 pathology45. These findings support a model in which secreted NSP3 interacts with fibrinogen subunits, triggering inflammatory cytokine responses and contributing to coagulation abnormalities in severe COVID-19 patients.

Fig. 5: NSP3 induces coagulation defects via fibrinogen.
figure 5

a NSP3 with signal peptide, processed by Sec translocon and cleaved by Signal Peptidase I (SPI); SPI sequence in red, with cleavage site and non-signal regions indicated. b Glycosylation sites (score > 0.5) are mapped by amino acid (AA) position based on NetNGlyc 1.0 and NetOGlyc 4.0. c TMHMM predicts four transmembrane regions. d IB of whole cell extracts (WCEs) and supernatants from NSP3-V5 HepG2 cells using anti-V5 antibody for NSP3, with FGA and GAPDH (1:1,000) as controls. e WCEs and immunoprecipitates (IPs) from NSP3-V5 HepG2 and HEK293 cells were probed with FGA, NSP3, or GAPDH antibodies, normalized to input FGA. f FGA IPs from COVID-19 (n = 4) and control (n = 2) plasma were probed with NSP3 and FGA antibodies. g Representative micrographs show (n = 7-10) NSP3 co-localization with FGA in SARS-CoV-2-infected but not mock Huh-7 cells; nuclei in blue. Scale bar, 10 μm. FGA fluorescence in infected vs. mock cells (n = 100, P = 2.1e−73, two-tailed paired t-test). h Coagulation was measured at 420 nm over time using varying NSP3 with fibrinogen, with thrombin-treated and untreated fibrinogen as controls. i Fibrinogen expression in indicated Huh-7 cells was measured by ELISA over time (n = 5 technical replicates; 3.2e−9  ≤ P  ≥ 9.7e−2, two-sided paired t-test). j IB of saliva WCEs using FGA antibody; mean FGA levels (n = 10) from respective IB with band intensities normalized to GAPDH (1.1e−22 ≤ P ≥ 2.9 e−2, one-tailed paired t-test). k FGA levels (P  =  2.5e−19, two-tailed paired t-test) in NSP3-V5 vs. empty vector HepG2 cells. Quantification (n = 10) based on WCEs from panel e using FGA antibody and GAPDH as LC. l IL-6 and CCL2 mRNA levels were measured in NSP3-transfected and untransfected M1 vs. M0 macrophages (n = 4 biological replicates; P = 7.6e−7 for IL-6, P = 5.4e−6 for CCL2; two-tailed paired t-test).  Data are mean ± SEM (g, i-l). IBs are representative (n = 2) with molecular weight markers (kDa). A clearer protein ladder from the same blot was overlaid (dashed line). Source data are provided as a Source Data file.

NSP3 binds to and de-ISGylates IFIT5 to evade host innate immunity

Our network revealed interactions between SARS-CoV-2 proteins (NSP1/2/3/5/6/8/9/10/12/14/15, ORF7A) and IFIT (IFN-induced protein with tetratricopeptide repeats) family members (IFIT1/2/3/5, IFIT1B) in saliva from SARS-CoV-2-infected individuals or M0 macrophages (Fig. 6a). Supporting this, NSP2 has been reported to interact with IFIT5 in HEK293 cells46, and ClusterONE identified MPC 053 linking NSP3, 5 and 10 with IFIT5 (Supplementary Data 23). Evidence shows SARS-CoV-2-infected macrophages secrete ISG15 (ubiquitin-like interferon-stimulated gene 15) via NSP3, with extracellular free, non-conjugated ISG15 acts as a cytokine, to exacerbate inflammation47. While immune evasion by SARS-CoV-2 proteins is well-studied48,49, NSP3’s role in suppressing IFIT5-mediated immunity via its interaction with IFIT5 remains unclear.

Fig. 6: NSP3 antagonizes IFIT5 to evade innate immunity.
figure 6

a IFITs targeted by SARS-CoV-2 proteins via known and newly identified interactions by AP-MS (M0 macrophages) and CF-MS (SARS-CoV-2-infected patient saliva). Scatter plot shows SARS-CoV-2 interactions with IFITs (RF > 0.75). IFIT5 (b) or NSP3 (c) IPs from indicated samples were probed with NSP3 or IFIT antibodies. d qRT-PCR of IFIT5 and NSP3 transcripts in IFIT5 knockdown (KD) and non-targeting (NT) siRNA M1 macrophages, with or without NSP3, and in NSP3 M1 cells overexpressing (OE) IFIT5-VA; M0 served as negative control (data are mean ± SEM; n = 3-4 biological replicates; 7.2e−7 ≤ P  ≥  1.0e−4; two-sided paired t-test). e ISG15 IB of WCEs from M1 and NSP3-V5 M1 macrophages shows ISGylation/de-ISGylation; NSP3 and GAPDH (LC) were probed with V5 and LC antibodies. A clearer protein ladder from the same blot is overlaid on the membrane (dashed line). f IFIT5 IPs probed with ISG15 show IFIT5 ISGylation in M1 macrophages, and reduced in NSP3-V5 cells. IFIT5 levels verified by IFIT5 antibody. IFN-β by ELISA (g), IRF3 and NF-κB target gene expression by qRT-PCR (h), and NF-κB activity by ELISA (i) were measured in samples from panel d (data are mean ± SEM; n = 3-6 biological replicates; 6.5e−5 ≤ P  ≥  4.4e−2 or significance with asterisks by one or two-sided paired t-tests). j Structural docking of the ISG15-IFIT5 interface with NSP3 (PDB ID: 6YVA); full data available in Source file. k IPs of wild-type (WT) and mutant IFIT5-VA with or without NSP3-V5 in M1 macrophages were probed using ISG15 (k-1), V5 (NSP3, k-2), and FLAG (IFIT5-VA, k-3) antibodies. Plot k-2 shows NSP3-V5 band intensities (Lanes 3, 5, 7, 9) matched to corresponding IFIT5-FLAG samples, normalized to WT IFIT5-VA transfected with NSP3-V5. Plot k-3 shows IFIT5-FLAG intensities (Lanes 2, 4, 6, 8) for WT and mutants, normalized to WT IFIT5-VA (Lane 2). Data are mean ± SEM (n = 10; 4.6e−24 ≤ P  ≥   8.0e−21; 1.5e−19 ≤ P  ≥  1.4 e−13 by one or two-sided paired t-tests). Representative IBs (n = 2) include molecular weight markers (kDa). Source data are provided as a Source Data file.

We confirmed NSP3’s interaction with IFIT family proteins via co-IP in NSP3-transfected M1 macrophages, saliva from individuals infected with ancestral or variant SARS-CoV-2 strains, plasma from COVID-19 patients, and SARS-CoV-2-infected A549-hACE2-TMPRSS2 (A549-AT) cells at various time points (Fig. 6b, c and Supplementary Figs. 5a, b). IFIT5 mRNA levels were increased in M1 macrophages and further enhanced by NSP3 compared to M0 macrophages (Fig. 6d). Given ISG15’s role in modifying IFITs via ISGylation50, a key antiviral defense against SARS-CoV-247 and other viruses51,52, and NSP3’s de-ISGylating activity in suppressing antiviral responses49, we examined whether NSP3 inhibits IFIT5 activation via de-ISGylation. While cellular ISGylation was induced in M1 macrophages, total protein ISGylation decreased following NSP3 transfection (Fig. 6e). Similarly, ISGylation increased in A549-AT cells after IFN-γ stimulation but declined 12 h post-SARS-CoV-2infection (Supplementary Fig. 5c). To determine if IFIT5 is an ISG15 target, we immunoprecipitated IFIT5 from M1 macrophages (Fig. 6f) and IFN-γ activated A549-AT cells (Supplementary Fig. 5d) and identified it among higher molecular weight ISGylated proteins, confirming its ISGylation. However, IFIT5 ISGylation decreased in SARS-CoV-2-infected A549-AT cells at 12 h post-infection (Supplementary Fig. 5d), which is due to NSP3, as a similar reduction (Fig. 6f) was observed in NSP3-transfected M1 macrophages.

