Combined targeting of glioblastoma stem cells of different cellular states disrupts malignant progression

Abstract

Glioblastoma (GBM) is the most lethal primary brain tumor with intra-tumoral hierarchy of glioblastoma stem cells (GSCs). The heterogeneity of GSCs within GBM inevitably leads to treatment resistance and tumor recurrence. Molecular mechanisms of different cellular state GSCs remain unclear. Here, we find that classical (CL) and mesenchymal (MES) GSCs are enriched in reactive immune region and high CL-MES signature informs poor prognosis in GBM. Through integrated analyses of GSCs RNA sequencing and single-cell RNA sequencing datasets, we identify specific GSCs targets, including MEOX2 for the CL GSCs and SRGN for the MES GSCs. MEOX2-NOTCH and SRGN-NFκB axes play important roles in promoting proliferation and maintaining stemness and subtype signatures of CL and MES GSCs, respectively. In the tumor microenvironment, MEOX2 and SRGN mediate the resistance of CL and MES GSCs to macrophage phagocytosis. Using genetic and pharmacologic approaches, we identify FDA-approved drugs targeting MEOX2 and SRGN. Combined CL and MES GSCs targeting demonstrates enhanced efficacy, both in vitro and in vivo. Our results highlighted a therapeutic strategy for the elimination of heterogeneous GSCs populations through combinatorial targeting of MEOX2 and SRGN in GSCs.

Fig. 1. MEOX2 and SRGN are enriched in CL and MES GSCs, respectively.

a Composition of four cellular states GSC-like tumor cells (SOX2⁺ in conjunction with CD133⁺/CD15⁺/OCT4⁺) in each GBM sample from scRNA-seq datasets. b Kaplan–Meier survival curves are shown for GBM patients in our in-housed generated scRNA-seq dataset, based on combination of GSC-like tumor cells proportion as indicated (high: >50%, low: <50%). c Diagram depicting the screening workflow to identify targets for CL/AC-like and MES/MES-like subtype GSCs. Created in BioRender. Wang, X. (2025) https://BioRender.com/f42j224. d Kaplan–Meier survival curve of GBM patients from TCGA (HG-U133A) dataset, based on MEOX2 and SRGN expression (high: >50%, low: <50%). e Bubble plot showing proportions and normalized expression levels of MEOX2 and SRGN in different cellular state GSC-like tumor cells from scRNA-seq datasets. f MEOX2 (left) and SRGN (right) gene expression in the GSCs (n = 3)/DGCs (n = 3) RNA-seq dataset (in-house generated). *p < 0.05. g MEOX2 and SRGN protein levels were measured by western blotting in CL/MES DGCs (n = 3) and CL/MES GSCs (n = 3). Tubulin was used as the loading control. h MEOX2 and SRGN protein levels were measured by western blotting in NSCs (n = 2) and GSCs (n = 10). Tubulin was used as the loading control. Immunofluorescence images showing co-staining of DAPI (blue), SOX2 (green), CD45 (red) with MEOX2 (yellow, i) or SRGN (purple, j) in glioblastoma samples. Scale bar, 2 mm (left), 100 μm (middle and right). The correlations between MEOX2/SRGN, SOX2, and CD45 expressions are shown. The gray area indicates the 95% CI of the regression. k Pan-cancer analysis of MEOX2 (left) and SRGN (right) expression levels in 24 different tumors from the TCGA dataset. **p < 0.01. Box plots show median (center line), 25th–75th percentiles (box bounds), and whiskers extend to minimum and maximum values within 1.5 times the interquartile range. Statistics: (b, d) Log-rank test. e, f, k Unpaired Student’s t test for two-group comparison. i, j The Pearson correlation test. Images: g, h Representative blots (n = 3). i, j Representative images (n = 30). Source data are provided as a Source Data file.

