BIOM-65. A METHYLATION-SIGNATURE OF RADIATION RESPONSE IN GLIOBLASTOMA

Abstract Methylation signatures based on Infinium beadchip arrays have become central to brain tumor diagnosis and classification. We proposed to develop a signature that could predict intrinsic radiosensitivity in anticipation of adaptive treatments. We utilized Infinium Epic methylation data from 27 patient-derived glioma stem-like cell lines (GSCs) with well-characterized radiation response based on clonogenic survival and limiting dilution assays. GSCs were categorized as radioresistant (n=11), intermediate (n=7), or radiosensitive (n=9). We evaluated multiple modeling approaches using test set validation, including generalized linear and binomial models, random forest, LASSO, and elastic net. Initial feature selection was performed by identifying methylation probe sites with the largest differences in median beta-values between radiation-resistant and radiation-sensitive groups. We then applied LASSO logistic regression for further feature refinement. Using the selected features, we constructed a generalized logarithmic model that predicts radiation response based on a specific subset of methylation probes. Notably, all selected methylation probes have an absolute median beta-value difference between the radioresistant and radiosensitive groups >0.5, with 40% of probe sites demonstrating increased methylation and 60% of probe sites demonstrating decreased methylation with radioresistance. We validated this model in a set of subtotally-resected patients with available methylation profiles at our institution. Clinical radiation response was stratified per RANO criteria and then further quantified based on the percent reduction in the T1-gadolinium enhancing post-surgical volume at the post-radiation MRI, using >25% reduction as the radiosensitive categorization. We assessed the ability of our model to predict response in the clinical subset using receiver operative characteristic and area under the curve determination. In summary, we have developed a methylation signature of radiation response using GSC clonogenic survival, validated in an independent clinical dataset.