To assess whether NSP3 mediates the decrease in IFIT5 ISGylation, we depleted IFIT5 in NSP3-challenged M1 macrophages and observed a greater reduction in IFIT5 mRNA levels compared to NSP3 transfection alone (Fig. 6d and Supplementary Fig. 5e). Silencing IFIT5 in M1 cells with NSP3 significantly reduced NSP3 transcript levels, indicating that NSP3 depends on IFIT5 (Fig. 6d). Given NSP3’s role in regulating IFN and NF-κB pathways49, we evaluated its effect on IFN-β production, IRF3 (IFN regulatory factor-3)-activated INF-I (IFNA17, B1) and INF-II (IFNG), and NF-κB-mediated pro-inflammatory cytokine (IL-5) gene expression in M1 macrophages with or without IFIT5. NSP3-transfection reduced IFN-β release and mRNA levels of IRF3- or NF-κB-responsive genes, while IFIT5 overexpression enhanced these responses. This finding is supported by the observation that IFIT5 knockdown diminished IFN-β production and IRF3/NF-κB-driven gene transcription in NSP3-transfected cells (Fig. 6g, h). NF-κB activity was also declined in NSP3-transfected M1 macrophages regardless of IFIT5, but the ectopic expression of IFIT5 restored NF-κB levels (Fig. 6i). These results establish NSP3 as a negative regulator and IFIT5 as a positive regulator of IRF3 and NF-κB activation in innate immune response.

Since IFIT5 potentiates IFN and IRF3 signaling in antiviral defense, we probed whether IFIT5 serves as a key link between MAVS and TBK1 in the signalosome. Immunoprecipitation of IFIT5 from M1 macrophage lysates, with and without NSP3, revealed enhanced binding of IFIT5 to key IFN signaling effectors, including IκB kinase (IKK α/β/γ/ε), MAVS, TBK1, and IRF3, in the presence of NSP3 (Supplementary Fig. 5f, g). These factors are crucial for IRF3 and NF-κB pathway activation, suggesting IFIT5 bridges MAVS and TBK1. In contrast, IFIT5 depletion reduced NSP3’s binding to these effectors, but overexpression of IFIT5 in NSP3-transfected M1 macrophages restored effector interactions (Supplementary Fig. 5h, i), suggesting that IFIT5 binding is critical for NSP3 to suppress IRF3 and NF-κB activation.

Although human IFIT proteins share 40-56% sequence identity (Supplementary Fig. 5j), IFIT5 and its orthologs exhibit high similarity (72-99%; Supplementary Fig. 5k), indicating a shared ancestral gene. Molecular docking using AlphaFold structures of ISG15 and IFIT5 with the NSP3 crystal structure identified three conserved lysine residues (K160, K197, K206) on IFIT5 at the interaction interface with ISG15 and NSP3 (Fig. 6j). This supports the IFIT5-ISG15-NSP3 association and ISG15’s ability to conjugate lysines through ISGylation53. To confirm these lysines at IFIT5-ISG15 interface as ISGylation sites, we generated three single-site IFIT5 mutants by substituting lysine residues with arginine (K160R, K197R, K206R). Mutants VA-IFIT5 K197R and K206R showed reduced ISGylation with or without NSP3, while K160R exhibited wild-type IFIT5 ISGylation levels, but only in response to NSP3 (Fig. 6k-1), indicating K197 and K206 as key ISGylation sites on IFIT5. Notably, these IFIT5 mutants impacted SARS-CoV-2 infection, affecting viral replication and titers in supernatants and lysates in CRISPR IFIT5 knockout and CRISPR-resistant IFIT5 single-site mutants (K197R, K206R) in IFIT5 knockouts at 4, 8, and 12 h post-infection in A549-AT cells. However, these effects were not seen in IFIT5 knockout cells complemented with wild-type CRISPR-resistant IFIT5 (Supplementary Fig. 5l, m).

We then assessed how these mutations affect IFIT5-NSP3 binding by immunoprecipitating wild-type and mutant VA-IFIT5 from M1 macrophages, while probing for NSP3. The IFIT5 mutants disrupted IFIT5-NSP3 binding at the interface, albeit with lower affinity than wild-type IFIT5, as shown by quantified band intensities (Fig. 6k-2). This disruption was specific, as IFIT5 mutants showed slightly higher expression levels than the wild-type IFIT5 (Fig. 6k-3), suggesting ISGylation affects IFIT5 stability. These findings support a model where NSP3 interacts with ISG15 and IFIT proteins to suppress ISG15-dependent ISGylation of IFIT5 at K197 and K206. This suppression frees ISG15 to enhance pro-inflammatory cytokine secretion linked to COVID-19 cytokine storms. Furthermore, while IFIT5 acts as an adaptor linking IFN effectors to activate IRF3 and NF-κB, NSP3 inhibits these pathways in the presence of IFIT5.

Deep learning-based peptides impede ancestral SARS-CoV-2 and variant replication

While vaccines boost immunity and protect against COVID-19, their efficacy may be reduced by emerging SARS-CoV-2 variants. Antiviral peptides (AVPs), on the other hand, offer prophylactic and therapeutic benefits due to their specificity and low toxicity. AVPs target viral proteins like the S protein to block host receptor binding (e.g., ACE2, TMPRSS2), inhibit viral infection or replication, disrupt PPIs, and compete for binding by mimicking protein surfaces54. Currently, 21 synthetic peptides are under implementation for COVID-19 treatment55, indicating the potential for discovering effective peptides as promising therapeutics for this virus and future outbreaks. Since the S glycoprotein RBD binds the human ACE2 receptor for viral entry56, we postulated that AVPs blocking this interaction could counter emerging variants. Using the In-Silico Protein Synthesizer (InSiPS11), we designed synthetic peptides targeting the RBD or S1/S2 cleavage site of the S protein, critical for viral entry. InSiPS generates random sequences and predicts high-affinity peptides without requiring 3D structural input (Supplementary Fig. 6a). Among 27 peptides designed (15 targeting RBD, 12 at the S1/S2; Fig. 7a), three RBD-binding peptides reduced SARS-CoV-2 RNA by 54-95% and S protein expression in Huh-7 cells at 0.27-2.7 µM, as confirmed by real-time PCR and immunofluorescence (Fig. 7b and Supplementary Fig. 6b).

Fig. 7: Inhibitory peptides reduce ancestral and variant SARS-CoV-2 infectivity.
figure 7

a SARS-CoV-2 S protein structure shows domains and InSiPS-derived peptides (arrows) targeting the receptor-binding ___domain (RBD) and S1/S2 protease cleavage site. SS, single sequence; NTD, N-terminal ___domain; SD1, subdomain 1; SD2, subdomain 2; S2’ (black arrow), protease cleavage site; FP, fusion peptide; HR1, heptad repeat 1; CH, central helix; CD, connector ___domain; HR2, heptad repeat 2; TM, transmembrane ___domain; CT, cytoplasmic tail. b qRT-PCR of viral RNA in Huh-7 cells treated with SARS-CoV-2 and RBD-binding antiviral peptides (AVPs) at specified concentrations were compared to virus-only or vehicle (DMSO) controls (data are mean ± SEM (n = 3–13 biological samples; P  =  8.3e−5 ≤ P  ≥  2.7e−2 or P ≤  1.1e−4 by one or two-sided, paired or unpaired t-test). c mRNA expression of markers in SARS-CoV-2-infected Huh-7-ACE2 cells treated with AVPs measured by qRT-PCR and compared to untreated controls (n = 3 biological replicates; 6.7e−8 ≤ P  ≥  2.0e−4 by two-sided, paired t-test). d Plaque assays in Vero E6-TMPRSS2 cells show dose-dependent AVP inhibition of ancestral and variant SARS-CoV-2 (n = 2 biological replicates per dose). e Surface plasmon resonance (SPR, right) shows ACE2 (positive control) or AVP binding to His-tagged S-RBD (wild-type, Delta with T478K L452R mutations, and Omicron with 16 mutations). SPR (left) reveals AVPs inhibit ACE2 binding to wild-type or mutated S-RBD. f SEC-HPLC/MS-based hierarchical clustering of co-fractionated protein profiles in SARS-CoV-2-infected Huh-7 cells treated with AVPs vs. mock (uninfected/untreated) control. g Heatmap shows AVP-treated SARS-CoV-2-infected Huh-7 cells exhibit protein coelution profiles more similar to mock (uninfected/untreated) control than to untreated virus-infected cells. h Viral-host protein pairs (n = 48) show altered correlations in AVP-treated SARS-CoV-2-infected Huh-7 cells, but not in those treated with AVP 1 or AVP 2 (PCC < 0.25). PCC differences for each co-eluted human protein pair were calculated across samples, with two-sided Z-transformed P-values adjusted for FDR ≤ 5e−2 using the BH method. Source data are provided as a Source Data file.

