ID #999 A compendium of tumor-enriched splice junctions in the Pediatric Brain Tumor Atlas

Abstract

Alternative splicing is frequently dysregulated in pediatric central nervous system (CNS) tumors and is increasingly being recognized as a potential source of tumor-specific antigens for immunotherapy. However, the tumor-specific splicing landscape across pediatric CNS cancers remains uncharacterized. We leveraged RNA-sequencing data from pediatric CNS tumors in the Open Pediatric Cancer Project to identify tumor-enriched splice junctions (TESJs). Using rMATS-turbo, we quantified splice junction expression in 2,203 tumors and cell lines from 1,757 pediatric patients and compared them to a reference cohort of normal brains spanning fetal to young adult stages (GTEx <40 years, Evo-Devo atlas, and pediatric normal brains; total N = 250). TESJs were defined as junctions with counts per million fold-change and signal-to-noise ratio > 5 in tumors relative to either 1) all reference brain samples (tumor-specific) or 2) postnatal reference brains only (oncofetal). We identified 121,016 TESJs across 1,248 primary tumors, of which only 22,906 (18.9%) were recurrent within a histology (>5% of tumors), underscoring the high degree of splicing heterogeneity in pediatric CNS tumors. Notably, 2,253 recurrent TESJs (10.9%) used at least one unannotated splice site. Clustering tumors by the 5,000 most variable TESJs recapitulated histologic classifications and further delineated molecular subtypes within low-grade gliomas, medulloblastomas, and ependymomas. Expression of numerous splicing factors correlated with TESJ expression in a histology-specific manner, implicating splicing factor dysregulation as a key driver of tumor-specific splicing. For example, in high-grade gliomas (HGGs), expression of CELF3, CELF4, and CELF5 was inversely correlated with both overall TESJ burden and expression of 1,034 recurrent HGG TESJs. In summary, we present a comprehensive, junction-level resource of tumor-enriched splicing events to enable discovery of splice-driven tumor biology and candidate tumor-specific proteoforms.