A frozen human wound-repair reference reveals heterogeneous tumor niches and conditional immunotherapy associations
Purpose Tumors are often framed as unresolved wounds, but the extent to which independently defined physiological repair programs are recontextualized across cancers remains uncertain. Methods We derived five lineage-associated modules and a composite score from 58,823 cells in a longitudinal human skin-wound cohort before examining tumors. Frozen modules were evaluated in paired hepatocellular carcinoma and lung adenocarcinoma single-cell datasets, spatial and cell-resolved non-small-cell lung cancer cohorts, and four pretreatment immune-checkpoint-inhibitor cohorts. Patient- or section-level inference and prespecified sensitivity analyses were retained. Results Myeloid repair modules showed the most consistent tumor-associated increase in paired cancers; other lineage effects varied by context. Repair-high non-small-cell lung cancer regions combined epithelial/proliferative, extracellular-matrix fibroblast, and C1QC/broad-myeloid programs, whereas repair-cytotoxic neighborhood correlations were split across sections. In six Prime 5K Xenium patients, repair-high regions were fibroblast enriched (+0.239 high-minus-low fraction; 6/6 concordant; exact sign-flip P=0.031; FDR=0.047) and T/NK depleted (-0.151; 6/6; FDR=0.047). Direct ligand-receptor neighbor effects were not reproducible. Repair_FBI showed an adverse ICI-response direction, but the random-effects estimate was imprecise (OR=1.69, 95% CI 0.81-3.51). Conclusion Tumors reuse elements of physiological repair to form heterogeneous multicellular niches. Fibroblast remodeling is the most stable candidate; the evidence supports conditional association, not universal immune exclusion, a senescence-defined state, or a generlizable ICI biomarker.