Peptide treatment also reduced mRNA levels of proinflammatory cytokines (IL-1β, IL-6, TNF-α, PGE2) and the inflammatory marker (eNOS) in SARS-CoV-2-infected Huh-7-ACE2 cells compared to untreated controls (Fig. 7c). Peptides decreased infectivity of ancestral and variant SARS-CoV-2 strains in Vero E6-TMPRSS2 cells dose-dependently by plaque assays (Fig. 7d), and inhibited S protein activity in HEK293T cells over time using a pseudovirus entry assay (Supplementary Fig. 6c). Similarly, peptide-treated SARS-CoV-2-infected Huh-7 cells showed marked inhibition of cytoplasmic S protein localization (Supplementary Fig. 6d), confirming their ability to block S protein entry. The peptides showed specificity, with no toxicity (in Vero E6-TMPRSS2 or A549-AT cells (Supplementary Fig. 6e), and no RNA-level changes in Huh-7 cells transfected with other human coronaviruses (HCoV-229E, OC43; Supplementary Fig. 6f). ProtParam analysis revealed favorable physicochemical properties of these AVPs, including an aliphatic index of 66.4-87.4, a half-life of 1.4-100 hrs, and negative GRAVY scores (Supplementary Fig. 6g), suggesting they are thermally stable and hydrophilic.

To confirm that peptides block SARS-CoV-2 or variant S protein RBD binding to ACE2, surface plasmon resonance was performed using C-terminal histidine-tagged recombinant wild-type and mutated (Delta variant with T478K L452R mutations, Omicron variant with 16 mutations) S RBD proteins, expressed in Expi293 cells. The peptides exhibited high-affinity binding (KD = ~ 30.5 ± 3.5 nM; Fig. 7e) to all S RBDs, and effectively inhibited S-RBD:ACE2 interactions. In silico modeling of SARS-CoV-2 RBD-ACE2 crystal structures57 showed peptides bind to conserved S RBD residues (N487, T500, Y449/453/489/505) and SARS-CoV-2 or variant-specific residues (E484, F456, L455, N501, Q493, Y473), forming interfaces with ACE2 (RBD:E484-ACE2:K31-AVP2:S4/T3; RBD:Y453-ACE2:H34-AVP3:Y18; Supplementary Fig. 7a). These findings underscore peptide residues’ role in disrupting S-RBD:ACE2 binding.

Using a CF-MS framework similar to that applied to SARS-CoV-2-infected patients’ saliva (Fig. 3a), we evaluated whether altered co-eluting human proteins in SARS-CoV-2-infected Huh-7 cells could be restored by treating them with peptides AVP1 and AVP2. SEC-MS analysis of 384 fractions from soluble protein extracts showed that the hierarchical clustering of 1189 human proteins profiles (Fig. 7f; Supplementary Data 26) from peptide-treated cells had elution patterns more similar to mock (uninfected) than virus-infected cells. PCC analysis of co-eluted human protein profiles identified 4539 protein pairs grouped into three clusters, with positively (cluster 1) and negatively (cluster 3) correlated PPI pairs in peptide-treated cells, resembling those in mock cells more closely than in untreated SARS-CoV-2-infected cells (Fig. 7g; Supplementary Data 27). GO cellular component analysis of clusters 1 and 3 showed enrichment for proteins in endosomal sorting, pre-mRNA splicing, and mitochondrial biogenesis (Supplementary Fig. 7b). These findings suggest peptides restore host protein assemblies hijacked by SARS-CoV-2, reflecting antiviral efficacy by targeting virus-dependent host pathways.

To examine changes in viral-host PPIs, we compared correlation scores of protein pairs in SCoV-2-infected Huh-7 cells treated with or without peptides. Applying a 5% FDR, we generated a differential viral-host PPI network (Fig. 7h; Supplementary Data 28), identifying 48 protein pairs highly correlated in untreated SCoV-2-infected Huh-7 cells but weakly correlated in peptide-treated cells (Fig. 7h). For example, the strong correlation (r = 0.6) between SARS-CoV-2 ORF9B and the mitochondrial antiviral signaling protein MAVS in untreated SARS-CoV-2-infected cells reflected their physical coupling and role in suppressing IFN production57. However, in peptide-treated cells, this association was poorly correlated (r = -0.08 to 0.06), suggesting that peptides disrupt ORF9B-MAVS binding, thereby preserving host innate immune signaling cascades.

Discussion

Ancestral SARS-CoV-2 and its variants impact various organs, tissues, and bodily fluids, enabling viral transmission and immune evasion, potentially leading to multiorgan failure and death. Despite extensive studies on SARS-CoV-2-human PPIs using experimental methods4,5,6,7, our understanding of context-dependent host interactions with SARS-CoV-2 and its variants at physiological infection sites remains incomplete. To address this, we employed AP-MS across eight cell lines from five mammalian organs susceptible to SARS-CoV-2 infection. This analysis revealed hundreds of PPIs, shared or unique to different organ cell lines, including those with the S protein of SARS-CoV-2 and its variants. Rigorous validation using complementary analyses, such as CF-MS profiling of patient saliva carrying SARS-CoV-2 or its variants, mapped reconfigured protein assemblies during infection.

Differential network analysis of saliva from individuals infected with SARS-CoV-2 and its variants revealed notable variations in viral protein abundance and host responses, particularly in inflammatory or immune pathways like the immunoglobulin kappa variable cluster. Certain host interacting proteins from immune cells, epithelial, myeloid, and connective tissues in the salivary glands were markedly induced during variant infections. The merged network showed NSP6, S, and ORF3A/7A relying heavily on host factors, while other SARS-CoV-2 proteins (E, M, N, NSPs, S, ORFs) linked to M1 macrophage- or microglia-specific human proteins were associated with neuroinflammation or neurological disorders. Overall, identifying viral proteins dependent on host factors altered during SARS-CoV-2 or variant infections is essential for understanding COVID-19 pathophysiology, developing host-directed therapies, and uncovering mechanisms of action. Mapping SARS-CoV-2 ancestral and variant-host PPIs revealed 173 human protein subunits within predicted MPCs targeted by at least one of the 7626 cataloged drugs. This includes fostamatinib, a phase 3 oral spleen tyrosine kinase inhibitor for COVID-19, which targets 25 kinases associated with NSPs, ORFs, M and N SARS-CoV-2 proteins in MPCs. Additionally, 13 drugs target 10 or more subunits within 52 MPCs connected to various SARS-CoV-2 proteins. While further investigation is needed, targeting these host factors may disrupt their interactions with viral proteins, potentially aiding in SARS-CoV-2 or variant infection treatment.

Among the context-specific host-viral PPIs identified, our probabilistic network revealed previously unnoticed associations in SARS-CoV-2 and variant biology, supported by biochemical, genetic, and cell-based assays. Key findings include the interaction of the secreted NSP3 viral protein with human fibrinogen complex subunits (FGA, FGB, FGG), crucial for initiating inflammatory cytokine responses and causing coagulation abnormalities in COVID-19 patients. Our inquiry into NSP3 binding with IFN-induced proteins, particularly IFIT5, in LPS-IFN-γ-activated or NSP3-transfected M1 macrophages and saliva from ancestral or variant SARS-CoV-2-infected individuals, revealed significant findings. Specifically: (1) NSP3 reduces ISGylation of IFIT5; (2) IFIT5 acts as an adaptor linking MAVS, TBK1, and IFN effectors to trigger IRF3 and NF-κB pathways, suppressed by NSP3-IFIT5 interaction; (3) IFIT5 lysine residues (K160, K197, K206) interface with ISG15 and NSP3; and (4) NSP3 inhibits IFIT5 ISGylation at K197 and K206, releasing free ISG15 to promote pro-inflammatory cytokine secretion, driving cytokine storms linked to COVID-19 severity. These insights into NSP3 dependency on IFIT5 reveal a key immune evasion mechanism and identify a promising molecular target for antiviral development.

