An immunosuppressive vascular niche drives macrophage polarization and immunotherapy resistance in glioblastoma

Abstract

Cancer immunity is subjected to spatiotemporal regulation by leukocyte interaction with neoplastic and stromal cells, contributing to immune evasion and immunotherapy resistance. Here, we identify a distinct mesenchymal-like population of endothelial cells (ECs) that form an immunosuppressive vascular niche in glioblastoma (GBM). We reveal a spatially restricted, Twist1/SATB1-mediated sequential transcriptional activation mechanism, through which tumor ECs produce osteopontin to promote immunosuppressive macrophage (Mφ) phenotypes. Genetic or pharmacological ablation of Twist1 reverses Mφ-mediated immunosuppression and enhances T cell infiltration and activation, leading to reduced GBM growth and extended mouse survival, and sensitizing tumor to chimeric antigen receptor T immunotherapy. Thus, these findings uncover a spatially restricted mechanism controlling tumor immunity and suggest that targeting endothelial Twist1 may offer attractive opportunities for optimizing cancer immunotherapy.

Fig. 1.. A mesenchymal-like population of ECs with Twist1 expression induces Mφ immunosuppression in GBM.

(A to C) Tumor cell spheres, derived from genetically engineered mouse GBM tumors in Ntv-a;Ink4a-Arf^−/−;Pten^fl/fl;LSL-Luc mice, were transplanted into Rosa-LSL-tdTomato;Cdh5-Cre^ERT2 mice. Tumors were excised and analyzed by scRNA-seq (n = 3 mice). (A) Experimental procedures. (B) Uniform Manifold Approximation and Projection (UMAP) analysis of transcriptome gene signature assigned cells into different clusters. (C) Immunosuppressive score in different cell clusters was analyzed on the basis of the average expression of immunosuppressive molecules. (D) Human peripheral blood mononuclear cell (PBMC)–derived Mφs were incubated with ECs derived from human normal brain or GBM tumors, or GBM tumor cells, followed by flow cytometry analysis. Left: Representative sortings. Right: Quantified results (n = 3 human samples, means ± SEM). Statistical analysis by one-way ANOVA. (E to I) GBM was induced in Rosa-LSL-tdTomato;Cdh5-Cre^ERT2 mice, followed by scRNA-seq analysis (n = 3 mice). [(E) to (G)] Genes with altered expression were identified in tumor ECs with high immunosuppressive scores, compared with ECs with low immunosuppressive scores. (E) Top up-regulated and down-regulated genes. (F) Normalized enrichment scores (NES) were calculated for top enriched pathway analysis. (G) Enrichment analysis of EMT pathway. (H) Expression distribution of endothelial- and mesenchymal-associated genes in two tdTomato⁺ EC populations. (I) The expression distribution of EMT-associated transcriptional factors was analyzed in all cell clusters. Left: UMAP analysis of Twist1 expression. Right: Expression profiles of Snai1, Snai2, Twist1/2, and Zeb1/2.

Fig. 2.. Genetic ablation of Twist1 in ECs inhibits Mφ immunosuppression in vitro and in vivo.

(A to C) Human GBM ECs were treated with siRNA targeting Twist1, Snail, or control sequence. (A) EC lysates were immunoblotted. [(B) and (C)] Treated ECs were incubated with human Mφs and analyzed by flow cytometry (n = 3 human samples, means ± SEM). Statistical analysis by one-way ANOVA. (B) CD206⁺ cells were measured in CD11b⁺ Mφs. Left: Representative sortings. Right: Quantified results. (C) IL-10 expression in CD11b⁺ Mφs. (D and E) Twist1^fl/fl mice were crossed with Cdh5-Cre^ERT2 mice to generate Twist1^fl/fl (control) or Cdh5-Cre^ERT2;Twist1^fl/fl (Twist1-ΔEC) mice, followed by treatment with tamoxifen at 2 weeks old. (D) Schematic approach. (E) ECs were isolated from mouse aortas. Brain tissue and ECs were subjected to immunoblot analysis. (F to L) GBM was induced in control and Twist1-ΔEC recipient mice. Tumors were excised and analyzed by CyTOF. (F) Schematic approach. (G) representative CyTOF sortings. [(H) to (L)] Quantified CyTOF results (n = 5 mice, means ± SEM). Analyses for (H) ECs, (I) immune cells, (J) T effector and regulatory cells, (K) exhausted T cells, and (L) M2-polarized Mφs. [(H) to (J) and (L)] Statistical analysis by two-tailed Student’s t test. (K) Statistical analysis by two-way ANOVA.

