OS11.3 Machine learning for better prognostic stratification and driver genes identification in 1p/19q-codeleted grade III gliomas

Abstract

BACKGROUND: Although major advances have been accomplished over the last years in their characterization, 1p/19q-codeleted anaplastic gliomas have variable clinical behavior and mysterious 1p/19q-codeletion “driven” molecular oncogenesis. We have recently shown that the most common alteration (i.e. 9p21.3 allelic loss detected in 42% of cases) is an independent prognostic factor in this tumor type. However, less frequent Genomic Copy Number Variations (CNV) may also have clinical value and may shed light on molecular oncogenesis of this tumor type. METHODS: A cohort of 197, 1p/19q-codeleted grade III gliomas was collected as part of the French POLA network. Clinical, pathological and molecular information were recorded and the patients were clinically followed up. CNV analysis was performed using SNP arrays. Computational biology and machine learning analysis were applied to identify less common additional CNV events associated with overall survival and other clinical-pathological variables RESULTS: Recurrent chromosomal events were identified in chromosomes 4, 9, and 11. 46 focal amplification events and 22 focal deletion events were identified. 12 of the focal events overlapped with known cancer related genomic regions. 24 focal CNV areas were associated with survival and five of them were significantly associated with survival after multivariate analysis. 9/24 signals, detected in 3 to 26 percent of patients, were validated using an external cohort of gliomas of the Cancer Genome Atlas. Five of the validated signals contain a cancer related gene or MIR: CDKN2A deletion, SS18L1 amplification, RHOA/MIR191 copy-neutral loss of heterozigosity, FGFR3 amplification, and ARNT amplification. The CNV profile enables better survival prediction compared to clinical risk assessment. CONCLUSIONS: Several recurrent CNV events are characteristics for 1p/19q-codeleted grade III gliomas. These genomic regions are associated with survival and enable better survival prediction. More important, they may help identifying potential genes for understanding oncogenesis and for personalized therapy.