Molecular predictors of outcome and response to bevacizumab (BEV) based on analysis of RTOG 0825, a phase III trial comparing chemoradiation (CRT) with and without BEV in patients with newly diagnosed glioblastoma (GBM).

Abstract

LBA2010 Background: RTOG 0825 evaluated the addition of BEV to standard CRT in the treatment of GBM and included molecular stratification that assessed the degree of mesenchymal (MES) gene enrichment. We investigated the ability of the MES signature to predict response to BEV. Methods: Sufficient FFPE tissue for molecular analysis was available for 650 registered, eligible patients. TaqMan PCR was performed prospectively using the molecular stratifier on all patients and an expanded 43 member MES set on 234 cases. A subset of specimens was subjected to whole genome expression profiling (GEP). Predictive models were evaluated for their ability to predict survival (overall, OS, and progression-free, PFS) in the training cohort of the BEV arm after adjusting for prognostic factors and treatment arm. Unsupervised clustering of GEP data was used to identify molecular subsets and gene set enrichment analysis (GSEA) performed to evaluate for MES enrichment. Results: We observed a significant association between increasing MES signature and worse PFS and OS in the BEV arm (p=0.036 and p=0.032, respectively). Based on the association between high MES expression and poor outcome in the BEV arm, we sought to optimize a predictor using an expanded set of MES genes. Unbiased gene selection from a total of 43 genes followed by radial basis machine modeling identified a 10-gene predictor of outcome in the BEV arm (p<0.001/HR 4.04 for OS and p<0.001/HR 2.21 for PFS). To further support the association of MES enrichment and poor outcome in the BEV arm, we performed GEP. Unsupervised clustering identified a subtype of tumors with MES enrichment associated with the 10-gene predictor survival probability (p=0.0124, t-test). Ongoing studies will determine the extent to which this represents a predictive marker for BEV. Conclusions: We developed a 10-gene predictor specific to BEV treatment and suitable for FFPE that may serve to identify subsets of patients with newly diagnosed GBM who benefit from BEV. Supported by RTOG grant U10 CA21661, CCOP grant U10 CA37422, and Brain SPORE P50 CA127001 from the NCI and Genentech. Clinical trial information: NCT00884741.