Our deep learning-driven InSiPS analysis designed synthetic inhibitory peptides targeting the RBD or S1/S2 cleavage region of the S protein, identifying three potent RBD-binding peptides (AVP1-3) that strongly inhibit infectivity of both ancestral and variant SARS-CoV-2 strains by blocking RBD-ACE2 binding, critical for viral entry. The identification of 48 highly correlated viral-host interacting protein pairs in SARS-CoV-2-infected Huh-7 cells, absent with AVP treatment, further supports the antiviral activity of these peptides in disrupting viral-host PPIs, interrupting the SARS-CoV-2 lifecycle, and preserving host innate immunity. Together, this study offers a comprehensive biochemical map of the SARS-CoV-2 and variant interactome across organ or immune-derived cell lines and biofluids. The host-directed interventions and antiviral candidates presented here advance understanding of COVID pathophysiology and inform pan-viral therapeutic development. By providing host-viral PPI data via a public repository (MassIVE: MSV000092774, MSV000096435, MSV000097310) and a web portal (https://babulab-uofr.shinyapps.io/scov2db/), we offer this resource to support further exploration of context-specific interactions critical to the distinct biology of SARS-CoV-2 and other viruses.

Methods

Saliva and blood plasma collection

Saliva specimens were collected from COVID-19 patients infected with the ancestral SARS-CoV-2 strain (n = 16), or the Alpha (n = 4), Delta (n = 2), and Omicron (n = 10) variants during three consecutive waves. Additionally, blood samples (n = 6) were taken from SARS-CoV-2 patients in the first wave. All COVID-19 cases were verified via nasopharyngeal or throat swabs, with SARS-CoV-2 RNA detected by real-time quantitative polymerase chain reaction (qRT-PCR) and variants identified via whole-genome sequencing. COVID-19 patients were admitted to Regina General Hospital, Canada, from June 2020 to December 2022. Control blood and saliva samples were obtained from 6-10 healthy volunteers without clinical symptoms for four weeks, confirmed negative for SARS-CoV-2 RNA by qRT-PCR. The study, approved by the Saskatchewan Health Authority research ethics board, adhered to University of Regina ethical standards, with written informed consent obtained from all participants. For blood collection, BD Vacutainer tubes (Thermo Fisher Scientific) treated with lithium heparin were used for plasma separation. Stimulated whole saliva were collected into sterile polystyrene tubes from both healthy and COVID-19 patients after chewing paraffin, sugar-free, or sugared chewing gum for five min, then cooled on ice, and processed for cellular assays.

Cell lines, viruses, plasmids, and reagents

Mammalian cell lines and coronaviruses (HCoV-OC43, 229E) were obtained from the American Type Culture Collection, USA. Vero E6 cells expressing the transmembrane serine protease TMPRSS2 (Vero E6-TMPRSS2) were sourced from the Japanese Collection of Research Bioresources Cell Bank (JCRB1819), and the human alveolar epithelial cell line A549, expressing angiotensin-converting enzyme 2 (ACE2) and TMPRSS2 (A549-AT) was from InvivoGen. Ancestral SARS-CoV-2 (VIDO/01/2020) was provided by Darryl Falzarano from VIDO-INTERVAC, and variants Alpha (B.1.1.7), Beta (B.1.351), Gamma (P.1), Delta (B.1.617.2), and Omicron (B.1.1.529) were isolated from COVID-19 patients. Viruses were propagated at a low multiplicity of infection (MOI = 0.0001) in HCoV-OC43 and 229E-infected Huh-7 cells, and ancestral or variant SARS-CoV-2-infected Huh-7-ACE2, Vero E6-TMPRSS2, and A549-AT cells with media containing 2% heat-inactivated fetal bovine serum (FBS). Viral stocks were stored at −80 °C and titrated via TCID50 (50% tissue culture infective dose) or plaque-forming assays. Expression plasmids for SARS-CoV-2 open reading frames and an eGFP control in the pLVX-EF1alpha-2xStrep-IRES-Puro vector5, were gifts from Nevan Krogan (Addgene 141367-141381; 141383-141395). Gateway-compatible entry plasmids (pDONR207, 223) for S (Addgene 152988), NSP3 (Addgene 141257), and NSP16 (Addgene 141269) were from Frederick Roth (Supplementary Note 1). Detailed information on antibodies, plasmids, oligonucleotides, and reagents is provided in Supplementary Data 29. All work with ancestral or variant SARS-CoV-2 strains was conducted in biosafety level 3 facilities at the University of Alberta, University of Toronto, University of Pennsylvania, University of Saskatchewan/VIDO-INTERVAC in Saskatoon, and Canada’s National Microbiology Laboratory in Winnipeg.

Cell culture

HEK293T, Huh-7, HepG2, N9 microglia, Huh-7 expressing ACE2 receptor (Huh-7-ACE2), Vero (Vero E6, 81, E6-TMPRSS2), and A549-AT cells were cultured in high-glucose Dulbecco’s modified Eagle medium (DMEM) with 2–4 mM L-glutamine. Caco-2 cells were grown in Eagle’s minimum essential medium (MEM) with 2 mM L-glutamine, while HULEC-5a cells were maintained in MCDB131 medium with 10 ng/mL epidermal growth factor, 1 µg/mL hydrocortisone, and 10 mM glutamine. Human monocytic THP-1 cells were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium with 2 mM L-glutamine and differentiated into M0 macrophages by treating them with RPMI medium supplemented with 150 nM phorbol 12-myristate 13-acetate for 24 h, followed by incubation in RPMI medium without supplements for an additional 24 h. To induce M1 polarization, M0 macrophages were stimulated with 20 ng/mL IFN-γ and 100 ng/mL LPS for 48 h in RPMI medium. Expi293F cells (Thermo Fisher Scientific) were cultured in Expi293 expression medium. All cell lines were maintained at 37 °C and 5% CO2 in their respective media with antibiotics (100 U/mL penicillin, 100 µg/mL streptomycin) and 10% heat-inactivated FBS, except for Caco-2 cells, which were cultured with 20% FBS.

Lentivector production and transduction into mammalian cells

HEK293T cells were seeded at 4 ×106 cells per 10 cm dish. At 60% confluence, 7.5 µg psPAX2 packaging plasmid (Addgene 12260), 750 ng pMD2.G envelope plasmid (Addgene 12259), 7.5 µg lentiviral expression plasmid (Supplementary Note 1), and 15 µL PLUS reagent (Lipofectamine LTX kit) were mixed in 750 µL Opti-MEM I reduced serum medium (Thermo Fisher Scientific). In another sterile tube, 52.5 μL Lipofectamine LTX was mixed with 697.5 μL Opti-MEM I. The contents of both tubes were combined, incubated at room temperature for 15 min, added to HEK293T cells, and incubated at 37 °C in 5% CO2 in DMEM without antibiotics. After 16 h, the medium was replaced with DMEM containing 100 U/mL penicillin, 100 µg/mL streptomycin, and 30% FBS. Lentiviral particles were collected at 24 and 48 h post-transfection and filtered through a 0.4 µm low protein-binding polyethersulfone membrane syringe filter.

