ID #633 Parallel in vitro and in vivo CRISPR screening reveals distinct dependency landscapes in pediatric glioma

Abstract

CRISPR dependency screens have been instrumental in identifying therapeutic vulnerabilities in cancer, but in vitro studies can be confounded by culture-specific artifacts. In vivo patient-derived xenograft (PDX) CRISPR screening platforms offer a complementary approach to identify context-dependent dependencies and capture tumor–microenvironment interactions. Here, we performed parallel in vitro and PDX-based in vivo reduced-representation CRISPR screens in an H3 K27M diffuse midline glioma (DMG) or an H3 G34R diffuse hemispheric glioma (DHG). Libraries targeted candidate genes nominated through hidden-network analysis of K27M-mutant tumor transcriptomes (n = 437 genes) and curated gene sets related to neuronal differentiation and development (n = 451 genes), with common essential genes included as internal controls. Quality control analyses demonstrated depletion of common essential genes and strong concordance in relative gene depletion or enrichment across targets shared between the two libraries. Principal component analysis revealed distinct dependency profiles between in vitro and in vivo conditions, indicating context-specific genetic dependencies. Notably, transcription factor dependencies were significantly reduced in vivo, suggesting that tumor–microenvironment interactions may buffer transcription factor loss. In contrast, dependencies unique to in vivo G34R tumors were enriched for ion channel genes, highlighting vulnerabilities not evident in cells from the PDX model grown in vitro. Shared in vitro and in vivo dependencies included known drivers such as PDGFRA and SOX2, as well as genes involved in ubiquitination and cholesterol biosynthesis. Ongoing studies aim to validate these dependencies in vivo and elucidate the mechanistic basis of these vulnerabilities.