Abstract
BACKGROUND: Gliosarcoma, a rare and aggressive variant of IDH-wildtype glioblastoma, is characterized by striking intra-tumoral heterogeneity with alternating glial and mesenchymal differentiation. The mesenchymal component is often enriched at recurrence and associated with therapy resistance, suggesting treatment-driven clonal selection. Despite its poor prognosis and distinct histology, gliosarcoma remains underexplored at the molecular level, with no systematic profiling of its cellular or spatial architecture - limiting progress toward targeted therapies. METHODS: Single-nucleus RNA sequencing (snRNA-seq) was performed on eight FFPE gliosarcoma samples, complemented by spatial transcriptomic profiling (10X Visium) in four matched cases. Multimodal integration enabled high-resolution mapping of malignant and non-malignant compartments. To enhance spatial context extraction and to model tumor-microenvironment interactions, advanced computational approaches, including graph neural networks (GNNs) and convolutional neural networks (CNNs), are being developed. RESULTS: Integration of snRNA-seq and spatial transcriptomic data revealed a distinct population of stromal-like malignant cells enriched for mesenchymal signatures. Compared to conventional glioblastoma, gliosarcoma samples exhibited a reduced presence of neural progenitor cell (NPC-like) and oligodendrocyte progenitor cell (OPC-like) states, suggesting early dominance or selective expansion of the mesenchymal compartment. Spatial annotation of the tumor microenvironment revealed immune-rich niches populated by disease-associated microglia (DIMs), damage-associated macrophages (DAMs), and macrophages (MACs). Preliminary spatial modeling indicates potential interactions between immune population and mesenchymal-like tumor cells that may support therapy-resistant niches. CONCLUSION: This ongoing study integrates single-nucleus and spatial transcriptomic profiling to characterize mesenchymal transition and resistance in gliosarcoma. Early findings highlight a distinct mesenchymal-like malignant population and associated immune environments that may drive resistance. Further computational modeling aims to uncover key cellular interactions and molecular programs, that could inform novel therapeutic strategies.