A Novel Method for Rapid Molecular Subgrouping of Medulloblastoma

Purpose: The classification of medulloblastoma into WNT, SHH, group 3, and group 4 subgroups has become of critical importance for patient risk stratification and subgroup-tailored clinical trials. Here, we aimed to develop a simplified, clinically applicable classification approach that can be implemented in the majority of centers treating patients with medulloblastoma. Experimental Design: We analyzed 1,577 samples comprising previously published DNA methylation microarray data (913 medulloblastomas, 457 non-medulloblastoma tumors, 85 normal tissues), and 122 frozen and formalin-fixed paraffin-embedded medulloblastoma samples. Biomarkers were identified applying stringent selection filters and Linear Discriminant Analysis (LDA) method, and validated using DNA methylation microarray data, bisulfite pyrosequencing, and direct-bisulfite sequencing. Results: Using a LDA-based approach, we developed and validated a prediction method (EpiWNT-SHH classifier) based on six epigenetic biomarkers that allowed for rapid classification of medulloblastoma into the clinically relevant subgroups WNT, SHH, and non-WNT/non-SHH with excellent concordance (>99%) with current gold-standard methods, DNA methylation microarray, and gene signature profiling analysis. The EpiWNT-SHH classifier showed high prediction capacity using both frozen and formalin-fixed material, as well as diverse DNA methylation detection methods. Similarly, we developed a classifier specific for group 3 and group 4 tumors, based on five biomarkers (EpiG3-G4) with good discriminatory capacity, allowing for correct assignment of more than 92% of tumors. EpiWNT-SHH and EpiG3-G4 methylation profiles remained stable across tumor primary, metastasis, and relapse samples. Conclusions: The EpiWNT-SHH and EpiG3-G4 classifiers represent a new simplified approach for accurate, rapid, and cost-effective molecular classification of single medulloblastoma DNA samples, using clinically applicable DNA methylation detection methods. Clin Cancer Res; 24(6); 1355–63. ©2018 AACR.