Fig. 3.. EC-specific Twist1 knockout inhibits tumor growth and improves animal survival.

GBM was induced in control and Twist1-ΔEC recipient mice. (A) Schematic approach. (B and C) Animal survival was monitored after injection. MS, median survival. Statistical analysis by log-rank. (B) GL261 model (n = 7 to 9 mice). (C) RACS model (n = 7 to 8 mice). (D to G) Tumor growth was analyzed by whole-body bioluminescence imaging. [(D) and (E)] Tumor volume in individual mice. [(F) and (G)] Quantified results on different days (means ± SEM). Statistical analysis by two-way ANOVA. [(D) and (F)] GL261 model (n = 7 to 9 mice). [(E) and (G)] RACS model (n = 7 to 8 mice).

Fig. 4.. Twist1 inhibition reverses Mφ immunosuppression in vitro and in vivo.

(A and B) Human PBMC-derived Mφs were incubated with or without human GBM ECs pretreated with or without harmine, and analyzed by flow cytometry for (A) CD206 and (B) IL-10 expression in CD11b⁺ Mφs. Left: Representative sortings. Right: Quantified results (n = 3 humans, means ± SEM). Statistical analysis by one-way ANOVA. (C to H) GBM was induced in immunocompetent mice, followed by saline or harmine treatment. (C) Experimental procedure. [(D) and (E)] Tumor-derived single-cell suspensions were analyzed by CyTOF. (D) Representative t-distributed stochastic neighbor embedding (tSNE) cell distribution. (E) Quantitative results (means ± SEM, n = 3 to 4 mice). Statistical analysis by two-tailed Student’s t test. [(F) to (H)] Tumor-derived cells were analyzed by flow cytometry analysis for (F) CD206 and [(G) and (H)] IL-10 and arginase 1 (Arg1) expression in tumor Mφs and myeloid cells. (G) Left: Representative cell sortings. Right: Quantitative results (n = 6 mice). Statistical analysis by [(E) to (G)] two-tailed Student’s t test or (H) two-way ANOVA.

Fig. 5.. Twist1 inhibition sensitizes GBM to CAR T immunotherapy.

GBM was induced in immunocompetent mice by transplantation of Egfrviii⁺;fLuc⁺ RCAS or GL261 tumor cells, followed by treatment with saline, harmine, or siRNA LNP, plus treatment adoptive CAR T cell transfer. (A) Experimental procedure. (B and C) RCAS model with harmine treatment (n = 10 to 15 mice). (B) Animal survival was monitored and analyzed by log-rank test. (C) Tumor volumes were analyzed by bioluminescence imaging. (D and E) GL261 model with harmine treatment (n = 8 mice). (D) Animal survival was monitored and analyzed by log-rank test. (E) Tumor volumes were analyzed by bioluminescence imaging. (F) Knockdown efficiency of siRNA LNP. Mouse ECs were treated with LNPs containing control or Twist1 siRNA, and cell lysates were immunoblotted. (G and H) GL261 model with siRNA LNP treatment (n = 12 to 15 mice) (harmine). Animal survival was monitored and analyzed by log-rank test. (H) Tumor volumes were analyzed by bioluminescence imaging.

Fig. 6.. Endothelial twist1 promotes alternative Mφ polarization via OPN.

