Glioblastoma's malignant states are not scattered at random: a developmental-to-injury-and-hypoxia trajectory layers around the necrotic core

Glioblastoma heterogeneity can be written as a directional, spatially ordered malignant cell trajectory rather than a list of unrelated cell states. Earlier molecular classification sorted tumours into subtypes such as developmental-like and mesenchymal-like, and these states were known to coexist within one tumour and to show recurrent spatial arrangement, but they had not been linked into a cross-tumour cellular trajectory. A paired single-nucleus and spatial atlas of 12 IDH-wild-type primary tumours, 57 sampling sites, more than one million nuclei and 97 Visium sections showed developmental-like states near infiltrated regions and gliosis and hypoxia states forming layered zones around necrosis.

The trajectory appears in space as a stepwise transition from developmental-like to gliosis to hypoxia, matched to distinct tissue microenvironments. Earlier spatial studies placed hypoxic mesenchymal-like states at the necrotic core and developmental-like states at the infiltrating edge, but did not quantify the adjacency order among states. A network built from pairwise spatial distances of tissue microenvironments across 97 Visium sections ordered them along a main axis from the infiltrating edge to the necrotic core; 17 Xenium sections validated this zonation at single-cell resolution.

Genetically distinct subclones within the same tumour each span developmental-like and gliosis-hypoxia states, and subclones intermingle finely rather than occupying separate territories. Earlier views of glioblastoma clonal spatial structure were coarser, and whether clones correspond to different cell states was unclear. Multiome copy-number inference showed most subclones contained both developmental-like and gliosis-hypoxia cells, with minority states averaging 15% of each clone (range 3% to 49%); laser-capture microdissection whole-genome sequencing supported clonal intermixing at roughly 100 by 100 micrometre resolution, and SpaceTree reported high clonal entropy in Visium data.

The trajectory is experimentally recapitulated in patient-derived organoids and is coupled to regionalized myeloid microenvironment signalling. Earlier trajectory inference came mainly from static snapshots of patient tissue, without evidence tracking the same cell-state transition over time in vitro. Two developmental-like patient lines were labelled and engrafted into glioblastoma organoids and tracked to day 28 with 5,101-plex Xenium spatial transcriptomics, where developmental-like cells gave rise to gliosis-hypoxia cells over time; myeloid ligands were enriched differently in developmental-like versus gliosis-hypoxia microenvironments.

The conclusions rest on 57 sampling sites from 12 IDH-wild-type primary glioblastoma tumours, 97 Visium sections and 17 Xenium sections, with organoid validation using two patient-derived lines tracked to day 28; the intended readers are tumour biologists and spatial-omics researchers.

Organoid tracking is at population level and does not provide single-cell-resolution lineage tracing; the study does not test whether malignant cells can revert from gliosis and hypoxia to developmental-like states, and a companion study predicts limited reversion from gene regulatory networks.