Abstract / Summary
Lung adenocarcinoma is shaped by both tumor-cell programs and the immune-stromal microenvironment, yet compact transcriptomic markers rarely capture these components in a biologically interpretable and technically reproducible manner. A four-gene score spanning B-cell, cytotoxic T-cell, Wnt-regulatory, and lymphatic-remodeling signals may provide a parsimonious summary of this ecological balance. To evaluate a fixed four-gene transcriptomic score for overall survival in lung adenocarcinoma and to examine its relationship with pathologic stage, common driver alterations, and observed anti-PD-1 response. RNA-seq expression and clinicopathologic data from TCGA-LUAD were obtained through UCSC Xena. The fixed predictor set comprised CD79A, DKK1, VEGFC, and CD8A. Standardized expression values were entered into a penalized multivariable Cox model. Prognostic performance was assessed by Kaplan–Meier analysis and five-fold cross-validation with fold-specific scaling. Stage and EGFR/KRAS/ALK subgroups were exploratory. The fixed coefficient directions were additionally examined in the pretreatment anti-PD-1 cohort GSE126044. Among 501 tumors (182 deaths), higher DKK1 and VEGFC expression was associated with higher hazard, whereas CD79A was associated with lower hazard; CD8A was not independently significant in the joint model. The continuous score was associated with overall survival (HR 2.84, 95% CI 2.16–3.75; P < 0.001), and the median-split groups differed by log-rank testing ( P < 0.001). Mean held-out Harrell C-index was 0.673. The score increased modestly with pathologic stage (Spearman r = 0.17, P < 0.001) but did not differ across available driver groups ( P = 0.600). In GSE126044 (5 responders and 11 non-responders), the score showed a non-significant tendency toward higher values in non-responders (Mann–Whitney P = 0.069; descriptive AUC = 0.80). The four-gene score was associated with survival in TCGA-LUAD and showed moderate internal discrimination. Its relationship with anti-PD-1 response remains hypothesis-generating because the treated cohort was small and the group comparison was not statistically significant. Prospective, spatially resolved, and platform-harmonized validation is required.