The invasion phenotypes of glioblastoma depend on plastic and reprogrammable cell states.

Glioblastoma (GBM) is the most common primary brain cancer. It causes death mainly by local invasion via several routes, including infiltration of white matter tracts and penetration of perivascular spaces. However, the pathways that mediate these invasion routes are only partly known. Here, we conduct an integrative study to identify cell states and central drivers of route-specific invasion in GBM. Combining single-cell profiling and spatial protein detection in patient-derived xenograft models and clinical tumor samples, we demonstrate a close association between the differentiation state of GBM cells and their choice of invasion route. Computational modeling identifies ANXA1 as a driver of perivascular involvement in GBM cells with mesenchymal differentiation and the transcription factors RFX4 and HOPX as orchestrators of growth and differentiation in diffusely invading GBM cells. Ablation of these targets in tumor cells alters their invasion route, redistributes the cell states, and extends survival in xenografted mice. Our results define a close association between GBM cell differentiation states and invasion routes, identify functional biomarkers of route-specific invasion, and point toward targeted modulation of specific invasive cell states as a therapeutic strategy in GBM.