Robust prognostic gene expression signatures in bladder cancer and lung adenocarcinoma depend on cell cycle related genes.

Few prognostic biomarkers are approved for clinical use primarily because their initial performance cannot be repeated in independent datasets. We posited that robust biomarkers could be obtained by identifying deregulated biological processes shared among tumor types having a common etiology. We performed a gene set enrichment analysis in 20 publicly available gene expression datasets comprising 1968 patients having one of the three most common tobacco-related cancers (lung, bladder, head and neck) and identified cell cycle related genes as the most consistently prognostic class of biomarkers in bladder (BL) and lung adenocarcinoma (LUAD). We also found the prognostic value of 13 of 14 published BL and LUAD signatures were dependent on cell cycle related genes, supporting the importance of cell cycle related biomarkers for prognosis. Interestingly, no prognostic gene classes were identified in squamous cell lung carcinoma or head and neck squamous cell carcinoma. Next, a specific 31 gene cell cycle proliferation (CCP) signature, previously derived in prostate tumors was evaluated and found predictive of outcome in BL and LUAD cohorts in univariate and multivariate analyses. Specifically, CCP score significantly enhanced the predictive ability of multivariate models based on standard clinical variables for progression in BL patients and survival in LUAD patients in multiple cohorts. We then generated random CCP signatures of various sizes and found sets of 10-15 genes had robust performance in these BL and LUAD cohorts, a finding that was confirmed in an independent cohort. Our work characterizes the importance of cell cycle related genes in prognostic signatures for BL and LUAD patients and identifies a specific signature likely to survive additional validation.