Fig. 2. MEOX2 and SRGN are essential for GSCs proliferation, self-renewal, and stemness in CL and MES GSCs respectively.

a, b Cell viability of indicated GSCs with shCONT, shMEOX2, or shSRGN measured by CellTiter-Glo assay (n = 6 per group). Data are presented as mean ± SD. c, e The number of spheres formed by CL and MES GSCs transduced with shCONT or shRNAs targeting MEOX2 or SRGN was quantified (n = 6 per group). Data are presented as mean ± SD. Box plots show median (center line), 25th–75th percentiles (box bounds), and whiskers extend to minimum and maximum values within 1.5 times the interquartile range. ns: non-significant. d, f In vitro limited dilution assay of CL and MES GSCs transduced with shCONT, shMEOX2-1/2 or shSRGN-1/2, respectively (n = 30 per group). χ2 test. g, j Heatmap of SOX2 and ID1 expression in MEOX2 and SRGN knockout RNA-seq data. h, k Quantitative RT-PCR assessment of SOX2 and ID1 mRNA levels in CL GSCs (3028, RKI) and MES GSCs (2907, 839) expressing shCONT, shMEOX2-1/2 or shSRGN-1/2, respectively. Data are presented as mean ± SD. n = 3 per group. i, l IB analysis of SOX2 and ID1 levels in CL GSCs (3028, RKI) and MES GSCs (2907, 839) transduced with shMEOX2, shSRGN, or control shRNA, respectively. m, o Bioluminescence imaging and quantification of orthotopic xenografts derived from luciferase-expressing 3028 (CL) and 2907 (2907) GSCs transduced with shCONT, shMEOX2, or shSRGN. Data are presented as mean ± SD. n = 3 per group. n, p Kaplan–Meier survival curves of mice bearing intracranial tumors derived from 3028 (CL) and 2907 (MES) GSCs (n = 6 per group) from the respective groups above. Statistics: a, b, c, e, m, o one-way ANOVA with Dunnett’s multiple-comparison test. d, f χ² test for two-group comparison. h, k Unpaired Student’s t test for two-group comparison. n, p Log-rank test. Images: i, l Representative blots (n = 3). Source data are provided as a Source Data file.

Fig. 3. MEOX2 functions through transcriptional activation of the Notch signaling pathway in CL GSCs.

a The UMAP plot of 3028 CL GSCs annotated by four cellular states. b Proportion changes of four cellular states cells in 3028 CL GSCs transduced with MEOX2 sgRNA or control sgRNA. c Diagram depicting the screening workflow to identify the MEOX2 downstream mechanism. Created in BioRender. Wang, X. (2025) https://BioRender.com/f42j224. d Gene set enrichment analysis (GSEA) of Hallmark, KEGG, and Reactome gene sets after MEOX2 knockout in 3028 (CL) GSCs. Normalized enrichment score (NES) and p values are shown by heatmap. e Violin-box plots showing the notch signaling score (Hallmark and KEGG) of the GSCs scRNA-seq data comparing sgCONT (n = 7078) vs. sgMEOX2 (n = 3168). Data are aggregated from technical replicates. f KEGG pathway enrichment analysis of MEOX2 ChIP-seq peak genes. g Input or MEOX2 binding signals of NOTCH1, NOTCH2, JAG1, and PSEN1 from Chip-seq on 3028 (CL) GSCs. h NOTCH1 overexpression rescued MEOX2 knockdown-induced cell viability decrease in CL GSCs (n = 6). Data are mean ± SD. i Sphere formation of CL GSCs with shCONT, shMEOX2, or shMEOX2 plus NOTCH1 overexpression (n = 6). Data are mean ± SD. j IB analysis of SOX2 and ID1 levels in 3028 (CL) and RKI (CL) GSCs transduced with NOTCH1 shRNA or control shRNA. k Bioluminescence images of mice with orthotopic xenografts from luciferase-expressing 3028 (CL) cells, showing tumor growth under shMEOX2 or shMEOX2 + NOTCH1 treatment (left). Luciferase signal intensities are shown (right). Data are mean ± SD; n = 3 biologically replicates per group. l Kaplan–Meier survival curves of mice bearing intracranial tumors derived from 3028 CL GSCs (n = 6) from the three groups in (k). Statistics: d The nominal p-value was calculated based on GSEA. e Unpaired Student’s t-test for two-group comparison. h, i, k one-way ANOVA with Dunnett’s multiple-comparison test. l Log-rank test. e, i Box plots show median (center line), 25th–75th percentiles (box bounds), and whiskers extend to minimum and maximum values within 1.5 times the interquartile range. Images: j Representative blots (n = 3). Source data are provided as a Source Data file.

