A comparative study of COVID-19 transcriptional signatures between clinical samples and preclinical cell models in the search for disease master regulators and drug repositioning candidates

SARS-CoV-2 has caused a global pandemic with high morbidity and mortality. Although vaccination has reduced severe outcomes, emerging variants (e.g., Omicron) with increased transmissibility and immune evasion sustain ongoing disease burden and highlight the need for effective outpatient therapeutics. Many repurposed drugs identified via empirical or in vitro strategies have failed to translate clinically, partly due to the complexity of COVID-19 pathophysiology and limitations of preclinical models. The study aims to characterize host transcriptional signatures of severe COVID-19 from clinical lung autopsies, infer disease master regulators (MRs), and use these to prioritize drug repositioning candidates via Connectivity Map (CMap). It further evaluates how well preclinical cell models and patients’ nasopharyngeal swabs recapitulate severe lung signatures by comparing differential expression, gene ontology (GO), MRs, and drug predictions, to guide translational applicability of models.

The article notes that while antivirals such as Remdesivir and Paxlovid show promise, their efficacy may be undermined by viral mutations conferring resistance in vitro or potentially in the clinic. Prior drug repositioning efforts often relied on empirical screening or theoretical models, with mixed clinical success, underscoring the need for holistic, systems-level approaches. Immunomodulators (e.g., corticosteroids like dexamethasone and cytokine inhibitors such as tocilizumab) have demonstrated benefit in severe COVID-19. Antibiotics and psychoanaleptics have been proposed for their anti-inflammatory or antiviral effects, although evidence remains inconsistent and concerns about antimicrobial resistance persist. Targeting host transcriptional changes may offer variant-agnostic therapeutic strategies and improve translation across coronaviruses. The study builds on master regulator-based approaches previously applied in cancer and neurodegenerative diseases for mechanism elucidation and drug repurposing.

- Data sources: Public RNA-seq datasets from GEO. Clinical severe lung autopsies (GSE155241: 3 COVID-19 deceased vs 3 healthy), nasopharyngeal swabs (GSE152075: 430 SARS-CoV-2 positive vs 54 negative), human cell lines (GSE147507: A549 cells infected with SARS-CoV-2 at MOI 0.2 and 2; NHBE cells at MOI 2, 24 h), and Vero cells (GSE159316: MOI 0.01, 24 h and 48 h). - Processing: For GSE159316, GSE147507, GSE155241, raw data QC via fastqcr and quantification with Salmon; GSE152075 used provided counts. Differential expression performed with DESeq2 (R 4.1.0); DEGs defined at FDR q ≤ 0.05. - Gene Ontology: Enrichment of overrepresented Biological Processes using ClueGO (Cytoscape 3.8.2), hypergeometric test with q ≤ 0.05. - Master Regulator analysis: Built upon a transcription factor-centered healthy lung regulatory network (from GSE23546; De Bastiani & Klamt 2019) inferred with RTN, mapping TFs to regulons. Enrichment of each TF regulon for DEGs assessed by hypergeometric test (q ≤ 0.05). Only TFs whose regulons had ≥100 target gene hits were considered biologically relevant MRs. - Connectivity Map (CMap) drug repositioning: Implemented with PharmacoGx. For each dataset, MR regulon target profiles were compared to perturbational signatures from drug-treated cell lines using GSEA-derived connectivity scores. Candidates were drugs with negative connectivity scores (opposing the disease signature); significance assessed by permutation test (n = 1000), q ≤ 0.05. - Similarity analyses: Compared GO terms, MR sets, and CMap-predicted drugs between models (cell lines, nasopharyngeal swabs) and clinical severe lung samples using Jaccard index and Fisher’s exact tests (GeneOverlap). Intersections documented in supplemental files; UpSet plots generated via Intervene.

