Abstract 5597: Molecular classification of glioblastoma multiforme using isoform-level gene expression signatures.

Glioblastomamultiforme (GBM) is one of the most deadly cancers in adult and even with aggressive therapy prognosis is poor with a median survival of about 15 months. Recent large scale genomic studies have shed light on the molecular mechanisms of cancer and shown the existence of distinct molecular sub-groups of GBM whose identification is the key to develop specific therapeutic regimen for each GBM group. Since transcriptome studies, by us and others, have discovered that majority of genes produce multiple protein isoforms, which could be involved in multiple functional pathways, we hypothesized that the isoform-level expression profiling will generate better classification to identify themolecular sub-groups of GBM. To test this hypothesis, we performed isoform-level analysis of the exon array expression data for GBM patient samples from the TCGA data portal, and discovered that isoform-level analysis identifies 2.5 fold more differentially expressed transcript variants than differentially expressed genes captured by gene-level analysis, indicating that isoform-level expression profiling is more sensitive in identifying molecular changes among GBM patients. Next, we applied consensus non-negative matrix factorization (NMF) clustering method, based on isoform-level expression ofmost variable isoforms and effectively grouped the GBM samples into 4 sub-groups with significant (p=0.0103) survival differences between the groups. In contrast, though clustering based on gene-level expression produced four homogenous groups there was no significant survival difference among the sub-groups. Based on the prognostic value of the molecular sub-groups, our goal was to build a classifier that can assign each GBM patient a molecular sub-group.We compared the prediction accuracy of a gene based vs an isoform based classifier to identify sub-group and found that isoform based classifier is a better predictor (85%vs90%). Using the Random forest feature selection we have build a classifier based on the expression of 147 isoforms that is ∼91% accurate and have developed a high-throughput RT-qPCR assay to measure the expression of these discriminatory isoforms. We have successfully validated the classifier to identify the molecular sub-group in an independent cohort of GBM patient samples from the Human Brain Tumor Tissue bank at University of Pennsylvania.Our study has led to the development of a classification assay for GBM patient sub-grouping, which can be useful in developing targeted drug therapy for each sub-group, and suggests that isoform based expression analysis can lead to better molecular classification of cancer, a requirement for the quest of personalized therapy. Citation Format: Sharmistha Pal, Yingtao Bi, Lukasz Macyszyn, Louise C. Showe, Donald M. O9Rourke, Ramana V. Davuluri. Molecular classification of glioblastoma multiforme using isoform-level gene expression signatures. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 5597. doi:10.1158/1538-7445.AM2013-5597 Note: This abstract was not presented at the AACR Annual Meeting 2013 because the presenter was unable to attend.