Tumor-suppressor signature for robust prognosis and prediction in adenocarcinoma non-small cell lung cancer

Abstract

Non-small cell lung cancer (NSCLC) remains the leading cause of cancer-related mortality, partly due to limited early detection strategies and incomplete understanding of tumor-suppressive mechanisms. In our previous work, we identified 26 tumor suppressor (TS) genes and characterized their biological functions and regulatory networks. We systematically evaluated these TS genes across multiple microarray datasets by analyzing differential expression patterns, correlations with oncogenes, tumor-associated genes, and PD-1-related immune genes, as well as somatic mutation frequencies. A weighted scoring algorithm was used to construct a TS gene signature. Patients were stratified using z-score normalization, and univariable and multivariable Cox proportional hazards models were applied across multiple adenocarcinoma (ADC) cohorts. Prognostic performance was assessed using Kaplan-Meier analysis and AUC metrics. The 26 TS genes were consistently down-regulated in tumors and showed strong negative correlations with oncogenes, particularly in advanced stages. TS genes also exhibited stage-dependent correlations with PD-1-associated immune genes, with chemotaxis/cytokine-signaling genes behaving TS-like, while PDCD1 and SIT1 showed oncogene-like patterns. Somatic mutations were detected in only 32% of LUAD samples for TS genes, compared with 68% for oncogenes. Across seven independent ADC cohorts, high TS-signature expression was associated with significantly reduced risk of death and recurrence/relapse. The TS signature outperformed several published prognostic signatures and demonstrated robust predictive accuracy, with AUC values exceeding 0.7 in multiple datasets and >0.8 for relapse prediction in GSE30219. Across seven independent cohorts, high TS signature expression was consistently associated with significantly reduced risk of death or recurrence/relapse in ADC patients. Patients with high TS signature expression exhibited markedly improved survival probabilities compared with those with low expression. When benchmarked against several established prognostic signatures using AUC metrics, our TS signature demonstrated superior robustness and predictive accuracy for ADC prognosis.