Lentiviral supernatants were transduced into mammalian cells at 10% confluence with an MOI of 0.5 for 48 h using filter-sterilized (0.22 µm syringe filter) polybrene (10 µg/mL) to enhance infection. After 48 h, cells were transferred to 10 cm dishes and selected with puromycin (2 µg/ mL) for three days. For pLX304-hACE2 in Huh-7 cells, blasticidin (5 µg/mL) was used for two weeks to ensure stable expression of VA-tagged ancestral SARS-CoV-2 or variants, pLEX-307-LgBiT, or pcDNA3.1-SARS2-Spike-HiBiT. VA-tag expression was verified by immunoblotting using an anti-Strep antibody. Stable transformants were expanded to five 10 cm dishes, reaching 80% confluency ( ~5 × 107 cells). Cells were harvested in batches, yielding up to six biological replicates, and pellets were either processed immediately or stored at -80 °C for future use.

Affinity purification and SEC fractionation of saliva or plasma protein extracts

Fresh or frozen cell pellets were crosslinked using dithiobis succinimidyl propionate as described in studies on human-SARS-CoV-2 PPIs58,59. Crosslinked cells expressing Strep or VA-tagged SARS-CoV-2 proteins were lysed in a buffer containing 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 0.5% Nonidet-P40 (NP-40), and protease/phosphatase inhibitors. For VA-tagged SARS-CoV-2 ‘S’ bait proteins, 1% n-dodecyl-β-D-maltoside (DDM) was replaced with NP-40. Lysates were chilled on ice for 30 min, centrifuged at 13,000 x g for 15 min at 4 °C, and the supernatant was incubated with 30 µL MagStrep type3 Strep-Tactin beads for purification as detailed previously5. Proteins were eluted with 50 mM D-biotin (Sigma), dried with a Savant Speedvac, and then denatured, reduced, and alkylated following established protocol12. Strep-tagged protein expression was confirmed via immunoblotting with an anti-Strep antibody. Proteins were digested overnight with sequencing-grade trypsin (Promega) at room temperature on a rocking shaker, and digestion was quenched with 0.1% (v/v) formic acid. Peptides were purified using Zip-Tip C18 tips (Millipore)12, air-dried, resuspended in 0.1% formic acid, and analyzed by liquid chromatography-tandem MS/MS (LC-MS/MS), as detailed in Supplementary Note 2.

Saliva or plasma samples from healthy individuals, patients infected with SARS-CoV-2 or its variants, or antiviral peptide-treated Huh-7 cells (with and without SARS-CoV-2 infection) were mixed with an equal volume of SEC buffer (25% v/v acetonitrile, 50 mM triethylamine phosphate, 50 mM sodium perchlorate)30, lysed by syringe with a 23-gauge needle, and centrifuged at 3000 x g for 10 min at 4 °C to remove debris. The supernatant was filtered through a 0.22-μm polyethersulfone filter. About 500 μg of protein per sample was separated using an Agilent 1100 high performance liquid chromatography system with a binary pump and SEC buffer on a Yarra SEC-4000 column (300 × 7.8 mm i.d., 3 μm). Proteins were eluted at 0.5 mL/min for 40 min, monitored at 280 nm, and collected in 84-96 fractions. HPLC fractions were acid-precipitated and digested with sequencing-grade trypsin, and resuspended in 0.1% formic acid, following the above described protocol. Scoring of PPIs from AP-MS and CF-MS datasets, along with the construction of differential networks comparing saliva from SARS-CoV-2-infected patients to healthy subjects, and between the ancestral virus and its variants, are detailed in Supplementary Note 3.

MPC prediction

High-confidence viral-host PPIs from the merged network were analyzed using the ClusterONE algorithm60 to delineate complex membership. The algorithm’s settings were optimized using various parameter combinations (density, d; overlap, o) and complementary evaluation metrics (accuracy, maximum matching ratio, overlap) as previously described30. Predicted complex sets were benchmarked against CORUM human protein assemblies interacting with SARS-CoV-2 proteins in the BioGRID, at various density (0.20-0.70) and overlap (0.5–0.8) settings. The parameter combination yielding the highest composite score (i.e., sum of overlap, accuracy, and maximal matching ratio) was deemed as the most effective for predicting MPCs. Only clusters involving four or more human proteins interacting with SARS-CoV-2 proteins were considered (Supplementary Data 23). All viral-host or host-host PPIs, and MPCs were visualized using Cytoscape (ver.3.9.1) software and are accessible through our web portal (https://babulab-uofr.shinyapps.io/scov2db/).

Whole-cell salivary proteomics

Stimulated saliva (1 mL) from healthy individuals and patients infected with SARS-CoV-2 or its variants was mixed with 200 µL of lysis buffer (6 M urea, 2 M thiourea, protease inhibitor cocktail (PIC), 0.05% Protease MaxTM surfactant) and incubated on ice for 30 min. The mixture was centrifuged at 3000 x g for 10 min at 4 °C, and the supernatant was filtered through a 0.22-μm filter. About 50 µg of protein per sample was trypsin digested for LC-MS/MS analysis.

Authenticating viral-host or host-host PPIs

We evaluated our viral-host and host-host PPIs from AP-MS and CF-MS methods against known interactions in BioGRID. SARS-CoV-2-human PPIs were also assessed against non-human cell lines using InParanoid orthology predictions with default BLASTP and BLOSUM62 settings. We examined the overlap of SARS-CoV-2-human PPIs with those identified in a previous interactome study of human proteins interacting between pathogen pairs (references in Supplementary Data 5). The co-localization of viral-host PPIs was assessed using localization data from prior studies15,61 (with additional references reviewed in Grand, 202361) and the Human Protein Atlas62, considering only main subcellular locations where proteins are localized in the cell. Furthermore, host interacting protein pairs within the salivary glands during infection were investigated using data from oral atlases that include human minor salivary glands and gingival/palatal mucosae2.

Enrichment analysis

We identified the most enriched terms in the viral-host or host-host PPIs using protein subunits from MPCs from databases like CORUM, Reactome, WikiPathways, and GO categories using the g:Profiler R package. Enrichment analysis utilized a hypergeometric test (P ≤ 5.0e−2) with Benjamini-Hochberg FDR correction. For Gene ontology analysis, we prioritized terms with fewer than 500 proteins, focusing on those with the lowest adjusted P-values to highlight hits.

Site-directed mutagenesis

The pDONR223 Gateway-compatible entry plasmid encoding IFIT5 cDNA (HsCD00512366) from the human ORFeome collection was obtained and sequence-verified with M13 forward and reverse primers at the DNASU Plasmid Repository. Catalytically inactive IFIT5 single mutants (Lys160Arg, Lys197Arg, Lys206Arg) were generated in the pDONR223 plasmid using the Phusion site-directed mutagenesis kit (Thermo Fisher Scientific) by a PCR-based method with primers listed in Supplementary Data 29. Modified pDONR223 plasmids, each containing one of the IFIT5 single mutants, were recombined with the pLD-puro-CcVA destination vector using the Gateway cloning LR II enzyme mix. IFIT5 mutant construct accuracy was verified by Sanger DNA sequencing with CMV forward and cPPT reverse primers at Toronto’s TCAG DNA sequencing facility. To produce mutated SARS-CoV-2 S-RBD proteins, Delta variant double mutations (T478K, L452R) were engineered into the pcDNA3.1-SARS-CoV-2-S-RBD-8xHis-tag mammalian expression plasmid (Addgene 145145) using the QuikChange Lightning mutagenesis kit (Agilent) with primers from Supplementary Data 29. Omicron mutations (G339D, S371L, S373P, S375F, K417N, N440K, G446S, S477N, T478K, E484A, Q493R, Q498R, G496S, N501Y, Y505H, T547K) in the SARS-CoV-2 S-RBD protein (amino acids 319–591) were synthesized by Genscript and cloned into the pcDNA 3.1 plasmid with a C-terminal mouse Fc and 6xHis-tag.

