Multi-cohort transcriptomics and donor-aware single-cell analyses prioritize S100A8 within an emergency-myeloid, HLA-DR-low state in sepsis.

Sepsis is heterogeneous, and transcripts that distinguish sepsis from health may lack clinical specificity. We tested whether S100A8 is reproducibly associated with sepsis and maps to a donor-aware myeloid state defined independently of S100A8. A discovery whole-blood cohort was analyzed on the log2 robust multi-array average scale using maximum-probe aggregation, limma, weighted gene co-expression network analysis (WGCNA), and leakage-controlled feature prioritization. The analysis identified 1,508 differentially expressed genes; 265 overlapped the strongest sepsis-positive WGCNA module. The network retained all 802 samples and 11,719 genes; the black-module eigengene correlated with sepsis status (r = 0.409; false-discovery rate 3.72 × 10 - 33), and S100A8 belonged to this module (module membership 0.700). S100A8 was upregulated (log2 fold change 1.87; false-discovery rate 2.15 × 10 - 171). Discrimination against clinical comparators was contrast dependent, and outcome analyses did not support a stable prognostic association. Donor-level monocyte pseudo-bulk analysis supported increased S100A8 and decreased HLA-DRA but included only two control and four sepsis donors. Across single-cell datasets, S100A8 tracked an emergency-myeloid program and inversely tracked antigen presentation. In THP-1 cells, S100A8 knockdown attenuated lipopolysaccharide-associated reductions in CD74 and HLA-DRα. Triptolide retained a similar NF-κB phosphorylation effect after S100A8 knockdown, providing no evidence that this effect required S100A8. S100A8 is therefore prioritized as a marker of a sepsis-associated emergency-myeloid, HLA-DR-low state, not as an established diagnostic, prognostic, direct drug-binding, or therapeutic target.