Gene-expression signature predicts postoperative recurrence in stage I non-small cell lung cancer patients

Abstract

About 30% stage I non-small cell lung cancer (NSCLC) patients undergoing resection will recur. Robust prognostic markers are required to better manage therapy options. The purpose of this study is to develop and validate a novel gene-expression signature that can predict tumor recurrence of stage I NSCLC patients. Cox proportional hazards regression analysis was performed to identify recurrence-related genes and a partial Cox regression model was used to generate a gene signature of recurrence in the training dataset -142 stage I lung adenocarcinomas without adjunctive therapy from the Director's Challenge Consortium. Four independent validation datasets, including GSE5843, GSE8894, and two other datasets provided by Mayo Clinic and Washington University, were used to assess the prediction accuracy by calculating the correlation between risk score estimated from gene expression and real recurrence-free survival time and AUC of time-dependent ROC analysis. Pathway-based survival analyses were also performed. 104 probesets correlated with recurrence in the training dataset. They are enriched in cell adhesion, apoptosis and regulation of cell proliferation. A 51-gene expression signature was identified to distinguish patients likely to develop tumor recurrence (Dxy = -0.83, P<1e-16) and this signature was validated in four independent datasets with AUC >85%. Multiple pathways including leukocyte transendothelial migration and cell adhesion were highly correlated with recurrence-free survival. The gene signature is highly predictive of recurrence in stage I NSCLC patients, which has important prognostic and therapeutic implications for the future management of these patients.

Figure 1. Survival analyses of the training set of 142 stage I denocarcinomas.

(A) Kaplan-Meier survival curves for two groups of patients with stage IA or IB. (B) Kaplan-Meier survival curves for the two groups of patients defined by having positive (high risk) or negative (low risk) risk scores of recurrence-free survival. The risk scores were estimated with 15 principle components based on the model using 51 recurrence-free survival-related genes. (C) The area under the curve (AUC) of time-dependent ROC analysis for survival models based on stage information or 51-gene expression data respectively. Time is indicated in months on the x-axis, cumulative survival is indicated on the y-axis. Tick marks, patients whose data were censored at last follow-up.

Figure 2. Validation of the 51-gene signature in four independent datasets.

Kaplan-Meier survival analysis was performed in low (full red line) and high (dashed blue line) risk patient groups defined by the 51-gene classifier. AUC for survival models based on stage (dashed red line) or 51-gene classifier (full black line) was also compared. The testing dataset GSE8894 do not have available stage information and all patients in the WUSTL dataset are stage IB. So the time dependent ROC using stage information in these two datasets could not be calculated; all set at 0.5 instead. Tick marks, patients whose data were censored at last follow-up.