ID #497 Whole Genome Enzymatic Methylation Sequencing for Pediatric Central Nervous System Tumor Classification

Abstract

DNA methylation profiling for classification of central nervous system (CNS) tumors currently relies on Infinium (EPIC) arrays that interrogate a fraction of CpG loci, require high DNA input that precludes cfDNA applications, and are subject to manufacturer-driven platform changes. Methylation sequencing offers a scalable alternative to EPIC arrays that is not constrained by fixed feature sets or high input requirements and supports high-resolution copy number profiling. We assessed the clinical feasibility of whole-genome enzymatic methylation sequencing (WG-EMSeq) for CNS tumor classification.

We constructed a neural network CNS classifier model from publicly available array data (N = 7372) to perform hierarchical (methylation family and class) prediction. For WG-EMSeq, DNA from 163 pediatric CNS tumors, including medulloblastomas, ependymomas, and gliomas, was extracted from fresh frozen (N = 136) and FFPE (N = 27) samples; enzymatic conversion was performed using the NEBNext Enzymatic Methyl-seq workflow. Libraries were sequenced to a median depth of 34X. Paired-end reads were aligned with DRAGEN and resulting methylation profiles were used to generate methylation family and class predictions.

Predicted methylation families and classes were compared to reference labels based on pathologic analysis, methylation array profiling, and/or chromosomal microarray. WG-EMSeq classifications achieved a macro-averaged one-versus-rest AUROC=0.96. 144 of 163 (88%) WG-EMSeq samples matched a methylation family. The prediction accuracy of family-matched samples was 90% (sensitivity=0.90, specificity=0.99, F1=0.91). For class concordance, samples without a family match were excluded. 71% of samples matched to a methylation class (sensitivity=0.92, specificity=0.99, F1=0.98). High classification accuracy was observed for medulloblastomas, ependymomas, and high-grade gliomas, with reduced accuracy in low-grade gliomas. WG-EMSeq copy number profiles were highly concordant with patient-matched chromosomal microarrays. Classifier predictions were consistent down to 1ng DNA input, indicating WG-EMSeq may be viable for cfDNA applications. Clinical validation of WG-EMSeq is currently in progress as an alternative to arrays for CNS tumor classification.