Fig. 4. SRGN promotes NF-κB signaling in MES GSCs post-transcriptionally by protecting NFKB1 from proteasomal degradation.

a The UMAP plot of 2907 MES GSCs annotated by four cellular states. b Cellular states distribution in 2907 MES GSCs with SRGN or control sgRNA. c NF-κB signaling score (Hallmark and KEGG) of the GSCs scRNA-seq data comparing sgCONT (n = 15,109) vs. sgSRGN (n = 24,993). Data are aggregated from technical replicates. d GO analysis of 670 SRGN-interacting proteins identified by IP-MS. e MES GSCs lysates underwent SRGN/NFKB1 IP followed by IB, with IgG as control. f Representative images of SRGN and NFKB1 immunofluorescence in 2907 (MES) and 839 (MES) GSCs. Scale bar, 20 μm. g MES GSCs co-transfected with SRGN shRNA and His-Ub were treated with MG132 (20 μM, 8 h), followed by NFKB1 IP and His/SRGN IB. h MES GSCs co-transfected with SRGN shRNA and His-Ub were treated with MG132 (20 μM, 8 h), followed by NFKB1 IP and His/SRGN IB. i NFKB1 ubiquitylation assay in MES GSCs with indicated His-Ub constructs. j MES GSCs expressing Ub WT or Ub K48R were cultured with shCONT or shSRGN (72 h) and analyzed by NFKB1/SRGN IB. k NFKB1 overexpression rescued shSRGN-reduced viability in MES GSCs (n = 6). Data are presented as mean ± SD. l Cell counts of MES GSCs with shCONT, shSRGN, or shSRGN+NFKB1 (n = 6). Data are presented as mean ± SD. m IB analysis of SOX2 and ID1 levels in 2907 (MES) and 839 (MES) GSCs transduced with NFKB1 shRNA or control shRNA. n Bioluminescence images of MES GSCs xenografts with shSRGN or shSRGN+NFKB1 and quantification. Data shown as mean ± SD from 3 biological replicates. o Survival curves of mice with intracranial tumors from MES GSCs (n = 6 per group). Statistics: c Unpaired Student’s t-test for two-group comparison. k, l, n one-way ANOVA with Dunnett’s multiple-comparison test. o Log-rank test. c, l Box plots show median (center line), 25th–75th percentiles (box bounds), and whiskers extend to minimum and maximum values within 1.5 times the interquartile range. Images: e, g–j, m Representative blots (n = 3). f Representative images (n = 3). Source data are provided as a Source Data file.

Fig. 5. MEOX2 and SRGN regulate the resistance of CL and MES GSCs to macrophage phagocytosis, respectively.

a Stacked bar plots indicate the cell composition of CL^lowMES^low (n = 109) and CL^highMES^high (n = 109) GBM in the TCGA GBM dataset based on the single-cell-based deconvolution method, CIBERSORTx. b Comparison of the proportion of cell composition between CL^lowMES^low (n = 109) and CL^highMES^high (n = 109) GBM in the TCGA GBM dataset. c Correlation coefficient between CL-MES signatures and various types of immune cell signatures in TCGA GBM dataset (n = 528). d Heatmap of anti-phagocytosis genes expression between four cellular state GSC-like tumor cells in scRNAseq datasets. e Heatmap of anti-phagocytosis genes expression between three subtypes GSCs in GSCs RNA-seq datasets (GSE119834). f Heatmap of anti-phagocytosis genes expression in MEOX2 knockout RNA-seq data. g Heatmap of anti-phagocytosis genes expression in SRGN knockout RNA-seq data. h Representative flow cytometry plots depicting the phagocytosis of GSCs transduced with shCONT, shMEOX2, shSRGN, or shMEOX2 + shSRGN co-cultured with macrophages. Created in BioRender. Wang, X. (2025) https://BioRender.com/f42j224. i Flow cytometry quantifies phagocytosis of CL + MES GSCs treated with shMEOX2, shSRGN, or dual treatment versus control. Data are presented as mean ± SD. n = 3 per group. j Experimental design to assess in vivo effects of combinatorial targeting of CL and MES on xenograft of mixed CL (3028) and MES (2907) GSCs. Created in BioRender. Wang, X. (2025) https://BioRender.com/f42j224. k Bioluminescence images show the effect of combined treatment with 75 mg/kg RG-4733 and 50 mg/kg PDTC every two days on tumor growth in mice bearing mixed CL and MES xenografts derived from luciferase-expressing 3028 (CL) and 2907 (MES) GSCs. Time points indicate days post-intracranial injection (left), with luciferase signal intensities shown (right). Data are presented as mean ± SD. n = 3 biologically independent mice per group. l, Kaplan–Meier survival curves for mice bearing subtype-mixed orthotopic tumors (3028 CL and 2907 MES) (n = 6) after combined treatment of 75 mg/kg RG-4733 every two days and 50 mg/kg PDTC every two days. Statistics: b, i Unpaired Student’s t test for two-group comparison. k one-way ANOVA with Dunnett’s multiple-comparison test. l Log-rank test. Source data are provided as a Source Data file.

