EPCO-08. METHYLATION-BASED CLASSIFIER IMPROVES MENINGIOMA PROGNOSTICATION

Abstract

Abstract Meningiomas are the most common central nervous system tumors and, while typically benign, a subset exhibits aggressive behavior or recurrence, necessitating repeated interventions. The current WHO classification, which comprises 15 subtypes across three grades based on histopathology, often fails to predict recurrence accurately. Although prior studies have linked DNA methylation patterns to tumor behavior, their clinical utility in forecasting recurrence remains unclear. We hypothesized that DNA methylation profiling could improve prognostication and recurrence risk stratification in meningiomas. We analyzed DNA methylation data (Illumina EPICv1) from 193 primary meningiomas across all WHO grades (24% recurrence), integrating in-house and public datasets. Data preprocessing was conducted using the Sesame pipeline. Methylation subtypes were identified through unsupervised clustering (K-means, PCA), supported by multiple model selection algorithms to determine the optimal number of clusters. Copy number alterations (CNAs) were inferred from the methylation data, and recurrence-free survival was assessed using Kaplan-Meier analysis. A random forest classifier was developed using the top 100 differentially methylated CpG sites (Wilcoxon, p < 0.05), trained on a discovery cohort (n=136), and validated both internally (n=57) and externally (n=506) and comparing classifications to published molecular subgroups. We identified three distinct methylation subtypes with unique clinical and molecular features. The hypermethylated subtype showed the shortest median recurrence-free survival (~80 days), the highest recurrence rate, and was enriched in males, WHO grade 2 tumors, non-skull base locations, and high CN losses (p = 0.002). In contrast, the two hypomethylated subtypes had longer recurrence-free intervals (~155 and 118 days), lower recurrence rates (29% and 24%), and were enriched in females, WHO grade 1 tumors, and had fewer CNAs. The classifier accurately predicted recurrence risk, with 89% and 97% concordance in high- and low-risk groups in the external validation. These findings support the clinical integration of methylation-based classification to improve meningioma risk assessment and management.

Ähnliche Arbeiten