Abstract BACKGROUND Gliosarcoma (GS) is a rare, aggressive variant of Glioblastoma (GB), representing approximately 2% of all GB cases. It is characterized by its biphasic histological growth pattern with glial and sarcomatous areas. Little is known about GS pathogenesis or the underlying reasons for its distinct clinical features compared to GB, including higher rates of extracranial metastasis and skull invasion. Given the similarity in mutational profiles and chromosomal alterations between GS and GB, we hypothesize that GS arises from a unique interplay between tumor cells and the tumor microenvironment (TME), promoting a mesenchymal phenotype and the development of sarcomatous niches responsible for its clinical behavior. MATERIAL AND METHODS Spatial Transcriptomics (10x Genomics) und methylation profiling was applied to 15 GS samples from 13 patients. We used non-negative matrix factorization (NMF) and single cell deconvolution using large reference single cell RNA sequencing datasets to confidently identify conserved spatial features in our data. To identify the most discerning characteristics between GS and GB, we used a graph-based neural network approach. RESULTS Using a reference-based DNA methylation classifier, 12 of the 13 tumors (92%) were identified as belonging to the mesenchymal (MES) GB subgroup. While transcriptional profiles of GS samples were heterogeneous, NMF revealed four recurrent transcriptional metaprograms. In addition to hypoxia- and glial-associated programs, we identified an extracellular matrix (ECM) remodeling program correlated with sarcomatous areas and enriched for vascular endothelial and smooth muscle cells. Lastly, an immune response program spatially associated with mesenchymal-like tumor cells and proliferating tumor-associated mononuclear cells. Subsequent cell-cell communication analysis uncovered potential signaling interactions and highlighted the importance of TGFβ- and MAPK-pathways for the formation of these distinct niches. A classifier based on the graph-based neural network successfully distinguished GB from GS subregions. CONCLUSION In this study, we performed the first in-depth spatial characterization of GS, identifying specific ECM remodeling and immune reactive spatial features unique for the TME of GS. Our findings provide insights into the transcriptional programs shaping GS tissue architecture, which could be crucial for understanding its biology and eventually aid the design of targeted therapies.
Mühlbauer et al. (Wed,) studied this question.