Cross-species comparison reveals therapeutic vulnerabilities halting glioblastoma progression

Abstract

The growth of a tumor is tightly linked to the distribution of its cells along a continuum of activation states. Here, we systematically decode the activation state architecture (ASA) in a glioblastoma (GBM) patient cohort through comparison to adult murine neural stem cells. Modelling of these data forecasts how tumor cells organize to sustain growth and identifies the rate of activation as the main predictor of growth. Accordingly, patients with a higher quiescence fraction exhibit improved outcomes. Further, DNA methylation arrays enable ASA-related patient stratification. Comparison of healthy and malignant gene expression dynamics reveals dysregulation of the Wnt-antagonist SFRP1 at the quiescence to activation transition. SFRP1 overexpression renders GBM quiescent and increases the overall survival of tumor-bearing mice. Surprisingly, it does so through reprogramming the tumor's stem-like methylome into an astrocyte-like one. Our findings offer a framework for patient stratification with prognostic value, biomarker identification, and therapeutic avenues to halt GBM progression.

Fig. 1. ASA in the fetal and adult mammalian brain.

a Schematic representation of the embryonic origins of adult v-SVZ astrocytes (qNSCs) and parenchymal astrocytes along with their lineage potential in adulthood under homeostatic and injury conditions, respectively. Created in BioRender. Kaya, O. (2025) https://BioRender.com/np19rtk. b UMAP embedding of integrated WT mouse v-SVZ NSC lineage scRNA-seq^,, (n = 6 replicates), showing NSC differentiation trajectory through quiescence, activation, and differentiation (QAD) stages. c Proportion of QAD-stage cells during human cortical development, with trend of Q-cell proportions shown below. Stage assignment by QAD-gene AUCell score. d Distribution of age at diagnosis for n = 399 primary GBMs in TCGA and Wu et al.. e Non-parametric QAD-gene set scoring of multi-modal expression measurements of GBMs from (d). Gene set prevalence denotes the proportion of GBMs having a positive GSVA score. f Left: schematic of GBM patient-derived xenograft (PDX) generation. mCherry ubiquitously labels tumor cells. Center: UMAP of single-cell GBM PDX (n = 4 replicates) showing QAD-gene set AUCell scores. Cells are colored by the maximum-scoring state, and enrichment relates to the margin between the maximum and second-highest scores. Right: Representative immunofluorescence images of PDX GBM cells with distinct neuron- and astrocyte-like morphologies (dashed arrows). Scale bars 25 µm. g Aitchison compositional distance between QAD-stage frequencies in GBM PDX from (f), young/old adult v-SVZ from (a), and late second trimester from (c). v-SVZ ventricular-subventricular zone, q/aNSC quiescent/active neural stem cell. Source data are provided as a Source Data file.

Fig. 2. ptalign decodes the ASA of GBMs using the adult NSC lineage as a reference.

a Left: UMAP embedding of integrated WT mouse v-SVZ NSC lineage scRNA-seq (n = 6 replicates), showing NSC differentiation trajectory through QAD stages. Center: v-SVZ lineage pseudotime acts as a reference into which query GBM cells are mapped by ptalign. Colored lines depict pseudotime-binned cells linked to their average position in UMAP. Right: UMAP of GBM PDX (n = 4 replicates) with ptalign-derived QAD-stages. Cycling cells (gray) are excluded from pseudotime analysis. Pie charts indicate QAD-stage proportions. b A pseudotime similarity metric derived from expression correlation along the v-SVZ lineage captures different stages and transitions by their characteristic similarity profiles. c In ptalign, a neural network is trained to predict v-SVZ lineage pseudotime based on similarity profiles derived from the masked reference. d Query tumor cell pseudotimes are assigned based on neural network predictions of v-SVZ similarity profiles. QAD-stage is derived based on the ptalign pseudotime value. e ptalign performance is quantified by DTW, while a permutation framework tests for alignment robustness. DTW values represent the transcriptome correlation of reference and tumor cells binned in pseudotime. f ptalign outlines a comparative view of tumor hierarchies, mapping patient samples within a single reference trajectory and enabling their comparison in that context. g scRNA-seq UMAP of n = 51 primary GBMs (Supplementary Data 2 and Supplementary Fig. 4), colored by patient. h Ternary plot of GBMs from (g) arranged by ptalign QAD-stage proportion, unveiling the underlying QAD-stage heterogeneity. Source data are provided as a Source Data file.

