Spatial tissue architecture as a unifying principle of tumor and microenvironmental states in adult gliomas

Adult gliomas exhibit substantial molecular and pathological variation. Prior single-cell studies have characterized intratumoral heterogeneity; in glioblastoma, tumor cells adopt lineage-associated programs-such as oligodendrocyte- and neuro- progenitor (OPC, NPC), astrocyte (AC) and mesenchymal (MES) like states. However, how these states recur across other gliomas remains unclear. Given their shared environment, defining conserved versus entity-specific tumor characteristics is essential for understanding glioma heterogeneity. Here, we analyzed 310 tumor cores from 284 patients spanning oligodendroglioma, astrocytoma, glioblastoma, and ependymoma, using single-cell spatial transcriptomics with a 344-gene panel (10x Xenium). This approach yielded a dataset of 2.8 million cells, including 18 tumor microenvironment (TME) cell types. To further probe local interactions, we profiled a validation cohort of astrocytomas using a 5,096-genes panel and multiplexed immunofluorescence (57 proteins). Our data revealed 9 recurrent tumor states across gliomas, organized into neighborhoods shaped by local cellular interactions. These formed structured layers, with OPC-/NPC-like tumors near cortical interfaces and AC-like regions forming a continuum with MES1-gliosis. Neighborhoods were largely subtype-specific suggesting that tissue organization and not cell composition alone, underlies glioma heterogeneity. Consistently, methylation classification-central to subtype definition in gliomas-stratified the cohort into 13 classes with distinct neighborhood profiles, including characteristic architectures in RTK1, RTK2, and MES glioblastomas. Neighborhoods also showed distinct TME variation: P2RY12⁺ microglia were enriched in AC-like and MES1-gliosis regions, while CD163⁺ myeloid cells concentrated in MES1-gliosis and MES2-hypoxia niches. These immune patterns corresponded to neighborhood-specific inflammatory programs and at the entity level to myeloid-dominated inflammation in ependymoma and low inflammation in oligodendroglioma. Beyond neighborhood definition, we evaluated their diagnostic potential. In astrocytoma, AC-like enrichment marked favorable prognosis, whereas MES1-like prevalence defined high-risk patients. Given this, we asked whether neighborhoods show histopathological correlates in routine H&E. Morphology embeddings extracted with histopathology foundation models recovered clusters mirroring niche composition, indicating that neighborhoods can be inferred from H&E morphology and incorporated into diagnostic workflows. Collectively, these findings link glioma microanatomy heterogeneity by connecting tumor states to their TME. By capturing spatially dependent programs, our work provides a framework for integrating spatial architecture into biological understanding, diagnosis, and risk stratification.