Reconstructing the single-cell spatiotemporal dynamics of glioblastoma invasion

Abstract

Abstract Glioblastoma invasion into healthy brain tissue remains a major barrier to effective treatment, but the dynamic single-cell behaviors underlying this process remain difficult to measure across patient models. Here, we introduce GlioTrace, an ex vivo imaging and computational framework for real-time tracking of glioblastoma invasion in patient-derived xenograft brain slices. Combining whole-specimen confocal microscopy with machine learning-based morphology classification and probabilistic modeling of state transitions, GlioTrace reconstructs how tumor cells switch between invasion phenotypes and how these dynamics are shaped by local tissue context and therapy. Across six patient-derived models, GlioTrace reveals that invasion plasticity is patient-specific, spatially organized, and modulated by vascular proximity and microglial interaction. Targeted therapies selectively alter phenotype composition and transition probabilities, whereas standard-of-care treatment enriches a highly motile, transition-connected branching state consistent with a treatment-tolerant invasive phenotype. Thus, GlioTrace makes glioblastoma invasion a measurable dynamic process for comparing patient-specific treatment responses in living brain tissue.

The authors' abstract, as published at the source. Nature Communications, 2026 · DOI ↗