Fig. 3. ASA informs GBM growth dynamics, prognosis, and DNA methylation-based stratification.

a, b Application of v-SVZ population models^, to GBM QAD-stage structure by ptalign, inferring descriptive parameters (a) and simulating growth to quantify growth rate among n = 51 primary scRNA-seq GBMs (b). c Left: Model simulated cell stages over time for the example tumor in (a). QAD populations reach steady-state equilibrium (tE) and grow until detection (tD, at 10¹¹ cells). Right: tE/tD ratios across tumors in (b). d, e Model simulated activation- and inferred growth rates for n = 51 primary GBMs. The dotted line shows a logarithmic fit of growth rate among GBMs (d). Tumors with high-residuals are shown in red, with the group-wise difference in growth rate and other model parameters shown in (e). Statistical significance was assessed by a two-sided t-test, precisely 0.11 and 0.36 for Growth rate and Amplification probability, respectively. f Predicted hazards with 90% confidence interval from a Cox model of n = 399 bulk GBMs from TCGA and Wu et al., with age and sex covariates of overall survival by GBM-QAD signature scores. P-value from multivariate Cox model with BH-correction. g PCA embedding of variable methylation sites for n = 83 GBMs from (f) with matched RNA and methylome measurements. Tumors are colored by their dominant RNA stage; gray circles have no clear dominant stage. Localized stage-enrichments are underscored with ellipses. h Linear regression over ElasticNet predictions on a (n = 28 GBMs) holdout set, with 90% confidence interval of RNA QAD-scores predicted from methylomes in (g). Models were trained on a (n = 55) GBM cohort. Pearson correlations (r) are indicated. A^nc: active non-cycling; A^c: active cycling. Box plots (c, e) span the 25th to 75th percentile, with the median indicated. Whiskers extend to 1.5 times the interquartile range. Source data are provided as a Source Data file.

Fig. 4. Comparative analysis of expression dynamics reveals recurrently dysregulated pathways in GBM.

a Shared ptalign axis between GBMs and v-SVZ enables comparative assessment of healthy vs malignant expression dynamics by pairwise EMD. Mean EMD across GBMs informs gene dysregulation among the GBM cohort. b Genes ranked by mean EMD vs v-SVZ among n = 51 GBMs. Genes above the inflection point (n = 164, purple) are considered recurrently dysregulated. c GO enrichment of recurrently dysregulated genes from (b) with all considered genes as background. P-values by hypergeometric test with FDR-correction. d Zoomed view of recurrently dysregulated genes from (b) with key signaling pathway genes indicated. e v-SVZ expression bias for recurrently dysregulated genes from (b), colored by pathway genes from (d). P-values represent a one-sided permutation test for Q-bias. f Expression splines and EMD values for individual genes from (d). Individual, and 95% confidence intervals, of mean GBM expression dynamics are colored red, while v-SVZ dynamics are blue. g Log-normalized SFRP1 expression in the v-SVZ UMAP. Cycling cells are colored gray. Source data are provided as a Source Data file.

Fig. 5. Dysregulation of Wnt activity at the Q–A transition in GBM.

a TCF/Lef-EGFP construct reporting canonical Wnt signaling activity in v-SVZ NSCs from a transgenic mouse line (top) and by lentiviral vector (bottom) in PDX and PDA tumors. mCherry ubiquitously labels tumor cells. Created in BioRender. Kaya, O. (2025) https://BioRender.com/np19rtk. b TCF/Lef-EGFP activity quantified in QAD-stage v-SVZ NSCs (n = 1564 cells, n = 5 replicates) and T6 PDX (n = 1176 cells, n = 4 replicates), and PDA (n = 12,378 cells, n = 4 replicates) cells. Reporter activity was quantified by FACS and QAD-stage by scRNA-seq. Data presented as mean, error bars are standard deviation in the normalized Wnt-active cell proportion. c Representative immunofluorescence images of TCF/Lef-H2B::EGFP activity for v-SVZ lineage populations. Top: EGFP+ NSCs (arrow heads) lining LV identified by presence of GFAP, SOX2, and absence of S100B and DCX (scale bars 20 µm and 10 µm). Middle: EGFP- NBs (Dotted lines; scale bars 50 µm and 20 µm) in v-SVZ marked by the presence of DCX. Bottom: EGFP+ neurons (arrow heads) marked by the presence of NeuN in OB (scale bars 50 µm and 20 µm). d Representative spatial transcriptomics view of QAD-stage cells in a GBM PDX. Left: striatal section of clustered cells from QAD-stages overlaid with DAPI and mCherry immunofluorescence. Tumor transcripts are colored by QAD-stage with a white outline; mouse transcripts are colored by cell type. Right: QAD-stage cell EGFP fluorescence. Scale bars, 10 µm. e Mean fraction of EGFP+ spots for QAD-stage cells in n = 6 spatial transcriptomics ROIs. f Immunofluorescence image of GBM PDX tumor cells (mCherry) in a large ventricular outgrowth of the third ventricle. Scale bars 500 µm and 100 µm. Insets highlight tumor cells devoid of Wnt activity, which is regained upon entry to the brain parenchyma. NB neuroblast, STR striatum, LV lateral ventricle, OB olfactory bulb, CTX cortex, CC corpus callosum, 3V third ventricle. Each experiment was repeated independently at least three times. Source data are provided as a Source Data file.