IFIT5 knockdown and knockout

siRNA targeting human IFIT5 sourced from Horizon Discovery was used for gene knockdown in cells cultured from 48 h to 7 days, depending on the experiment. To transiently knockdown IFIT5 in M1 macrophages, 2 ×106 cells were seeded in 10 mL of RPMI medium in a 10 cm dish and transfected with IFIT5 siRNA using Lipofectamine RNAiMAX (Thermo Fisher Scientific), following the manufacturer’s instructions. Non-targeting siRNA served as a control. For IFIT5 knockdown in NSP3-expressing cells, 2 ×106 M1 macrophages in a 10 cm dish were transduced with V5- or VA-tagged NSP3 at an MOI of 0.4 for 5 days with blasticidin selection. The next day, IFIT5 was knockdown or overexpressed for 48 h in the same dish using Lipofectamine RNAiMAX or LTX as per the manufacturer’s protocol. Knockdown efficacy was verified by immunoblotting with tag- or protein-specific antibodies prior to qRT-PCR or immunoprecipitation.

IFIT5 knockout was generated using oligonucleotide pairs encoding 20-nt guide RNAs targeting IFIT5. sgRNAs were designed with Synthego and CHOPCHOP software and synthesized by Integrated DNA Technologies (USA). The sgRNA was annealed with its complementary ssDNA at 95 °C for 5 min, then cooled to 20 °C to form dsDNA. This dsDNA was ligated into a BamH1-digested lentiCRISPR v2-Blast (Addgene 83480) plasmid and transformed into SURE 2 supercompetent cells (Agilent). Positive transformants were selected with ampicillin and sequence verified at Toronto’s TCAG facility. The sequence-confirmed IFIT5-KO lentiCRISPR v2-Blast plasmid was co-transfected into A549-TMPRSS2-ACE2 cells with packaging and envelope plasmids, followed by seven days of blasticidin selection to establish stable IFIT5 knockout. To create CRISPR-resistant (CR) IFIT5 mutants (K206R, K197R), the IFIT5-VA overexpression plasmid with puromycin and the QuikChange II mutagenesis kit (Agilent) were used to engineer single-site mutations conferring resistance to CRISPR cleavage. Briefly, the IFIT5-VA CR wildtype and mutants were amplified with CR-IFIT5 primers using QuikSolution reagent and QuikChange lightning enzyme, following manufacturer’s instructions. The amplicon, containing non-mutated plasmids, was digested with DpnI restriction enzyme, and the mutated plasmids were transformed into SURE 2 competent cells, with positive transformants selected overnight using ampicillin. Sequence-verified IFIT5-VA CR wild-type and mutant plasmids were transfected into A549-TMPRSS2-ACE2 IFIT5 knockout cells with 72 h of blasticidin and puromycin selection.

Expression and purification of recombinant SARS-CoV-2 S protein RBD

Plasmid pcDNA3.1 (Addgene 145145) encoding SARS-CoV-2 S RBDs (amino acids 333-529) with C-terminal tags (8xHis for ancestral and Delta variant, and 6xHis for Omicron), was expressed in Expi293 cells. For transfection, 30 mL of Expi293 cells at 3 × 106 cells/mL in expression medium were prepared following the Expi293 system user guide (Thermo Fisher Scientific). Three days post-transfection, cells were harvested by centrifugation at 500 x g for 20 min at 4 °C, and the supernatant was discarded. Cell pellets were resuspended in 30 mL lysis buffer (50 mM Tris-HCl pH 7.8, 150 mM NaCl, 5% glycerol for ancestral/Delta; 50 mM Tris-HCl pH 7.5, 300 mM NaCl, 1 mM PMSF for Omicron) and incubated at 4 °C for 1 h. Cells were lysed using an Emulsiflex (Avestin) homogenizer at 30,000 psi, and the lysates were centrifuged at 20,000 x g for 30 min at 4 °C. The supernatant was incubated with HisPur Ni-NTA resin at 4 °C for 2 h. After washing with increasing concentrations of imidazole, His-tagged proteins were eluted in buffer containing 50 mM Tris-HCl (pH 8.0), 150 mM NaCl, 250 mM imidazole, and 2 mM DTT. Purity of the eluted proteins was confirmed by Coomassie-stained 10% SDS-PAGE, and pooled fractions were dialyzed at 4 °C for 6 h in dialysis buffer (50 mM Tris, pH 7.8, 150 mM NaCl, 2 mM DTT). Dialyzed proteins were concentrated, flash frozen, and stored at −80 °C.

Expression and purification of human ACE2

Full-length human ACE2 with C-terminal Myc and FLAG tags in pCMV6-entry vector (Origene RC208442) was expressed in Expi293 cells and purified as described previously63. Four days post-transfection, cell pellets were centrifuged at 500 x g for 20 min at 4 °C, resuspended in 30 mL ice-cold lysis buffer (50 mM Tris-HCl pH 7.5, 300 mM NaCl, 1 mM PMSF), lysed with an Emulsiflex homogenizer at 30,000 psi, and solubilized with 0.1% Triton X-100. Lysates were clarified by centrifugation at 20,000 x g for 30 min at 4 °C. The supernatant was treated with 0.05% DDM, and incubated with anti-FLAG resin in Tris-buffer saline (TBS) containing 0.05% DDM at 4 °C for 1 h. The resin was transferred to a gravity flow column, washed with TBS containing 0.05% DDM, and eluted with 0.1 M bicine (pH 3.5). Protein purity was confirmed by Coomassie-stained 10% SDS-PAGE. Pooled eluted fractions were dialyzed for 6 h at 4 °C in buffer containing 50 mM Tris-HCl (pH 8.0), 150 mM NaCl, and 5% Glycerol. The concentrated proteins were aliquoted, flash frozen in liquid nitrogen, and stored at −80 °C for later use.

RNA isolation and qRT-PCR

Total RNA was extracted from SARS-CoV-2 protein-treated cell lysates (with or without antiviral peptides), and siRNA knockdown or overexpressed genes in M1 macrophages or Huh-7 cells transduced with viral proteins using the Qiagen RNeasy kit with DNase treatment. Reverse transcription was performed using the oligo (dT) primers and RNase H + MMLV (Moloney Murine Leukemia Virus) reverse transcriptase. Target mRNA transcripts were quantified in triplicate using FastStart SYBR Green I dye (Roche) on a LightCycler 96 SW qRT-PCR instrument. Transcript abundance was calculated by the 2−ΔΔCT method, normalized to housekeeping genes (β-Tubulin, GAPDH), and expressed as fold change in target gene expression relative to the control. Primers were synthesized as single-stranded oligomers and purified by standard desalting (IDT, USA).

Enzyme-Linked Immunosorbent Assay (ELISA)

Total IFN-β in M1 macrophage supernatants was measured with the VeriKine Human Interferon Beta ELISA kit (PBL Assay Science), and fibrinogen levels in uninfected and SARS-CoV-2-infected Huh7 cell supernatants were assessed at various time points using the Human Fibrinogen ELISA Kit (Abcam), following manufacturer’s protocol. Diluted supernatants were incubated in assay diluent at room temperature for 1 h, followed by the addition of antibody and horseradish peroxidase. Samples were incubated with 3,3′,5,5′-tetramethylbenzidine (TMB) substrate for 60 min (IFN-β) or 15 min (fibrinogen) in the dark at room temperature. After adding the stop solution, absorbance at 450 nm was measured with a Multiskan FC microplate reader (Thermo Fisher Scientific). Triplicate samples were blank-corrected, and protein concentrations were determined using IFN-β and fibrinogen standards, with linear regression used to fit the standard curve. NF-κB p65 activity in M1 macrophages with silenced or overexpressed IFIT5, challenged with or without NSP3, was assessed using an ELISA-based NF-κB p65 kit (Abcam) as instructed. M0 macrophages served as controls. Cells (1 × 104/well) seeded in a 24-well plate were lysed in Abcam’s cell extraction buffer and centrifuged at 10,000 x g for 15 min at 4 °C. Supernatants were transferred to a 96-well plate and incubated with an NF-κB antibody for 1 h at room temperature. After three 5-min phosphate-buffered saline with Tween 20 (PBS-T) washes, TMB substrate was added, and absorbance at 600 nm was recorded every 40 s for 20 min using a Multiskan FC plate reader. NF-κB activity was quantified with a four-parameter logistic regression model.