Fig. 6. Nilotinib and DHEC have inhibitory effects on MEOX2, while paliperidone and risperidone have inhibitory effects on SRGN.

a Workflow of experimental strategy for finding MEOX2 and SRGN targeting drugs. Created in BioRender. Wang, X. (2025) https://BioRender.com/f42j224. b MEOX2 (residues 187-246) was docked with nilotinib and DHEC, and SRGN (residues 28-60) was docked with paliperidone and risperidone. c Dose-response curves for nilotinib, DHEC, paliperidone, and risperidone in CL GSCs (3028 and RKI) and MES GSCs (2907 and 839). Data are presented as mean ± SD. n = 6 biologically independent replicates per group. **p < 0.01. d Effects of nilotinib and DHEC on Notch signaling pathway genes (NOTCH1, NOTCH2, JAG1, KAT2B, MAML2, and PSEN1) expression in 3028 (CL) and RKI (CL) GSCs. Data are presented as mean ± SD from three independent experiments. e IB analysis of NFKB1 expression in 2907 (MES) and 839 (MES) GSCs treated with paliperidone or risperidone. f Bio-layer interferometry (BLI) was used to detect the equilibrium dissociation constant (K_d) between nilotinib with MEOX2, DHEC with MEOX2, paliperidone with SRGN, and risperidone with SRGN. g (Left) Schematic representation of MEOX2 truncated protein. (Right) Coomassie blue-stained SDS-PAGE gel of truncated MEOX2 protein. h Bio-layer interferometry (BLI) was used to detect the equilibrium dissociation constant (K_d) between nilotinib and DHEC with truncated MEOX2 protein. i (Left) Schematic representation of SRGN WT and truncated mutants with binding region indicated (amino acids 28-60). (Right) Representative Coomassie blue-stained SDS-PAGE gel showing purified truncated SRGN protein. j Bio-layer interferometry (BLI) was used to detect the equilibrium dissociation constant (K_d) between paliperidone and risperidone with truncated SRGN protein. Statistics: c, d Unpaired Student’s t test for two-group comparison. Images: e Representative blots (n = 3). g, i Representative images (n = 3). Source data are provided as a Source Data file.

Fig. 7. In vivo therapeutic efficacy of combined pharmacological inhibition of MEOX2 and SRGN in a subtype-mixed glioblastoma model.

a Four tumor cellular states mapped in UMAP coordinates from WL1 scRNA-seq data. b Summary of mice medium survival in the in vivo limiting dilution assay. c, e, g Bioluminescence images of mice bearing xenografts derived from luciferase-expressing WL1 GSCs with the indicated numbers of cells are shown (left). In vivo bioluminescence imaging and intensities of luciferase signal in mice are shown (right). Data are presented as mean ± SD. n = 3 biologically independent mice per group. d, f, h Kaplan–Meier survival curves of mice implanted with the indicated numbers of WL1 GSCs (n = 6) are shown. i Experimental design to assess in vivo effects of nilotinib/DHEC and paliperidone/risperidone on xenograft of mixed CL (EGFR⁺) and MES (CD44⁺) WL1 GSCs. Created in BioRender. Wang, X. (2025) https://BioRender.com/f42j224. j, l Bioluminescence images of mice bearing mixed CL and MES xenografts derived from luciferase-expressing WL1 GSCs, showing the effect of combined treatments (nilotinib + paliperidone or DHEC + risperidone) on tumor growth. Time points indicate days after intracranial injection of mixed GSC populations (left). Intensities of luciferase signal in mice are shown (right). Data are presented as mean ± SD. n = 3 biologically independent mice per group. k, m Kaplan–Meier survival curves for mice bearing subtype-mixed orthotopic tumors (WL1) after combined treatments of nilotinib (25 mg/kg) + paliperidone (10 mg/kg) or DHEC (15 mg/kg) + risperidone (10 mg/kg). n = 6 mice per group. Statistics: c, e, g Unpaired Student’s t test for two-group comparison. d, f, h, k, m Log-rank test. j, l one-way ANOVA with Dunnett’s multiple-comparison test. Source data are provided as a Source Data file.