Fig. 6. Targeted disruption of ASA by SFRP1 renders the tumor quiescent.

a SFRP1-OE lentiviral construct used to generate GBM PDXs. mCherry ubiquitously labels tumor cells. Created in BioRender. Kaya, O. (2025) https://BioRender.com/np19rtk. b Kaplan–Meier curve of mice reaching endpoint post injection among three batches of n = 6 for control and SFRP1-OE mice each. P-value from log-rank test, precisely 1.3 × 10⁻⁴. c Proportion of QAD-stage cells identified by ptalign in SFRP1-OE (n = 3 replicates) and control (n = 4 replicates) scRNA-seq. Bars present mean, error bars standard deviation. d Selected GSEA enrichments from genes ranked by DEseq2 log fold-change between pseudobulked SFRP1-OE and control from (c). P-values from GSEA enrichment are FDR-adjusted. e Parameter estimates from population models of GBM dynamics (Fig. 3), including activation- and inferred growth rates, as well as self-renewal and amplification probabilities for T6 control and SFRP1-OE GBM PDXs from (c) among n = 51 primary GBMs. The rank percentile for each parameter is indicated. f Representative immunofluorescence images of GBM cells in a control and SFRP1-OE (e) PDX brain. Scale bars 100 µm, in insets 25 µm. g Entire spatial transcriptomics ROI depicting similar regions in SFRP1-OE (g) and control PDX brains. Transcripts were associated with segmented nuclei to assign species and QAD-stage. Pie charts indicate the sum of QAD-stage cells by brain region across ROIs. ROI region of interest, CTX cortex, CC corpus callosum, LV lateral ventricle, V-outgr. ventricular outgrowth, SN septal nuclei, STR striatum. Source data are provided as a Source Data file.

Fig. 7. Emergence of a conserved astrocyte-like methylome in GBM cells by SFRP1.

a Cross-species interrogation of murine v-SVZ cell type VMRs from Kremer et al.. Example NSC and astrocyte VMR methylation profiles are depicted, with methylation at the corresponding human locus quantified in SFRP1-overexpressing (OE) and control GBM WGBS. b Left: Mean methylation of v-SVZ cell type-specific VMRs from ref. in SFRP1-OE and control WGBS (n = 3 technical replicates each). Differentially methylated regions are highlighted by genotype. Right: v-SVZ cell type VMRs in a Gaussian KDE over the vertical axis of the scatterplot. P-value by one-sided hypergeometric test, for Astro. precisely 5.8 × 10⁻⁵. c Selected v-SVZ astrocyte VMR overlapping an NFIB-promoter in SFRP1-OE and control. Points depict mean CpG methylation among replicates, lines comprise a 10-CpG moving average. d Proportion of NFIB+ (left) and GFAP+ (right) Q-cells among control and SFRP1-OE samples. P-value by two-sided t-test, precisely 1.4 × 10⁻⁴ and 3.4 × 10⁻⁵ for NFIB, GFAP, respectively. e Schematic representation of the lineage potential in NSC-like GBM cells in control (above) vs lineage restrictions imposed by the remodeled methylome of the expanded astrocyte-like GBM cells upon SFRP1-OE (bottom). Adapted from ref. , Springer Nature, Inc. VMR: variably methylated region. Source data are provided as a Source Data file.