Systems immunology of transcriptional responses to viral infection identifies conserved antiviral pathways across macaques and humans

The study addresses how well non-human primate (NHP; macaque) models recapitulate human antiviral immune responses at the transcriptional level across diverse RNA viruses. Human data for highly lethal or emerging viruses are sparse and early infection time points are difficult to capture ethically and logistically. Macaques enable controlled challenge studies with longitudinal sampling, but the extent of cross-species conservation of antiviral responses remains unclear. Building on prior work that identified a conserved human pan-viral transcriptional response (the meta-virus signature, MVS) that distinguishes infection and associates with severity, the authors test whether this response is conserved in macaques, define virus- and family-conserved responses in NHPs, characterize temporal dynamics across viral families, and evaluate translatability of macaque-derived signatures to human cohorts. The overarching hypothesis is that macaques and humans share conserved antiviral pathways detectable in blood transcriptomes that can generalize across viral families and disease stages.

The introduction situates the work in the context of frequent emergence of RNA viruses (Flaviviridae, Coronaviridae, Orthomyxoviridae, etc.) and the utility of macaque models for pathogenesis, vaccines, and therapeutics when human challenge studies are infeasible. Prior research by the group identified a conserved human meta-virus signature (MVS) spanning multiple respiratory viruses that predicts infection and severity. Interferon-stimulated genes (ISGs) are known to be conserved but do not capture the full systemic immune response; MVS integrates innate and adaptive cell-type signals. Limitations in available human transcriptomic data for deadly viruses (e.g., Lassa, Marburg) and early time points motivate leveraging NHP datasets. The authors also draw on literature implicating myeloid cells and type I interferon pathways as central to antiviral responses, crosstalk between RNA and DNA sensing pathways, and known variability in T cell responses and exhaustion across viral infections (e.g., COVID-19, Ebola, dengue).

- Data collection and preprocessing (NHP): 21 blood transcriptomic datasets (microarray and RNA-seq) from macaques (rhesus, cynomolgus, pig-tailed), totaling 743 samples from 198 animals infected with 13 viruses across five RNA virus families (Arenaviridae, Coronaviridae, Filoviridae, Flaviviridae, Orthomyxoviridae). Datasets included varied infection routes/doses and platforms. Time points were binned: T0 (pre-infection), T1 (days 1–2), T2 (days 3–5), T3 (days 6–8), T4 (days 9–13), T5 (day 14+). NHP datasets were processed individually (arrays used processed data; RNA-seq aligned to macaque genomes with Salmon, gene-level summarized via Tximport, normalized via DESeq2 VST). All genes were mapped to human orthologs for integrative analyses. - Baseline comparability: Pairwise correlations of mean/median expression across 3,055 shared genes showed no baseline expression differences across macaque species beyond between-dataset variability. - Human datasets: 47 datasets comprising 5,345 blood/PBMC samples across 14+ viral infections; processed with log2/quantile normalization as needed; some collections co-normalized using COCONUT where applicable; otherwise analyzed individually to minimize gene loss. - Scoring and signatures: The previously validated human MVS score (geometric mean of overexpressed genes minus underexpressed genes) was computed per sample. Additional gene sets: ISGs, MS1 (MDSC-like), T cell activation, and MHC class II were analyzed. - Single-cell RNA-seq: Analyzed macaque Ebola scRNA-seq blood dataset (GSE158390; 56,929 cells, 17 macaques) with existing cell-type annotations; per-cell MVS and module scores computed, averaged by cell type and day post infection. Additional human scRNA-seq PBMC datasets for severe COVID-19 and severe dengue were processed with Seurat/scanpy, annotated, and differential expression performed per infected individual vs healthy by cell type; BTMs enrichment applied to up- and downregulated gene sets. - Unbiased DEG analyses and meta-signatures: Within each viral family, differentially expressed genes at each time bin vs T0 were identified (FDR < 0.05, |effect size| > 0.1) to define peak-response bins. Signatures per viral family and an across-family Viral Response Signature (VRS) were derived using multi-cohort meta-analysis (combining effect sizes and p values). Thresholds were tuned (effect size ≥ 0.6, FDR 0.05–0.0001) to yield ~200-gene signatures. For VRS, leave-one-dataset-out meta-analysis mitigated bias from large datasets. Signature scores were scaled within datasets. - Temporal dynamics: Mixed-effects models (lmerTest) modeled MVS over time with virus-family interactions: MVS ~ Time + Time^2 + Virus_Family + Time*Virus_Family + Time^2*Virus_Family + (1 + Time | Subject). Human challenge studies (influenza, rhinovirus, RSV) and NHP datasets were analyzed for kinetics; comparisons between species for Orthomyxoviridae performed. - Pathway analysis: Blood Transcriptional Modules (BTMs) overrepresentation analyses were performed separately on up- and downregulated signature genes and on DEGs at peak time points; Bonferroni-adjusted p values. Circos plots visualized shared enriched modules across viral families. - Performance evaluation: AUROC used to distinguish infected vs healthy; associations with severity/risk factors evaluated via ANCOVA, Wilcoxon tests, and Jonckheere-Terpstra trend tests; correlations via Spearman. - Software: R (ggplot2, ComplexHeatmap, lmerTest, Seurat), scanpy, Salmon, DESeq2, Tximport.

