Is the hypoxic-core / wound-rim niche real in PFA tumors?
ModeledSpeculative
?
Question
PFA tumors are theorized to have a stereotyped hypoxic core surrounded by a wound-healing rim. If true, that coupled niche could be a drug target. Does the pattern show up in bulk transcriptomics?
!
What we found
Yes — strongly. Hypoxia and wound-healing pathway scores correlate at r = 0.88 within PFA samples vs r = 0.72 in non-PFA EPN. The Δ of +0.16 passes our gate. The biology is consistent with the published story.
⚠
Caveat
Bulk correlation cannot prove SPATIAL coupling — only that the two signals are co-elevated in the same tumors. The decisive test is Visium spatial transcriptomics. Also: Cabozantinib (the named drug in the original hypothesis) ranks 68/92 for PFA — the niche may be real but Cabozantinib may not be the right drug.
→
Next step
Ingest Donson 2022 Visium. Parser is already written + unit-tested. Spatial autocorrelation (bivariate Moran's I) at the spot level confirms or refutes the coupling. If confirmed, design a PFA PDX experiment with a CNS-penetrant multi-RTK alternative.
Scroll for the data, methods, and per-sample detail.
PFA1 hypoxia-wound niche → Cabozantinib
Do PFA samples (n=124) show coupled hypoxia × EMT signal that non-PFA EPN doesn't? Niche-based test — independent of the falsification framework verdict.
PFA samples show hypoxia × EMT bulk-correlation above gate AND meaningfully higher than non-PFA EPN. Justifies Donson 2022 Visium ingestion to confirm at spatial resolution.
PFA correlation
0.883
n=124
Non-PFA EPN correlation
0.720
n=246
Delta (PFA − non-PFA)
0.163
gate: ≥ 0.15
Cabozantinib median rank (PFA)
68/92
top-half = ≤ 46
Hypoxia × EMT bulk coupling
MeasuredDerivedReference
Why bevacizumab failed in PFA — one-axis blockade vs a braided secretome
Bevacizumab is a monoclonal antibody against VEGF-A. In PFA, that's a single thread of a coupled niche: the hypoxic core simultaneously secretes HGF (→ MET), GAS6 (→ AXL), PDGF (→ PDGFRβ on stromal/pericyte cells), and FGFs (→ FGFR2). Knock out VEGF and the wound-rim signalling continues through the other RTK axes — the tumor adapts within weeks. That's the published clinical-failure pattern, and it's the geometric reason a CNS-penetrant multi-RTK like Cabozantinib is a more biology-faithful candidate: it covers VEGFR2, MET, and AXL at once.
Plain-English version: Bevacizumab pulled one rope; the niche has five ropes tying it together. Cabozantinib pulls three of them at the same time — still not all five, but enough that the bypass routes the tumor uses for bevacizumab escape (HGF/MET and GAS6/AXL specifically) get blocked too.
RTK target coverage across KIRhub's multi-RTK panel
MeasuredDerived
Each row is a drug; each column is one of the five RTK axes the PFA niche signals through. Filled portion = inhibition (%) at 1 µM (i.e. 100 − residual activity). The visual story: Cabozantinib, Crizotinib, and Tepotinib collapse the MET + AXL axes the niche relies on for bevacizumab-escape; Pazopanib and Lenvatinib pick up FGFR2/PDGFRβ that the others miss — there is no single drug in our panel that covers all five axes.
Filled bar = % inhibition at 1.0 µM (lower residual activity = more filled). "*" indicates we fell back to the best-inhibited variant measurement because no wild-type kinase measurement was available for that drug × target pair.
How this test runs (caveats)
PRD §EPN-2 prefers a bespoke wound signature (POSTN/COL1A1/HGF/TGFB1/PDGFB) from raw RNA + Visium spatial autocorrelation. Bulk pathway correlation is a weaker proxy — strong here only refutes the spatial coupling cheaply if it fails, justifies Donson 2022 Visium ingestion if it holds.
A high bulk correlation is necessary but not sufficient for the spatial coupling the hypothesis predicts: bulk averaging can show correlation if both signals are co-elevated in the same samples even when they're spatially separate within the tissue. Visium is the decisive test.
# Platform
Saifudeen, A., et al. (2026). KIRhub: a falsification-first research workbench for translational oncology. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03090-8
# Data source
McFerrin, L. G., et al. (2018). Oncoscape: a tool for interactive cancer genomics data analysis. Nature Genetics.
Arora, S., et al. (2026). A pan-pediatric brain tumor reference map: medulloblastoma and ependymoma in shared UMAP space. Neuro-Oncology.
# Datasets used in this insight
Oncoscape pediatric brain tumor compendium (n=1,358)
# Provenance
insight_id: pfa1-niche