Abstract B026: Utility of DNA Methylation-Based Diagnostic Classifiers for Cohort Selection in Pediatric Neuro-Oncology Research
Patrick J. Beck, Jessica B. FosterAbstract
Background
/Purpose: DNA methylation-based central nervous system (CNS) tumor classifiers have become increasingly useful for pediatric CNS malignancies, complementing traditional histological and molecular tumor diagnostics. Here, we investigate additional roles for methylation-based classifiers to aid in cohort selection for computational research.
Methods:
Multi-omic data for a total of 153 samples comprising pediatric high-grade glioma (pHGG, n = 57), ependymoma (EPN, n = 35), medulloblastoma (MB, n = 29), atypical teratoid rhabdoid tumor (ATRT, n = 22), and craniopharyngioma (CP, n = 10) were obtained through The Open Pediatric Cancer Project (OpenPedCan). Harmonized diagnoses, derived by following disease-specific OpenPedCan subtyping algorithms that combine tumor histology with characteristic clinical, genomic, transcriptomic, and/or epigenetic features, were compared to diagnoses predicted by the Heidelberg v12.8.1 (German Cancer Research Center) and Bethesda v3.0 (National Cancer Institute) CNS tumor DNA methylation classifiers.
Results:
Classification performance was grossly equal between the Heidelberg and Bethesda classifiers. Following the OpenPedCan subtyping algorithms for pHGG, EPN, MB, and CP, methylation classification contributed to the harmonized diagnosis only when specific histological/molecular features were inconclusive or unavailable, which applied to 21-37% of samples, yielding a definitive diagnosis in ∼75% of those cases. Methylation classification provided more precise subtype diagnoses for histone H3 wild type (H3WT) pHGGs compared to traditional diagnostic approaches. ATRT molecular subtyping relied solely on methylation classification, which it achieved successfully for 73% of ATRT samples. Ten samples in the cohort methylation-classified as “control tissue” or “inflammatory microenvironment,” four of which would not have been identified if strictly following the OpenPedCan algorithm. One EPN sample methylation-classified as H3WT diffuse midline glioma, which was ultimately supported by the identification of EZHIP overexpression, TP53 inactivation, and PDGFRA copy number gain on subsequent analysis. Another EPN sample methylation-classified as CNS tumor with BCOR internal tandem duplication; further genomic analysis confirmed the presence of this alteration. Both samples lacked transcriptional enrichment for ciliary, microtubular, and differentiation signatures expected for cells of ependymal origin.
Conclusions:
This study highlights that methylation-based CNS tumor classifiers can be useful within research specimens for suggesting alternative diagnoses when standard tissue histology or molecular markers are ambiguous or imprecise, providing additional layers of diagnostic granularity for specific CNS tumor subtypes, and identifying and excluding samples with limited tumor cells that are more consistent with normal/inflammatory tissue controls. DNA methylation offers selection of more homogeneous analysis cohorts, ultimately elevating the validity of pediatric neuro-oncology research.
Citation Format:
Patrick J. Beck, Jessica B. Foster. Utility of DNA Methylation-Based Diagnostic Classifiers for Cohort Selection in Pediatric Neuro-Oncology Research [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Bridging Discovery and Clinical Impact in Pediatric Cancer; 2026 Sep 22-25; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2026;86(18_Suppl_1):Abstract nr B026.