EVOLUTIONARY SELECTION OF KEY ONCOGENIC ALTERATIONS IN PATIENT-DERIVED MODELS OF PAEDIATRIC DIFFUSE HIGH GRADE GLIOMA (PDHGG) SUBTYPES IN VITRO AND IN VIVO

Abstract PDHGG are a diverse group of childhood brain tumours comprising multiple subgroups carrying distinct molecular drivers. Patient-derived models accurately recapitulating this underlying biology are critical for mechanistic/preclinical studies aimed at improving patient outcome, however their behaviour over time in the environments in which they are propagated, and how this relates to the human disease, is largely unknown. To explore this, we collected 94 models of PDHGG established as 2D/3D stem cell cultures in vitro, and generated patient-derived xenografts (PDX) in 33/62 specimens implanted orthotopically in vivo. We carried out exome/targeted sequencing, methylation profiling and RNAseq to profile cells through their first 25 passages in culture, and sequential implantation from p0-p2 in mice. In 15/83 cultures, we observed enrichment of gene expression signatures of non-malignant cells over the first 5 passages, with concurrent depletion of somatic mutations/CNAs, excluding them from further study. Validated models retained tumour-matched genotypes, CNAs and driver alterations including H3.3G34R, H3.3/H3.1K27M, BRAF and ACVR1 over time, however subclonal alterations underwent selection in culture which profoundly altered their response to targeted drug treatment. In 6/7 PDGFRA-mutant models, activating mutations were selected against between p5-20 in 2D and/or 3D, whilst MAPK pathway mutations in NF1/PIK3R1 similarly diverged over 15 passages under different growth conditions, resulting in isogenic models with differential signalling, in vivo tumorigenicity, and in vitro sensitivity to multiple MEK inhibitors. In PDXs, serial xenografting reduced the time to tumour formation by up to half, with a concomitant shift in clonal architecture. Multi-region sequencing of diffusely-infiltrating tumours showed selection for alterations such as PIK3CA/NF1 at distant sites, with evidence for convergent evolution of subclonal mutations, as in human tumours. Understanding the evolutionary dynamics of targetable/predictive alterations in PDHGG model systems is key to developing new and effective therapeutic interventions in this highly heterogenous disease.