CNSC-20. BRAIN NEURODEVELOPMENT-DERIVED POLYGENIC BIOMARKER FOR PROGNOSTIC STRATIFICATION OF PEDIATRIC AND ADULT BRAIN TUMORS: THE FETAL-TO-ADULT INDEX (FTAI)

Abstract Despite advances in brain tumor classification and molecular diagnostics, tools for early-stage, transcriptome-based prognostication remain limited, especially in low-resource settings. Polygenic prognostic models may better capture biological heterogeneity than single-gene approaches. We developed a fetal-to-adult index (FTAI), a transcriptomic score derived from normal human brain neurodevelopment, to enable biologically informed risk stratification across pediatric and adult brain tumors. Spatiotemporal transcriptomic data from the Allen Brain Atlas were used to curate fetal and adult gene sets, with 10-, 20-, and 30-gene panels tested via gene set variation analysis (GSVA). Principal component analysis confirmed clear separation by developmental stage. Using curated sets of genes highly expressed during fetal brain development (SOX2, PAX6, OLIG2, etc.) and genes with predominant expression in the adult brain (GFAP, S100B, SYN1, etc.) we confirmed developmental gene expression separation, with ~40% of variance explained by PCA (PC1). FTAI correlated strongly with chronological age—high in infancy and decreasing with age. FTAI was computed across 1,001 tumor samples, including pediatric brain tumors (PBTs, n=404), lower-grade gliomas (LGGs, n=522), and glioblastomas (GBMs, n=75), showing significant differences across groups (p=6.9x10-11). Higher FTAI was associated with worse overall survival in both PBTs and LGGs, as demonstrated by Kaplan–Meier survival analyses (PBTs: p<1.0x10-3; LGGs: p=1.2x10-2). In multivariable Cox and logistic models adjusted for age, sex, and race, each 0.1 increase in FTAI was associated with significantly worse prognosis (PBTs: HR 2.2 [p=1.5x10-3], aOR 2.8 [p=1.1x10-3]; LGGs: HR 1.86 [p=3.6x10-4], aOR 2.6 [p=1.4x10-3]), but not in GBMs (p>0.05). Machine learning models (random forest, XGBoost) using FTAI predicted survival with AUCs of 0.72. While post-hoc and sensitivity analyses revealed no significant association within individual pediatric tumor subtypes, likely due to limited sample sizes, this underscores its potential as a broadly applicable prognostic biomarker. Our findings support the FTAI as a static, polygenic, developmentally reflective index that enables early prognostic stratification across diverse brain tumor types, with applicability in both comprehensive and targeted transcriptomic workflows.