3023 Background: Pediatric central nervous system (CNS) cancers often pose unique challenges including tumor ‘invisibility’, where surgical resection is restricted due to the sensitive tumor location and tissue biopsy is not always feasible. Detecting cancer associated mutations and copy number variations (CNV) at diagnosis is increasingly important, as the WHO classification of pediatric CNS cancers has incorporated molecular signatures with tumor grade. To achieve CNS tumor molecular ‘visibility’, we previously established a liquid biopsy platform for detecting single nucleotide variants in circulating tumor DNA (ctDNA). However, our method was limited by the restricted number of genes that can be monitored and the inability to detect genomic events including CNVs. To address this, we developed a deep sequencing liquid biopsy approach to profile alterations across selected genes. Our platform provides an opportunity for multi-gene monitoring, to assess tumor subclonal evolution and response to treatment in the absence of repeat tissue biopsies. Methods: We tested the performance of our platform using paired tissue, CSF, and plasma/serum from 10 children with diffuse midline glioma (DMG). ctDNA was analyzed using the TruSight Oncology 500 (TSO500) ctDNA targeted panel covering 523 genes. Matched tumor, CSF, and blood were assessed for concordance and sequencing results were compared to digital droplet PCR (ddPCR) detection of H3K27M mutation. Results: The median exons with ³500X coverage was 96% for 7 CSF samples with optimal input (³60ng), 0.01% for 3 CSF samples with < 5ng input, and 74.5% for plasma/serum samples. ctDNA was more readily detectable in CSF, yet concordance between paired tumor, CSF and plasma/serum was observed. DMG associated mutations in genes including H3F3A, HIST1H3B, TP53, and ACVR1 were detected in ctDNA. Of 9 H3K27M mutations identified in tumor, 8 were present in CSF and 3 in plasma/serum, for a positive percent agreement of 89% and 33%, respectively, with the tumor results. Among CSF samples, H3.3K27M was detected in 6/6 cases, and H3.1K27M in 2/3 cases, with variant allele frequencies comparable to ddPCR results. CNVs including PDGFRA/B and MDM4 amplifications were present in CSF and confirmed by analysis of paired tumor. Additional events, including PIK3CA p.E545Q, PPM1D truncation, and KRAS amplification, were detected in CSF but absent from paired tumor, indicating tissue heterogeneity. Strategies to optimize ctDNA detection, including optimization of ctDNA isolation and adjustment of library QC metrics, were identified. Conclusions: This proof-of-concept study demonstrates the feasibility of our high depth, targeted sequencing approach for detecting clinically relevant mutations in ctDNA from children with CNS tumors. This approach may aid in diagnosis of CNS tumor molecular subtype, and monitoring of tumor evolution and response to therapy in serially collected ctDNA.