EPCO-55. SPATIAL HETEROGENEITY IN THE TUMOR CORE AND INFILTRATING MARGIN OF<i>IDH</i>WT GLIOBLASTOMA

Abstract We integrated single-cell RNA-sequencing, Visium spatial transcriptomic, and Xenium in situ transcriptomic data to identify differences in cell composition between the tumor core and infiltrating edge in IDHWT glioblastoma (GBM). Using CellTrek, an algorithm that maps single cells to their most-likely locations in spatial transcriptomic data, we identified extensive spatial organization in GBM. Cell mapping revealed distinct regions enriched for tumor, immune, endothelial, neuronal, and glial cells. These were associated with the location of the tissue sample – edge regions were enriched for neurons and glia (as expected), while core regions were enriched for tumor cells and infiltrating immune cells, particularly macrophages. Within the core samples, the tumor and immune cells tended to occupy distinct spatial niches. Spatial clustering of mapped tumor cells across all samples revealed five spatially segregated tumor cell states, each associated with specific biological processes, including (1) histone methylation, (2) proliferation, (3) wound-healing, EMT, and antigen presentation, (4) hypoxia response, and (5) stem cell differentiation. These tumor cell states exhibited preferential colocalization with specific non-malignant cell types. For instance, antigen-presenting/wound-healing tumor cells tended to colocalize with diverse immune cells, while stem-like tumor cells tended to reside in neuron-rich regions at the tumor edge. Next, to identify ligand-receptor interactions governing this spatial organization, we analyzed known ligand-receptor pairs within all three data types. This revealed known and novel mediators of tumor cell interactions with vasculature, lymphoid cells, myeloid cells, neurons, and glia. While some interactions were shared across all tumor cell states (e.g. all tumor cell states interacted with myeloid cells via APP/CD74 interactions), others were specific to particular tumor cell states (e.g. wound-healing/antigen-presenting tumor cells interacted with macrophages via CSF1/CSF1R, while proliferating and stem-like tumor cells interacted with neurons through multiple known NRXN/NLGN interactions). These interactions suggest potential targets for clinical disruption of the tumor ecosystem.