Sequencing based dissection of inflammation associated natural killer cell remodeling in lung adenocarcinoma identifies state biomarkers and therapeutic vulnerabilities

Background Chronic inflammation may shift tissue-protective immunity toward a tumor-permissive microenvironment, but natural killer (NK)-cell remodeling in lung adenocarcinoma (LUAD) remains incompletely characterized. We investigated whether NK-cell depletion and state redistribution are reproducible tumor-associated features with biomarker or therapeutic relevance. Methods Two patient-matched single-cell RNA-sequencing cohorts (GSE131907 and E-MTAB-13526) were analyzed using patient-level paired tests. We assessed NK-cell abundance within the immune compartment, CD16-positive terminal and CD16-negative tissue-adapted fractions within NK cells, and pseudobulk expression. Chromatin accessibility was evaluated using GSE304741 GEX–ATAC data. TCGA-LUAD, Visium spatial transcriptomics, and six pretreatment immune checkpoint inhibitor (ICI) cohorts provided external assessment. Western blots in NK-92 cells examined SMAD2 phosphorylation and perforin/granzyme B expression after TGF-β1 stimulation with or without SB431542. Results In GSE131907, NK-cell abundance and terminal-state fractions declined in all 10 matched patients, whereas tissue-adapted fractions increased (all P = 0.00293); these patterns were independently replicated. Pseudobulk effects showed moderate genome-wide concordance (Spearman’s ρ = 0.350), with 112/116 discovery-significant genes directionally concordant and 58 significant in both cohorts (FDR <0.05). In GSE304741, 15/19 prespecified promoters showed differential accessibility, with strong RNA–chromatin concordance (ρ = 0.851). TGF-β, hypoxia, MAPK, EGFR, and estrogen-related signatures changed concordantly across analytical contexts. TCGA-LUAD supported reduced inferred NK abundance and terminal cytotoxic programs. In three LUAD/nonsquamous NSCLC ICI cohorts, clinical benefit was associated with higher terminal cytotoxic activity (Hedges’ g = 0.445, P = 0.0319) and lower abundance-adjusted state balance (g = −0.487, P = 0.0192), but not NK abundance. Visium showed a nonsignificant NK-abundance decrease in four of five patients; NK-state estimates were method-dependent. Qualitative NK-92 immunoblots suggested increased SMAD2 phosphorylation and reduced perforin/granzyme B after TGF-β1, attenuated by SB431542. Conclusion LUAD exhibits reproducible NK-cell depletion accompanied by loss of terminal-state representation and relative enrichment of tissue-adapted states. Terminal cytotoxic programs and abundance-adjusted state balance warrant investigation as biopsy-based biomarkers, whereas KLRC1/NKG2A and TGF-β–hypoxia signaling remain therapeutic hypotheses requiring further validation. These findings do not establish a longitudinal premalignant trajectory.