Co-IP and immunoblotting

Saliva and blood plasma from healthy and SARS-CoV-2-infected individuals, SARS-CoV-2-infected A549-TMPRSS2-ACE2 cells, along with various cell lines (HepG2, THP-1 M0 or M1 macrophages, HEK293, Caco-2, and HULEC-5a) transfected with wild-type or mutant VA-IFIT5 (with or without V5/VA-NSP3 or Strep-S/ORF3A/7 A/NSP5) were used for immunoblotting and co-IP experiments. For co-IP, Tag-or protein-specific polyclonal antibodies raised in rabbit for protein A magnetic microbeads, or in mouse for protein G microbeads (Miltenyi) were used if validated by Antibodypedia or commercial vendors. After 48 h of transfection, samples were lysed in RIPA buffer (150 mM NaCl, 50 mM Tris-HCl pH 7.5, 0.1% sodium dodecyl sulfate (SDS), 1% Na deoxycholate, 1% NP-40, 1 mM EDTA) with 1X protease inhibitor cocktail, centrifuged at 13,000 x g for 20 min at 4 °C, and supernatants were incubated with antibodies for 2 h at 4 °C and μMACS protein A or G magnetic microbeads overnight at 4 °C. Bead suspensions were passed through μMACS columns pre-equilibrated with RIPA and PIC, washed twice with RIPA containing 0.1% detergents and PIC, and finally with detergent-free buffer. Proteins bound to the beads were eluted with pre-heated 2x Laemmli buffer at 95 °C, separated by Bis-Tris SDS-PAGE, and transferred onto nitrocellulose membranes using the iBlot system (Thermo Fisher Scientific). Immunoreactive proteins were detected using IRDye 680RD goat anti-rabbit IgG or IRDye 800 CW conjugated goat anti-mouse IgG secondary antibodies (LI-COR), and the protein bands were visualized with an Odyssey Fc imaging system (LI-COR). Bands were digitized and quantified using ImageJ, with pixel intensities measured and normalized to loading or positive controls. Immunoblots were converted to grayscale using Adobe Photoshop (ver. 26.5.0), with linear exposure or contrast adjustments applied as necessary. Blot images were cropped for visual clarity. In instances where the original protein ladder was of low quality, a clearer ladder from the same blot was overlaid for presentation, and is indicated by a dashed line. All original, unprocessed Western blot images are provided in the Source Data corresponding to each figure.

Immunofluorescence and confocal imaging

Approximately 30,000 HEK293 cells expressing plasmids for SARS-CoV-2 ancestral or variant S proteins-Strep, and Huh-7 cells infected with and without SARS-CoV-2, were seeded onto 18 mm glass coverslips in 6-well plates. After 24 h at 37 °C, cells were washed with PBS, fixed with 4% paraformaldehyde for 20 min at room temperature, permeabilized with 0.2% Triton X-100 in PBS for 5 min, and blocked with PBS-T containing 5% FBS for 2 h at room temperature. Cells were incubated overnight at 4 °C with primary antibodies against Strep-tag (LSBio, C387305), S (Sino Biological, 40150-R007), NSP3 (GeneTex, GTX135589; Cell signaling, 88086S), FGA (Abcam, ab34269), E-cadherin (Abcam, ab40772), or Golgin-97 (Thermo Fisher Scientific, PA5-30048) in PBS supplemented with 5% FBS. After three 10 min PBS washes, cells were incubated with Goat anti-mouse Alexa Fluor 488 (Invitrogen, A32723) and/or Donkey anti-rabbit Alexa Fluor 647 (Invitrogen, A32795) secondary antibodies for 2 h at room temperature in PBS-T with 5% FBS. For ER staining, cells were treated with 100 nM ER-Tracker blue-white DPX (Thermo Fisher Scientific, E12353) in DMEM with 10% FBS for 30 min at 37 °C in the dark, followed by fixation, permeabilization, and antibody labeling. Coverslips were mounted in 100% glycerol, or mounting medium with DAPI, and images were captured using a Zeiss LSM 900 confocal microscope with a 63X, 1.4 NA oil objective, and Airyscan detector. Images were processed using Zen Blue software (Zeiss), and FGA fluorescent intensity was quantified in ImageJ. Brightness and contrast were uniformly adjusted across entire images in Adobe Photoshop, and cropped for presentation. All original fluorescence microscopy images have been deposited on Figshare.

Detection of SARS-CoV-2 NSP3 protein secretion

HepG2 cells expressing NSP3-V5 tag were seeded at 60% confluency in DMEM with 10% FBS. After 48 h, cell pellets and supernatants were collected and clarified by centrifugation at 700 x g for 5 min to remove debris. Both supernatants and pellets were analyzed by immunoblotting using an anti-V5 antibody. FGA was used as a positive control, and Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as a negative control for secreted protein. Blots were imaged using the Odyssey Fc imager (LI-COR) and quantified with ImageJ plugin.

Coagulation kinetic assay

Thrombin (positive control) was obtained from Sigma-Aldrich, human plasma fibrinogen from Millipore Sigma, and partial NSP3 recombinant protein (papain-like protease ___domain, amino acids 746-1060) with an N-terminal 10xHis-tag from CUSABIO. Briefly, fibrinogen diluted in PBS to 1 mg/mL was treated with 0.25 U thrombin or varying NSP3 concentrations, while fibrinogen in PBS without thrombin or NSP3 served as the negative control. Enzyme kinetics were monitored at 37 °C in a 96-well plate using a Multiskan FC plate reader, with absorbance at 425 nm recorded every minute for 2 h or until thrombin reached a plateau, indicating complete coagulation.

Peptide design and synthesis

All 15 RBD and 12 S1/S2-targeting-synthetic peptides of the SARS-CoV-2 S protein were computationally designed using InSiPS, as described previously11. Peptides were custom-synthesized by Shanghai Royobiotech at >95% purity and characterized using reverse-phase HPLC and MS analysis as per the company’s quality control standards. Lyophilized peptides were dissolved in dimethylsulfoxide (DMSO) to a final concentration of 0.5% (v/v) before use.

SARS-CoV-2 ‘S’pseudotyped lentivirus production

For transfection, 7.5 µg psPAX2 packaging plasmid, 750 ng pcDNA3.1-SARS2-Spike or E484K Spike-HiBiT envelope plasmid, and 7.5 µg pLV-eGFP (Addgene 36083) were used. Lentiviral supernatant was concentrated via sucrose cushion centrifugation to produce high-titer lentivirus particles. Briefly, 30 mL of supernatant was layered onto a 20% sucrose buffer (50 mM Tris-HCl, pH 7.4; 100 mM NaCl; 0.5 mM EDTA) at a 4:1 v/v ratio and centrifuged at 8000 x g for 3 h at 4 °C. The pellet was resuspended in 150 µL PBS, and pseudotype virus infectivity was assessed in HEK293T cells 24 h post-infection by counting GFP-positive cells under a microscope.

SARS-CoV-2 S-mediated pseudovirus entry assay

We developed a nanoluciferase complementation bioreporter-based cell assay to evaluate peptides targeting the RBD and S1/S2 regions of the S protein, aimed at blocking ACE2-RBD binding. HEK293T cells expressing pLEX-307-LgBiT were seeded in a black 96-well plate at 5 × 104 cells per well. After 24 h, these cells were treated with pseudotyped lentivirus (either pcDNA3.1-SARS2-Spike-HiBiT or pcDNA3.1-SARS2-E484K Spike-HiBiT), with or without 15 RBD or 12 S1/S2 peptides. Subsequently, 100 µL of furimazine (Chemshuttle), diluted in DMEM to 10 µM, was added to each well. Luminescence was recorded every 30 s for 1 h using an EnSpire multimode plate reader (Perkin Elmer), and results were presented in relative units over time.

