225 Markers and Mechanisms of Disease Progression in IDH-Mutant Astrocytomas

Abstract

Abstract INTRODUCTION IDH-mutant astrocytomas (astrocytoma) represent an important subset of glioma with unpredictable clinical course. To explain variability in clinical course, we leveraged genomic data from The Cancer Genome Atlas (TCGA) and utilized machine-learning approaches to model clinical outcomes and identify mortality risk factors in astrocytomas. Next, we examined the utility of clinical features of astrocytoma progression to predict survival, including: 1) contrast-enhancement on magnetic resonance imaging (MR); 2) increased cellular density; and 3) increased cellular proliferation. METHODS Patients with astrocytomas were selected from TCGA. Machine-learning and Bayesian optimization was used to identify molecular features associated with clinical outcomes. Pre-operative MR was available for 74 patients and reviewed for the presence of unequivocal contrast-enhancement (CE+). Cellular density was quantified from digital histopathological slides. Overall survival (OS) differences were assessed using log-rank tests and Kaplan-Meier curves. RESULTS >Our machine-learning algorithm ranked genetic alterations associated with clinical outcomes. Increased protein expression of CDKN1A (ranked #2) and YAP1 (#8) and mutations in MAP-3-kinase (#9) were associated with increased mortality risk. 2007 WHO grade and histologic subtype were not associated with OS (P = 0.13 and P = 0.54, respectively). Patients with CE- tumors (n = 41) outnumbered those CE+ patients (n = 33); there was no significant difference in OS (P = 0.39). Increased cellular density did not predict survival (P = 0.82). Increased MKI67 expression, a proliferation marker, was associated with poor survival (high = 63.5 vs. low = 87.4 months, P = 0.012). Given their similarity to glioblastoma, we investigated VEGFA in astrocytomas. Increased VEGFA expression was associated with worse OS (high = 62.9 vs. low = 105.1 months, P = 0.014). CONCLUSION This investigation demonstrates the strength of machine learning in uncovering biological processes in glioma and revealed key genetic alterations that may help predict aggressive behavior in astrocytomas. The survival associations between cellular proliferation and VEGFA marks the first step in stratifying risk for astrocytomas in the molecular era.