Abstract 2996: The prognostic value of a cell cycle proliferation gene signature in bladder cancer

Abstract

Abstract An important aspect of personalized medicine involves the identification of high risk patients likely to benefit from neoadjuvant or adjuvant therapy. In the case of bladder cancer, this entails identifying the 20-30% of patients with non-muscle invasive (NMI) tumors that will progress to muscle-invasive (MI) disease, and the approximately 50% of patients diagnosed with MI tumors who do not survive past 5 years. In breast cancer, independently derived prognostic gene signatures are enriched with cell cycle proliferation (CCP) genes. Recently, a CCP score obtained from the average expression level of 31 genes was found to predict recurrence free and disease specific survival (DSS) in prostate cancer. Here, we assess the importance of CCP correlated genes for prognosis in bladder cancer and the prognostic value of the CCP score in 8 publicly available and heterogeneous microarray datasets (N = 762), with samples obtained from both cystectomy and transurethral resection of the bladder (TURBT) and profiled on 7 microarray platforms. Clusters of correlated and predictive genes for stage, grade, and DSS were identified using an iterative subtractive method. A gene set enrichment analysis on the predictive clusters found that CCP was the only consistently enriched class of genes (P < 0.05) for each clinical variable. We then analyzed the extent in which the performance of predictive gene signatures, composed of the top 10-100 genes in step sizes of 10, depended on genes that correlated with the CCP gene signature. When the gene expression data was adjusted for the 31 CCP genes, the predictive ability of the gene signatures was significantly and negatively impaired in most datasets. We next evaluated the CCP score for its prognostic value. Without optimizing the CCP signature for bladder cancer, we found the CCP score to be significantly higher in MI compared to NMI tumors in 3/4 cystectomy datasets (P < 0.05), with AUC ranging from 0.62 to 0.84, and in 2/4 TURBT datasets (P < 0.05) with AUC ranging from 0.60 to 0.75. CCP score was significantly higher in high grade tumors compared to low grade tumors in 3/4 cystectomy datasets (P < 0.05), with AUC ranging from 0.63 to 0.88, and in all 5 TURBT datasets (P < 0.05), with AUC ranging from 0.76 to 0.91. Most notably, although nodal staging was the most consistent predictor of DSS in cystectomy datasets (P < 0.05), CCP score was the only clinical variable marginally predictive of DSS (P < 0.10) in multiple cystectomy datasets when only preoperative variables were considered. In TURBT datasets, CCP score was the only variable predictive of DSS (P < 0.05) in multiple datasets. Finally, CCP score was predictive of NMI to MI progression in low grade tumors (AUC = 0.73, P = 0.020). Altogether, these results indicate that the prognostic value of gene signatures in bladder cancer depend on genes that correlate with CCP and that optimization of the CCP signature may yield clinically relevant prognostic markers in bladder cancer. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 2996. doi:1538-7445.AM2012-2996