Enhancing Molecular Classifying Accuracy of Pediatric CNS Tumors: A Dual-Classifier Approach Using DNA Methylation Profiling

BackgroundAccurate classification of pediatric central nervous system (CNS) tumors is critical for optimal treatment yet remains challenging due to the limitations of traditional histopathological methods. DNA methylation profiling has gained attention as a promising tool for the molecular classification of CNS tumors. However, despite its potential clinical value, methylation classifiers remain limited to the research settings. This study aims to assess the use of two DNA methylation-based classifiers for CNS tumor diagnostics, with the eventual goal of integrating them into clinical practice and the impact of technical factors such as fixation methods, DNA quantity, and array choice, while exploring the utility of visualization tools (UMAP/t-SNE) and the integration of molecular data for resolving diagnostic ambiguities. MethodsWe analyzed 96 pediatric pathology tissue samples, including 75 CNS tumors, 10 with CNS non-tumoral lesions and 11 with non-CNS tumors, performing 130 methylation analyses. DNA from both formalin-fixed paraffin-embedded (FFPE) and fresh frozen (FF) tissues were used for methylation profiling using the Illumina MethylationEPIC V1 and V2 arrays. The performance of two DNA methylation-based classifiers (Heidelberg and NIH) was evaluated by comparing the classification results with histopathological diagnoses. Technical variables that may affect quality such as DNA quantity, extraction method, and sample fixation were also investigated. ResultsBoth classifiers demonstrated an 88% concordance with histopathological diagnoses in CNS tumors. Methylation profiling refined the histological diagnoses in 54.66% of cases and contributed to molecular subtyping in 52% of CNS tumor cases. The analysis in a small percentage of cases (5.33%) exhibited conflicting diagnoses, emphasizing the need for cautious interpretation and re-evaluation of the cases of uncertainty. Interestingly, both classifiers also identified CNS non-tumor tissues from tumor cases, although they misclassified some normal tissues and malformations as CNS tumors. Technical factors, including DNA quantity and sample fixation, had minimal impact on classifier performance. ConclusionThis study highlights the potential of DNA methylation profiling as a complementary diagnostic tool in pediatric CNS tumor classification, paving the way for its integration into routine clinical practice. To the best of our knowledge, this is the first publication comparing two DNA methylation classifiers in a pediatric CNS tumor cohort. While the classifiers show promise for improving diagnostic accuracy, especially in complex or undiagnosed cases, they should be used as a complementary tool to histopathologic classification. Further research is needed to validate their integration into clinical practice, including refining technical protocols, addressing limitations, and evaluating long-term clinical outcomes.