Methylation profiling and alternative classification approaches in a glioma-enriched FFPE stereotaxic biopsy cohort

DNA methylation profiling of CNS tumors supports and refines histological diagnoses. Tumor DNA methylation signatures allow for an alignment with reference methylation classes and copy number profiling required for molecular grading. Overall, and especially in cases with limited tissue like in stereotaxic (STX) biopsies, there is a need for informed triage to most promising analysis techniques. Here, we investigated the diagnostic potential and accuracy of DNA methylation profiling and molecular grading in STX samples. FFPE samples of 237 patients were subjected to DNA methylation analysis including the Heidelberg (v11b4), Bethesda (v2) and EpiDiP brain tumor classifiers, copy number profiling and tumor deconvolution using MethylCIBERSORT. 90.5% of samples (n = 232) were allocated DNA methylation classes, with 61.6% reaching Heidelberg classifier scores of > 0.84. Indicatory copy number variations (CNVs) were detectable in 96.9% of cases. Recursive partitioning supported by a resampling-based validation with repeated random subsampling pointed towards brain tumor classifier dependent minimum analyte inputs to reach robust calibrated scores for DNA methylation class allocation. Successful integrative diagnostics furthermore depended on pre-classifier H&E diagnoses with 96.2% of H&E malignant diffuse gliomas, 91.5% of H&E diffuse gliomas but only 60% of H&E circumscribed gliomas classifiable using the Heidelberg brain tumor classifier. When the alternative classifiers EpiDiP and Bethesda were used, higher DNA input amounts ameliorated the classification robustness. Nevertheless, diagnostics of high-grade gliomas in midline location and lower-grade glial/glioneuronal tumors posed a challenge. Our findings demonstrate the broad applicability of DNA methylation profiling and molecular grading to small brain tumor specimens under the caveat of certain tumors and locations potentially benefiting from additional molecular as well as computational approaches to increase methylation signals relevant for classification in the future.