Infection with authentic SARS-CoV-2

Huh-7 or Huh-7-ACE2 cells were seeded at 5 × 104 cells/mL in 96-well plates using serum-free DMEM. These cells were infected with an SARS-CoV-2 isolate at an MOI of 0.0001 pfu/cell, with or without specified doses of RBD-targeting peptides for 1 h at 37 °C. Post-infection, cells were washed three times with PBS and cultured in DMEM with 2% FBS. At 48 h post-infection, viral RNA was extracted from the supernatant for qRT-PCR, using primers to quantify intracellular SARS-CoV-2 RNA. mRNA levels of proinflammatory cytokines and inflammatory marker were also measured by qRT-PCR using KiCqStart SYBR Green Primers (Sigma-Aldrich, KSPQ12012). The efficacy of the peptides in inhibiting SARS-CoV-2 infectivity was evaluated using a plaque-forming assay. To assess the effect of RBD-targeting peptides on SARS-CoV-2 S protein expression, immunofluorescence was performed on SARS-CoV-2-infected Huh-7 cells treated with or without peptides. Cells were stained with an S protein antibody and Hoechst 33342 for nuclear staining. Fluorescence was measured using an EnSpire plate reader (Perkin Elmer) at an emission of 461 nm and excitation at 358 nm. S protein fluorescence in peptide-treated cells, normalized to untreated cells, were reported relative to DMSO control. The peptides’ inhibition of S protein entry was also assessed using immunofluorescence with infectious SARS-CoV-2 particles, with a primary antibody against the S protein (Sino Biological, 40150-R007) at a 1:400 dilution and a Goat anti-rabbit Alexa Fluor 488 (Abcam Ab150077) secondary antibody.

To assess viral replication and titers, IFIT5 knockout, along with IFIT5-VA CR wild-type and mutants in IFIT5 knockout A549-TMPRSS2-ACE2 cells infected with ancestral SARS-CoV-2 at MOI of 0.1 pfu/cell, were collected at various time points. Total RNA was extracted using the TRIzol method. Supernatants were used to determine viral titers, while cell pellets were resuspended in 30 µL of RNAase free water for viral replication analysis. qRT-PCR was conducted using the Luna Probe One-Step RT-qPCR Mix (New England Biolabs) with N1 and N2 primer-probe sets, following the manufacturer’s instructions. All reactions were performed in triplicate on an Applied Biosystems 7500 Fast Real-Time PCR detection system with automatic threshold and baseline settings. A standard curve was generated using a series of 5-fold dilutions starting from a 106 copies/uL standard. Using the standard curves for the N1 and N2 genes, mean cycle threshold (Ct) values from the samples were converted into absolute viral copy numbers.

IC50 determination of SARS-CoV-2 peptides by plaque assay

Vero E6-TMPRSS2 cells were seeded in 12-well plates at 5 ×105 cells/well and infected with ancestral or variant SARS-CoV-2 strains at MOI 0.0001 pfu/cell. The infection medium included DMEM with 1x non-essential amino acids, 10 mM HEPES, 2% FBS, and 50 U/mL penicillin and streptomycin, and varying doses of RBD or S1/S2-targeting peptides. After 1 h, the medium was replaced with MEM containing 10 mM HEPES and 1.2% Avicel RC-591 (DuPont), with suitable peptide dose. After 48 h, cells were fixed in 10% formaldehyde and stained with 0.5% (w/v) crystal violet. Plaques were counted and data plotted as percent inhibition vs. Log10 [peptide] in GraphPad Prism (ver. 10.1.1.323). IC50 values were calculated using a non-linear regression model.

Cytotoxicity of SARS-CoV-2 peptide inhibitors

Cell viability was evaluated using the CellTiter-Glo ATP content assay kit (Promega) or the Cell Counting Kit-8 (CCK-8, Sigma-Aldrich) following manufacturer’s guidelines. A549-AT and Vero E6-TMPRSS2 cells were seeded at 5 × 103 cells/well in 96-well plates and incubated overnight. RBD-targeting peptides of S protein, dissolved in DMSO, were applied at 5 µM and 50 µM. After a 24 h incubation at 37 °C with 5% CO2, the substrate was added, and cell viability was assessed by measuring ATP activity or dye uptake relative to DMSO control.

Surface plasmon resonance (SPR)

SPR kinetic experiments were conducted on an OpenSPR Rev4 system (Nicoya Lifesciences) to measure the binding affinities of InSiPS-generated peptides to wild-type and mutant SARS-CoV-2 S RBD proteins, including the Delta variant (T478K, L452R mutations), and the Omicron variant with multiple RBD mutations, each tagged with either 6x or 8x-His tags. Recombinant wild-type or mutant S RBD-His proteins were immobilized on an NTA sensor chip (Nicoya Lifesciences), and antiviral peptides were injected at specified concentrations into the SPR system. To evaluate the inhibition of SARS-CoV-2 S protein’s binding to ACE2, recombinant ACE2-FLAG protein was immobilized on a carboxyl sensor chip, and antiviral peptides along with SARS-CoV-2 wild-type or mutated S RBD-His proteins were injected into the SPR system. Measurements were taken at a flow rate of 40 μL/min with an immobilization time of 600 s at 20 °C using a running buffer of 10 mM HEPES (pH 7.5), 150 mM NaCl, 3 mM EDTA, and 0.005% (v/v) surfactant P20 (pH 7.4). Steady-state responses were plotted against time on a sensorgram to generate binding curves, and kinetic parameters (ka and kd) were analyzed by TraceDrawer ver. 1.9.2 software (Ridgeview instruments) to estimate the dissociation constant (KD).

Structural modeling

To assess whether candidate peptides bind the same SARS-CoV-2 S RBD residues and interface with the human ACE2 receptor, we employed a molecular docking strategy with trRosetta for predicting the 3D structures of SARS-CoV-2 peptide inhibitors. These structures were docked onto the trimeric S protein RBD (PDB ID: 7CWL) using ZDOCK server (ver. 3.0.2) to generate 10 scoring models and identify interface residues between the SARS-CoV-2 peptide inhibitors and the S protein trimeric RBD. The best model was chosen based on the lowest binding free energy and interaction specificity (i.e., salt bridges, van der Waals forces, and hydrogen bonds). Interface residues of the trimeric SARS-CoV-2 S protein-peptide inhibitor were mapped onto the crystal structure of the S RBD-ACE2 complex64. Similarly, we identified interface residues in the NSP3-IFIT5-ISG15 complex using the resolved NSP3 structure (PDB ID: 6YVA) for docking AlphaFold structures of ISG15 and IFIT5 via Z-Dock. Interface residues across all studied protein-peptide inhibitors and complexes were visually examined using PyMOL (ver. 3.4).

Web portal overview

The interactive web portal (https://babulab-uofr.shinyapps.io/scov2db/) provides access to SARS-CoV-2-human PPI data from AP-MS and CF-MS experiments. The portal is built with Shiny for Python and hosted on Shiny-Apps server. Multiprotein complexes are visualized using Pyvis and includes five sections: Project description, AP-MS, CF-MS, Merged network, Multiprotein complex, and Contact. The AP-MS tab allows users to explore high-confidence SARS-CoV-2–human PPIs, including those identified in wild-type (WT) and variant (Var) strains. Each SARS-CoV-2 bait protein is linked to its human prey protein, accompanied by a brief description, a direct link to the UniProt database, and information on the cell line source of the interaction. Additionally, the ‘Cell line score’ button allows users to retrieve the predicted interaction score for each cell line. Users can also extract high-confidence interactions from AP-MS studies for specific or all human cell lines. In the Spike WT/Var-human PPI tab, they can filter spike-human PPIs across all variants and wild-type strains or focus on specific viral strains. The CF-MS tab provides high-confidence SARS-CoV-2-human PPIs obtained from the saliva of COVID-19 patients infected with wild-type SARS-CoV-2, along with their predicted interaction scores. The Merged network tab presents all high-confidence SARS-CoV-2-human PPIs from both AP-MS and CF-MS datasets, with a “More info” button providing details on the source of each interaction. The Multiprotein complex tab displays predicted protein complexes containing both SARS-CoV-2 and human proteins, listing the complex ID, total number of subunits, and viral and host proteins forming the complex. A “Show network” button allows users to visualize and zoom in on each complex, highlighting SARS-CoV-2 and human proteins as well as drug targets as nodes, while the edges represent viral-host PPIs. Lastly, the Contact tab provides details on whom to reach out to for inquiries, additional data requests, or technical support.

Reporting summary

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