Multidimensional profiling of heterogeneity in supratentorial ependymomas
Summary
This study presents a comprehensive, single-cell and spatial multi-omics atlas of supratentorial ependymomas (ST-EPNs). The authors profiled 42 tumour specimens (snRNA-/scRNA-seq) across all defined ST-EPN molecular subgroups (including canonical and non-canonical ZFTA fusions and ST-YAP1), and performed high-resolution spatial transcriptomics (10x Xenium) on 56 sections from ZFTA-RELA tumours. They integrated transcriptomics, spatial mapping, patient-derived models, cocultures and live imaging to define recurrent malignant cell states (eight metaprograms), their developmental alignments, spatial organisation, morphology and migratory behaviours.
The key findings are: distinct ST-EPN molecular subgroups map to different prenatal cortical developmental timepoints and show subgroup-specific mixes of malignant cell states; ZFTA-RELA tumours commonly contain neuroepithelial-like, neuronal-like and ependymal-like states, whereas other ZFTA clusters and ST-YAP1 show different lineage biases (for example, ZFTA cluster 3 is enriched for embryonic-like cells and ST-YAP1 for ependymal-like cells). Spatial analyses reveal two global architectures—”structured” (high compartmentalisation) and “disorganised” (scattered states)—and a spatial coherence score that correlates positively with mesenchymal/hypoxia states and negatively with embryonic-neuronal-like cells. Local niches and 26 stable spatial clusters (CellCharter) reveal conserved and sample‑specific microenvironments. Morphology and live imaging show cell-state-specific tumour microtube (TM) phenotypes and migratory modes: neuronal-like cells adopt long TMs and saltatory migration reminiscent of immature neurons, while a neuroepithelial-like-2 population is both highly migratory and highly proliferative, implicating division of labour between states in tumour progression. PDX and coculture models (including human iPS-derived neurons/astrocytes) recapitulate many of these states and behaviours.
Key Points
- ST-EPN molecular subgroups are transcriptionally distinct and map to different prenatal cortical developmental timepoints.
- Eight recurrent metaprograms (cell states) were identified: cycling, mesenchymal/hypoxia, embryonic-like, neuroepithelial-like, radial glial-like, embryonic-neuronal-like, neuronal-like and ependymal-like.
- ZFTA-RELA tumours typically contain neuroepithelial-like, neuronal-like and ependymal-like cells; ZFTA cluster 3 is enriched for embryonic-like cells; ST-YAP1 aligns to ependymal-like identity.
- High-resolution spatial transcriptomics defines two global tumour organisations—structured versus disorganised—quantified by a spatial coherence score linked to state composition.
- Mesenchymal/hypoxia states associate with organised, compartmentalised tumours; embryonic-neuronal-like cells correlate with disorganisation and scattering.
- Local spatial niches and 26 CellCharter spatial clusters reveal both conserved microenvironments (myeloid-/endothelial-rich) and tumour-specific malignant niches.
- Morphomolecular mapping shows cell-state–specific tumour microtube morphologies and behaviours: neuronal-like cells are highly migratory with long TMs; neuroepithelial-like-2 cells combine high proliferation and migration.
- PDX and coculture systems (rat and human iPSC-derived neurons/astrocytes) best recapitulate patient cell states, supporting model use for functional follow-up.
Context and relevance
This paper fills a gap by profiling non-canonical ST-ZFTA subgroups alongside canonical ZFTA-RELA and ST-YAP1 tumours, linking molecular subgroup, developmental origin and single-cell states to spatial architecture and cellular behaviour. The work highlights hypoxia/mesenchymal programmes as central to tumour structuring and reveals neuronal-like invasive phenotypes that mirror developmental migration. For researchers and clinicians interested in paediatric neuro-oncology, tumour microenvironment interactions, or single-cell/spatial methodology, the dataset and the analytical framework are highly relevant and actionable: they point to state-specific vulnerabilities (for example, hypoxia-driven niches, invasion-associated neuronal-like cells) and emphasise the need for subgroup-aware therapeutic strategies and clinical outcome correlations.
Why should I read this?
Short version — if you care about what actually makes these childhood brain tumours tick (who the cell types are, where they sit, and how they move), this paper is a neat, data-rich map. It’s got single-cell, spatial and live imaging all tied together, so you can see not just gene signatures but the real-world arrangement and behaviour of tumour cells. If you’re into therapy development or understanding invasion at the resection margin, it’s worth a read: there are clear leads here on hypoxia-driven structure and an invasive neuronal‑like subpopulation that could matter for recurrence.
Author style
Punchy take: this is a high‑resolution atlas that distinguishes developmental origins, cell-state composition and spatial logic across ST-EPN subgroups. For translational folks, it amplifies the point that one-size-fits-all approaches are unlikely to work—therapies will need to consider which cell states and spatial niches dominate a patient’s tumour.
Implications and next steps
Key next steps include correlating these cell-state and spatial patterns with clinical outcomes and therapy responses across all ST-EPN subgroups, experimentally probing hypoxia-driven organisational mechanisms, and testing interventions that target invasive neuronal-like behaviour or state plasticity (for example blocking TM dynamics or hypoxia pathways). The authors provide accessible sequencing and spatial datasets (GEO accessions) and code to enable follow-up analyses.