- Human MVS conserves in macaques: Across all 21 NHP datasets, MVS scores significantly increased at peak infection vs baseline (p adj < 0.001) and distinguished infected from uninfected with AUROC ≥ ~0.8 across viruses and families. - Severity associations in NHPs: MVS scores associated with known severity risk factors (p < 0.04), including unvaccinated vs vaccinated Machupo infection, Ebola Mayinga vs Makona strains, old vs young influenza-infected macaques, and live vs inactivated influenza. - Cell-type drivers: In macaque Ebola scRNA-seq, MVS was preferentially expressed in myeloid cells across time, paralleling human COVID-19 findings. Longitudinally, myeloid ISGs increased (r = 0.93, p < 1e-4), MS1 (MDSC-like) genes increased (r = 0.69, p < 1e-4), and MHC class II decreased (r = -0.61, p < 6e-4). T cell activation genes were downregulated in CD4 and CD8 T cells (r < -0.5, p < 0.008). - Temporal dynamics differ by viral family: In human challenge data, MVS peaked earlier for influenza and rhinovirus (days 1–5) and later for RSV (days 3–7). In NHPs, Orthomyxoviridae peaked early (1–3 DPI) while Arenaviridae and Filoviridae showed later peaks (e.g., largest differences at T3). Mixed-effects modeling showed significant virus-family interactions with time (e.g., Arenaviridae, Filoviridae vs Orthomyxoviridae p ≤ 0.01); species had no significant effect when comparing human and NHP Orthomyxoviridae. - Unbiased NHP-derived signatures: A macaque Viral Response Signature (VRS) and family-specific signatures were derived. Despite minimal gene overlap across family-specific signatures, enriched BTMs overlapped extensively, highlighting conserved innate/myeloid pathway activation. Each family-specific signature generalized to other families (AUROC ≥ 0.75). - Translation to humans: The NHP-derived VRS robustly distinguished symptomatic human viral infections across 20 datasets (n = 3,183), including SARS-CoV-2, Ebola, dengue (all p adj < 0.0001), and even chikungunya (p adj < 0.0001), which was not in discovery. VRS also detected acute non-ssRNA infections (adenovirus, rotavirus), latent EBV (but not latent HCMV), and chronic HIV, HBV, and HCV (p ≤ 0.05–0.01 depending on cohort). - VRS vs MVS: Although sharing only 17 genes (Jaccard index ~0.03), VRS and MVS scores correlated across NHP (r ≥ 0.54, p < 1.4e-8) and human data (r ≥ 0.42, p < 1.1e-8). Both showed upregulated innate/antiviral modules and downregulated adaptive/lymphoid modules; VRS additionally captured signal transduction and eIF3-related downregulation. - Lymphoid/T cell differences by virus: Module 4 (lymphoid) relationships to VRS varied by family. In macaques, Module 4 inversely tracked VRS over time for non-Flaviviridae but showed inconsistent relation in Flaviviridae. In humans, the negative correlation between Module 4 and VRS was weaker in dengue and Ebola than in chikungunya or SARS-CoV-2. Severity analyses showed Module 4 negatively associated with severity in SARS-CoV-2 and chikungunya but not significantly in dengue. Human scRNA-seq indicated CD8 T effector/cytotoxic genes (e.g., GZMB, NKG7, PRF1) were upregulated in severe dengue but downregulated in severe COVID-19; BTMs reflected upregulated T cell activation/differentiation in severe dengue vs downregulation in severe COVID-19.

The study demonstrates that macaque blood transcriptomic responses to acute RNA virus infection robustly recapitulate conserved human antiviral programs. The validated human MVS generalized to NHPs and associated with markers of severe disease, supporting macaques as a faithful model for human antiviral immunology. Temporal kinetics of the conserved response differ by viral families in both species, implying virus-specific biology (replication site/dynamics, immune evasion, tissue tropism, and latency) shapes the timing of systemic responses. An unbiased NHP-derived VRS captured conserved innate/myeloid pathways and generalized across diverse human infections, including DNA viruses and chronic infections, highlighting broad shared host responses and crosstalk among nucleic acid sensing pathways. Differences in lymphoid/T cell module behavior across viruses, particularly between Flaviviridae and respiratory/hemorrhagic viruses, suggest virus-specific T cell dynamics (activation vs exhaustion) may contribute to disease heterogeneity and inform vaccine or immunomodulatory strategies. Overall, the findings validate NHP models for predicting human responses to emerging pathogens and for identifying conserved antiviral targets.

This work provides the largest cross-species, cross-virus transcriptomic analysis to date, showing: (1) a conserved antiviral transcriptional response shared by macaques and humans; (2) distinct temporal kinetics by viral family; (3) an NHP-derived Viral Response Signature that generalizes to acute and chronic human infections across multiple virus classes; and (4) virus-dependent differences in T cell/lymphoid responses, especially between dengue and COVID-19. These insights support using macaque models for pandemic preparedness and to guide timing of diagnostics and host-directed therapies. Future studies should: expand datasets to additional viruses and variants; integrate tissue-specific and longitudinal single-cell/multi-omic profiling; dissect antigen-specific T cell subsets and functions across viruses; and evaluate conserved pathways (e.g., myeloid/IFN programs, translational control) as targets for broad-spectrum immunomodulatory interventions.

- Microarray coverage: Not all genes in target gene sets were measured in some microarray datasets; analyses used available subsets. Prior work suggests redundancy among signature genes mitigates this limitation. - Ortholog mapping: Macaque data were mapped to human orthologs, potentially excluding macaque-specific genes without clear human homologs that may influence antiviral dynamics. - Sample source: Only blood/PBMC transcriptomes were analyzed; site-specific/tissue responses were not assessed. - Viral representation: Some viral families had limited species/variant representation; family-level patterns may not generalize to all members. Certain viruses differ in pathogenicity across species (e.g., HIV in humans vs SIV in macaques). - Data availability: scRNA-seq comparisons across species for the same virus were not possible due to limited datasets; cross-virus comparisons (e.g., macaque Ebola vs human COVID-19) were used instead. - Heterogeneity: Technical and biological heterogeneity (platforms, infection routes/doses, sample processing) was incorporated but may introduce residual confounding despite meta-analytic approaches.