Abstract
Abstract Glioblastoma multiforme (GBM) is the most aggressive form of primary brain tumor, with no curative treatment options. Multiple studies have characterized at single cell resolution the GBM as being composed of transcriptional cell states interconnected with components in the tumor immune microenvironment (TME). Our group proposed and validated the first single cell guided functional classification of GBM in four tumor-intrinsic cell states which informed clinical outcome and delivered therapeutic options. However, single cell technologies lack the spatial relationships among the cell states of GBM and between GBM cell states and TME. Spatially resolved transcriptomic technologies are emerging as powerful tools to reconstruct the spatial architecture of a tissue. We performed spatial transcriptomics of multicellular regions of interest (ROI) in 8 IDH wild-type GBM with both CosMx Spatial Molecular Imager, which analyzes 1,000 RNA probes and 64 proteins at single cell resolution, and GeoMx Digital Spatial Profiler which profiles the whole transcriptome (~18,000 genes) at ROI resolution. We integrated the two platforms to define single cell states and non-malignant cells and developed a computational deconvolution approach for CosMx spatial data which utilized GeoMx ROI resolved transcriptomic profiles as a-priori information to predict cell type abundances. Spatial deconvolution of CosMx derived single cells revealed spatial segregation of the tumor cell clones and cellular states and highlighted recurrent patterns of cell states, distinct TME cell types associated with coherent histopathological features across multiple samples. Our studies established a scalable approach to resolve the transcriptional heterogeneity of GBM and reconstruct the architecture of GBM cell states and tumor microenvironment.