Construction and cross-cohort evaluation of a shared metabolic reprogramming–related gene signature in esophageal and lung squamous cell carcinoma

Abstract

We investigated directionally consistent metabolic alterations in esophageal squamous cell carcinoma (ESCC) and lung squamous cell carcinoma (LUSC) and developed a pooled prognostic risk score. TCGA and GTEx data were used to identify shared differentially expressed metabolic genes. Candidates were screened by cancer-type-stratified Cox regression and LASSO; the eight largest absolute coefficients among 10 genes retained at lambda.min formed the final score. Apparent TCGA performance was evaluated with a 1,000-resample full-pipeline bootstrap. Cancer-balanced, cross-cancer, and GEO analyses assessed robustness and transportability. Fourteen candidates derived from 137 strict shared metabolic aberrant genes yielded an eight-gene score. Higher scores were associated with poorer overall survival in TCGA-LUSC (HR 2.71, 95% CI 1.87–3.93) and the merged cohort (HR 2.72, 95% CI 1.92–3.85), but the association was borderline in TCGA-ESCC (P = 0.058). The optimism-corrected C-index was 0.652 and the corrected 3-year calibration slope was 0.841. Cancer-balanced and bidirectional analyses preserved the adverse direction but showed reduced stability. In GSE53622, cohort-standardized analysis supported transportability (HR 1.82, P = 0.032), whereas strict fixed-parameter transfer was attenuated (HR 1.39, P = 0.218). The complete model was not evaluable in GSE73403 because ADH1B was unavailable; the exploratory reduced seven-gene score showed a borderline association (P = 0.051). The pooled score showed moderate optimism-corrected performance and limited cross-cohort transportability. It did not establish equivalent effects across ESCC and LUSC or a platform-independent clinical model. Prospective and functional validation are required.