Variant allelic fraction of driver mutations predicts success of genomic methylation classification in CNS tumors.

e14035 Background: Whole genome CpG DNA methylation profiling is proving to be an extremely valuable tool in the diagnostic and prognostic classification of central nervous system (CNS) tumors, and is especially adept at discriminating between very different tumors, or tumor subtypes, that are histologically similar. However, like any bulk assay, reliability of methylation-based classification depends on sufficient tumor cellularity, as admixed nonneoplastic elements (e.g., neurons, glial cells, inflammatory cells) have their own unique methylation profiles, and thus can skew the results. Currently, it is recommended that at least 70% of the analyzed specimen be tumor; however, this is based on subjective estimates from light microscopy. Many neoplastic entities have well-known driver mutations that occur in virtually 100% of tumor cells, and next-generation sequencing (NGS) assays can detect those mutations and report their relative amounts in the form of Variant Allelic Fraction (VAF). Since NGS and methylation profiling are now frequently being done on the same tumor block, we sought to determine whether driver mutation VAF affects the accuracy of methylation profiling, and whether VAF can help establish more rigorous cutoffs for quality assurance. Methods: Using NGS and Infinium Epic850K methylation arrays, we evaluated 62 CNS neoplasms representing a range of entities, including TERT promoter-mutant glioblastoma, IDH-mutant astrocytoma, IDH-mutant oligodendroglioma, SHH-driven medulloblastoma, and CTNNB1-driven adamantinomatous craniopharyngioma. Results: VAFs of each driver mutation ranged between 1-60%. Forty-four of 62 cases (71%) had a methylation classification score ≥0.9, the most widely accepted cutoff for a successful result. A fit-of-mixture analysis via CutoffFinder (PMID: 23251644) suggested that the optimal VAF cutoff is 21%. Forty-one of 46 (89%) cases with a VAF of 21% or higher had a methylation classification score ≥0.9, whereas only 3/16 (19%) at or below 21% were classifiable with methylation profiling ( P< 0.0001 by Fisher’s exact test). Nonlinear regression of VAF versus methylation score by least squares fit produced R2= 0.54, with sum of squares = 2.0. Conclusions: These data indicate that VAF can serve as a useful predictor of classification success via methylation profiling, and should be taken into account when interpreting methylation profiling results.