(A to D) GBM ECs isolated from patients (n = 3) were transduced with lentivirus encoding control or Twist1 CRISPR/sgRNA, followed by RNA-seq analysis. (A) Analysis of Twist1/2 expression. Top: Heatmap of gene expression. Bottom: Quantified results (means ± SEM). Statistical analysis by two-way ANOVA. (B) Analysis of top regulated genes with differentiated expression (DE; fold change > 0.5) and P < 0.05. (C) Reactome analysis of top regulated genes. (D) to (F) Genetically engineered mouse GBM tumors were analyzed by scRNA-seq (n = 3). (D) Overlap in top regulated genes identified in bulk RNA-seq analysis of human GBM ECs treated with control or Twist1 sgRNA (left) and scRNA-seq analysis of mouse GBM ECs with high or low expression of Twist1 (right). [(E) and (F)] Analysis of mouse scRNA-seq data. (E) OPN (Spp1) expression in different cell clusters was analyzed. (F) Potential interaction pathways between ECs and monocytes/Mφs were predicted by ligand-receptor analysis. (G) Human GBM ECs were treated with control or Twist1 siRNA. Cell lysates and medium were immunoblotted. (H) Human PBMC-derived Mφs were treated with OPN at different doses, followed by flow cytometry analysis. Left: Representative cell sortings. Right: Quantified results (means ± SEM, n = 3). Statistical analysis by one-way ANOVA. (I) Human Mφs were treated with GBM EC–derived conditioned medium in the presence or absence of anti-OPN neutralizing antibody, and analyzed by flow cytometry. Left: Representative cell sortings. Right: Quantified results (means ± SEM, n = 3). Statistical analysis by Student’s t test. (J and K) GBM ECs were treated with control or OPN siRNA. (J) Cell lysates were immunoblotted. (K) Treated ECs were incubated with human Mφs and analyzed by flow cytometry. Left: Representative cell sortings. Right: Quantified results (means ± SEM, n = 3). Statistical analysis by Student’s t test.

Fig. 7.. Twist1-dependent SATB1 expression induces OPN transcription in GBM ECs.

(A) Human GBM ECs were subjected to CUT&RUN analysis with anti-Twist1 and control immunoglobulin G (IgG) antibodies. Pileup signals for Twist1 binding to the SPP1 gene region were calculated with IgG normalization (pooled from three EC samples). (B) ECs were subjected to chromatin immunoprecipitation (ChIP) analysis with anti-Twist1 or IgG antibody. Top: Different primers targeting the SPP1 gene were used. Bottom: Quantified results (means ± SEM, n = 3). Statistical analysis by two-way ANOVA. (C and D) ECs were subjected to CUT&RUN analysis with anti-Twist1 and control IgG antibodies. (C) ECs were transduced with lentivirus encoding control or Twist1 CRISPR/sgRNA and subjected to RNA-seq analysis. The top 50 down-regulated transcriptional factors were shown with CUT&RUN pileup signals (n = 3). (D) Pileup signals for Twist1 binding to the SATB1 gene region were calculated with IgG normalization (pooled from three EC samples). (E) ECs were transduced with lentivirus encoding control or Twist1 CRISPR/sgRNA and subjected to RNA-seq analysis for SATB1 expression. Top: Heatmap of gene expression. Bottom: Quantified results (means ± SEM, n = 3). Statistical analysis by Student’s t test. (F) ECs were treated with control or Twist1 siRNA. Cell lysates were immunoblotted. (G) ECs were subjected to ChIP analysis with anti-Twist1 or IgG antibodies and analyzed by ChIP analysis. Top: Different primers targeting the SATB1 gene were used. Bottom: Quantified results (means ± SEM, n = 3). Statistical analysis by two-way ANOVA. (H) ECs were treated with control or SATB1 siRNA. Cell lysates were immunoblotted. (I) ECs were subjected to ChIP analysis with anti-SATB1 or IgG antibodies and analyzed by ChIP analysis. Top: Different primers targeting the SPP1 gene were used. Bottom: Quantified results (means ± SEM, n = 3). Statistical analysis by two-way ANOVA. (J) A schematic model. Mesenchymal-like ECs form an immunosuppressive niche, inducing Mφ polarization via Twist1/SATB1-mediated OPN expression.