P13.20.A SPATIAL HETEROGENEITY OF METHYLATION CLASSES IN MENINGIOMAS: IMPACT ON CLASSIFICATION ACCURACY AND CLINICAL PROGNOSIS

Abstract BACKGROUND Meningiomas are the most frequent primary central nervous system tumors and are predominantly benign. Although the risk of tumor recurrence generally correlates with malignancy, some patients with histologically benign meningiomas also experience recurrence. Methylation classification promises to improve risk prediction in meningiomas. However, the relationship between methylation patterns and clinical behavior remains incompletely understood. We hypothesize that intra-tumoral heterogeneity may confound methylation classification and influence clinical outcomes. MATERIAL AND METHODS To investigate our hypotheses, we applied an attention-based multi-instance learning model named Hetairos previously validated to predict methylation classes, to infer methylation class distribution on hematoxylin and eosin (H&E) stained whole slide images. We validated the in-silico predictions by molecular analyses, including methylation classification on selectively sampled tissue and immunohistochemistry (IHC) staining. For quantification, we defined intra-tumoral heterogeneity as the presence and abundance of multiple predicted methylation classes within individual tumors. We correlated the extent of heterogeneity with clinical outcomes such as recurrence rates and progression-free survival in a cohort of over 200 meningioma patients across different centers with tumors of benign or intermediate methylation classification. RESULTS We successfully validated the predictions of Hetairos by confirming the presence of heterogeneous tumor regions through comparative molecular methylation profiling and IHC staining of cases displaying high intra-tumoral heterogeneity. By stratifying patients into purely benign versus heterogeneous benign/intermediate groups based on in-silico methylation prediction, we found that tumors with high intra-tumoral heterogeneity experienced significantly higher recurrence rates and shorter progression-free survival compared to homogeneous ones. These findings suggest that spatial heterogeneity in methylation classification has clinical significance and may explain recurrence in tumors initially classified as benign. CONCLUSION Our analyses confirm that some meningiomas show high intra-tumoral heterogeneity for methylation classes, here spatially illustrated with an attention-based image model, interfering with accurate methylation-based classification that can lead to a clinically relevant misdiagnosis towards a more benign tumor phenotype. Recognizing and quantifying heterogeneity could improve risk stratification and guide postoperative management. Further validation in larger cohorts is warranted to establish the assessment of presence and spatial latitude for multiple methylation classes as a prognostically relevant factor for meningioma patients.