Impact of Tissue Sampling Strategy on the Diagnostic Accuracy of DNA Methylation Profiling in Central Nervous System Tumors.

DNA methylation-based classification is a key adjunct to histopathology in CNS tumor diagnostics. Although both fresh-frozen and FFPE tissues are used, the impact of sampling strategy on classifier performance remains insufficiently characterized. We analyzed 179 CNS specimens (162 tumors and 17 controls), including 42 FFPE and 137 fresh-frozen samples with ≥ 50% tumor cellularity. Fresh-frozen cases were stratified into histology-validated and blind sampling groups. Genome-wide DNA methylation profiling was performed using Illumina EPIC/EPIC v2.0 arrays and classified using the DKFZ Brain Tumor Classifier, with independent validation using the NIH Methylscape platform. Results were categorized as confirmed, refined, changed, or unclassified and correlated with histopathological diagnoses. FFPE samples showed high diagnostic yield with minimal unclassified results (2%). Among fresh-frozen specimens, histology-validated samples demonstrated robust classification, whereas blind sampling showed higher discordance and unclassified rates (22%). Discordant results were confined to the blind-sampling group and were most consistent with sampling bias and variable tumor representation rather than technical limitations. Methylation profiling confirmed or refined diagnoses in most cases and led to clinically relevant reclassification in a subset. Classifier performance appeared to be influenced more by sampling strategy rather than preservation method, with histology-guided sampling associated with the most reliable diagnostic outcomes.