- Severe lung autopsies (GSE155241): 299 DEGs (q ≤ 0.05). Most altered included HLA-DQB1, TNFSF12/TNFSF13, FAP, CYP1A1 (|log2FC| > 4), and downregulated DENND11, EIF3CL (log2FC < −4). Enriched processes involved juxtacrine signaling, ECM remodeling, cytoskeleton, aggrephagy, innate immune response, cytokine production, steroid response, DNA damage, and signaling via integrin, GPCR, NFκB, RAS/MAPK/ERK pathways. - Severe COVID-19 master regulators: Six TFs with enriched regulons—TAL1, TEAD4, EPAS1 (HIF2A), ATOH8, ERG, ARNTL2—linked to inflammation regulation, cell morphogenesis, cytoskeletal dynamics, and hypoxic responses. - CMap for severe COVID-19 lung: 52 drugs with negative connectivity (q ≤ 0.05). Main ATC classes: corticosteroids (e.g., beclometasone, betamethasone, flunisolide, fluocinonide, fluorometholone), antibiotics (cephalosporins, lincosamides/macrolides/streptogramins, fluoroquinolones; e.g., cefalotin, cefixime, clindamycin, erythromycin, lomefloxacin, imipenem), psychoanaleptics (e.g., amoxapine, nialamide, piracetam), cardiovascular agents (bisoprolol, verapamil), antihistamines (cetirizine), antifungals (griseofulvin, sertaconazole), antiparasitics (clofazimine, emetine, diethylcarbamazine), and others (propylthiouracil, sulfinpyrazone, cimetidine, oxybutynin, puromycin). - Preclinical models and nasopharyngeal swabs DEGs: • A549 MOI 2: 7,494 DEGs; top: EGR1, BHLHE41, PPP4R4, KRT4, EPHX1, UPK1B (|log2FC| > 3). • A549 MOI 0.2: 3,874 DEGs; top: EGR1, ZNF354B, FOSB, LY6E, CA9, CDH2 (|log2FC| > 4). • NHBE (MOI 2, 24 h): 884 DEGs; top: CXCL5, CCL20, CXCL8, MTRNR2L3, SPATA13, TRAPPC3 (|log2FC| > 1). • Vero 24 h: 2,609 DEGs; top: OASL, CXCL10, G0S2, PENK, MMP10, SPP1 (|log2FC| > 2). • Vero 48 h: 1,859 DEGs; top: MRC1, OASL, RYR1, PENK, S100A2, ID3 (|log2FC| > 2). • Nasopharyngeal swabs: 5,396 DEGs; top: CASP17P, PCSK1N, AL022578.1, IGHG1, IGHG3, IGHM (|log2FC| > 4). • Only four genes were consistently identified as DEGs across all models (MAFF, CSRNP1, NFKBIA, DUSP1), with no uniform directionality. - GO enrichment: All models showed immune regulation, cell death, and stress responses; dataset-specific processes included oxidative/redox metabolism (A549), nitric oxide biosynthesis and acute-phase response (NHBE), stress granules and p53 signaling (Vero 24 h), chromatin and lipid metabolism (Vero 48 h), ER stress and juxtacrine signaling (swabs). - MR counts per dataset: A549 MOI 2: 33; A549 MOI 0.2: 28; NHBE: 4; Vero 24 h: 20; Vero 48 h: 16; swabs: 29. - CMap candidates per dataset: NHBE: 118; A549 MOI 2: 101; A549 MOI 0.2: 75; Vero 24 h: 30; Vero 48 h: 98; swabs: 105. - Similarity with severe lung autopsies: NHBE showed the highest similarity among cell models—Jaccard GO 0.13, MR 0.11, CMap 0.02. NHBE and severe lungs shared >200 GO terms, one shared MR (TEAD4), and four drugs (nialamide, amoxapine, bisoprolol, puromycin), though drug overlap was not statistically significant. - Similarity with nasopharyngeal swabs: A549 MOI 2 had highest similarity—Jaccard GO 0.33, MR 0.35, CMap 0.17; >800 shared GO terms, 18 shared MRs, and five shared drugs (stachydrine, mycophenolic acid, tolmetin, GW-8510, staurosporine). These five drugs were also predicted across all other cell models. Consensus MRs between swabs and all models (except NHBE) included YBX1, ATF2, GABPA, TRIM28, GOLGB1, and MAFG; NHBE shared only MAFG.

The study identifies master regulators underlying severe COVID-19 lung pathology, emphasizing dysregulated inflammation, cytoskeletal remodeling, and hypoxic adaptation. MRs TAL1, TEAD4, ARNTL2, and ERG relate to cytokine signaling and inflammatory balance, while EPAS1 (HIF2A) and ATOH8 link to morphogenesis and hypoxia-associated pathways, aligning with known hyperinflammation and tissue damage in severe disease. The MR-centric CMap approach reprioritized therapeutics that could invert the disease transcriptional program. Notably, many predicted drugs fall into clinically relevant classes that have shown benefit or mechanistic plausibility (e.g., corticosteroids, selected cardiovascular agents), supporting the validity of a host-targeted strategy that is less sensitive to viral variants. Comparative analyses revealed that NHBE cells best approximate the lung autopsy transcriptional landscape at the level of GO and MR similarity, whereas A549 models more closely reflect upper airway (swab) signatures. Vero cells, while useful for virology, were less concordant with human clinical responses. These insights guide selection of preclinical models according to the clinical context and endpoint (mechanism vs. drug screening), and support the utility of MR-based repositioning to capture disease-level regulatory perturbations.

This work proposes and applies a systems-level, MR-driven framework to characterize severe COVID-19 host responses in lung tissue and prioritize drug repositioning candidates capable of reversing disease signatures. It highlights immunomodulatory agents as particularly relevant, and suggests additional candidates across antibiotics, psychoanaleptics, beta-blockers, and calcium channel blockers for further evaluation. The translational assessment indicates NHBE cells are more suitable for modeling lung tissue biology, while A549 cells better mirror upper airway epithelial responses; however, no cellular model yielded statistically reliable overlap in drug predictions with severe lung clinical data. Future work should expand clinical sample sizes, validate MR targets experimentally, and conduct rigorous preclinical and clinical testing of prioritized candidates to improve translational success.

- Limited number of severe lung autopsy samples (3 cases vs 3 controls) constrains statistical power and generalizability. - In silico MR inference and CMap predictions lack experimental validation in this study. - Heterogeneity across datasets (platforms, infection MOI, timepoints, species) may affect comparability. - Drug overlap (CMap) between NHBE and severe lungs showed low Jaccard similarity and lacked statistical significance. - Vero cell findings have limited translational relevance to human lung responses. - Antibiotic and psychoanaleptic repositioning remains controversial due to mixed clinical evidence and antimicrobial resistance concerns.