JS09.7.A GENETIC DRIVES PROMOTE DISTINCT MICROENVIRONMENT ORGANIZATION IN GLIOBLASTOMA

Abstract Genomic drivers in glioblastoma (GB) are associated with specific transcriptional states. Within tumour microenvironment architecture however, malignant transcriptional states are largely determined by local cues such as hypoxia or immune infiltration. To what extent genomic drivers affect the TME architecture remains unknown. To investigate the effect of different genomic drivers on TME architecture in GB, we used a replication-competent avian-like sarcoma virus system to introduce EGFRvIII expression, NF1 loss and PDGFB overexpression in a tp53-/- mouse model with Nestin promoter-dependent tumor virus A expression. These genetic alterations recapitulated known classical, mesenchymal, and proneural transcriptomic GB subtypes. We leveraged single cell RNA sequencing and combined spatial transcriptomics and proteomics data at regional (NanoString GeoMx) and single cell (10x Visium HD) level to acquire an in-depth characterization of the differences in the TME organization. Our results revealed intricate dependencies of the GB TME architecture on genetic backgrounds. At single cell level, NF1- and EGFRvIII- driven tumours overrepresented mesenchymal- and astrocyte-like states, whereas PDGFB-driven tumours showed increased neuronal progenitor states. Regardless these evident transcriptional differences, cellular neighbourhoods within the TME locally showed conserved spatial compositions across all samples, with nuanced distinctions between genomic drivers; our integration of spatial transcriptomics and proteomics data showed that EGFRvIII-driven tumours contained complex perivascular immune neighbourhoods with high clustering of plasma B and antigen presenting cells. NF1 driven tumours contained strong perivascular mesenchymal tumour cell enrichment with wound healing features. PDGFB driven tumours increased neuronal signalling with markedly decreased immune infiltration. To extrapolate our findings to human glioblastoma, we leveraged a cohort of 50 spatial transcriptomics samples with known genomic alterations in EGFR, NF1 or PDGFB. Using a graph neural network approach, we were able to reliably predict which cellular neighbourhoods represented genomic alterations. Strikingly, both in mouse and human settings, EGFR, NF1 and PDGFB alterations showed enrichment of immune, tissue reactive and neurodevelopmental organizational structures. By employing sophisticated model systems alongside high dimensional spatial interrogations, we showed how the TME organization is dependent on a complex interplay between cancer cell intrinsic features and its TME. While spatial tumour architecture largely relies on external factors, genomic drivers impose preferred TME structures across mouse and human glioblastomas that could be important in future therapy designs.