Abstract 4817: A five-gene signature predictive of survival in lung adenocarcinoma but not in squamous cell carcinoma

Abstract

Abstract Identification of novel molecular markers for lung carcinogenesis, outcome and response to therapy is expected to improve the clinical management of lung cancer such as non-small cell lung cancer (NSCLC). Previously, we have derived gene expression signatures indicative of differential gene expression among cells constituting an in vitro model of human lung carcinogenesis and relevant to survival in NSCLC. In this study we assess the prognostic efficacy of our previously described six genes by using several prediction algorithms and a leave-one-out-cross-validation (LOOCV) strategy as well as risk-score prediction models. The NCI Director's Challenge datasets (n=443) were used as a training set and gene expression data of adenocarcinomas from the Duke and Harvard cohorts (DH cohort; n=183) were pooled as a validation set. In addition, two independent published datasets comprised of 130 and 58 lung squamous cell carcinomas (SCCs) served as a SCC validation cohort. A Five-gene in vitro lung carcinogenesis model signature (FILM) classifier was derived and found to be superior in prediction as the lowest specificity or sensitivity of the six prediction algorithms was 0.943. Importantly, all six prediction algorithms showed that the overall survival of all-stage or stage-I only human lung adenocarcinoma patients in the DH validation cohort that expressed FILM was significantly poorer than that of patients predicted to lack the signature. Moreover, no differences in overall survival between lung SCCs predicted to express or lack FILM were observed demonstrating the prognostic specificity of this classifier for lung adenocarcinomas. We then developed a risk score-prediction model for lung adenocarcinoma based on the Cox regression coefficients and expression of FILM genes. Lung adenocarcinoma patients identified to be at high risk based on the FILM risk model exhibited significantly worse survival (p=5.4 × 10-7, 100 months follow-up) than patients at low risk. For validation of the FILM risk model, Cox regression coefficients and the dichotomization cut-off threshold generated from the training cohort (n=443) were directly applied to the DH validation cohort (n=183). All stages or stage-I only lung adenocarcinoma patients in the validation cohort and predicted to be at high risk displayed significantly worse survival than patients predicted to be at low risk by the FILM risk model (p=0.0006 and p=0.0005 of the log-rank test, respectively). Our findings highlight a novel five-gene signature which, although derived originally from an in vitro cell model, is highly effective in predicting survival of lung adenocarcinoma patients. Studies to validate the effectiveness of FILM in predicting, in particular, the response of lung adenocarcinoma patients to various therapies are highly warranted. Supported by DOD grant W81XWH-04-1-0142 and NCI lung cancer SPORE (P50 CA70907). Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 101st Annual Meeting of the American Association for Cancer Research; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr 4817.